xref: /openbmc/linux/mm/mm_init.c (revision c1a6c536fb088c01d6bdce77731d89ad5e1734c6)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * mm_init.c - Memory initialisation verification and debugging
4  *
5  * Copyright 2008 IBM Corporation, 2008
6  * Author Mel Gorman <mel@csn.ul.ie>
7  *
8  */
9 #include <linux/kernel.h>
10 #include <linux/init.h>
11 #include <linux/kobject.h>
12 #include <linux/export.h>
13 #include <linux/memory.h>
14 #include <linux/notifier.h>
15 #include <linux/sched.h>
16 #include <linux/mman.h>
17 #include <linux/memblock.h>
18 #include <linux/page-isolation.h>
19 #include <linux/padata.h>
20 #include <linux/nmi.h>
21 #include <linux/buffer_head.h>
22 #include <linux/kmemleak.h>
23 #include <linux/kfence.h>
24 #include <linux/page_ext.h>
25 #include <linux/pti.h>
26 #include <linux/pgtable.h>
27 #include <linux/swap.h>
28 #include <linux/cma.h>
29 #include "internal.h"
30 #include "slab.h"
31 #include "shuffle.h"
32 
33 #include <asm/setup.h>
34 
35 #ifdef CONFIG_DEBUG_MEMORY_INIT
36 int __meminitdata mminit_loglevel;
37 
38 /* The zonelists are simply reported, validation is manual. */
39 void __init mminit_verify_zonelist(void)
40 {
41 	int nid;
42 
43 	if (mminit_loglevel < MMINIT_VERIFY)
44 		return;
45 
46 	for_each_online_node(nid) {
47 		pg_data_t *pgdat = NODE_DATA(nid);
48 		struct zone *zone;
49 		struct zoneref *z;
50 		struct zonelist *zonelist;
51 		int i, listid, zoneid;
52 
53 		BUILD_BUG_ON(MAX_ZONELISTS > 2);
54 		for (i = 0; i < MAX_ZONELISTS * MAX_NR_ZONES; i++) {
55 
56 			/* Identify the zone and nodelist */
57 			zoneid = i % MAX_NR_ZONES;
58 			listid = i / MAX_NR_ZONES;
59 			zonelist = &pgdat->node_zonelists[listid];
60 			zone = &pgdat->node_zones[zoneid];
61 			if (!populated_zone(zone))
62 				continue;
63 
64 			/* Print information about the zonelist */
65 			printk(KERN_DEBUG "mminit::zonelist %s %d:%s = ",
66 				listid > 0 ? "thisnode" : "general", nid,
67 				zone->name);
68 
69 			/* Iterate the zonelist */
70 			for_each_zone_zonelist(zone, z, zonelist, zoneid)
71 				pr_cont("%d:%s ", zone_to_nid(zone), zone->name);
72 			pr_cont("\n");
73 		}
74 	}
75 }
76 
77 void __init mminit_verify_pageflags_layout(void)
78 {
79 	int shift, width;
80 	unsigned long or_mask, add_mask;
81 
82 	shift = 8 * sizeof(unsigned long);
83 	width = shift - SECTIONS_WIDTH - NODES_WIDTH - ZONES_WIDTH
84 		- LAST_CPUPID_SHIFT - KASAN_TAG_WIDTH - LRU_GEN_WIDTH - LRU_REFS_WIDTH;
85 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_widths",
86 		"Section %d Node %d Zone %d Lastcpupid %d Kasantag %d Gen %d Tier %d Flags %d\n",
87 		SECTIONS_WIDTH,
88 		NODES_WIDTH,
89 		ZONES_WIDTH,
90 		LAST_CPUPID_WIDTH,
91 		KASAN_TAG_WIDTH,
92 		LRU_GEN_WIDTH,
93 		LRU_REFS_WIDTH,
94 		NR_PAGEFLAGS);
95 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_shifts",
96 		"Section %d Node %d Zone %d Lastcpupid %d Kasantag %d\n",
97 		SECTIONS_SHIFT,
98 		NODES_SHIFT,
99 		ZONES_SHIFT,
100 		LAST_CPUPID_SHIFT,
101 		KASAN_TAG_WIDTH);
102 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_pgshifts",
103 		"Section %lu Node %lu Zone %lu Lastcpupid %lu Kasantag %lu\n",
104 		(unsigned long)SECTIONS_PGSHIFT,
105 		(unsigned long)NODES_PGSHIFT,
106 		(unsigned long)ZONES_PGSHIFT,
107 		(unsigned long)LAST_CPUPID_PGSHIFT,
108 		(unsigned long)KASAN_TAG_PGSHIFT);
109 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodezoneid",
110 		"Node/Zone ID: %lu -> %lu\n",
111 		(unsigned long)(ZONEID_PGOFF + ZONEID_SHIFT),
112 		(unsigned long)ZONEID_PGOFF);
113 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_usage",
114 		"location: %d -> %d layout %d -> %d unused %d -> %d page-flags\n",
115 		shift, width, width, NR_PAGEFLAGS, NR_PAGEFLAGS, 0);
116 #ifdef NODE_NOT_IN_PAGE_FLAGS
117 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
118 		"Node not in page flags");
119 #endif
120 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
121 	mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
122 		"Last cpupid not in page flags");
123 #endif
124 
125 	if (SECTIONS_WIDTH) {
126 		shift -= SECTIONS_WIDTH;
127 		BUG_ON(shift != SECTIONS_PGSHIFT);
128 	}
129 	if (NODES_WIDTH) {
130 		shift -= NODES_WIDTH;
131 		BUG_ON(shift != NODES_PGSHIFT);
132 	}
133 	if (ZONES_WIDTH) {
134 		shift -= ZONES_WIDTH;
135 		BUG_ON(shift != ZONES_PGSHIFT);
136 	}
137 
138 	/* Check for bitmask overlaps */
139 	or_mask = (ZONES_MASK << ZONES_PGSHIFT) |
140 			(NODES_MASK << NODES_PGSHIFT) |
141 			(SECTIONS_MASK << SECTIONS_PGSHIFT);
142 	add_mask = (ZONES_MASK << ZONES_PGSHIFT) +
143 			(NODES_MASK << NODES_PGSHIFT) +
144 			(SECTIONS_MASK << SECTIONS_PGSHIFT);
145 	BUG_ON(or_mask != add_mask);
146 }
147 
148 static __init int set_mminit_loglevel(char *str)
149 {
150 	get_option(&str, &mminit_loglevel);
151 	return 0;
152 }
153 early_param("mminit_loglevel", set_mminit_loglevel);
154 #endif /* CONFIG_DEBUG_MEMORY_INIT */
155 
156 struct kobject *mm_kobj;
157 EXPORT_SYMBOL_GPL(mm_kobj);
158 
159 #ifdef CONFIG_SMP
160 s32 vm_committed_as_batch = 32;
161 
162 void mm_compute_batch(int overcommit_policy)
163 {
164 	u64 memsized_batch;
165 	s32 nr = num_present_cpus();
166 	s32 batch = max_t(s32, nr*2, 32);
167 	unsigned long ram_pages = totalram_pages();
168 
169 	/*
170 	 * For policy OVERCOMMIT_NEVER, set batch size to 0.4% of
171 	 * (total memory/#cpus), and lift it to 25% for other policies
172 	 * to easy the possible lock contention for percpu_counter
173 	 * vm_committed_as, while the max limit is INT_MAX
174 	 */
175 	if (overcommit_policy == OVERCOMMIT_NEVER)
176 		memsized_batch = min_t(u64, ram_pages/nr/256, INT_MAX);
177 	else
178 		memsized_batch = min_t(u64, ram_pages/nr/4, INT_MAX);
179 
180 	vm_committed_as_batch = max_t(s32, memsized_batch, batch);
181 }
182 
183 static int __meminit mm_compute_batch_notifier(struct notifier_block *self,
184 					unsigned long action, void *arg)
185 {
186 	switch (action) {
187 	case MEM_ONLINE:
188 	case MEM_OFFLINE:
189 		mm_compute_batch(sysctl_overcommit_memory);
190 		break;
191 	default:
192 		break;
193 	}
194 	return NOTIFY_OK;
195 }
196 
197 static int __init mm_compute_batch_init(void)
198 {
199 	mm_compute_batch(sysctl_overcommit_memory);
200 	hotplug_memory_notifier(mm_compute_batch_notifier, MM_COMPUTE_BATCH_PRI);
201 	return 0;
202 }
203 
204 __initcall(mm_compute_batch_init);
205 
206 #endif
207 
208 static int __init mm_sysfs_init(void)
209 {
210 	mm_kobj = kobject_create_and_add("mm", kernel_kobj);
211 	if (!mm_kobj)
212 		return -ENOMEM;
213 
214 	return 0;
215 }
216 postcore_initcall(mm_sysfs_init);
217 
218 static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __initdata;
219 static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __initdata;
220 static unsigned long zone_movable_pfn[MAX_NUMNODES] __initdata;
221 
222 static unsigned long required_kernelcore __initdata;
223 static unsigned long required_kernelcore_percent __initdata;
224 static unsigned long required_movablecore __initdata;
225 static unsigned long required_movablecore_percent __initdata;
226 
227 static unsigned long nr_kernel_pages __initdata;
228 static unsigned long nr_all_pages __initdata;
229 static unsigned long dma_reserve __initdata;
230 
231 static bool deferred_struct_pages __meminitdata;
232 
233 static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
234 
235 static int __init cmdline_parse_core(char *p, unsigned long *core,
236 				     unsigned long *percent)
237 {
238 	unsigned long long coremem;
239 	char *endptr;
240 
241 	if (!p)
242 		return -EINVAL;
243 
244 	/* Value may be a percentage of total memory, otherwise bytes */
245 	coremem = simple_strtoull(p, &endptr, 0);
246 	if (*endptr == '%') {
247 		/* Paranoid check for percent values greater than 100 */
248 		WARN_ON(coremem > 100);
249 
250 		*percent = coremem;
251 	} else {
252 		coremem = memparse(p, &p);
253 		/* Paranoid check that UL is enough for the coremem value */
254 		WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
255 
256 		*core = coremem >> PAGE_SHIFT;
257 		*percent = 0UL;
258 	}
259 	return 0;
260 }
261 
262 bool mirrored_kernelcore __initdata_memblock;
263 
264 /*
265  * kernelcore=size sets the amount of memory for use for allocations that
266  * cannot be reclaimed or migrated.
267  */
268 static int __init cmdline_parse_kernelcore(char *p)
269 {
270 	/* parse kernelcore=mirror */
271 	if (parse_option_str(p, "mirror")) {
272 		mirrored_kernelcore = true;
273 		return 0;
274 	}
275 
276 	return cmdline_parse_core(p, &required_kernelcore,
277 				  &required_kernelcore_percent);
278 }
279 early_param("kernelcore", cmdline_parse_kernelcore);
280 
281 /*
282  * movablecore=size sets the amount of memory for use for allocations that
283  * can be reclaimed or migrated.
284  */
285 static int __init cmdline_parse_movablecore(char *p)
286 {
287 	return cmdline_parse_core(p, &required_movablecore,
288 				  &required_movablecore_percent);
289 }
290 early_param("movablecore", cmdline_parse_movablecore);
291 
292 /*
293  * early_calculate_totalpages()
294  * Sum pages in active regions for movable zone.
295  * Populate N_MEMORY for calculating usable_nodes.
296  */
297 static unsigned long __init early_calculate_totalpages(void)
298 {
299 	unsigned long totalpages = 0;
300 	unsigned long start_pfn, end_pfn;
301 	int i, nid;
302 
303 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
304 		unsigned long pages = end_pfn - start_pfn;
305 
306 		totalpages += pages;
307 		if (pages)
308 			node_set_state(nid, N_MEMORY);
309 	}
310 	return totalpages;
311 }
312 
313 /*
314  * This finds a zone that can be used for ZONE_MOVABLE pages. The
315  * assumption is made that zones within a node are ordered in monotonic
316  * increasing memory addresses so that the "highest" populated zone is used
317  */
318 static void __init find_usable_zone_for_movable(void)
319 {
320 	int zone_index;
321 	for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
322 		if (zone_index == ZONE_MOVABLE)
323 			continue;
324 
325 		if (arch_zone_highest_possible_pfn[zone_index] >
326 				arch_zone_lowest_possible_pfn[zone_index])
327 			break;
328 	}
329 
330 	VM_BUG_ON(zone_index == -1);
331 	movable_zone = zone_index;
332 }
333 
334 /*
335  * Find the PFN the Movable zone begins in each node. Kernel memory
336  * is spread evenly between nodes as long as the nodes have enough
337  * memory. When they don't, some nodes will have more kernelcore than
338  * others
339  */
340 static void __init find_zone_movable_pfns_for_nodes(void)
341 {
342 	int i, nid;
343 	unsigned long usable_startpfn;
344 	unsigned long kernelcore_node, kernelcore_remaining;
345 	/* save the state before borrow the nodemask */
346 	nodemask_t saved_node_state = node_states[N_MEMORY];
347 	unsigned long totalpages = early_calculate_totalpages();
348 	int usable_nodes = nodes_weight(node_states[N_MEMORY]);
349 	struct memblock_region *r;
350 
351 	/* Need to find movable_zone earlier when movable_node is specified. */
352 	find_usable_zone_for_movable();
353 
354 	/*
355 	 * If movable_node is specified, ignore kernelcore and movablecore
356 	 * options.
357 	 */
358 	if (movable_node_is_enabled()) {
359 		for_each_mem_region(r) {
360 			if (!memblock_is_hotpluggable(r))
361 				continue;
362 
363 			nid = memblock_get_region_node(r);
364 
365 			usable_startpfn = PFN_DOWN(r->base);
366 			zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
367 				min(usable_startpfn, zone_movable_pfn[nid]) :
368 				usable_startpfn;
369 		}
370 
371 		goto out2;
372 	}
373 
374 	/*
375 	 * If kernelcore=mirror is specified, ignore movablecore option
376 	 */
377 	if (mirrored_kernelcore) {
378 		bool mem_below_4gb_not_mirrored = false;
379 
380 		for_each_mem_region(r) {
381 			if (memblock_is_mirror(r))
382 				continue;
383 
384 			nid = memblock_get_region_node(r);
385 
386 			usable_startpfn = memblock_region_memory_base_pfn(r);
387 
388 			if (usable_startpfn < PHYS_PFN(SZ_4G)) {
389 				mem_below_4gb_not_mirrored = true;
390 				continue;
391 			}
392 
393 			zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
394 				min(usable_startpfn, zone_movable_pfn[nid]) :
395 				usable_startpfn;
396 		}
397 
398 		if (mem_below_4gb_not_mirrored)
399 			pr_warn("This configuration results in unmirrored kernel memory.\n");
400 
401 		goto out2;
402 	}
403 
404 	/*
405 	 * If kernelcore=nn% or movablecore=nn% was specified, calculate the
406 	 * amount of necessary memory.
407 	 */
408 	if (required_kernelcore_percent)
409 		required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
410 				       10000UL;
411 	if (required_movablecore_percent)
412 		required_movablecore = (totalpages * 100 * required_movablecore_percent) /
413 					10000UL;
414 
415 	/*
416 	 * If movablecore= was specified, calculate what size of
417 	 * kernelcore that corresponds so that memory usable for
418 	 * any allocation type is evenly spread. If both kernelcore
419 	 * and movablecore are specified, then the value of kernelcore
420 	 * will be used for required_kernelcore if it's greater than
421 	 * what movablecore would have allowed.
422 	 */
423 	if (required_movablecore) {
424 		unsigned long corepages;
425 
426 		/*
427 		 * Round-up so that ZONE_MOVABLE is at least as large as what
428 		 * was requested by the user
429 		 */
430 		required_movablecore =
431 			roundup(required_movablecore, MAX_ORDER_NR_PAGES);
432 		required_movablecore = min(totalpages, required_movablecore);
433 		corepages = totalpages - required_movablecore;
434 
435 		required_kernelcore = max(required_kernelcore, corepages);
436 	}
437 
438 	/*
439 	 * If kernelcore was not specified or kernelcore size is larger
440 	 * than totalpages, there is no ZONE_MOVABLE.
441 	 */
442 	if (!required_kernelcore || required_kernelcore >= totalpages)
443 		goto out;
444 
445 	/* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
446 	usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
447 
448 restart:
449 	/* Spread kernelcore memory as evenly as possible throughout nodes */
450 	kernelcore_node = required_kernelcore / usable_nodes;
451 	for_each_node_state(nid, N_MEMORY) {
452 		unsigned long start_pfn, end_pfn;
453 
454 		/*
455 		 * Recalculate kernelcore_node if the division per node
456 		 * now exceeds what is necessary to satisfy the requested
457 		 * amount of memory for the kernel
458 		 */
459 		if (required_kernelcore < kernelcore_node)
460 			kernelcore_node = required_kernelcore / usable_nodes;
461 
462 		/*
463 		 * As the map is walked, we track how much memory is usable
464 		 * by the kernel using kernelcore_remaining. When it is
465 		 * 0, the rest of the node is usable by ZONE_MOVABLE
466 		 */
467 		kernelcore_remaining = kernelcore_node;
468 
469 		/* Go through each range of PFNs within this node */
470 		for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
471 			unsigned long size_pages;
472 
473 			start_pfn = max(start_pfn, zone_movable_pfn[nid]);
474 			if (start_pfn >= end_pfn)
475 				continue;
476 
477 			/* Account for what is only usable for kernelcore */
478 			if (start_pfn < usable_startpfn) {
479 				unsigned long kernel_pages;
480 				kernel_pages = min(end_pfn, usable_startpfn)
481 								- start_pfn;
482 
483 				kernelcore_remaining -= min(kernel_pages,
484 							kernelcore_remaining);
485 				required_kernelcore -= min(kernel_pages,
486 							required_kernelcore);
487 
488 				/* Continue if range is now fully accounted */
489 				if (end_pfn <= usable_startpfn) {
490 
491 					/*
492 					 * Push zone_movable_pfn to the end so
493 					 * that if we have to rebalance
494 					 * kernelcore across nodes, we will
495 					 * not double account here
496 					 */
497 					zone_movable_pfn[nid] = end_pfn;
498 					continue;
499 				}
500 				start_pfn = usable_startpfn;
501 			}
502 
503 			/*
504 			 * The usable PFN range for ZONE_MOVABLE is from
505 			 * start_pfn->end_pfn. Calculate size_pages as the
506 			 * number of pages used as kernelcore
507 			 */
508 			size_pages = end_pfn - start_pfn;
509 			if (size_pages > kernelcore_remaining)
510 				size_pages = kernelcore_remaining;
511 			zone_movable_pfn[nid] = start_pfn + size_pages;
512 
513 			/*
514 			 * Some kernelcore has been met, update counts and
515 			 * break if the kernelcore for this node has been
516 			 * satisfied
517 			 */
518 			required_kernelcore -= min(required_kernelcore,
519 								size_pages);
520 			kernelcore_remaining -= size_pages;
521 			if (!kernelcore_remaining)
522 				break;
523 		}
524 	}
525 
526 	/*
527 	 * If there is still required_kernelcore, we do another pass with one
528 	 * less node in the count. This will push zone_movable_pfn[nid] further
529 	 * along on the nodes that still have memory until kernelcore is
530 	 * satisfied
531 	 */
532 	usable_nodes--;
533 	if (usable_nodes && required_kernelcore > usable_nodes)
534 		goto restart;
535 
536 out2:
537 	/* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
538 	for (nid = 0; nid < MAX_NUMNODES; nid++) {
539 		unsigned long start_pfn, end_pfn;
540 
541 		zone_movable_pfn[nid] =
542 			roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
543 
544 		get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
545 		if (zone_movable_pfn[nid] >= end_pfn)
546 			zone_movable_pfn[nid] = 0;
547 	}
548 
549 out:
550 	/* restore the node_state */
551 	node_states[N_MEMORY] = saved_node_state;
552 }
553 
554 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
555 				unsigned long zone, int nid)
556 {
557 	mm_zero_struct_page(page);
558 	set_page_links(page, zone, nid, pfn);
559 	init_page_count(page);
560 	page_mapcount_reset(page);
561 	page_cpupid_reset_last(page);
562 	page_kasan_tag_reset(page);
563 
564 	INIT_LIST_HEAD(&page->lru);
565 #ifdef WANT_PAGE_VIRTUAL
566 	/* The shift won't overflow because ZONE_NORMAL is below 4G. */
567 	if (!is_highmem_idx(zone))
568 		set_page_address(page, __va(pfn << PAGE_SHIFT));
569 #endif
570 }
571 
572 #ifdef CONFIG_NUMA
573 /*
574  * During memory init memblocks map pfns to nids. The search is expensive and
575  * this caches recent lookups. The implementation of __early_pfn_to_nid
576  * treats start/end as pfns.
577  */
578 struct mminit_pfnnid_cache {
579 	unsigned long last_start;
580 	unsigned long last_end;
581 	int last_nid;
582 };
583 
584 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
585 
586 /*
587  * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
588  */
589 static int __meminit __early_pfn_to_nid(unsigned long pfn,
590 					struct mminit_pfnnid_cache *state)
591 {
592 	unsigned long start_pfn, end_pfn;
593 	int nid;
594 
595 	if (state->last_start <= pfn && pfn < state->last_end)
596 		return state->last_nid;
597 
598 	nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
599 	if (nid != NUMA_NO_NODE) {
600 		state->last_start = start_pfn;
601 		state->last_end = end_pfn;
602 		state->last_nid = nid;
603 	}
604 
605 	return nid;
606 }
607 
608 int __meminit early_pfn_to_nid(unsigned long pfn)
609 {
610 	static DEFINE_SPINLOCK(early_pfn_lock);
611 	int nid;
612 
613 	spin_lock(&early_pfn_lock);
614 	nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
615 	if (nid < 0)
616 		nid = first_online_node;
617 	spin_unlock(&early_pfn_lock);
618 
619 	return nid;
620 }
621 
622 int hashdist = HASHDIST_DEFAULT;
623 
624 static int __init set_hashdist(char *str)
625 {
626 	if (!str)
627 		return 0;
628 	hashdist = simple_strtoul(str, &str, 0);
629 	return 1;
630 }
631 __setup("hashdist=", set_hashdist);
632 
633 static inline void fixup_hashdist(void)
634 {
635 	if (num_node_state(N_MEMORY) == 1)
636 		hashdist = 0;
637 }
638 #else
639 static inline void fixup_hashdist(void) {}
640 #endif /* CONFIG_NUMA */
641 
642 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
643 static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
644 {
645 	pgdat->first_deferred_pfn = ULONG_MAX;
646 }
647 
648 /* Returns true if the struct page for the pfn is initialised */
649 static inline bool __meminit early_page_initialised(unsigned long pfn, int nid)
650 {
651 	if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
652 		return false;
653 
654 	return true;
655 }
656 
657 /*
658  * Returns true when the remaining initialisation should be deferred until
659  * later in the boot cycle when it can be parallelised.
660  */
661 static bool __meminit
662 defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
663 {
664 	static unsigned long prev_end_pfn, nr_initialised;
665 
666 	if (early_page_ext_enabled())
667 		return false;
668 	/*
669 	 * prev_end_pfn static that contains the end of previous zone
670 	 * No need to protect because called very early in boot before smp_init.
671 	 */
672 	if (prev_end_pfn != end_pfn) {
673 		prev_end_pfn = end_pfn;
674 		nr_initialised = 0;
675 	}
676 
677 	/* Always populate low zones for address-constrained allocations */
678 	if (end_pfn < pgdat_end_pfn(NODE_DATA(nid)))
679 		return false;
680 
681 	if (NODE_DATA(nid)->first_deferred_pfn != ULONG_MAX)
682 		return true;
683 	/*
684 	 * We start only with one section of pages, more pages are added as
685 	 * needed until the rest of deferred pages are initialized.
686 	 */
687 	nr_initialised++;
688 	if ((nr_initialised > PAGES_PER_SECTION) &&
689 	    (pfn & (PAGES_PER_SECTION - 1)) == 0) {
690 		NODE_DATA(nid)->first_deferred_pfn = pfn;
691 		return true;
692 	}
693 	return false;
694 }
695 
696 static void __meminit init_reserved_page(unsigned long pfn, int nid)
697 {
698 	pg_data_t *pgdat;
699 	int zid;
700 
701 	if (early_page_initialised(pfn, nid))
702 		return;
703 
704 	pgdat = NODE_DATA(nid);
705 
706 	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
707 		struct zone *zone = &pgdat->node_zones[zid];
708 
709 		if (zone_spans_pfn(zone, pfn))
710 			break;
711 	}
712 	__init_single_page(pfn_to_page(pfn), pfn, zid, nid);
713 }
714 #else
715 static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
716 
717 static inline bool early_page_initialised(unsigned long pfn, int nid)
718 {
719 	return true;
720 }
721 
722 static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
723 {
724 	return false;
725 }
726 
727 static inline void init_reserved_page(unsigned long pfn, int nid)
728 {
729 }
730 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
731 
732 /*
733  * Initialised pages do not have PageReserved set. This function is
734  * called for each range allocated by the bootmem allocator and
735  * marks the pages PageReserved. The remaining valid pages are later
736  * sent to the buddy page allocator.
737  */
738 void __meminit reserve_bootmem_region(phys_addr_t start,
739 				      phys_addr_t end, int nid)
740 {
741 	unsigned long start_pfn = PFN_DOWN(start);
742 	unsigned long end_pfn = PFN_UP(end);
743 
744 	for (; start_pfn < end_pfn; start_pfn++) {
745 		if (pfn_valid(start_pfn)) {
746 			struct page *page = pfn_to_page(start_pfn);
747 
748 			init_reserved_page(start_pfn, nid);
749 
750 			/* Avoid false-positive PageTail() */
751 			INIT_LIST_HEAD(&page->lru);
752 
753 			/*
754 			 * no need for atomic set_bit because the struct
755 			 * page is not visible yet so nobody should
756 			 * access it yet.
757 			 */
758 			__SetPageReserved(page);
759 		}
760 	}
761 }
762 
763 /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */
764 static bool __meminit
765 overlap_memmap_init(unsigned long zone, unsigned long *pfn)
766 {
767 	static struct memblock_region *r;
768 
769 	if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
770 		if (!r || *pfn >= memblock_region_memory_end_pfn(r)) {
771 			for_each_mem_region(r) {
772 				if (*pfn < memblock_region_memory_end_pfn(r))
773 					break;
774 			}
775 		}
776 		if (*pfn >= memblock_region_memory_base_pfn(r) &&
777 		    memblock_is_mirror(r)) {
778 			*pfn = memblock_region_memory_end_pfn(r);
779 			return true;
780 		}
781 	}
782 	return false;
783 }
784 
785 /*
786  * Only struct pages that correspond to ranges defined by memblock.memory
787  * are zeroed and initialized by going through __init_single_page() during
788  * memmap_init_zone_range().
789  *
790  * But, there could be struct pages that correspond to holes in
791  * memblock.memory. This can happen because of the following reasons:
792  * - physical memory bank size is not necessarily the exact multiple of the
793  *   arbitrary section size
794  * - early reserved memory may not be listed in memblock.memory
795  * - memory layouts defined with memmap= kernel parameter may not align
796  *   nicely with memmap sections
797  *
798  * Explicitly initialize those struct pages so that:
799  * - PG_Reserved is set
800  * - zone and node links point to zone and node that span the page if the
801  *   hole is in the middle of a zone
802  * - zone and node links point to adjacent zone/node if the hole falls on
803  *   the zone boundary; the pages in such holes will be prepended to the
804  *   zone/node above the hole except for the trailing pages in the last
805  *   section that will be appended to the zone/node below.
806  */
807 static void __init init_unavailable_range(unsigned long spfn,
808 					  unsigned long epfn,
809 					  int zone, int node)
810 {
811 	unsigned long pfn;
812 	u64 pgcnt = 0;
813 
814 	for (pfn = spfn; pfn < epfn; pfn++) {
815 		if (!pfn_valid(pageblock_start_pfn(pfn))) {
816 			pfn = pageblock_end_pfn(pfn) - 1;
817 			continue;
818 		}
819 		__init_single_page(pfn_to_page(pfn), pfn, zone, node);
820 		__SetPageReserved(pfn_to_page(pfn));
821 		pgcnt++;
822 	}
823 
824 	if (pgcnt)
825 		pr_info("On node %d, zone %s: %lld pages in unavailable ranges",
826 			node, zone_names[zone], pgcnt);
827 }
828 
829 /*
830  * Initially all pages are reserved - free ones are freed
831  * up by memblock_free_all() once the early boot process is
832  * done. Non-atomic initialization, single-pass.
833  *
834  * All aligned pageblocks are initialized to the specified migratetype
835  * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related
836  * zone stats (e.g., nr_isolate_pageblock) are touched.
837  */
838 void __meminit memmap_init_range(unsigned long size, int nid, unsigned long zone,
839 		unsigned long start_pfn, unsigned long zone_end_pfn,
840 		enum meminit_context context,
841 		struct vmem_altmap *altmap, int migratetype)
842 {
843 	unsigned long pfn, end_pfn = start_pfn + size;
844 	struct page *page;
845 
846 	if (highest_memmap_pfn < end_pfn - 1)
847 		highest_memmap_pfn = end_pfn - 1;
848 
849 #ifdef CONFIG_ZONE_DEVICE
850 	/*
851 	 * Honor reservation requested by the driver for this ZONE_DEVICE
852 	 * memory. We limit the total number of pages to initialize to just
853 	 * those that might contain the memory mapping. We will defer the
854 	 * ZONE_DEVICE page initialization until after we have released
855 	 * the hotplug lock.
856 	 */
857 	if (zone == ZONE_DEVICE) {
858 		if (!altmap)
859 			return;
860 
861 		if (start_pfn == altmap->base_pfn)
862 			start_pfn += altmap->reserve;
863 		end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
864 	}
865 #endif
866 
867 	for (pfn = start_pfn; pfn < end_pfn; ) {
868 		/*
869 		 * There can be holes in boot-time mem_map[]s handed to this
870 		 * function.  They do not exist on hotplugged memory.
871 		 */
872 		if (context == MEMINIT_EARLY) {
873 			if (overlap_memmap_init(zone, &pfn))
874 				continue;
875 			if (defer_init(nid, pfn, zone_end_pfn)) {
876 				deferred_struct_pages = true;
877 				break;
878 			}
879 		}
880 
881 		page = pfn_to_page(pfn);
882 		__init_single_page(page, pfn, zone, nid);
883 		if (context == MEMINIT_HOTPLUG)
884 			__SetPageReserved(page);
885 
886 		/*
887 		 * Usually, we want to mark the pageblock MIGRATE_MOVABLE,
888 		 * such that unmovable allocations won't be scattered all
889 		 * over the place during system boot.
890 		 */
891 		if (pageblock_aligned(pfn)) {
892 			set_pageblock_migratetype(page, migratetype);
893 			cond_resched();
894 		}
895 		pfn++;
896 	}
897 }
898 
899 static void __init memmap_init_zone_range(struct zone *zone,
900 					  unsigned long start_pfn,
901 					  unsigned long end_pfn,
902 					  unsigned long *hole_pfn)
903 {
904 	unsigned long zone_start_pfn = zone->zone_start_pfn;
905 	unsigned long zone_end_pfn = zone_start_pfn + zone->spanned_pages;
906 	int nid = zone_to_nid(zone), zone_id = zone_idx(zone);
907 
908 	start_pfn = clamp(start_pfn, zone_start_pfn, zone_end_pfn);
909 	end_pfn = clamp(end_pfn, zone_start_pfn, zone_end_pfn);
910 
911 	if (start_pfn >= end_pfn)
912 		return;
913 
914 	memmap_init_range(end_pfn - start_pfn, nid, zone_id, start_pfn,
915 			  zone_end_pfn, MEMINIT_EARLY, NULL, MIGRATE_MOVABLE);
916 
917 	if (*hole_pfn < start_pfn)
918 		init_unavailable_range(*hole_pfn, start_pfn, zone_id, nid);
919 
920 	*hole_pfn = end_pfn;
921 }
922 
923 static void __init memmap_init(void)
924 {
925 	unsigned long start_pfn, end_pfn;
926 	unsigned long hole_pfn = 0;
927 	int i, j, zone_id = 0, nid;
928 
929 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
930 		struct pglist_data *node = NODE_DATA(nid);
931 
932 		for (j = 0; j < MAX_NR_ZONES; j++) {
933 			struct zone *zone = node->node_zones + j;
934 
935 			if (!populated_zone(zone))
936 				continue;
937 
938 			memmap_init_zone_range(zone, start_pfn, end_pfn,
939 					       &hole_pfn);
940 			zone_id = j;
941 		}
942 	}
943 
944 #ifdef CONFIG_SPARSEMEM
945 	/*
946 	 * Initialize the memory map for hole in the range [memory_end,
947 	 * section_end].
948 	 * Append the pages in this hole to the highest zone in the last
949 	 * node.
950 	 * The call to init_unavailable_range() is outside the ifdef to
951 	 * silence the compiler warining about zone_id set but not used;
952 	 * for FLATMEM it is a nop anyway
953 	 */
954 	end_pfn = round_up(end_pfn, PAGES_PER_SECTION);
955 	if (hole_pfn < end_pfn)
956 #endif
957 		init_unavailable_range(hole_pfn, end_pfn, zone_id, nid);
958 }
959 
960 #ifdef CONFIG_ZONE_DEVICE
961 static void __ref __init_zone_device_page(struct page *page, unsigned long pfn,
962 					  unsigned long zone_idx, int nid,
963 					  struct dev_pagemap *pgmap)
964 {
965 
966 	__init_single_page(page, pfn, zone_idx, nid);
967 
968 	/*
969 	 * Mark page reserved as it will need to wait for onlining
970 	 * phase for it to be fully associated with a zone.
971 	 *
972 	 * We can use the non-atomic __set_bit operation for setting
973 	 * the flag as we are still initializing the pages.
974 	 */
975 	__SetPageReserved(page);
976 
977 	/*
978 	 * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer
979 	 * and zone_device_data.  It is a bug if a ZONE_DEVICE page is
980 	 * ever freed or placed on a driver-private list.
981 	 */
982 	page->pgmap = pgmap;
983 	page->zone_device_data = NULL;
984 
985 	/*
986 	 * Mark the block movable so that blocks are reserved for
987 	 * movable at startup. This will force kernel allocations
988 	 * to reserve their blocks rather than leaking throughout
989 	 * the address space during boot when many long-lived
990 	 * kernel allocations are made.
991 	 *
992 	 * Please note that MEMINIT_HOTPLUG path doesn't clear memmap
993 	 * because this is done early in section_activate()
994 	 */
995 	if (pageblock_aligned(pfn)) {
996 		set_pageblock_migratetype(page, MIGRATE_MOVABLE);
997 		cond_resched();
998 	}
999 
1000 	/*
1001 	 * ZONE_DEVICE pages are released directly to the driver page allocator
1002 	 * which will set the page count to 1 when allocating the page.
1003 	 */
1004 	if (pgmap->type == MEMORY_DEVICE_PRIVATE ||
1005 	    pgmap->type == MEMORY_DEVICE_COHERENT)
1006 		set_page_count(page, 0);
1007 }
1008 
1009 /*
1010  * With compound page geometry and when struct pages are stored in ram most
1011  * tail pages are reused. Consequently, the amount of unique struct pages to
1012  * initialize is a lot smaller that the total amount of struct pages being
1013  * mapped. This is a paired / mild layering violation with explicit knowledge
1014  * of how the sparse_vmemmap internals handle compound pages in the lack
1015  * of an altmap. See vmemmap_populate_compound_pages().
1016  */
1017 static inline unsigned long compound_nr_pages(struct vmem_altmap *altmap,
1018 					      struct dev_pagemap *pgmap)
1019 {
1020 	if (!vmemmap_can_optimize(altmap, pgmap))
1021 		return pgmap_vmemmap_nr(pgmap);
1022 
1023 	return VMEMMAP_RESERVE_NR * (PAGE_SIZE / sizeof(struct page));
1024 }
1025 
1026 static void __ref memmap_init_compound(struct page *head,
1027 				       unsigned long head_pfn,
1028 				       unsigned long zone_idx, int nid,
1029 				       struct dev_pagemap *pgmap,
1030 				       unsigned long nr_pages)
1031 {
1032 	unsigned long pfn, end_pfn = head_pfn + nr_pages;
1033 	unsigned int order = pgmap->vmemmap_shift;
1034 
1035 	__SetPageHead(head);
1036 	for (pfn = head_pfn + 1; pfn < end_pfn; pfn++) {
1037 		struct page *page = pfn_to_page(pfn);
1038 
1039 		__init_zone_device_page(page, pfn, zone_idx, nid, pgmap);
1040 		prep_compound_tail(head, pfn - head_pfn);
1041 		set_page_count(page, 0);
1042 
1043 		/*
1044 		 * The first tail page stores important compound page info.
1045 		 * Call prep_compound_head() after the first tail page has
1046 		 * been initialized, to not have the data overwritten.
1047 		 */
1048 		if (pfn == head_pfn + 1)
1049 			prep_compound_head(head, order);
1050 	}
1051 }
1052 
1053 void __ref memmap_init_zone_device(struct zone *zone,
1054 				   unsigned long start_pfn,
1055 				   unsigned long nr_pages,
1056 				   struct dev_pagemap *pgmap)
1057 {
1058 	unsigned long pfn, end_pfn = start_pfn + nr_pages;
1059 	struct pglist_data *pgdat = zone->zone_pgdat;
1060 	struct vmem_altmap *altmap = pgmap_altmap(pgmap);
1061 	unsigned int pfns_per_compound = pgmap_vmemmap_nr(pgmap);
1062 	unsigned long zone_idx = zone_idx(zone);
1063 	unsigned long start = jiffies;
1064 	int nid = pgdat->node_id;
1065 
1066 	if (WARN_ON_ONCE(!pgmap || zone_idx != ZONE_DEVICE))
1067 		return;
1068 
1069 	/*
1070 	 * The call to memmap_init should have already taken care
1071 	 * of the pages reserved for the memmap, so we can just jump to
1072 	 * the end of that region and start processing the device pages.
1073 	 */
1074 	if (altmap) {
1075 		start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
1076 		nr_pages = end_pfn - start_pfn;
1077 	}
1078 
1079 	for (pfn = start_pfn; pfn < end_pfn; pfn += pfns_per_compound) {
1080 		struct page *page = pfn_to_page(pfn);
1081 
1082 		__init_zone_device_page(page, pfn, zone_idx, nid, pgmap);
1083 
1084 		if (pfns_per_compound == 1)
1085 			continue;
1086 
1087 		memmap_init_compound(page, pfn, zone_idx, nid, pgmap,
1088 				     compound_nr_pages(altmap, pgmap));
1089 	}
1090 
1091 	pr_debug("%s initialised %lu pages in %ums\n", __func__,
1092 		nr_pages, jiffies_to_msecs(jiffies - start));
1093 }
1094 #endif
1095 
1096 /*
1097  * The zone ranges provided by the architecture do not include ZONE_MOVABLE
1098  * because it is sized independent of architecture. Unlike the other zones,
1099  * the starting point for ZONE_MOVABLE is not fixed. It may be different
1100  * in each node depending on the size of each node and how evenly kernelcore
1101  * is distributed. This helper function adjusts the zone ranges
1102  * provided by the architecture for a given node by using the end of the
1103  * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
1104  * zones within a node are in order of monotonic increases memory addresses
1105  */
1106 static void __init adjust_zone_range_for_zone_movable(int nid,
1107 					unsigned long zone_type,
1108 					unsigned long node_end_pfn,
1109 					unsigned long *zone_start_pfn,
1110 					unsigned long *zone_end_pfn)
1111 {
1112 	/* Only adjust if ZONE_MOVABLE is on this node */
1113 	if (zone_movable_pfn[nid]) {
1114 		/* Size ZONE_MOVABLE */
1115 		if (zone_type == ZONE_MOVABLE) {
1116 			*zone_start_pfn = zone_movable_pfn[nid];
1117 			*zone_end_pfn = min(node_end_pfn,
1118 				arch_zone_highest_possible_pfn[movable_zone]);
1119 
1120 		/* Adjust for ZONE_MOVABLE starting within this range */
1121 		} else if (!mirrored_kernelcore &&
1122 			*zone_start_pfn < zone_movable_pfn[nid] &&
1123 			*zone_end_pfn > zone_movable_pfn[nid]) {
1124 			*zone_end_pfn = zone_movable_pfn[nid];
1125 
1126 		/* Check if this whole range is within ZONE_MOVABLE */
1127 		} else if (*zone_start_pfn >= zone_movable_pfn[nid])
1128 			*zone_start_pfn = *zone_end_pfn;
1129 	}
1130 }
1131 
1132 /*
1133  * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
1134  * then all holes in the requested range will be accounted for.
1135  */
1136 unsigned long __init __absent_pages_in_range(int nid,
1137 				unsigned long range_start_pfn,
1138 				unsigned long range_end_pfn)
1139 {
1140 	unsigned long nr_absent = range_end_pfn - range_start_pfn;
1141 	unsigned long start_pfn, end_pfn;
1142 	int i;
1143 
1144 	for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
1145 		start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
1146 		end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
1147 		nr_absent -= end_pfn - start_pfn;
1148 	}
1149 	return nr_absent;
1150 }
1151 
1152 /**
1153  * absent_pages_in_range - Return number of page frames in holes within a range
1154  * @start_pfn: The start PFN to start searching for holes
1155  * @end_pfn: The end PFN to stop searching for holes
1156  *
1157  * Return: the number of pages frames in memory holes within a range.
1158  */
1159 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
1160 							unsigned long end_pfn)
1161 {
1162 	return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
1163 }
1164 
1165 /* Return the number of page frames in holes in a zone on a node */
1166 static unsigned long __init zone_absent_pages_in_node(int nid,
1167 					unsigned long zone_type,
1168 					unsigned long zone_start_pfn,
1169 					unsigned long zone_end_pfn)
1170 {
1171 	unsigned long nr_absent;
1172 
1173 	/* zone is empty, we don't have any absent pages */
1174 	if (zone_start_pfn == zone_end_pfn)
1175 		return 0;
1176 
1177 	nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
1178 
1179 	/*
1180 	 * ZONE_MOVABLE handling.
1181 	 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
1182 	 * and vice versa.
1183 	 */
1184 	if (mirrored_kernelcore && zone_movable_pfn[nid]) {
1185 		unsigned long start_pfn, end_pfn;
1186 		struct memblock_region *r;
1187 
1188 		for_each_mem_region(r) {
1189 			start_pfn = clamp(memblock_region_memory_base_pfn(r),
1190 					  zone_start_pfn, zone_end_pfn);
1191 			end_pfn = clamp(memblock_region_memory_end_pfn(r),
1192 					zone_start_pfn, zone_end_pfn);
1193 
1194 			if (zone_type == ZONE_MOVABLE &&
1195 			    memblock_is_mirror(r))
1196 				nr_absent += end_pfn - start_pfn;
1197 
1198 			if (zone_type == ZONE_NORMAL &&
1199 			    !memblock_is_mirror(r))
1200 				nr_absent += end_pfn - start_pfn;
1201 		}
1202 	}
1203 
1204 	return nr_absent;
1205 }
1206 
1207 /*
1208  * Return the number of pages a zone spans in a node, including holes
1209  * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
1210  */
1211 static unsigned long __init zone_spanned_pages_in_node(int nid,
1212 					unsigned long zone_type,
1213 					unsigned long node_start_pfn,
1214 					unsigned long node_end_pfn,
1215 					unsigned long *zone_start_pfn,
1216 					unsigned long *zone_end_pfn)
1217 {
1218 	unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
1219 	unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
1220 
1221 	/* Get the start and end of the zone */
1222 	*zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
1223 	*zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
1224 	adjust_zone_range_for_zone_movable(nid, zone_type, node_end_pfn,
1225 					   zone_start_pfn, zone_end_pfn);
1226 
1227 	/* Check that this node has pages within the zone's required range */
1228 	if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
1229 		return 0;
1230 
1231 	/* Move the zone boundaries inside the node if necessary */
1232 	*zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
1233 	*zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
1234 
1235 	/* Return the spanned pages */
1236 	return *zone_end_pfn - *zone_start_pfn;
1237 }
1238 
1239 static void __init reset_memoryless_node_totalpages(struct pglist_data *pgdat)
1240 {
1241 	struct zone *z;
1242 
1243 	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) {
1244 		z->zone_start_pfn = 0;
1245 		z->spanned_pages = 0;
1246 		z->present_pages = 0;
1247 #if defined(CONFIG_MEMORY_HOTPLUG)
1248 		z->present_early_pages = 0;
1249 #endif
1250 	}
1251 
1252 	pgdat->node_spanned_pages = 0;
1253 	pgdat->node_present_pages = 0;
1254 	pr_debug("On node %d totalpages: 0\n", pgdat->node_id);
1255 }
1256 
1257 static void __init calculate_node_totalpages(struct pglist_data *pgdat,
1258 						unsigned long node_start_pfn,
1259 						unsigned long node_end_pfn)
1260 {
1261 	unsigned long realtotalpages = 0, totalpages = 0;
1262 	enum zone_type i;
1263 
1264 	for (i = 0; i < MAX_NR_ZONES; i++) {
1265 		struct zone *zone = pgdat->node_zones + i;
1266 		unsigned long zone_start_pfn, zone_end_pfn;
1267 		unsigned long spanned, absent;
1268 		unsigned long real_size;
1269 
1270 		spanned = zone_spanned_pages_in_node(pgdat->node_id, i,
1271 						     node_start_pfn,
1272 						     node_end_pfn,
1273 						     &zone_start_pfn,
1274 						     &zone_end_pfn);
1275 		absent = zone_absent_pages_in_node(pgdat->node_id, i,
1276 						   zone_start_pfn,
1277 						   zone_end_pfn);
1278 
1279 		real_size = spanned - absent;
1280 
1281 		if (spanned)
1282 			zone->zone_start_pfn = zone_start_pfn;
1283 		else
1284 			zone->zone_start_pfn = 0;
1285 		zone->spanned_pages = spanned;
1286 		zone->present_pages = real_size;
1287 #if defined(CONFIG_MEMORY_HOTPLUG)
1288 		zone->present_early_pages = real_size;
1289 #endif
1290 
1291 		totalpages += spanned;
1292 		realtotalpages += real_size;
1293 	}
1294 
1295 	pgdat->node_spanned_pages = totalpages;
1296 	pgdat->node_present_pages = realtotalpages;
1297 	pr_debug("On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1298 }
1299 
1300 static unsigned long __init calc_memmap_size(unsigned long spanned_pages,
1301 						unsigned long present_pages)
1302 {
1303 	unsigned long pages = spanned_pages;
1304 
1305 	/*
1306 	 * Provide a more accurate estimation if there are holes within
1307 	 * the zone and SPARSEMEM is in use. If there are holes within the
1308 	 * zone, each populated memory region may cost us one or two extra
1309 	 * memmap pages due to alignment because memmap pages for each
1310 	 * populated regions may not be naturally aligned on page boundary.
1311 	 * So the (present_pages >> 4) heuristic is a tradeoff for that.
1312 	 */
1313 	if (spanned_pages > present_pages + (present_pages >> 4) &&
1314 	    IS_ENABLED(CONFIG_SPARSEMEM))
1315 		pages = present_pages;
1316 
1317 	return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
1318 }
1319 
1320 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1321 static void pgdat_init_split_queue(struct pglist_data *pgdat)
1322 {
1323 	struct deferred_split *ds_queue = &pgdat->deferred_split_queue;
1324 
1325 	spin_lock_init(&ds_queue->split_queue_lock);
1326 	INIT_LIST_HEAD(&ds_queue->split_queue);
1327 	ds_queue->split_queue_len = 0;
1328 }
1329 #else
1330 static void pgdat_init_split_queue(struct pglist_data *pgdat) {}
1331 #endif
1332 
1333 #ifdef CONFIG_COMPACTION
1334 static void pgdat_init_kcompactd(struct pglist_data *pgdat)
1335 {
1336 	init_waitqueue_head(&pgdat->kcompactd_wait);
1337 }
1338 #else
1339 static void pgdat_init_kcompactd(struct pglist_data *pgdat) {}
1340 #endif
1341 
1342 static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
1343 {
1344 	int i;
1345 
1346 	pgdat_resize_init(pgdat);
1347 	pgdat_kswapd_lock_init(pgdat);
1348 
1349 	pgdat_init_split_queue(pgdat);
1350 	pgdat_init_kcompactd(pgdat);
1351 
1352 	init_waitqueue_head(&pgdat->kswapd_wait);
1353 	init_waitqueue_head(&pgdat->pfmemalloc_wait);
1354 
1355 	for (i = 0; i < NR_VMSCAN_THROTTLE; i++)
1356 		init_waitqueue_head(&pgdat->reclaim_wait[i]);
1357 
1358 	pgdat_page_ext_init(pgdat);
1359 	lruvec_init(&pgdat->__lruvec);
1360 }
1361 
1362 static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
1363 							unsigned long remaining_pages)
1364 {
1365 	atomic_long_set(&zone->managed_pages, remaining_pages);
1366 	zone_set_nid(zone, nid);
1367 	zone->name = zone_names[idx];
1368 	zone->zone_pgdat = NODE_DATA(nid);
1369 	spin_lock_init(&zone->lock);
1370 	zone_seqlock_init(zone);
1371 	zone_pcp_init(zone);
1372 }
1373 
1374 static void __meminit zone_init_free_lists(struct zone *zone)
1375 {
1376 	unsigned int order, t;
1377 	for_each_migratetype_order(order, t) {
1378 		INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
1379 		zone->free_area[order].nr_free = 0;
1380 	}
1381 
1382 #ifdef CONFIG_UNACCEPTED_MEMORY
1383 	INIT_LIST_HEAD(&zone->unaccepted_pages);
1384 #endif
1385 }
1386 
1387 void __meminit init_currently_empty_zone(struct zone *zone,
1388 					unsigned long zone_start_pfn,
1389 					unsigned long size)
1390 {
1391 	struct pglist_data *pgdat = zone->zone_pgdat;
1392 	int zone_idx = zone_idx(zone) + 1;
1393 
1394 	if (zone_idx > pgdat->nr_zones)
1395 		pgdat->nr_zones = zone_idx;
1396 
1397 	zone->zone_start_pfn = zone_start_pfn;
1398 
1399 	mminit_dprintk(MMINIT_TRACE, "memmap_init",
1400 			"Initialising map node %d zone %lu pfns %lu -> %lu\n",
1401 			pgdat->node_id,
1402 			(unsigned long)zone_idx(zone),
1403 			zone_start_pfn, (zone_start_pfn + size));
1404 
1405 	zone_init_free_lists(zone);
1406 	zone->initialized = 1;
1407 }
1408 
1409 #ifndef CONFIG_SPARSEMEM
1410 /*
1411  * Calculate the size of the zone->blockflags rounded to an unsigned long
1412  * Start by making sure zonesize is a multiple of pageblock_order by rounding
1413  * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
1414  * round what is now in bits to nearest long in bits, then return it in
1415  * bytes.
1416  */
1417 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
1418 {
1419 	unsigned long usemapsize;
1420 
1421 	zonesize += zone_start_pfn & (pageblock_nr_pages-1);
1422 	usemapsize = roundup(zonesize, pageblock_nr_pages);
1423 	usemapsize = usemapsize >> pageblock_order;
1424 	usemapsize *= NR_PAGEBLOCK_BITS;
1425 	usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
1426 
1427 	return usemapsize / 8;
1428 }
1429 
1430 static void __ref setup_usemap(struct zone *zone)
1431 {
1432 	unsigned long usemapsize = usemap_size(zone->zone_start_pfn,
1433 					       zone->spanned_pages);
1434 	zone->pageblock_flags = NULL;
1435 	if (usemapsize) {
1436 		zone->pageblock_flags =
1437 			memblock_alloc_node(usemapsize, SMP_CACHE_BYTES,
1438 					    zone_to_nid(zone));
1439 		if (!zone->pageblock_flags)
1440 			panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n",
1441 			      usemapsize, zone->name, zone_to_nid(zone));
1442 	}
1443 }
1444 #else
1445 static inline void setup_usemap(struct zone *zone) {}
1446 #endif /* CONFIG_SPARSEMEM */
1447 
1448 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
1449 
1450 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
1451 void __init set_pageblock_order(void)
1452 {
1453 	unsigned int order = MAX_ORDER;
1454 
1455 	/* Check that pageblock_nr_pages has not already been setup */
1456 	if (pageblock_order)
1457 		return;
1458 
1459 	/* Don't let pageblocks exceed the maximum allocation granularity. */
1460 	if (HPAGE_SHIFT > PAGE_SHIFT && HUGETLB_PAGE_ORDER < order)
1461 		order = HUGETLB_PAGE_ORDER;
1462 
1463 	/*
1464 	 * Assume the largest contiguous order of interest is a huge page.
1465 	 * This value may be variable depending on boot parameters on IA64 and
1466 	 * powerpc.
1467 	 */
1468 	pageblock_order = order;
1469 }
1470 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
1471 
1472 /*
1473  * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
1474  * is unused as pageblock_order is set at compile-time. See
1475  * include/linux/pageblock-flags.h for the values of pageblock_order based on
1476  * the kernel config
1477  */
1478 void __init set_pageblock_order(void)
1479 {
1480 }
1481 
1482 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
1483 
1484 /*
1485  * Set up the zone data structures
1486  * - init pgdat internals
1487  * - init all zones belonging to this node
1488  *
1489  * NOTE: this function is only called during memory hotplug
1490  */
1491 #ifdef CONFIG_MEMORY_HOTPLUG
1492 void __ref free_area_init_core_hotplug(struct pglist_data *pgdat)
1493 {
1494 	int nid = pgdat->node_id;
1495 	enum zone_type z;
1496 	int cpu;
1497 
1498 	pgdat_init_internals(pgdat);
1499 
1500 	if (pgdat->per_cpu_nodestats == &boot_nodestats)
1501 		pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1502 
1503 	/*
1504 	 * Reset the nr_zones, order and highest_zoneidx before reuse.
1505 	 * Note that kswapd will init kswapd_highest_zoneidx properly
1506 	 * when it starts in the near future.
1507 	 */
1508 	pgdat->nr_zones = 0;
1509 	pgdat->kswapd_order = 0;
1510 	pgdat->kswapd_highest_zoneidx = 0;
1511 	pgdat->node_start_pfn = 0;
1512 	pgdat->node_present_pages = 0;
1513 
1514 	for_each_online_cpu(cpu) {
1515 		struct per_cpu_nodestat *p;
1516 
1517 		p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
1518 		memset(p, 0, sizeof(*p));
1519 	}
1520 
1521 	/*
1522 	 * When memory is hot-added, all the memory is in offline state. So
1523 	 * clear all zones' present_pages and managed_pages because they will
1524 	 * be updated in online_pages() and offline_pages().
1525 	 */
1526 	for (z = 0; z < MAX_NR_ZONES; z++) {
1527 		struct zone *zone = pgdat->node_zones + z;
1528 
1529 		zone->present_pages = 0;
1530 		zone_init_internals(zone, z, nid, 0);
1531 	}
1532 }
1533 #endif
1534 
1535 /*
1536  * Set up the zone data structures:
1537  *   - mark all pages reserved
1538  *   - mark all memory queues empty
1539  *   - clear the memory bitmaps
1540  *
1541  * NOTE: pgdat should get zeroed by caller.
1542  * NOTE: this function is only called during early init.
1543  */
1544 static void __init free_area_init_core(struct pglist_data *pgdat)
1545 {
1546 	enum zone_type j;
1547 	int nid = pgdat->node_id;
1548 
1549 	pgdat_init_internals(pgdat);
1550 	pgdat->per_cpu_nodestats = &boot_nodestats;
1551 
1552 	for (j = 0; j < MAX_NR_ZONES; j++) {
1553 		struct zone *zone = pgdat->node_zones + j;
1554 		unsigned long size, freesize, memmap_pages;
1555 
1556 		size = zone->spanned_pages;
1557 		freesize = zone->present_pages;
1558 
1559 		/*
1560 		 * Adjust freesize so that it accounts for how much memory
1561 		 * is used by this zone for memmap. This affects the watermark
1562 		 * and per-cpu initialisations
1563 		 */
1564 		memmap_pages = calc_memmap_size(size, freesize);
1565 		if (!is_highmem_idx(j)) {
1566 			if (freesize >= memmap_pages) {
1567 				freesize -= memmap_pages;
1568 				if (memmap_pages)
1569 					pr_debug("  %s zone: %lu pages used for memmap\n",
1570 						 zone_names[j], memmap_pages);
1571 			} else
1572 				pr_warn("  %s zone: %lu memmap pages exceeds freesize %lu\n",
1573 					zone_names[j], memmap_pages, freesize);
1574 		}
1575 
1576 		/* Account for reserved pages */
1577 		if (j == 0 && freesize > dma_reserve) {
1578 			freesize -= dma_reserve;
1579 			pr_debug("  %s zone: %lu pages reserved\n", zone_names[0], dma_reserve);
1580 		}
1581 
1582 		if (!is_highmem_idx(j))
1583 			nr_kernel_pages += freesize;
1584 		/* Charge for highmem memmap if there are enough kernel pages */
1585 		else if (nr_kernel_pages > memmap_pages * 2)
1586 			nr_kernel_pages -= memmap_pages;
1587 		nr_all_pages += freesize;
1588 
1589 		/*
1590 		 * Set an approximate value for lowmem here, it will be adjusted
1591 		 * when the bootmem allocator frees pages into the buddy system.
1592 		 * And all highmem pages will be managed by the buddy system.
1593 		 */
1594 		zone_init_internals(zone, j, nid, freesize);
1595 
1596 		if (!size)
1597 			continue;
1598 
1599 		setup_usemap(zone);
1600 		init_currently_empty_zone(zone, zone->zone_start_pfn, size);
1601 	}
1602 }
1603 
1604 void __init *memmap_alloc(phys_addr_t size, phys_addr_t align,
1605 			  phys_addr_t min_addr, int nid, bool exact_nid)
1606 {
1607 	void *ptr;
1608 
1609 	if (exact_nid)
1610 		ptr = memblock_alloc_exact_nid_raw(size, align, min_addr,
1611 						   MEMBLOCK_ALLOC_ACCESSIBLE,
1612 						   nid);
1613 	else
1614 		ptr = memblock_alloc_try_nid_raw(size, align, min_addr,
1615 						 MEMBLOCK_ALLOC_ACCESSIBLE,
1616 						 nid);
1617 
1618 	if (ptr && size > 0)
1619 		page_init_poison(ptr, size);
1620 
1621 	return ptr;
1622 }
1623 
1624 #ifdef CONFIG_FLATMEM
1625 static void __init alloc_node_mem_map(struct pglist_data *pgdat)
1626 {
1627 	unsigned long __maybe_unused start = 0;
1628 	unsigned long __maybe_unused offset = 0;
1629 
1630 	/* Skip empty nodes */
1631 	if (!pgdat->node_spanned_pages)
1632 		return;
1633 
1634 	start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
1635 	offset = pgdat->node_start_pfn - start;
1636 	/* ia64 gets its own node_mem_map, before this, without bootmem */
1637 	if (!pgdat->node_mem_map) {
1638 		unsigned long size, end;
1639 		struct page *map;
1640 
1641 		/*
1642 		 * The zone's endpoints aren't required to be MAX_ORDER
1643 		 * aligned but the node_mem_map endpoints must be in order
1644 		 * for the buddy allocator to function correctly.
1645 		 */
1646 		end = pgdat_end_pfn(pgdat);
1647 		end = ALIGN(end, MAX_ORDER_NR_PAGES);
1648 		size =  (end - start) * sizeof(struct page);
1649 		map = memmap_alloc(size, SMP_CACHE_BYTES, MEMBLOCK_LOW_LIMIT,
1650 				   pgdat->node_id, false);
1651 		if (!map)
1652 			panic("Failed to allocate %ld bytes for node %d memory map\n",
1653 			      size, pgdat->node_id);
1654 		pgdat->node_mem_map = map + offset;
1655 	}
1656 	pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
1657 				__func__, pgdat->node_id, (unsigned long)pgdat,
1658 				(unsigned long)pgdat->node_mem_map);
1659 #ifndef CONFIG_NUMA
1660 	/*
1661 	 * With no DISCONTIG, the global mem_map is just set as node 0's
1662 	 */
1663 	if (pgdat == NODE_DATA(0)) {
1664 		mem_map = NODE_DATA(0)->node_mem_map;
1665 		if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
1666 			mem_map -= offset;
1667 	}
1668 #endif
1669 }
1670 #else
1671 static inline void alloc_node_mem_map(struct pglist_data *pgdat) { }
1672 #endif /* CONFIG_FLATMEM */
1673 
1674 /**
1675  * get_pfn_range_for_nid - Return the start and end page frames for a node
1676  * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
1677  * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
1678  * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
1679  *
1680  * It returns the start and end page frame of a node based on information
1681  * provided by memblock_set_node(). If called for a node
1682  * with no available memory, the start and end PFNs will be 0.
1683  */
1684 void __init get_pfn_range_for_nid(unsigned int nid,
1685 			unsigned long *start_pfn, unsigned long *end_pfn)
1686 {
1687 	unsigned long this_start_pfn, this_end_pfn;
1688 	int i;
1689 
1690 	*start_pfn = -1UL;
1691 	*end_pfn = 0;
1692 
1693 	for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
1694 		*start_pfn = min(*start_pfn, this_start_pfn);
1695 		*end_pfn = max(*end_pfn, this_end_pfn);
1696 	}
1697 
1698 	if (*start_pfn == -1UL)
1699 		*start_pfn = 0;
1700 }
1701 
1702 static void __init free_area_init_node(int nid)
1703 {
1704 	pg_data_t *pgdat = NODE_DATA(nid);
1705 	unsigned long start_pfn = 0;
1706 	unsigned long end_pfn = 0;
1707 
1708 	/* pg_data_t should be reset to zero when it's allocated */
1709 	WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx);
1710 
1711 	get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
1712 
1713 	pgdat->node_id = nid;
1714 	pgdat->node_start_pfn = start_pfn;
1715 	pgdat->per_cpu_nodestats = NULL;
1716 
1717 	if (start_pfn != end_pfn) {
1718 		pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
1719 			(u64)start_pfn << PAGE_SHIFT,
1720 			end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
1721 
1722 		calculate_node_totalpages(pgdat, start_pfn, end_pfn);
1723 	} else {
1724 		pr_info("Initmem setup node %d as memoryless\n", nid);
1725 
1726 		reset_memoryless_node_totalpages(pgdat);
1727 	}
1728 
1729 	alloc_node_mem_map(pgdat);
1730 	pgdat_set_deferred_range(pgdat);
1731 
1732 	free_area_init_core(pgdat);
1733 	lru_gen_init_pgdat(pgdat);
1734 }
1735 
1736 /* Any regular or high memory on that node ? */
1737 static void __init check_for_memory(pg_data_t *pgdat)
1738 {
1739 	enum zone_type zone_type;
1740 
1741 	for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
1742 		struct zone *zone = &pgdat->node_zones[zone_type];
1743 		if (populated_zone(zone)) {
1744 			if (IS_ENABLED(CONFIG_HIGHMEM))
1745 				node_set_state(pgdat->node_id, N_HIGH_MEMORY);
1746 			if (zone_type <= ZONE_NORMAL)
1747 				node_set_state(pgdat->node_id, N_NORMAL_MEMORY);
1748 			break;
1749 		}
1750 	}
1751 }
1752 
1753 #if MAX_NUMNODES > 1
1754 /*
1755  * Figure out the number of possible node ids.
1756  */
1757 void __init setup_nr_node_ids(void)
1758 {
1759 	unsigned int highest;
1760 
1761 	highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
1762 	nr_node_ids = highest + 1;
1763 }
1764 #endif
1765 
1766 /*
1767  * Some architectures, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For
1768  * such cases we allow max_zone_pfn sorted in the descending order
1769  */
1770 static bool arch_has_descending_max_zone_pfns(void)
1771 {
1772 	return IS_ENABLED(CONFIG_ARC) && !IS_ENABLED(CONFIG_ARC_HAS_PAE40);
1773 }
1774 
1775 /**
1776  * free_area_init - Initialise all pg_data_t and zone data
1777  * @max_zone_pfn: an array of max PFNs for each zone
1778  *
1779  * This will call free_area_init_node() for each active node in the system.
1780  * Using the page ranges provided by memblock_set_node(), the size of each
1781  * zone in each node and their holes is calculated. If the maximum PFN
1782  * between two adjacent zones match, it is assumed that the zone is empty.
1783  * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
1784  * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
1785  * starts where the previous one ended. For example, ZONE_DMA32 starts
1786  * at arch_max_dma_pfn.
1787  */
1788 void __init free_area_init(unsigned long *max_zone_pfn)
1789 {
1790 	unsigned long start_pfn, end_pfn;
1791 	int i, nid, zone;
1792 	bool descending;
1793 
1794 	/* Record where the zone boundaries are */
1795 	memset(arch_zone_lowest_possible_pfn, 0,
1796 				sizeof(arch_zone_lowest_possible_pfn));
1797 	memset(arch_zone_highest_possible_pfn, 0,
1798 				sizeof(arch_zone_highest_possible_pfn));
1799 
1800 	start_pfn = PHYS_PFN(memblock_start_of_DRAM());
1801 	descending = arch_has_descending_max_zone_pfns();
1802 
1803 	for (i = 0; i < MAX_NR_ZONES; i++) {
1804 		if (descending)
1805 			zone = MAX_NR_ZONES - i - 1;
1806 		else
1807 			zone = i;
1808 
1809 		if (zone == ZONE_MOVABLE)
1810 			continue;
1811 
1812 		end_pfn = max(max_zone_pfn[zone], start_pfn);
1813 		arch_zone_lowest_possible_pfn[zone] = start_pfn;
1814 		arch_zone_highest_possible_pfn[zone] = end_pfn;
1815 
1816 		start_pfn = end_pfn;
1817 	}
1818 
1819 	/* Find the PFNs that ZONE_MOVABLE begins at in each node */
1820 	memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
1821 	find_zone_movable_pfns_for_nodes();
1822 
1823 	/* Print out the zone ranges */
1824 	pr_info("Zone ranges:\n");
1825 	for (i = 0; i < MAX_NR_ZONES; i++) {
1826 		if (i == ZONE_MOVABLE)
1827 			continue;
1828 		pr_info("  %-8s ", zone_names[i]);
1829 		if (arch_zone_lowest_possible_pfn[i] ==
1830 				arch_zone_highest_possible_pfn[i])
1831 			pr_cont("empty\n");
1832 		else
1833 			pr_cont("[mem %#018Lx-%#018Lx]\n",
1834 				(u64)arch_zone_lowest_possible_pfn[i]
1835 					<< PAGE_SHIFT,
1836 				((u64)arch_zone_highest_possible_pfn[i]
1837 					<< PAGE_SHIFT) - 1);
1838 	}
1839 
1840 	/* Print out the PFNs ZONE_MOVABLE begins at in each node */
1841 	pr_info("Movable zone start for each node\n");
1842 	for (i = 0; i < MAX_NUMNODES; i++) {
1843 		if (zone_movable_pfn[i])
1844 			pr_info("  Node %d: %#018Lx\n", i,
1845 			       (u64)zone_movable_pfn[i] << PAGE_SHIFT);
1846 	}
1847 
1848 	/*
1849 	 * Print out the early node map, and initialize the
1850 	 * subsection-map relative to active online memory ranges to
1851 	 * enable future "sub-section" extensions of the memory map.
1852 	 */
1853 	pr_info("Early memory node ranges\n");
1854 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
1855 		pr_info("  node %3d: [mem %#018Lx-%#018Lx]\n", nid,
1856 			(u64)start_pfn << PAGE_SHIFT,
1857 			((u64)end_pfn << PAGE_SHIFT) - 1);
1858 		subsection_map_init(start_pfn, end_pfn - start_pfn);
1859 	}
1860 
1861 	/* Initialise every node */
1862 	mminit_verify_pageflags_layout();
1863 	setup_nr_node_ids();
1864 	set_pageblock_order();
1865 
1866 	for_each_node(nid) {
1867 		pg_data_t *pgdat;
1868 
1869 		if (!node_online(nid)) {
1870 			pr_info("Initializing node %d as memoryless\n", nid);
1871 
1872 			/* Allocator not initialized yet */
1873 			pgdat = arch_alloc_nodedata(nid);
1874 			if (!pgdat)
1875 				panic("Cannot allocate %zuB for node %d.\n",
1876 				       sizeof(*pgdat), nid);
1877 			arch_refresh_nodedata(nid, pgdat);
1878 			free_area_init_node(nid);
1879 
1880 			/*
1881 			 * We do not want to confuse userspace by sysfs
1882 			 * files/directories for node without any memory
1883 			 * attached to it, so this node is not marked as
1884 			 * N_MEMORY and not marked online so that no sysfs
1885 			 * hierarchy will be created via register_one_node for
1886 			 * it. The pgdat will get fully initialized by
1887 			 * hotadd_init_pgdat() when memory is hotplugged into
1888 			 * this node.
1889 			 */
1890 			continue;
1891 		}
1892 
1893 		pgdat = NODE_DATA(nid);
1894 		free_area_init_node(nid);
1895 
1896 		/* Any memory on that node */
1897 		if (pgdat->node_present_pages)
1898 			node_set_state(nid, N_MEMORY);
1899 		check_for_memory(pgdat);
1900 	}
1901 
1902 	memmap_init();
1903 
1904 	/* disable hash distribution for systems with a single node */
1905 	fixup_hashdist();
1906 }
1907 
1908 /**
1909  * node_map_pfn_alignment - determine the maximum internode alignment
1910  *
1911  * This function should be called after node map is populated and sorted.
1912  * It calculates the maximum power of two alignment which can distinguish
1913  * all the nodes.
1914  *
1915  * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
1916  * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
1917  * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
1918  * shifted, 1GiB is enough and this function will indicate so.
1919  *
1920  * This is used to test whether pfn -> nid mapping of the chosen memory
1921  * model has fine enough granularity to avoid incorrect mapping for the
1922  * populated node map.
1923  *
1924  * Return: the determined alignment in pfn's.  0 if there is no alignment
1925  * requirement (single node).
1926  */
1927 unsigned long __init node_map_pfn_alignment(void)
1928 {
1929 	unsigned long accl_mask = 0, last_end = 0;
1930 	unsigned long start, end, mask;
1931 	int last_nid = NUMA_NO_NODE;
1932 	int i, nid;
1933 
1934 	for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
1935 		if (!start || last_nid < 0 || last_nid == nid) {
1936 			last_nid = nid;
1937 			last_end = end;
1938 			continue;
1939 		}
1940 
1941 		/*
1942 		 * Start with a mask granular enough to pin-point to the
1943 		 * start pfn and tick off bits one-by-one until it becomes
1944 		 * too coarse to separate the current node from the last.
1945 		 */
1946 		mask = ~((1 << __ffs(start)) - 1);
1947 		while (mask && last_end <= (start & (mask << 1)))
1948 			mask <<= 1;
1949 
1950 		/* accumulate all internode masks */
1951 		accl_mask |= mask;
1952 	}
1953 
1954 	/* convert mask to number of pages */
1955 	return ~accl_mask + 1;
1956 }
1957 
1958 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1959 static void __init deferred_free_range(unsigned long pfn,
1960 				       unsigned long nr_pages)
1961 {
1962 	struct page *page;
1963 	unsigned long i;
1964 
1965 	if (!nr_pages)
1966 		return;
1967 
1968 	page = pfn_to_page(pfn);
1969 
1970 	/* Free a large naturally-aligned chunk if possible */
1971 	if (nr_pages == MAX_ORDER_NR_PAGES && IS_MAX_ORDER_ALIGNED(pfn)) {
1972 		for (i = 0; i < nr_pages; i += pageblock_nr_pages)
1973 			set_pageblock_migratetype(page + i, MIGRATE_MOVABLE);
1974 		__free_pages_core(page, MAX_ORDER);
1975 		return;
1976 	}
1977 
1978 	/* Accept chunks smaller than MAX_ORDER upfront */
1979 	accept_memory(PFN_PHYS(pfn), PFN_PHYS(pfn + nr_pages));
1980 
1981 	for (i = 0; i < nr_pages; i++, page++, pfn++) {
1982 		if (pageblock_aligned(pfn))
1983 			set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1984 		__free_pages_core(page, 0);
1985 	}
1986 }
1987 
1988 /* Completion tracking for deferred_init_memmap() threads */
1989 static atomic_t pgdat_init_n_undone __initdata;
1990 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1991 
1992 static inline void __init pgdat_init_report_one_done(void)
1993 {
1994 	if (atomic_dec_and_test(&pgdat_init_n_undone))
1995 		complete(&pgdat_init_all_done_comp);
1996 }
1997 
1998 /*
1999  * Returns true if page needs to be initialized or freed to buddy allocator.
2000  *
2001  * We check if a current MAX_ORDER block is valid by only checking the validity
2002  * of the head pfn.
2003  */
2004 static inline bool __init deferred_pfn_valid(unsigned long pfn)
2005 {
2006 	if (IS_MAX_ORDER_ALIGNED(pfn) && !pfn_valid(pfn))
2007 		return false;
2008 	return true;
2009 }
2010 
2011 /*
2012  * Free pages to buddy allocator. Try to free aligned pages in
2013  * MAX_ORDER_NR_PAGES sizes.
2014  */
2015 static void __init deferred_free_pages(unsigned long pfn,
2016 				       unsigned long end_pfn)
2017 {
2018 	unsigned long nr_free = 0;
2019 
2020 	for (; pfn < end_pfn; pfn++) {
2021 		if (!deferred_pfn_valid(pfn)) {
2022 			deferred_free_range(pfn - nr_free, nr_free);
2023 			nr_free = 0;
2024 		} else if (IS_MAX_ORDER_ALIGNED(pfn)) {
2025 			deferred_free_range(pfn - nr_free, nr_free);
2026 			nr_free = 1;
2027 		} else {
2028 			nr_free++;
2029 		}
2030 	}
2031 	/* Free the last block of pages to allocator */
2032 	deferred_free_range(pfn - nr_free, nr_free);
2033 }
2034 
2035 /*
2036  * Initialize struct pages.  We minimize pfn page lookups and scheduler checks
2037  * by performing it only once every MAX_ORDER_NR_PAGES.
2038  * Return number of pages initialized.
2039  */
2040 static unsigned long  __init deferred_init_pages(struct zone *zone,
2041 						 unsigned long pfn,
2042 						 unsigned long end_pfn)
2043 {
2044 	int nid = zone_to_nid(zone);
2045 	unsigned long nr_pages = 0;
2046 	int zid = zone_idx(zone);
2047 	struct page *page = NULL;
2048 
2049 	for (; pfn < end_pfn; pfn++) {
2050 		if (!deferred_pfn_valid(pfn)) {
2051 			page = NULL;
2052 			continue;
2053 		} else if (!page || IS_MAX_ORDER_ALIGNED(pfn)) {
2054 			page = pfn_to_page(pfn);
2055 		} else {
2056 			page++;
2057 		}
2058 		__init_single_page(page, pfn, zid, nid);
2059 		nr_pages++;
2060 	}
2061 	return (nr_pages);
2062 }
2063 
2064 /*
2065  * This function is meant to pre-load the iterator for the zone init.
2066  * Specifically it walks through the ranges until we are caught up to the
2067  * first_init_pfn value and exits there. If we never encounter the value we
2068  * return false indicating there are no valid ranges left.
2069  */
2070 static bool __init
2071 deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone,
2072 				    unsigned long *spfn, unsigned long *epfn,
2073 				    unsigned long first_init_pfn)
2074 {
2075 	u64 j;
2076 
2077 	/*
2078 	 * Start out by walking through the ranges in this zone that have
2079 	 * already been initialized. We don't need to do anything with them
2080 	 * so we just need to flush them out of the system.
2081 	 */
2082 	for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) {
2083 		if (*epfn <= first_init_pfn)
2084 			continue;
2085 		if (*spfn < first_init_pfn)
2086 			*spfn = first_init_pfn;
2087 		*i = j;
2088 		return true;
2089 	}
2090 
2091 	return false;
2092 }
2093 
2094 /*
2095  * Initialize and free pages. We do it in two loops: first we initialize
2096  * struct page, then free to buddy allocator, because while we are
2097  * freeing pages we can access pages that are ahead (computing buddy
2098  * page in __free_one_page()).
2099  *
2100  * In order to try and keep some memory in the cache we have the loop
2101  * broken along max page order boundaries. This way we will not cause
2102  * any issues with the buddy page computation.
2103  */
2104 static unsigned long __init
2105 deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn,
2106 		       unsigned long *end_pfn)
2107 {
2108 	unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES);
2109 	unsigned long spfn = *start_pfn, epfn = *end_pfn;
2110 	unsigned long nr_pages = 0;
2111 	u64 j = *i;
2112 
2113 	/* First we loop through and initialize the page values */
2114 	for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) {
2115 		unsigned long t;
2116 
2117 		if (mo_pfn <= *start_pfn)
2118 			break;
2119 
2120 		t = min(mo_pfn, *end_pfn);
2121 		nr_pages += deferred_init_pages(zone, *start_pfn, t);
2122 
2123 		if (mo_pfn < *end_pfn) {
2124 			*start_pfn = mo_pfn;
2125 			break;
2126 		}
2127 	}
2128 
2129 	/* Reset values and now loop through freeing pages as needed */
2130 	swap(j, *i);
2131 
2132 	for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) {
2133 		unsigned long t;
2134 
2135 		if (mo_pfn <= spfn)
2136 			break;
2137 
2138 		t = min(mo_pfn, epfn);
2139 		deferred_free_pages(spfn, t);
2140 
2141 		if (mo_pfn <= epfn)
2142 			break;
2143 	}
2144 
2145 	return nr_pages;
2146 }
2147 
2148 static void __init
2149 deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
2150 			   void *arg)
2151 {
2152 	unsigned long spfn, epfn;
2153 	struct zone *zone = arg;
2154 	u64 i;
2155 
2156 	deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn);
2157 
2158 	/*
2159 	 * Initialize and free pages in MAX_ORDER sized increments so that we
2160 	 * can avoid introducing any issues with the buddy allocator.
2161 	 */
2162 	while (spfn < end_pfn) {
2163 		deferred_init_maxorder(&i, zone, &spfn, &epfn);
2164 		cond_resched();
2165 	}
2166 }
2167 
2168 /* An arch may override for more concurrency. */
2169 __weak int __init
2170 deferred_page_init_max_threads(const struct cpumask *node_cpumask)
2171 {
2172 	return 1;
2173 }
2174 
2175 /* Initialise remaining memory on a node */
2176 static int __init deferred_init_memmap(void *data)
2177 {
2178 	pg_data_t *pgdat = data;
2179 	const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
2180 	unsigned long spfn = 0, epfn = 0;
2181 	unsigned long first_init_pfn, flags;
2182 	unsigned long start = jiffies;
2183 	struct zone *zone;
2184 	int zid, max_threads;
2185 	u64 i;
2186 
2187 	/* Bind memory initialisation thread to a local node if possible */
2188 	if (!cpumask_empty(cpumask))
2189 		set_cpus_allowed_ptr(current, cpumask);
2190 
2191 	pgdat_resize_lock(pgdat, &flags);
2192 	first_init_pfn = pgdat->first_deferred_pfn;
2193 	if (first_init_pfn == ULONG_MAX) {
2194 		pgdat_resize_unlock(pgdat, &flags);
2195 		pgdat_init_report_one_done();
2196 		return 0;
2197 	}
2198 
2199 	/* Sanity check boundaries */
2200 	BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
2201 	BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
2202 	pgdat->first_deferred_pfn = ULONG_MAX;
2203 
2204 	/*
2205 	 * Once we unlock here, the zone cannot be grown anymore, thus if an
2206 	 * interrupt thread must allocate this early in boot, zone must be
2207 	 * pre-grown prior to start of deferred page initialization.
2208 	 */
2209 	pgdat_resize_unlock(pgdat, &flags);
2210 
2211 	/* Only the highest zone is deferred so find it */
2212 	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2213 		zone = pgdat->node_zones + zid;
2214 		if (first_init_pfn < zone_end_pfn(zone))
2215 			break;
2216 	}
2217 
2218 	/* If the zone is empty somebody else may have cleared out the zone */
2219 	if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2220 						 first_init_pfn))
2221 		goto zone_empty;
2222 
2223 	max_threads = deferred_page_init_max_threads(cpumask);
2224 
2225 	while (spfn < epfn) {
2226 		unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION);
2227 		struct padata_mt_job job = {
2228 			.thread_fn   = deferred_init_memmap_chunk,
2229 			.fn_arg      = zone,
2230 			.start       = spfn,
2231 			.size        = epfn_align - spfn,
2232 			.align       = PAGES_PER_SECTION,
2233 			.min_chunk   = PAGES_PER_SECTION,
2234 			.max_threads = max_threads,
2235 		};
2236 
2237 		padata_do_multithreaded(&job);
2238 		deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2239 						    epfn_align);
2240 	}
2241 zone_empty:
2242 	/* Sanity check that the next zone really is unpopulated */
2243 	WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
2244 
2245 	pr_info("node %d deferred pages initialised in %ums\n",
2246 		pgdat->node_id, jiffies_to_msecs(jiffies - start));
2247 
2248 	pgdat_init_report_one_done();
2249 	return 0;
2250 }
2251 
2252 /*
2253  * If this zone has deferred pages, try to grow it by initializing enough
2254  * deferred pages to satisfy the allocation specified by order, rounded up to
2255  * the nearest PAGES_PER_SECTION boundary.  So we're adding memory in increments
2256  * of SECTION_SIZE bytes by initializing struct pages in increments of
2257  * PAGES_PER_SECTION * sizeof(struct page) bytes.
2258  *
2259  * Return true when zone was grown, otherwise return false. We return true even
2260  * when we grow less than requested, to let the caller decide if there are
2261  * enough pages to satisfy the allocation.
2262  *
2263  * Note: We use noinline because this function is needed only during boot, and
2264  * it is called from a __ref function _deferred_grow_zone. This way we are
2265  * making sure that it is not inlined into permanent text section.
2266  */
2267 bool __init deferred_grow_zone(struct zone *zone, unsigned int order)
2268 {
2269 	unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
2270 	pg_data_t *pgdat = zone->zone_pgdat;
2271 	unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
2272 	unsigned long spfn, epfn, flags;
2273 	unsigned long nr_pages = 0;
2274 	u64 i;
2275 
2276 	/* Only the last zone may have deferred pages */
2277 	if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
2278 		return false;
2279 
2280 	pgdat_resize_lock(pgdat, &flags);
2281 
2282 	/*
2283 	 * If someone grew this zone while we were waiting for spinlock, return
2284 	 * true, as there might be enough pages already.
2285 	 */
2286 	if (first_deferred_pfn != pgdat->first_deferred_pfn) {
2287 		pgdat_resize_unlock(pgdat, &flags);
2288 		return true;
2289 	}
2290 
2291 	/* If the zone is empty somebody else may have cleared out the zone */
2292 	if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2293 						 first_deferred_pfn)) {
2294 		pgdat->first_deferred_pfn = ULONG_MAX;
2295 		pgdat_resize_unlock(pgdat, &flags);
2296 		/* Retry only once. */
2297 		return first_deferred_pfn != ULONG_MAX;
2298 	}
2299 
2300 	/*
2301 	 * Initialize and free pages in MAX_ORDER sized increments so
2302 	 * that we can avoid introducing any issues with the buddy
2303 	 * allocator.
2304 	 */
2305 	while (spfn < epfn) {
2306 		/* update our first deferred PFN for this section */
2307 		first_deferred_pfn = spfn;
2308 
2309 		nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn);
2310 		touch_nmi_watchdog();
2311 
2312 		/* We should only stop along section boundaries */
2313 		if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION)
2314 			continue;
2315 
2316 		/* If our quota has been met we can stop here */
2317 		if (nr_pages >= nr_pages_needed)
2318 			break;
2319 	}
2320 
2321 	pgdat->first_deferred_pfn = spfn;
2322 	pgdat_resize_unlock(pgdat, &flags);
2323 
2324 	return nr_pages > 0;
2325 }
2326 
2327 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
2328 
2329 #ifdef CONFIG_CMA
2330 void __init init_cma_reserved_pageblock(struct page *page)
2331 {
2332 	unsigned i = pageblock_nr_pages;
2333 	struct page *p = page;
2334 
2335 	do {
2336 		__ClearPageReserved(p);
2337 		set_page_count(p, 0);
2338 	} while (++p, --i);
2339 
2340 	set_pageblock_migratetype(page, MIGRATE_CMA);
2341 	set_page_refcounted(page);
2342 	__free_pages(page, pageblock_order);
2343 
2344 	adjust_managed_page_count(page, pageblock_nr_pages);
2345 	page_zone(page)->cma_pages += pageblock_nr_pages;
2346 }
2347 #endif
2348 
2349 void set_zone_contiguous(struct zone *zone)
2350 {
2351 	unsigned long block_start_pfn = zone->zone_start_pfn;
2352 	unsigned long block_end_pfn;
2353 
2354 	block_end_pfn = pageblock_end_pfn(block_start_pfn);
2355 	for (; block_start_pfn < zone_end_pfn(zone);
2356 			block_start_pfn = block_end_pfn,
2357 			 block_end_pfn += pageblock_nr_pages) {
2358 
2359 		block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
2360 
2361 		if (!__pageblock_pfn_to_page(block_start_pfn,
2362 					     block_end_pfn, zone))
2363 			return;
2364 		cond_resched();
2365 	}
2366 
2367 	/* We confirm that there is no hole */
2368 	zone->contiguous = true;
2369 }
2370 
2371 void __init page_alloc_init_late(void)
2372 {
2373 	struct zone *zone;
2374 	int nid;
2375 
2376 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
2377 
2378 	/* There will be num_node_state(N_MEMORY) threads */
2379 	atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
2380 	for_each_node_state(nid, N_MEMORY) {
2381 		kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
2382 	}
2383 
2384 	/* Block until all are initialised */
2385 	wait_for_completion(&pgdat_init_all_done_comp);
2386 
2387 	/*
2388 	 * We initialized the rest of the deferred pages.  Permanently disable
2389 	 * on-demand struct page initialization.
2390 	 */
2391 	static_branch_disable(&deferred_pages);
2392 
2393 	/* Reinit limits that are based on free pages after the kernel is up */
2394 	files_maxfiles_init();
2395 #endif
2396 
2397 	buffer_init();
2398 
2399 	/* Discard memblock private memory */
2400 	memblock_discard();
2401 
2402 	for_each_node_state(nid, N_MEMORY)
2403 		shuffle_free_memory(NODE_DATA(nid));
2404 
2405 	for_each_populated_zone(zone)
2406 		set_zone_contiguous(zone);
2407 
2408 	/* Initialize page ext after all struct pages are initialized. */
2409 	if (deferred_struct_pages)
2410 		page_ext_init();
2411 
2412 	page_alloc_sysctl_init();
2413 }
2414 
2415 #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
2416 /*
2417  * Returns the number of pages that arch has reserved but
2418  * is not known to alloc_large_system_hash().
2419  */
2420 static unsigned long __init arch_reserved_kernel_pages(void)
2421 {
2422 	return 0;
2423 }
2424 #endif
2425 
2426 /*
2427  * Adaptive scale is meant to reduce sizes of hash tables on large memory
2428  * machines. As memory size is increased the scale is also increased but at
2429  * slower pace.  Starting from ADAPT_SCALE_BASE (64G), every time memory
2430  * quadruples the scale is increased by one, which means the size of hash table
2431  * only doubles, instead of quadrupling as well.
2432  * Because 32-bit systems cannot have large physical memory, where this scaling
2433  * makes sense, it is disabled on such platforms.
2434  */
2435 #if __BITS_PER_LONG > 32
2436 #define ADAPT_SCALE_BASE	(64ul << 30)
2437 #define ADAPT_SCALE_SHIFT	2
2438 #define ADAPT_SCALE_NPAGES	(ADAPT_SCALE_BASE >> PAGE_SHIFT)
2439 #endif
2440 
2441 /*
2442  * allocate a large system hash table from bootmem
2443  * - it is assumed that the hash table must contain an exact power-of-2
2444  *   quantity of entries
2445  * - limit is the number of hash buckets, not the total allocation size
2446  */
2447 void *__init alloc_large_system_hash(const char *tablename,
2448 				     unsigned long bucketsize,
2449 				     unsigned long numentries,
2450 				     int scale,
2451 				     int flags,
2452 				     unsigned int *_hash_shift,
2453 				     unsigned int *_hash_mask,
2454 				     unsigned long low_limit,
2455 				     unsigned long high_limit)
2456 {
2457 	unsigned long long max = high_limit;
2458 	unsigned long log2qty, size;
2459 	void *table;
2460 	gfp_t gfp_flags;
2461 	bool virt;
2462 	bool huge;
2463 
2464 	/* allow the kernel cmdline to have a say */
2465 	if (!numentries) {
2466 		/* round applicable memory size up to nearest megabyte */
2467 		numentries = nr_kernel_pages;
2468 		numentries -= arch_reserved_kernel_pages();
2469 
2470 		/* It isn't necessary when PAGE_SIZE >= 1MB */
2471 		if (PAGE_SIZE < SZ_1M)
2472 			numentries = round_up(numentries, SZ_1M / PAGE_SIZE);
2473 
2474 #if __BITS_PER_LONG > 32
2475 		if (!high_limit) {
2476 			unsigned long adapt;
2477 
2478 			for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
2479 			     adapt <<= ADAPT_SCALE_SHIFT)
2480 				scale++;
2481 		}
2482 #endif
2483 
2484 		/* limit to 1 bucket per 2^scale bytes of low memory */
2485 		if (scale > PAGE_SHIFT)
2486 			numentries >>= (scale - PAGE_SHIFT);
2487 		else
2488 			numentries <<= (PAGE_SHIFT - scale);
2489 
2490 		if (unlikely((numentries * bucketsize) < PAGE_SIZE))
2491 			numentries = PAGE_SIZE / bucketsize;
2492 	}
2493 	numentries = roundup_pow_of_two(numentries);
2494 
2495 	/* limit allocation size to 1/16 total memory by default */
2496 	if (max == 0) {
2497 		max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2498 		do_div(max, bucketsize);
2499 	}
2500 	max = min(max, 0x80000000ULL);
2501 
2502 	if (numentries < low_limit)
2503 		numentries = low_limit;
2504 	if (numentries > max)
2505 		numentries = max;
2506 
2507 	log2qty = ilog2(numentries);
2508 
2509 	gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
2510 	do {
2511 		virt = false;
2512 		size = bucketsize << log2qty;
2513 		if (flags & HASH_EARLY) {
2514 			if (flags & HASH_ZERO)
2515 				table = memblock_alloc(size, SMP_CACHE_BYTES);
2516 			else
2517 				table = memblock_alloc_raw(size,
2518 							   SMP_CACHE_BYTES);
2519 		} else if (get_order(size) > MAX_ORDER || hashdist) {
2520 			table = vmalloc_huge(size, gfp_flags);
2521 			virt = true;
2522 			if (table)
2523 				huge = is_vm_area_hugepages(table);
2524 		} else {
2525 			/*
2526 			 * If bucketsize is not a power-of-two, we may free
2527 			 * some pages at the end of hash table which
2528 			 * alloc_pages_exact() automatically does
2529 			 */
2530 			table = alloc_pages_exact(size, gfp_flags);
2531 			kmemleak_alloc(table, size, 1, gfp_flags);
2532 		}
2533 	} while (!table && size > PAGE_SIZE && --log2qty);
2534 
2535 	if (!table)
2536 		panic("Failed to allocate %s hash table\n", tablename);
2537 
2538 	pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n",
2539 		tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size,
2540 		virt ? (huge ? "vmalloc hugepage" : "vmalloc") : "linear");
2541 
2542 	if (_hash_shift)
2543 		*_hash_shift = log2qty;
2544 	if (_hash_mask)
2545 		*_hash_mask = (1 << log2qty) - 1;
2546 
2547 	return table;
2548 }
2549 
2550 /**
2551  * set_dma_reserve - set the specified number of pages reserved in the first zone
2552  * @new_dma_reserve: The number of pages to mark reserved
2553  *
2554  * The per-cpu batchsize and zone watermarks are determined by managed_pages.
2555  * In the DMA zone, a significant percentage may be consumed by kernel image
2556  * and other unfreeable allocations which can skew the watermarks badly. This
2557  * function may optionally be used to account for unfreeable pages in the
2558  * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
2559  * smaller per-cpu batchsize.
2560  */
2561 void __init set_dma_reserve(unsigned long new_dma_reserve)
2562 {
2563 	dma_reserve = new_dma_reserve;
2564 }
2565 
2566 void __init memblock_free_pages(struct page *page, unsigned long pfn,
2567 							unsigned int order)
2568 {
2569 
2570 	if (IS_ENABLED(CONFIG_DEFERRED_STRUCT_PAGE_INIT)) {
2571 		int nid = early_pfn_to_nid(pfn);
2572 
2573 		if (!early_page_initialised(pfn, nid))
2574 			return;
2575 	}
2576 
2577 	if (!kmsan_memblock_free_pages(page, order)) {
2578 		/* KMSAN will take care of these pages. */
2579 		return;
2580 	}
2581 	__free_pages_core(page, order);
2582 }
2583 
2584 DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc);
2585 EXPORT_SYMBOL(init_on_alloc);
2586 
2587 DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free);
2588 EXPORT_SYMBOL(init_on_free);
2589 
2590 static bool _init_on_alloc_enabled_early __read_mostly
2591 				= IS_ENABLED(CONFIG_INIT_ON_ALLOC_DEFAULT_ON);
2592 static int __init early_init_on_alloc(char *buf)
2593 {
2594 
2595 	return kstrtobool(buf, &_init_on_alloc_enabled_early);
2596 }
2597 early_param("init_on_alloc", early_init_on_alloc);
2598 
2599 static bool _init_on_free_enabled_early __read_mostly
2600 				= IS_ENABLED(CONFIG_INIT_ON_FREE_DEFAULT_ON);
2601 static int __init early_init_on_free(char *buf)
2602 {
2603 	return kstrtobool(buf, &_init_on_free_enabled_early);
2604 }
2605 early_param("init_on_free", early_init_on_free);
2606 
2607 DEFINE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled);
2608 
2609 /*
2610  * Enable static keys related to various memory debugging and hardening options.
2611  * Some override others, and depend on early params that are evaluated in the
2612  * order of appearance. So we need to first gather the full picture of what was
2613  * enabled, and then make decisions.
2614  */
2615 static void __init mem_debugging_and_hardening_init(void)
2616 {
2617 	bool page_poisoning_requested = false;
2618 	bool want_check_pages = false;
2619 
2620 #ifdef CONFIG_PAGE_POISONING
2621 	/*
2622 	 * Page poisoning is debug page alloc for some arches. If
2623 	 * either of those options are enabled, enable poisoning.
2624 	 */
2625 	if (page_poisoning_enabled() ||
2626 	     (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) &&
2627 	      debug_pagealloc_enabled())) {
2628 		static_branch_enable(&_page_poisoning_enabled);
2629 		page_poisoning_requested = true;
2630 		want_check_pages = true;
2631 	}
2632 #endif
2633 
2634 	if ((_init_on_alloc_enabled_early || _init_on_free_enabled_early) &&
2635 	    page_poisoning_requested) {
2636 		pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, "
2637 			"will take precedence over init_on_alloc and init_on_free\n");
2638 		_init_on_alloc_enabled_early = false;
2639 		_init_on_free_enabled_early = false;
2640 	}
2641 
2642 	if (_init_on_alloc_enabled_early) {
2643 		want_check_pages = true;
2644 		static_branch_enable(&init_on_alloc);
2645 	} else {
2646 		static_branch_disable(&init_on_alloc);
2647 	}
2648 
2649 	if (_init_on_free_enabled_early) {
2650 		want_check_pages = true;
2651 		static_branch_enable(&init_on_free);
2652 	} else {
2653 		static_branch_disable(&init_on_free);
2654 	}
2655 
2656 	if (IS_ENABLED(CONFIG_KMSAN) &&
2657 	    (_init_on_alloc_enabled_early || _init_on_free_enabled_early))
2658 		pr_info("mem auto-init: please make sure init_on_alloc and init_on_free are disabled when running KMSAN\n");
2659 
2660 #ifdef CONFIG_DEBUG_PAGEALLOC
2661 	if (debug_pagealloc_enabled()) {
2662 		want_check_pages = true;
2663 		static_branch_enable(&_debug_pagealloc_enabled);
2664 
2665 		if (debug_guardpage_minorder())
2666 			static_branch_enable(&_debug_guardpage_enabled);
2667 	}
2668 #endif
2669 
2670 	/*
2671 	 * Any page debugging or hardening option also enables sanity checking
2672 	 * of struct pages being allocated or freed. With CONFIG_DEBUG_VM it's
2673 	 * enabled already.
2674 	 */
2675 	if (!IS_ENABLED(CONFIG_DEBUG_VM) && want_check_pages)
2676 		static_branch_enable(&check_pages_enabled);
2677 }
2678 
2679 /* Report memory auto-initialization states for this boot. */
2680 static void __init report_meminit(void)
2681 {
2682 	const char *stack;
2683 
2684 	if (IS_ENABLED(CONFIG_INIT_STACK_ALL_PATTERN))
2685 		stack = "all(pattern)";
2686 	else if (IS_ENABLED(CONFIG_INIT_STACK_ALL_ZERO))
2687 		stack = "all(zero)";
2688 	else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF_ALL))
2689 		stack = "byref_all(zero)";
2690 	else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF))
2691 		stack = "byref(zero)";
2692 	else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_USER))
2693 		stack = "__user(zero)";
2694 	else
2695 		stack = "off";
2696 
2697 	pr_info("mem auto-init: stack:%s, heap alloc:%s, heap free:%s\n",
2698 		stack, want_init_on_alloc(GFP_KERNEL) ? "on" : "off",
2699 		want_init_on_free() ? "on" : "off");
2700 	if (want_init_on_free())
2701 		pr_info("mem auto-init: clearing system memory may take some time...\n");
2702 }
2703 
2704 static void __init mem_init_print_info(void)
2705 {
2706 	unsigned long physpages, codesize, datasize, rosize, bss_size;
2707 	unsigned long init_code_size, init_data_size;
2708 
2709 	physpages = get_num_physpages();
2710 	codesize = _etext - _stext;
2711 	datasize = _edata - _sdata;
2712 	rosize = __end_rodata - __start_rodata;
2713 	bss_size = __bss_stop - __bss_start;
2714 	init_data_size = __init_end - __init_begin;
2715 	init_code_size = _einittext - _sinittext;
2716 
2717 	/*
2718 	 * Detect special cases and adjust section sizes accordingly:
2719 	 * 1) .init.* may be embedded into .data sections
2720 	 * 2) .init.text.* may be out of [__init_begin, __init_end],
2721 	 *    please refer to arch/tile/kernel/vmlinux.lds.S.
2722 	 * 3) .rodata.* may be embedded into .text or .data sections.
2723 	 */
2724 #define adj_init_size(start, end, size, pos, adj) \
2725 	do { \
2726 		if (&start[0] <= &pos[0] && &pos[0] < &end[0] && size > adj) \
2727 			size -= adj; \
2728 	} while (0)
2729 
2730 	adj_init_size(__init_begin, __init_end, init_data_size,
2731 		     _sinittext, init_code_size);
2732 	adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
2733 	adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
2734 	adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
2735 	adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
2736 
2737 #undef	adj_init_size
2738 
2739 	pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
2740 #ifdef	CONFIG_HIGHMEM
2741 		", %luK highmem"
2742 #endif
2743 		")\n",
2744 		K(nr_free_pages()), K(physpages),
2745 		codesize / SZ_1K, datasize / SZ_1K, rosize / SZ_1K,
2746 		(init_data_size + init_code_size) / SZ_1K, bss_size / SZ_1K,
2747 		K(physpages - totalram_pages() - totalcma_pages),
2748 		K(totalcma_pages)
2749 #ifdef	CONFIG_HIGHMEM
2750 		, K(totalhigh_pages())
2751 #endif
2752 		);
2753 }
2754 
2755 /*
2756  * Set up kernel memory allocators
2757  */
2758 void __init mm_core_init(void)
2759 {
2760 	/* Initializations relying on SMP setup */
2761 	build_all_zonelists(NULL);
2762 	page_alloc_init_cpuhp();
2763 
2764 	/*
2765 	 * page_ext requires contiguous pages,
2766 	 * bigger than MAX_ORDER unless SPARSEMEM.
2767 	 */
2768 	page_ext_init_flatmem();
2769 	mem_debugging_and_hardening_init();
2770 	kfence_alloc_pool_and_metadata();
2771 	report_meminit();
2772 	kmsan_init_shadow();
2773 	stack_depot_early_init();
2774 	mem_init();
2775 	mem_init_print_info();
2776 	kmem_cache_init();
2777 	/*
2778 	 * page_owner must be initialized after buddy is ready, and also after
2779 	 * slab is ready so that stack_depot_init() works properly
2780 	 */
2781 	page_ext_init_flatmem_late();
2782 	kmemleak_init();
2783 	ptlock_cache_init();
2784 	pgtable_cache_init();
2785 	debug_objects_mem_init();
2786 	vmalloc_init();
2787 	/* If no deferred init page_ext now, as vmap is fully initialized */
2788 	if (!deferred_struct_pages)
2789 		page_ext_init();
2790 	/* Should be run before the first non-init thread is created */
2791 	init_espfix_bsp();
2792 	/* Should be run after espfix64 is set up. */
2793 	pti_init();
2794 	kmsan_init_runtime();
2795 	mm_cache_init();
2796 }
2797