1 // SPDX-License-Identifier: GPL-2.0-only
3 * mm_init.c - Memory initialisation verification and debugging
5 * Copyright 2008 IBM Corporation, 2008
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 <linux/crash_dump.h>
34 #include <asm/setup.h>
36 #ifdef CONFIG_DEBUG_MEMORY_INIT
37 int __meminitdata mminit_loglevel;
39 /* The zonelists are simply reported, validation is manual. */
40 void __init mminit_verify_zonelist(void)
44 if (mminit_loglevel < MMINIT_VERIFY)
47 for_each_online_node(nid) {
48 pg_data_t *pgdat = NODE_DATA(nid);
51 struct zonelist *zonelist;
52 int i, listid, zoneid;
54 BUILD_BUG_ON(MAX_ZONELISTS > 2);
55 for (i = 0; i < MAX_ZONELISTS * MAX_NR_ZONES; i++) {
57 /* Identify the zone and nodelist */
58 zoneid = i % MAX_NR_ZONES;
59 listid = i / MAX_NR_ZONES;
60 zonelist = &pgdat->node_zonelists[listid];
61 zone = &pgdat->node_zones[zoneid];
62 if (!populated_zone(zone))
65 /* Print information about the zonelist */
66 printk(KERN_DEBUG "mminit::zonelist %s %d:%s = ",
67 listid > 0 ? "thisnode" : "general", nid,
70 /* Iterate the zonelist */
71 for_each_zone_zonelist(zone, z, zonelist, zoneid)
72 pr_cont("%d:%s ", zone_to_nid(zone), zone->name);
78 void __init mminit_verify_pageflags_layout(void)
81 unsigned long or_mask, add_mask;
83 shift = BITS_PER_LONG;
84 width = shift - SECTIONS_WIDTH - NODES_WIDTH - ZONES_WIDTH
85 - LAST_CPUPID_SHIFT - KASAN_TAG_WIDTH - LRU_GEN_WIDTH - LRU_REFS_WIDTH;
86 mminit_dprintk(MMINIT_TRACE, "pageflags_layout_widths",
87 "Section %d Node %d Zone %d Lastcpupid %d Kasantag %d Gen %d Tier %d Flags %d\n",
96 mminit_dprintk(MMINIT_TRACE, "pageflags_layout_shifts",
97 "Section %d Node %d Zone %d Lastcpupid %d Kasantag %d\n",
103 mminit_dprintk(MMINIT_TRACE, "pageflags_layout_pgshifts",
104 "Section %lu Node %lu Zone %lu Lastcpupid %lu Kasantag %lu\n",
105 (unsigned long)SECTIONS_PGSHIFT,
106 (unsigned long)NODES_PGSHIFT,
107 (unsigned long)ZONES_PGSHIFT,
108 (unsigned long)LAST_CPUPID_PGSHIFT,
109 (unsigned long)KASAN_TAG_PGSHIFT);
110 mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodezoneid",
111 "Node/Zone ID: %lu -> %lu\n",
112 (unsigned long)(ZONEID_PGOFF + ZONEID_SHIFT),
113 (unsigned long)ZONEID_PGOFF);
114 mminit_dprintk(MMINIT_TRACE, "pageflags_layout_usage",
115 "location: %d -> %d layout %d -> %d unused %d -> %d page-flags\n",
116 shift, width, width, NR_PAGEFLAGS, NR_PAGEFLAGS, 0);
117 #ifdef NODE_NOT_IN_PAGE_FLAGS
118 mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
119 "Node not in page flags");
121 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
122 mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
123 "Last cpupid not in page flags");
126 if (SECTIONS_WIDTH) {
127 shift -= SECTIONS_WIDTH;
128 BUG_ON(shift != SECTIONS_PGSHIFT);
131 shift -= NODES_WIDTH;
132 BUG_ON(shift != NODES_PGSHIFT);
135 shift -= ZONES_WIDTH;
136 BUG_ON(shift != ZONES_PGSHIFT);
139 /* Check for bitmask overlaps */
140 or_mask = (ZONES_MASK << ZONES_PGSHIFT) |
141 (NODES_MASK << NODES_PGSHIFT) |
142 (SECTIONS_MASK << SECTIONS_PGSHIFT);
143 add_mask = (ZONES_MASK << ZONES_PGSHIFT) +
144 (NODES_MASK << NODES_PGSHIFT) +
145 (SECTIONS_MASK << SECTIONS_PGSHIFT);
146 BUG_ON(or_mask != add_mask);
149 static __init int set_mminit_loglevel(char *str)
151 get_option(&str, &mminit_loglevel);
154 early_param("mminit_loglevel", set_mminit_loglevel);
155 #endif /* CONFIG_DEBUG_MEMORY_INIT */
157 struct kobject *mm_kobj;
160 s32 vm_committed_as_batch = 32;
162 void mm_compute_batch(int overcommit_policy)
165 s32 nr = num_present_cpus();
166 s32 batch = max_t(s32, nr*2, 32);
167 unsigned long ram_pages = totalram_pages();
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
175 if (overcommit_policy == OVERCOMMIT_NEVER)
176 memsized_batch = min_t(u64, ram_pages/nr/256, INT_MAX);
178 memsized_batch = min_t(u64, ram_pages/nr/4, INT_MAX);
180 vm_committed_as_batch = max_t(s32, memsized_batch, batch);
183 static int __meminit mm_compute_batch_notifier(struct notifier_block *self,
184 unsigned long action, void *arg)
189 mm_compute_batch(sysctl_overcommit_memory);
197 static int __init mm_compute_batch_init(void)
199 mm_compute_batch(sysctl_overcommit_memory);
200 hotplug_memory_notifier(mm_compute_batch_notifier, MM_COMPUTE_BATCH_PRI);
204 __initcall(mm_compute_batch_init);
208 static int __init mm_sysfs_init(void)
210 mm_kobj = kobject_create_and_add("mm", kernel_kobj);
216 postcore_initcall(mm_sysfs_init);
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;
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;
227 static unsigned long nr_kernel_pages __initdata;
228 static unsigned long nr_all_pages __initdata;
229 static unsigned long dma_reserve __initdata;
231 static bool deferred_struct_pages __meminitdata;
233 static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
235 static int __init cmdline_parse_core(char *p, unsigned long *core,
236 unsigned long *percent)
238 unsigned long long coremem;
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);
252 coremem = memparse(p, &p);
253 /* Paranoid check that UL is enough for the coremem value */
254 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
256 *core = coremem >> PAGE_SHIFT;
262 bool mirrored_kernelcore __initdata_memblock;
265 * kernelcore=size sets the amount of memory for use for allocations that
266 * cannot be reclaimed or migrated.
268 static int __init cmdline_parse_kernelcore(char *p)
270 /* parse kernelcore=mirror */
271 if (parse_option_str(p, "mirror")) {
272 mirrored_kernelcore = true;
276 return cmdline_parse_core(p, &required_kernelcore,
277 &required_kernelcore_percent);
279 early_param("kernelcore", cmdline_parse_kernelcore);
282 * movablecore=size sets the amount of memory for use for allocations that
283 * can be reclaimed or migrated.
285 static int __init cmdline_parse_movablecore(char *p)
287 return cmdline_parse_core(p, &required_movablecore,
288 &required_movablecore_percent);
290 early_param("movablecore", cmdline_parse_movablecore);
293 * early_calculate_totalpages()
294 * Sum pages in active regions for movable zone.
295 * Populate N_MEMORY for calculating usable_nodes.
297 static unsigned long __init early_calculate_totalpages(void)
299 unsigned long totalpages = 0;
300 unsigned long start_pfn, end_pfn;
303 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
304 unsigned long pages = end_pfn - start_pfn;
308 node_set_state(nid, N_MEMORY);
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
318 static void __init find_usable_zone_for_movable(void)
321 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
322 if (zone_index == ZONE_MOVABLE)
325 if (arch_zone_highest_possible_pfn[zone_index] >
326 arch_zone_lowest_possible_pfn[zone_index])
330 VM_BUG_ON(zone_index == -1);
331 movable_zone = zone_index;
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
340 static void __init find_zone_movable_pfns_for_nodes(void)
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;
351 /* Need to find movable_zone earlier when movable_node is specified. */
352 find_usable_zone_for_movable();
355 * If movable_node is specified, ignore kernelcore and movablecore
358 if (movable_node_is_enabled()) {
359 for_each_mem_region(r) {
360 if (!memblock_is_hotpluggable(r))
363 nid = memblock_get_region_node(r);
365 usable_startpfn = PFN_DOWN(r->base);
366 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
367 min(usable_startpfn, zone_movable_pfn[nid]) :
375 * If kernelcore=mirror is specified, ignore movablecore option
377 if (mirrored_kernelcore) {
378 bool mem_below_4gb_not_mirrored = false;
380 if (!memblock_has_mirror()) {
381 pr_warn("The system has no mirror memory, ignore kernelcore=mirror.\n");
385 if (is_kdump_kernel()) {
386 pr_warn("The system is under kdump, ignore kernelcore=mirror.\n");
390 for_each_mem_region(r) {
391 if (memblock_is_mirror(r))
394 nid = memblock_get_region_node(r);
396 usable_startpfn = memblock_region_memory_base_pfn(r);
398 if (usable_startpfn < PHYS_PFN(SZ_4G)) {
399 mem_below_4gb_not_mirrored = true;
403 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
404 min(usable_startpfn, zone_movable_pfn[nid]) :
408 if (mem_below_4gb_not_mirrored)
409 pr_warn("This configuration results in unmirrored kernel memory.\n");
415 * If kernelcore=nn% or movablecore=nn% was specified, calculate the
416 * amount of necessary memory.
418 if (required_kernelcore_percent)
419 required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
421 if (required_movablecore_percent)
422 required_movablecore = (totalpages * 100 * required_movablecore_percent) /
426 * If movablecore= was specified, calculate what size of
427 * kernelcore that corresponds so that memory usable for
428 * any allocation type is evenly spread. If both kernelcore
429 * and movablecore are specified, then the value of kernelcore
430 * will be used for required_kernelcore if it's greater than
431 * what movablecore would have allowed.
433 if (required_movablecore) {
434 unsigned long corepages;
437 * Round-up so that ZONE_MOVABLE is at least as large as what
438 * was requested by the user
440 required_movablecore =
441 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
442 required_movablecore = min(totalpages, required_movablecore);
443 corepages = totalpages - required_movablecore;
445 required_kernelcore = max(required_kernelcore, corepages);
449 * If kernelcore was not specified or kernelcore size is larger
450 * than totalpages, there is no ZONE_MOVABLE.
452 if (!required_kernelcore || required_kernelcore >= totalpages)
455 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
456 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
459 /* Spread kernelcore memory as evenly as possible throughout nodes */
460 kernelcore_node = required_kernelcore / usable_nodes;
461 for_each_node_state(nid, N_MEMORY) {
462 unsigned long start_pfn, end_pfn;
465 * Recalculate kernelcore_node if the division per node
466 * now exceeds what is necessary to satisfy the requested
467 * amount of memory for the kernel
469 if (required_kernelcore < kernelcore_node)
470 kernelcore_node = required_kernelcore / usable_nodes;
473 * As the map is walked, we track how much memory is usable
474 * by the kernel using kernelcore_remaining. When it is
475 * 0, the rest of the node is usable by ZONE_MOVABLE
477 kernelcore_remaining = kernelcore_node;
479 /* Go through each range of PFNs within this node */
480 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
481 unsigned long size_pages;
483 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
484 if (start_pfn >= end_pfn)
487 /* Account for what is only usable for kernelcore */
488 if (start_pfn < usable_startpfn) {
489 unsigned long kernel_pages;
490 kernel_pages = min(end_pfn, usable_startpfn)
493 kernelcore_remaining -= min(kernel_pages,
494 kernelcore_remaining);
495 required_kernelcore -= min(kernel_pages,
496 required_kernelcore);
498 /* Continue if range is now fully accounted */
499 if (end_pfn <= usable_startpfn) {
502 * Push zone_movable_pfn to the end so
503 * that if we have to rebalance
504 * kernelcore across nodes, we will
505 * not double account here
507 zone_movable_pfn[nid] = end_pfn;
510 start_pfn = usable_startpfn;
514 * The usable PFN range for ZONE_MOVABLE is from
515 * start_pfn->end_pfn. Calculate size_pages as the
516 * number of pages used as kernelcore
518 size_pages = end_pfn - start_pfn;
519 if (size_pages > kernelcore_remaining)
520 size_pages = kernelcore_remaining;
521 zone_movable_pfn[nid] = start_pfn + size_pages;
524 * Some kernelcore has been met, update counts and
525 * break if the kernelcore for this node has been
528 required_kernelcore -= min(required_kernelcore,
530 kernelcore_remaining -= size_pages;
531 if (!kernelcore_remaining)
537 * If there is still required_kernelcore, we do another pass with one
538 * less node in the count. This will push zone_movable_pfn[nid] further
539 * along on the nodes that still have memory until kernelcore is
543 if (usable_nodes && required_kernelcore > usable_nodes)
547 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
548 for (nid = 0; nid < MAX_NUMNODES; nid++) {
549 unsigned long start_pfn, end_pfn;
551 zone_movable_pfn[nid] =
552 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
554 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
555 if (zone_movable_pfn[nid] >= end_pfn)
556 zone_movable_pfn[nid] = 0;
560 /* restore the node_state */
561 node_states[N_MEMORY] = saved_node_state;
564 void __meminit __init_single_page(struct page *page, unsigned long pfn,
565 unsigned long zone, int nid)
567 mm_zero_struct_page(page);
568 set_page_links(page, zone, nid, pfn);
569 init_page_count(page);
570 page_mapcount_reset(page);
571 page_cpupid_reset_last(page);
572 page_kasan_tag_reset(page);
574 INIT_LIST_HEAD(&page->lru);
575 #ifdef WANT_PAGE_VIRTUAL
576 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
577 if (!is_highmem_idx(zone))
578 set_page_address(page, __va(pfn << PAGE_SHIFT));
584 * During memory init memblocks map pfns to nids. The search is expensive and
585 * this caches recent lookups. The implementation of __early_pfn_to_nid
586 * treats start/end as pfns.
588 struct mminit_pfnnid_cache {
589 unsigned long last_start;
590 unsigned long last_end;
594 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
597 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
599 static int __meminit __early_pfn_to_nid(unsigned long pfn,
600 struct mminit_pfnnid_cache *state)
602 unsigned long start_pfn, end_pfn;
605 if (state->last_start <= pfn && pfn < state->last_end)
606 return state->last_nid;
608 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
609 if (nid != NUMA_NO_NODE) {
610 state->last_start = start_pfn;
611 state->last_end = end_pfn;
612 state->last_nid = nid;
618 int __meminit early_pfn_to_nid(unsigned long pfn)
620 static DEFINE_SPINLOCK(early_pfn_lock);
623 spin_lock(&early_pfn_lock);
624 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
626 nid = first_online_node;
627 spin_unlock(&early_pfn_lock);
632 int hashdist = HASHDIST_DEFAULT;
634 static int __init set_hashdist(char *str)
638 hashdist = simple_strtoul(str, &str, 0);
641 __setup("hashdist=", set_hashdist);
643 static inline void fixup_hashdist(void)
645 if (num_node_state(N_MEMORY) == 1)
649 static inline void fixup_hashdist(void) {}
650 #endif /* CONFIG_NUMA */
652 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
653 static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
655 pgdat->first_deferred_pfn = ULONG_MAX;
658 /* Returns true if the struct page for the pfn is initialised */
659 static inline bool __meminit early_page_initialised(unsigned long pfn, int nid)
661 if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
668 * Returns true when the remaining initialisation should be deferred until
669 * later in the boot cycle when it can be parallelised.
671 static bool __meminit
672 defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
674 static unsigned long prev_end_pfn, nr_initialised;
676 if (early_page_ext_enabled())
679 * prev_end_pfn static that contains the end of previous zone
680 * No need to protect because called very early in boot before smp_init.
682 if (prev_end_pfn != end_pfn) {
683 prev_end_pfn = end_pfn;
687 /* Always populate low zones for address-constrained allocations */
688 if (end_pfn < pgdat_end_pfn(NODE_DATA(nid)))
691 if (NODE_DATA(nid)->first_deferred_pfn != ULONG_MAX)
694 * We start only with one section of pages, more pages are added as
695 * needed until the rest of deferred pages are initialized.
698 if ((nr_initialised > PAGES_PER_SECTION) &&
699 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
700 NODE_DATA(nid)->first_deferred_pfn = pfn;
706 static void __meminit init_reserved_page(unsigned long pfn, int nid)
711 if (early_page_initialised(pfn, nid))
714 pgdat = NODE_DATA(nid);
716 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
717 struct zone *zone = &pgdat->node_zones[zid];
719 if (zone_spans_pfn(zone, pfn))
722 __init_single_page(pfn_to_page(pfn), pfn, zid, nid);
725 static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
727 static inline bool early_page_initialised(unsigned long pfn, int nid)
732 static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
737 static inline void init_reserved_page(unsigned long pfn, int nid)
740 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
743 * Initialised pages do not have PageReserved set. This function is
744 * called for each range allocated by the bootmem allocator and
745 * marks the pages PageReserved. The remaining valid pages are later
746 * sent to the buddy page allocator.
748 void __meminit reserve_bootmem_region(phys_addr_t start,
749 phys_addr_t end, int nid)
751 unsigned long start_pfn = PFN_DOWN(start);
752 unsigned long end_pfn = PFN_UP(end);
754 for (; start_pfn < end_pfn; start_pfn++) {
755 if (pfn_valid(start_pfn)) {
756 struct page *page = pfn_to_page(start_pfn);
758 init_reserved_page(start_pfn, nid);
760 /* Avoid false-positive PageTail() */
761 INIT_LIST_HEAD(&page->lru);
764 * no need for atomic set_bit because the struct
765 * page is not visible yet so nobody should
768 __SetPageReserved(page);
773 /* If zone is ZONE_MOVABLE but memory is mirrored, it is an overlapped init */
774 static bool __meminit
775 overlap_memmap_init(unsigned long zone, unsigned long *pfn)
777 static struct memblock_region *r;
779 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
780 if (!r || *pfn >= memblock_region_memory_end_pfn(r)) {
781 for_each_mem_region(r) {
782 if (*pfn < memblock_region_memory_end_pfn(r))
786 if (*pfn >= memblock_region_memory_base_pfn(r) &&
787 memblock_is_mirror(r)) {
788 *pfn = memblock_region_memory_end_pfn(r);
796 * Only struct pages that correspond to ranges defined by memblock.memory
797 * are zeroed and initialized by going through __init_single_page() during
798 * memmap_init_zone_range().
800 * But, there could be struct pages that correspond to holes in
801 * memblock.memory. This can happen because of the following reasons:
802 * - physical memory bank size is not necessarily the exact multiple of the
803 * arbitrary section size
804 * - early reserved memory may not be listed in memblock.memory
805 * - non-memory regions covered by the contigious flatmem mapping
806 * - memory layouts defined with memmap= kernel parameter may not align
807 * nicely with memmap sections
809 * Explicitly initialize those struct pages so that:
810 * - PG_Reserved is set
811 * - zone and node links point to zone and node that span the page if the
812 * hole is in the middle of a zone
813 * - zone and node links point to adjacent zone/node if the hole falls on
814 * the zone boundary; the pages in such holes will be prepended to the
815 * zone/node above the hole except for the trailing pages in the last
816 * section that will be appended to the zone/node below.
818 static void __init init_unavailable_range(unsigned long spfn,
825 for (pfn = spfn; pfn < epfn; pfn++) {
826 if (!pfn_valid(pageblock_start_pfn(pfn))) {
827 pfn = pageblock_end_pfn(pfn) - 1;
830 __init_single_page(pfn_to_page(pfn), pfn, zone, node);
831 __SetPageReserved(pfn_to_page(pfn));
836 pr_info("On node %d, zone %s: %lld pages in unavailable ranges\n",
837 node, zone_names[zone], pgcnt);
841 * Initially all pages are reserved - free ones are freed
842 * up by memblock_free_all() once the early boot process is
843 * done. Non-atomic initialization, single-pass.
845 * All aligned pageblocks are initialized to the specified migratetype
846 * (usually MIGRATE_MOVABLE). Besides setting the migratetype, no related
847 * zone stats (e.g., nr_isolate_pageblock) are touched.
849 void __meminit memmap_init_range(unsigned long size, int nid, unsigned long zone,
850 unsigned long start_pfn, unsigned long zone_end_pfn,
851 enum meminit_context context,
852 struct vmem_altmap *altmap, int migratetype)
854 unsigned long pfn, end_pfn = start_pfn + size;
857 if (highest_memmap_pfn < end_pfn - 1)
858 highest_memmap_pfn = end_pfn - 1;
860 #ifdef CONFIG_ZONE_DEVICE
862 * Honor reservation requested by the driver for this ZONE_DEVICE
863 * memory. We limit the total number of pages to initialize to just
864 * those that might contain the memory mapping. We will defer the
865 * ZONE_DEVICE page initialization until after we have released
868 if (zone == ZONE_DEVICE) {
872 if (start_pfn == altmap->base_pfn)
873 start_pfn += altmap->reserve;
874 end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
878 for (pfn = start_pfn; pfn < end_pfn; ) {
880 * There can be holes in boot-time mem_map[]s handed to this
881 * function. They do not exist on hotplugged memory.
883 if (context == MEMINIT_EARLY) {
884 if (overlap_memmap_init(zone, &pfn))
886 if (defer_init(nid, pfn, zone_end_pfn)) {
887 deferred_struct_pages = true;
892 page = pfn_to_page(pfn);
893 __init_single_page(page, pfn, zone, nid);
894 if (context == MEMINIT_HOTPLUG)
895 __SetPageReserved(page);
898 * Usually, we want to mark the pageblock MIGRATE_MOVABLE,
899 * such that unmovable allocations won't be scattered all
900 * over the place during system boot.
902 if (pageblock_aligned(pfn)) {
903 set_pageblock_migratetype(page, migratetype);
910 static void __init memmap_init_zone_range(struct zone *zone,
911 unsigned long start_pfn,
912 unsigned long end_pfn,
913 unsigned long *hole_pfn)
915 unsigned long zone_start_pfn = zone->zone_start_pfn;
916 unsigned long zone_end_pfn = zone_start_pfn + zone->spanned_pages;
917 int nid = zone_to_nid(zone), zone_id = zone_idx(zone);
919 start_pfn = clamp(start_pfn, zone_start_pfn, zone_end_pfn);
920 end_pfn = clamp(end_pfn, zone_start_pfn, zone_end_pfn);
922 if (start_pfn >= end_pfn)
925 memmap_init_range(end_pfn - start_pfn, nid, zone_id, start_pfn,
926 zone_end_pfn, MEMINIT_EARLY, NULL, MIGRATE_MOVABLE);
928 if (*hole_pfn < start_pfn)
929 init_unavailable_range(*hole_pfn, start_pfn, zone_id, nid);
934 static void __init memmap_init(void)
936 unsigned long start_pfn, end_pfn;
937 unsigned long hole_pfn = 0;
938 int i, j, zone_id = 0, nid;
940 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
941 struct pglist_data *node = NODE_DATA(nid);
943 for (j = 0; j < MAX_NR_ZONES; j++) {
944 struct zone *zone = node->node_zones + j;
946 if (!populated_zone(zone))
949 memmap_init_zone_range(zone, start_pfn, end_pfn,
955 #ifdef CONFIG_SPARSEMEM
957 * Initialize the memory map for hole in the range [memory_end,
959 * Append the pages in this hole to the highest zone in the last
961 * The call to init_unavailable_range() is outside the ifdef to
962 * silence the compiler warining about zone_id set but not used;
963 * for FLATMEM it is a nop anyway
965 end_pfn = round_up(end_pfn, PAGES_PER_SECTION);
966 if (hole_pfn < end_pfn)
968 init_unavailable_range(hole_pfn, end_pfn, zone_id, nid);
971 #ifdef CONFIG_ZONE_DEVICE
972 static void __ref __init_zone_device_page(struct page *page, unsigned long pfn,
973 unsigned long zone_idx, int nid,
974 struct dev_pagemap *pgmap)
977 __init_single_page(page, pfn, zone_idx, nid);
980 * Mark page reserved as it will need to wait for onlining
981 * phase for it to be fully associated with a zone.
983 * We can use the non-atomic __set_bit operation for setting
984 * the flag as we are still initializing the pages.
986 __SetPageReserved(page);
989 * ZONE_DEVICE pages union ->lru with a ->pgmap back pointer
990 * and zone_device_data. It is a bug if a ZONE_DEVICE page is
991 * ever freed or placed on a driver-private list.
994 page->zone_device_data = NULL;
997 * Mark the block movable so that blocks are reserved for
998 * movable at startup. This will force kernel allocations
999 * to reserve their blocks rather than leaking throughout
1000 * the address space during boot when many long-lived
1001 * kernel allocations are made.
1003 * Please note that MEMINIT_HOTPLUG path doesn't clear memmap
1004 * because this is done early in section_activate()
1006 if (pageblock_aligned(pfn)) {
1007 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1012 * ZONE_DEVICE pages are released directly to the driver page allocator
1013 * which will set the page count to 1 when allocating the page.
1015 if (pgmap->type == MEMORY_DEVICE_PRIVATE ||
1016 pgmap->type == MEMORY_DEVICE_COHERENT)
1017 set_page_count(page, 0);
1021 * With compound page geometry and when struct pages are stored in ram most
1022 * tail pages are reused. Consequently, the amount of unique struct pages to
1023 * initialize is a lot smaller that the total amount of struct pages being
1024 * mapped. This is a paired / mild layering violation with explicit knowledge
1025 * of how the sparse_vmemmap internals handle compound pages in the lack
1026 * of an altmap. See vmemmap_populate_compound_pages().
1028 static inline unsigned long compound_nr_pages(struct vmem_altmap *altmap,
1029 struct dev_pagemap *pgmap)
1031 if (!vmemmap_can_optimize(altmap, pgmap))
1032 return pgmap_vmemmap_nr(pgmap);
1034 return VMEMMAP_RESERVE_NR * (PAGE_SIZE / sizeof(struct page));
1037 static void __ref memmap_init_compound(struct page *head,
1038 unsigned long head_pfn,
1039 unsigned long zone_idx, int nid,
1040 struct dev_pagemap *pgmap,
1041 unsigned long nr_pages)
1043 unsigned long pfn, end_pfn = head_pfn + nr_pages;
1044 unsigned int order = pgmap->vmemmap_shift;
1046 __SetPageHead(head);
1047 for (pfn = head_pfn + 1; pfn < end_pfn; pfn++) {
1048 struct page *page = pfn_to_page(pfn);
1050 __init_zone_device_page(page, pfn, zone_idx, nid, pgmap);
1051 prep_compound_tail(head, pfn - head_pfn);
1052 set_page_count(page, 0);
1055 * The first tail page stores important compound page info.
1056 * Call prep_compound_head() after the first tail page has
1057 * been initialized, to not have the data overwritten.
1059 if (pfn == head_pfn + 1)
1060 prep_compound_head(head, order);
1064 void __ref memmap_init_zone_device(struct zone *zone,
1065 unsigned long start_pfn,
1066 unsigned long nr_pages,
1067 struct dev_pagemap *pgmap)
1069 unsigned long pfn, end_pfn = start_pfn + nr_pages;
1070 struct pglist_data *pgdat = zone->zone_pgdat;
1071 struct vmem_altmap *altmap = pgmap_altmap(pgmap);
1072 unsigned int pfns_per_compound = pgmap_vmemmap_nr(pgmap);
1073 unsigned long zone_idx = zone_idx(zone);
1074 unsigned long start = jiffies;
1075 int nid = pgdat->node_id;
1077 if (WARN_ON_ONCE(!pgmap || zone_idx != ZONE_DEVICE))
1081 * The call to memmap_init should have already taken care
1082 * of the pages reserved for the memmap, so we can just jump to
1083 * the end of that region and start processing the device pages.
1086 start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
1087 nr_pages = end_pfn - start_pfn;
1090 for (pfn = start_pfn; pfn < end_pfn; pfn += pfns_per_compound) {
1091 struct page *page = pfn_to_page(pfn);
1093 __init_zone_device_page(page, pfn, zone_idx, nid, pgmap);
1095 if (pfns_per_compound == 1)
1098 memmap_init_compound(page, pfn, zone_idx, nid, pgmap,
1099 compound_nr_pages(altmap, pgmap));
1102 pr_debug("%s initialised %lu pages in %ums\n", __func__,
1103 nr_pages, jiffies_to_msecs(jiffies - start));
1108 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
1109 * because it is sized independent of architecture. Unlike the other zones,
1110 * the starting point for ZONE_MOVABLE is not fixed. It may be different
1111 * in each node depending on the size of each node and how evenly kernelcore
1112 * is distributed. This helper function adjusts the zone ranges
1113 * provided by the architecture for a given node by using the end of the
1114 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
1115 * zones within a node are in order of monotonic increases memory addresses
1117 static void __init adjust_zone_range_for_zone_movable(int nid,
1118 unsigned long zone_type,
1119 unsigned long node_end_pfn,
1120 unsigned long *zone_start_pfn,
1121 unsigned long *zone_end_pfn)
1123 /* Only adjust if ZONE_MOVABLE is on this node */
1124 if (zone_movable_pfn[nid]) {
1125 /* Size ZONE_MOVABLE */
1126 if (zone_type == ZONE_MOVABLE) {
1127 *zone_start_pfn = zone_movable_pfn[nid];
1128 *zone_end_pfn = min(node_end_pfn,
1129 arch_zone_highest_possible_pfn[movable_zone]);
1131 /* Adjust for ZONE_MOVABLE starting within this range */
1132 } else if (!mirrored_kernelcore &&
1133 *zone_start_pfn < zone_movable_pfn[nid] &&
1134 *zone_end_pfn > zone_movable_pfn[nid]) {
1135 *zone_end_pfn = zone_movable_pfn[nid];
1137 /* Check if this whole range is within ZONE_MOVABLE */
1138 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
1139 *zone_start_pfn = *zone_end_pfn;
1144 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
1145 * then all holes in the requested range will be accounted for.
1147 unsigned long __init __absent_pages_in_range(int nid,
1148 unsigned long range_start_pfn,
1149 unsigned long range_end_pfn)
1151 unsigned long nr_absent = range_end_pfn - range_start_pfn;
1152 unsigned long start_pfn, end_pfn;
1155 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
1156 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
1157 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
1158 nr_absent -= end_pfn - start_pfn;
1164 * absent_pages_in_range - Return number of page frames in holes within a range
1165 * @start_pfn: The start PFN to start searching for holes
1166 * @end_pfn: The end PFN to stop searching for holes
1168 * Return: the number of pages frames in memory holes within a range.
1170 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
1171 unsigned long end_pfn)
1173 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
1176 /* Return the number of page frames in holes in a zone on a node */
1177 static unsigned long __init zone_absent_pages_in_node(int nid,
1178 unsigned long zone_type,
1179 unsigned long zone_start_pfn,
1180 unsigned long zone_end_pfn)
1182 unsigned long nr_absent;
1184 /* zone is empty, we don't have any absent pages */
1185 if (zone_start_pfn == zone_end_pfn)
1188 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
1191 * ZONE_MOVABLE handling.
1192 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
1195 if (mirrored_kernelcore && zone_movable_pfn[nid]) {
1196 unsigned long start_pfn, end_pfn;
1197 struct memblock_region *r;
1199 for_each_mem_region(r) {
1200 start_pfn = clamp(memblock_region_memory_base_pfn(r),
1201 zone_start_pfn, zone_end_pfn);
1202 end_pfn = clamp(memblock_region_memory_end_pfn(r),
1203 zone_start_pfn, zone_end_pfn);
1205 if (zone_type == ZONE_MOVABLE &&
1206 memblock_is_mirror(r))
1207 nr_absent += end_pfn - start_pfn;
1209 if (zone_type == ZONE_NORMAL &&
1210 !memblock_is_mirror(r))
1211 nr_absent += end_pfn - start_pfn;
1219 * Return the number of pages a zone spans in a node, including holes
1220 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
1222 static unsigned long __init zone_spanned_pages_in_node(int nid,
1223 unsigned long zone_type,
1224 unsigned long node_start_pfn,
1225 unsigned long node_end_pfn,
1226 unsigned long *zone_start_pfn,
1227 unsigned long *zone_end_pfn)
1229 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
1230 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
1232 /* Get the start and end of the zone */
1233 *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
1234 *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
1235 adjust_zone_range_for_zone_movable(nid, zone_type, node_end_pfn,
1236 zone_start_pfn, zone_end_pfn);
1238 /* Check that this node has pages within the zone's required range */
1239 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
1242 /* Move the zone boundaries inside the node if necessary */
1243 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
1244 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
1246 /* Return the spanned pages */
1247 return *zone_end_pfn - *zone_start_pfn;
1250 static void __init reset_memoryless_node_totalpages(struct pglist_data *pgdat)
1254 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) {
1255 z->zone_start_pfn = 0;
1256 z->spanned_pages = 0;
1257 z->present_pages = 0;
1258 #if defined(CONFIG_MEMORY_HOTPLUG)
1259 z->present_early_pages = 0;
1263 pgdat->node_spanned_pages = 0;
1264 pgdat->node_present_pages = 0;
1265 pr_debug("On node %d totalpages: 0\n", pgdat->node_id);
1268 static void __init calculate_node_totalpages(struct pglist_data *pgdat,
1269 unsigned long node_start_pfn,
1270 unsigned long node_end_pfn)
1272 unsigned long realtotalpages = 0, totalpages = 0;
1275 for (i = 0; i < MAX_NR_ZONES; i++) {
1276 struct zone *zone = pgdat->node_zones + i;
1277 unsigned long zone_start_pfn, zone_end_pfn;
1278 unsigned long spanned, absent;
1279 unsigned long real_size;
1281 spanned = zone_spanned_pages_in_node(pgdat->node_id, i,
1286 absent = zone_absent_pages_in_node(pgdat->node_id, i,
1290 real_size = spanned - absent;
1293 zone->zone_start_pfn = zone_start_pfn;
1295 zone->zone_start_pfn = 0;
1296 zone->spanned_pages = spanned;
1297 zone->present_pages = real_size;
1298 #if defined(CONFIG_MEMORY_HOTPLUG)
1299 zone->present_early_pages = real_size;
1302 totalpages += spanned;
1303 realtotalpages += real_size;
1306 pgdat->node_spanned_pages = totalpages;
1307 pgdat->node_present_pages = realtotalpages;
1308 pr_debug("On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1311 static unsigned long __init calc_memmap_size(unsigned long spanned_pages,
1312 unsigned long present_pages)
1314 unsigned long pages = spanned_pages;
1317 * Provide a more accurate estimation if there are holes within
1318 * the zone and SPARSEMEM is in use. If there are holes within the
1319 * zone, each populated memory region may cost us one or two extra
1320 * memmap pages due to alignment because memmap pages for each
1321 * populated regions may not be naturally aligned on page boundary.
1322 * So the (present_pages >> 4) heuristic is a tradeoff for that.
1324 if (spanned_pages > present_pages + (present_pages >> 4) &&
1325 IS_ENABLED(CONFIG_SPARSEMEM))
1326 pages = present_pages;
1328 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
1331 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1332 static void pgdat_init_split_queue(struct pglist_data *pgdat)
1334 struct deferred_split *ds_queue = &pgdat->deferred_split_queue;
1336 spin_lock_init(&ds_queue->split_queue_lock);
1337 INIT_LIST_HEAD(&ds_queue->split_queue);
1338 ds_queue->split_queue_len = 0;
1341 static void pgdat_init_split_queue(struct pglist_data *pgdat) {}
1344 #ifdef CONFIG_COMPACTION
1345 static void pgdat_init_kcompactd(struct pglist_data *pgdat)
1347 init_waitqueue_head(&pgdat->kcompactd_wait);
1350 static void pgdat_init_kcompactd(struct pglist_data *pgdat) {}
1353 static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
1357 pgdat_resize_init(pgdat);
1358 pgdat_kswapd_lock_init(pgdat);
1360 pgdat_init_split_queue(pgdat);
1361 pgdat_init_kcompactd(pgdat);
1363 init_waitqueue_head(&pgdat->kswapd_wait);
1364 init_waitqueue_head(&pgdat->pfmemalloc_wait);
1366 for (i = 0; i < NR_VMSCAN_THROTTLE; i++)
1367 init_waitqueue_head(&pgdat->reclaim_wait[i]);
1369 pgdat_page_ext_init(pgdat);
1370 lruvec_init(&pgdat->__lruvec);
1373 static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
1374 unsigned long remaining_pages)
1376 atomic_long_set(&zone->managed_pages, remaining_pages);
1377 zone_set_nid(zone, nid);
1378 zone->name = zone_names[idx];
1379 zone->zone_pgdat = NODE_DATA(nid);
1380 spin_lock_init(&zone->lock);
1381 zone_seqlock_init(zone);
1382 zone_pcp_init(zone);
1385 static void __meminit zone_init_free_lists(struct zone *zone)
1387 unsigned int order, t;
1388 for_each_migratetype_order(order, t) {
1389 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
1390 zone->free_area[order].nr_free = 0;
1393 #ifdef CONFIG_UNACCEPTED_MEMORY
1394 INIT_LIST_HEAD(&zone->unaccepted_pages);
1398 void __meminit init_currently_empty_zone(struct zone *zone,
1399 unsigned long zone_start_pfn,
1402 struct pglist_data *pgdat = zone->zone_pgdat;
1403 int zone_idx = zone_idx(zone) + 1;
1405 if (zone_idx > pgdat->nr_zones)
1406 pgdat->nr_zones = zone_idx;
1408 zone->zone_start_pfn = zone_start_pfn;
1410 mminit_dprintk(MMINIT_TRACE, "memmap_init",
1411 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
1413 (unsigned long)zone_idx(zone),
1414 zone_start_pfn, (zone_start_pfn + size));
1416 zone_init_free_lists(zone);
1417 zone->initialized = 1;
1420 #ifndef CONFIG_SPARSEMEM
1422 * Calculate the size of the zone->blockflags rounded to an unsigned long
1423 * Start by making sure zonesize is a multiple of pageblock_order by rounding
1424 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
1425 * round what is now in bits to nearest long in bits, then return it in
1428 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
1430 unsigned long usemapsize;
1432 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
1433 usemapsize = roundup(zonesize, pageblock_nr_pages);
1434 usemapsize = usemapsize >> pageblock_order;
1435 usemapsize *= NR_PAGEBLOCK_BITS;
1436 usemapsize = roundup(usemapsize, BITS_PER_LONG);
1438 return usemapsize / BITS_PER_BYTE;
1441 static void __ref setup_usemap(struct zone *zone)
1443 unsigned long usemapsize = usemap_size(zone->zone_start_pfn,
1444 zone->spanned_pages);
1445 zone->pageblock_flags = NULL;
1447 zone->pageblock_flags =
1448 memblock_alloc_node(usemapsize, SMP_CACHE_BYTES,
1450 if (!zone->pageblock_flags)
1451 panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n",
1452 usemapsize, zone->name, zone_to_nid(zone));
1456 static inline void setup_usemap(struct zone *zone) {}
1457 #endif /* CONFIG_SPARSEMEM */
1459 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
1461 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
1462 void __init set_pageblock_order(void)
1464 unsigned int order = MAX_PAGE_ORDER;
1466 /* Check that pageblock_nr_pages has not already been setup */
1467 if (pageblock_order)
1470 /* Don't let pageblocks exceed the maximum allocation granularity. */
1471 if (HPAGE_SHIFT > PAGE_SHIFT && HUGETLB_PAGE_ORDER < order)
1472 order = HUGETLB_PAGE_ORDER;
1475 * Assume the largest contiguous order of interest is a huge page.
1476 * This value may be variable depending on boot parameters on powerpc.
1478 pageblock_order = order;
1480 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
1483 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
1484 * is unused as pageblock_order is set at compile-time. See
1485 * include/linux/pageblock-flags.h for the values of pageblock_order based on
1488 void __init set_pageblock_order(void)
1492 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
1495 * Set up the zone data structures
1496 * - init pgdat internals
1497 * - init all zones belonging to this node
1499 * NOTE: this function is only called during memory hotplug
1501 #ifdef CONFIG_MEMORY_HOTPLUG
1502 void __ref free_area_init_core_hotplug(struct pglist_data *pgdat)
1504 int nid = pgdat->node_id;
1508 pgdat_init_internals(pgdat);
1510 if (pgdat->per_cpu_nodestats == &boot_nodestats)
1511 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1514 * Reset the nr_zones, order and highest_zoneidx before reuse.
1515 * Note that kswapd will init kswapd_highest_zoneidx properly
1516 * when it starts in the near future.
1518 pgdat->nr_zones = 0;
1519 pgdat->kswapd_order = 0;
1520 pgdat->kswapd_highest_zoneidx = 0;
1521 pgdat->node_start_pfn = 0;
1522 pgdat->node_present_pages = 0;
1524 for_each_online_cpu(cpu) {
1525 struct per_cpu_nodestat *p;
1527 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
1528 memset(p, 0, sizeof(*p));
1532 * When memory is hot-added, all the memory is in offline state. So
1533 * clear all zones' present_pages and managed_pages because they will
1534 * be updated in online_pages() and offline_pages().
1536 for (z = 0; z < MAX_NR_ZONES; z++) {
1537 struct zone *zone = pgdat->node_zones + z;
1539 zone->present_pages = 0;
1540 zone_init_internals(zone, z, nid, 0);
1546 * Set up the zone data structures:
1547 * - mark all pages reserved
1548 * - mark all memory queues empty
1549 * - clear the memory bitmaps
1551 * NOTE: pgdat should get zeroed by caller.
1552 * NOTE: this function is only called during early init.
1554 static void __init free_area_init_core(struct pglist_data *pgdat)
1557 int nid = pgdat->node_id;
1559 pgdat_init_internals(pgdat);
1560 pgdat->per_cpu_nodestats = &boot_nodestats;
1562 for (j = 0; j < MAX_NR_ZONES; j++) {
1563 struct zone *zone = pgdat->node_zones + j;
1564 unsigned long size, freesize, memmap_pages;
1566 size = zone->spanned_pages;
1567 freesize = zone->present_pages;
1570 * Adjust freesize so that it accounts for how much memory
1571 * is used by this zone for memmap. This affects the watermark
1572 * and per-cpu initialisations
1574 memmap_pages = calc_memmap_size(size, freesize);
1575 if (!is_highmem_idx(j)) {
1576 if (freesize >= memmap_pages) {
1577 freesize -= memmap_pages;
1579 pr_debug(" %s zone: %lu pages used for memmap\n",
1580 zone_names[j], memmap_pages);
1582 pr_warn(" %s zone: %lu memmap pages exceeds freesize %lu\n",
1583 zone_names[j], memmap_pages, freesize);
1586 /* Account for reserved pages */
1587 if (j == 0 && freesize > dma_reserve) {
1588 freesize -= dma_reserve;
1589 pr_debug(" %s zone: %lu pages reserved\n", zone_names[0], dma_reserve);
1592 if (!is_highmem_idx(j))
1593 nr_kernel_pages += freesize;
1594 /* Charge for highmem memmap if there are enough kernel pages */
1595 else if (nr_kernel_pages > memmap_pages * 2)
1596 nr_kernel_pages -= memmap_pages;
1597 nr_all_pages += freesize;
1600 * Set an approximate value for lowmem here, it will be adjusted
1601 * when the bootmem allocator frees pages into the buddy system.
1602 * And all highmem pages will be managed by the buddy system.
1604 zone_init_internals(zone, j, nid, freesize);
1610 init_currently_empty_zone(zone, zone->zone_start_pfn, size);
1614 void __init *memmap_alloc(phys_addr_t size, phys_addr_t align,
1615 phys_addr_t min_addr, int nid, bool exact_nid)
1620 ptr = memblock_alloc_exact_nid_raw(size, align, min_addr,
1621 MEMBLOCK_ALLOC_ACCESSIBLE,
1624 ptr = memblock_alloc_try_nid_raw(size, align, min_addr,
1625 MEMBLOCK_ALLOC_ACCESSIBLE,
1628 if (ptr && size > 0)
1629 page_init_poison(ptr, size);
1634 #ifdef CONFIG_FLATMEM
1635 static void __init alloc_node_mem_map(struct pglist_data *pgdat)
1637 unsigned long start, offset, size, end;
1640 /* Skip empty nodes */
1641 if (!pgdat->node_spanned_pages)
1644 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
1645 offset = pgdat->node_start_pfn - start;
1647 * The zone's endpoints aren't required to be MAX_PAGE_ORDER
1648 * aligned but the node_mem_map endpoints must be in order
1649 * for the buddy allocator to function correctly.
1651 end = ALIGN(pgdat_end_pfn(pgdat), MAX_ORDER_NR_PAGES);
1652 size = (end - start) * sizeof(struct page);
1653 map = memmap_alloc(size, SMP_CACHE_BYTES, MEMBLOCK_LOW_LIMIT,
1654 pgdat->node_id, false);
1656 panic("Failed to allocate %ld bytes for node %d memory map\n",
1657 size, pgdat->node_id);
1658 pgdat->node_mem_map = map + offset;
1659 pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
1660 __func__, pgdat->node_id, (unsigned long)pgdat,
1661 (unsigned long)pgdat->node_mem_map);
1663 /* the global mem_map is just set as node 0's */
1664 if (pgdat == NODE_DATA(0)) {
1665 mem_map = NODE_DATA(0)->node_mem_map;
1666 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
1672 static inline void alloc_node_mem_map(struct pglist_data *pgdat) { }
1673 #endif /* CONFIG_FLATMEM */
1676 * get_pfn_range_for_nid - Return the start and end page frames for a node
1677 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
1678 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
1679 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
1681 * It returns the start and end page frame of a node based on information
1682 * provided by memblock_set_node(). If called for a node
1683 * with no available memory, the start and end PFNs will be 0.
1685 void __init get_pfn_range_for_nid(unsigned int nid,
1686 unsigned long *start_pfn, unsigned long *end_pfn)
1688 unsigned long this_start_pfn, this_end_pfn;
1694 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
1695 *start_pfn = min(*start_pfn, this_start_pfn);
1696 *end_pfn = max(*end_pfn, this_end_pfn);
1699 if (*start_pfn == -1UL)
1703 static void __init free_area_init_node(int nid)
1705 pg_data_t *pgdat = NODE_DATA(nid);
1706 unsigned long start_pfn = 0;
1707 unsigned long end_pfn = 0;
1709 /* pg_data_t should be reset to zero when it's allocated */
1710 WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx);
1712 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
1714 pgdat->node_id = nid;
1715 pgdat->node_start_pfn = start_pfn;
1716 pgdat->per_cpu_nodestats = NULL;
1718 if (start_pfn != end_pfn) {
1719 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
1720 (u64)start_pfn << PAGE_SHIFT,
1721 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
1723 calculate_node_totalpages(pgdat, start_pfn, end_pfn);
1725 pr_info("Initmem setup node %d as memoryless\n", nid);
1727 reset_memoryless_node_totalpages(pgdat);
1730 alloc_node_mem_map(pgdat);
1731 pgdat_set_deferred_range(pgdat);
1733 free_area_init_core(pgdat);
1734 lru_gen_init_pgdat(pgdat);
1737 /* Any regular or high memory on that node ? */
1738 static void __init check_for_memory(pg_data_t *pgdat)
1740 enum zone_type zone_type;
1742 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
1743 struct zone *zone = &pgdat->node_zones[zone_type];
1744 if (populated_zone(zone)) {
1745 if (IS_ENABLED(CONFIG_HIGHMEM))
1746 node_set_state(pgdat->node_id, N_HIGH_MEMORY);
1747 if (zone_type <= ZONE_NORMAL)
1748 node_set_state(pgdat->node_id, N_NORMAL_MEMORY);
1754 #if MAX_NUMNODES > 1
1756 * Figure out the number of possible node ids.
1758 void __init setup_nr_node_ids(void)
1760 unsigned int highest;
1762 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
1763 nr_node_ids = highest + 1;
1768 * Some architectures, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For
1769 * such cases we allow max_zone_pfn sorted in the descending order
1771 static bool arch_has_descending_max_zone_pfns(void)
1773 return IS_ENABLED(CONFIG_ARC) && !IS_ENABLED(CONFIG_ARC_HAS_PAE40);
1777 * free_area_init - Initialise all pg_data_t and zone data
1778 * @max_zone_pfn: an array of max PFNs for each zone
1780 * This will call free_area_init_node() for each active node in the system.
1781 * Using the page ranges provided by memblock_set_node(), the size of each
1782 * zone in each node and their holes is calculated. If the maximum PFN
1783 * between two adjacent zones match, it is assumed that the zone is empty.
1784 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
1785 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
1786 * starts where the previous one ended. For example, ZONE_DMA32 starts
1787 * at arch_max_dma_pfn.
1789 void __init free_area_init(unsigned long *max_zone_pfn)
1791 unsigned long start_pfn, end_pfn;
1795 /* Record where the zone boundaries are */
1796 memset(arch_zone_lowest_possible_pfn, 0,
1797 sizeof(arch_zone_lowest_possible_pfn));
1798 memset(arch_zone_highest_possible_pfn, 0,
1799 sizeof(arch_zone_highest_possible_pfn));
1801 start_pfn = PHYS_PFN(memblock_start_of_DRAM());
1802 descending = arch_has_descending_max_zone_pfns();
1804 for (i = 0; i < MAX_NR_ZONES; i++) {
1806 zone = MAX_NR_ZONES - i - 1;
1810 if (zone == ZONE_MOVABLE)
1813 end_pfn = max(max_zone_pfn[zone], start_pfn);
1814 arch_zone_lowest_possible_pfn[zone] = start_pfn;
1815 arch_zone_highest_possible_pfn[zone] = end_pfn;
1817 start_pfn = end_pfn;
1820 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
1821 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
1822 find_zone_movable_pfns_for_nodes();
1824 /* Print out the zone ranges */
1825 pr_info("Zone ranges:\n");
1826 for (i = 0; i < MAX_NR_ZONES; i++) {
1827 if (i == ZONE_MOVABLE)
1829 pr_info(" %-8s ", zone_names[i]);
1830 if (arch_zone_lowest_possible_pfn[i] ==
1831 arch_zone_highest_possible_pfn[i])
1834 pr_cont("[mem %#018Lx-%#018Lx]\n",
1835 (u64)arch_zone_lowest_possible_pfn[i]
1837 ((u64)arch_zone_highest_possible_pfn[i]
1838 << PAGE_SHIFT) - 1);
1841 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
1842 pr_info("Movable zone start for each node\n");
1843 for (i = 0; i < MAX_NUMNODES; i++) {
1844 if (zone_movable_pfn[i])
1845 pr_info(" Node %d: %#018Lx\n", i,
1846 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
1850 * Print out the early node map, and initialize the
1851 * subsection-map relative to active online memory ranges to
1852 * enable future "sub-section" extensions of the memory map.
1854 pr_info("Early memory node ranges\n");
1855 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
1856 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
1857 (u64)start_pfn << PAGE_SHIFT,
1858 ((u64)end_pfn << PAGE_SHIFT) - 1);
1859 subsection_map_init(start_pfn, end_pfn - start_pfn);
1862 /* Initialise every node */
1863 mminit_verify_pageflags_layout();
1864 setup_nr_node_ids();
1865 set_pageblock_order();
1867 for_each_node(nid) {
1870 if (!node_online(nid)) {
1871 /* Allocator not initialized yet */
1872 pgdat = arch_alloc_nodedata(nid);
1874 panic("Cannot allocate %zuB for node %d.\n",
1875 sizeof(*pgdat), nid);
1876 arch_refresh_nodedata(nid, pgdat);
1877 free_area_init_node(nid);
1880 * We do not want to confuse userspace by sysfs
1881 * files/directories for node without any memory
1882 * attached to it, so this node is not marked as
1883 * N_MEMORY and not marked online so that no sysfs
1884 * hierarchy will be created via register_one_node for
1885 * it. The pgdat will get fully initialized by
1886 * hotadd_init_pgdat() when memory is hotplugged into
1892 pgdat = NODE_DATA(nid);
1893 free_area_init_node(nid);
1895 /* Any memory on that node */
1896 if (pgdat->node_present_pages)
1897 node_set_state(nid, N_MEMORY);
1898 check_for_memory(pgdat);
1903 /* disable hash distribution for systems with a single node */
1908 * node_map_pfn_alignment - determine the maximum internode alignment
1910 * This function should be called after node map is populated and sorted.
1911 * It calculates the maximum power of two alignment which can distinguish
1914 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
1915 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
1916 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
1917 * shifted, 1GiB is enough and this function will indicate so.
1919 * This is used to test whether pfn -> nid mapping of the chosen memory
1920 * model has fine enough granularity to avoid incorrect mapping for the
1921 * populated node map.
1923 * Return: the determined alignment in pfn's. 0 if there is no alignment
1924 * requirement (single node).
1926 unsigned long __init node_map_pfn_alignment(void)
1928 unsigned long accl_mask = 0, last_end = 0;
1929 unsigned long start, end, mask;
1930 int last_nid = NUMA_NO_NODE;
1933 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
1934 if (!start || last_nid < 0 || last_nid == nid) {
1941 * Start with a mask granular enough to pin-point to the
1942 * start pfn and tick off bits one-by-one until it becomes
1943 * too coarse to separate the current node from the last.
1945 mask = ~((1 << __ffs(start)) - 1);
1946 while (mask && last_end <= (start & (mask << 1)))
1949 /* accumulate all internode masks */
1953 /* convert mask to number of pages */
1954 return ~accl_mask + 1;
1957 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1958 static void __init deferred_free_range(unsigned long pfn,
1959 unsigned long nr_pages)
1967 page = pfn_to_page(pfn);
1969 /* Free a large naturally-aligned chunk if possible */
1970 if (nr_pages == MAX_ORDER_NR_PAGES && IS_MAX_ORDER_ALIGNED(pfn)) {
1971 for (i = 0; i < nr_pages; i += pageblock_nr_pages)
1972 set_pageblock_migratetype(page + i, MIGRATE_MOVABLE);
1973 __free_pages_core(page, MAX_PAGE_ORDER);
1977 /* Accept chunks smaller than MAX_PAGE_ORDER upfront */
1978 accept_memory(PFN_PHYS(pfn), PFN_PHYS(pfn + nr_pages));
1980 for (i = 0; i < nr_pages; i++, page++, pfn++) {
1981 if (pageblock_aligned(pfn))
1982 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1983 __free_pages_core(page, 0);
1987 /* Completion tracking for deferred_init_memmap() threads */
1988 static atomic_t pgdat_init_n_undone __initdata;
1989 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1991 static inline void __init pgdat_init_report_one_done(void)
1993 if (atomic_dec_and_test(&pgdat_init_n_undone))
1994 complete(&pgdat_init_all_done_comp);
1998 * Returns true if page needs to be initialized or freed to buddy allocator.
2000 * We check if a current MAX_PAGE_ORDER block is valid by only checking the
2001 * validity of the head pfn.
2003 static inline bool __init deferred_pfn_valid(unsigned long pfn)
2005 if (IS_MAX_ORDER_ALIGNED(pfn) && !pfn_valid(pfn))
2011 * Free pages to buddy allocator. Try to free aligned pages in
2012 * MAX_ORDER_NR_PAGES sizes.
2014 static void __init deferred_free_pages(unsigned long pfn,
2015 unsigned long end_pfn)
2017 unsigned long nr_free = 0;
2019 for (; pfn < end_pfn; pfn++) {
2020 if (!deferred_pfn_valid(pfn)) {
2021 deferred_free_range(pfn - nr_free, nr_free);
2023 } else if (IS_MAX_ORDER_ALIGNED(pfn)) {
2024 deferred_free_range(pfn - nr_free, nr_free);
2030 /* Free the last block of pages to allocator */
2031 deferred_free_range(pfn - nr_free, nr_free);
2035 * Initialize struct pages. We minimize pfn page lookups and scheduler checks
2036 * by performing it only once every MAX_ORDER_NR_PAGES.
2037 * Return number of pages initialized.
2039 static unsigned long __init deferred_init_pages(struct zone *zone,
2041 unsigned long end_pfn)
2043 int nid = zone_to_nid(zone);
2044 unsigned long nr_pages = 0;
2045 int zid = zone_idx(zone);
2046 struct page *page = NULL;
2048 for (; pfn < end_pfn; pfn++) {
2049 if (!deferred_pfn_valid(pfn)) {
2052 } else if (!page || IS_MAX_ORDER_ALIGNED(pfn)) {
2053 page = pfn_to_page(pfn);
2057 __init_single_page(page, pfn, zid, nid);
2064 * This function is meant to pre-load the iterator for the zone init.
2065 * Specifically it walks through the ranges until we are caught up to the
2066 * first_init_pfn value and exits there. If we never encounter the value we
2067 * return false indicating there are no valid ranges left.
2070 deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone,
2071 unsigned long *spfn, unsigned long *epfn,
2072 unsigned long first_init_pfn)
2077 * Start out by walking through the ranges in this zone that have
2078 * already been initialized. We don't need to do anything with them
2079 * so we just need to flush them out of the system.
2081 for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) {
2082 if (*epfn <= first_init_pfn)
2084 if (*spfn < first_init_pfn)
2085 *spfn = first_init_pfn;
2094 * Initialize and free pages. We do it in two loops: first we initialize
2095 * struct page, then free to buddy allocator, because while we are
2096 * freeing pages we can access pages that are ahead (computing buddy
2097 * page in __free_one_page()).
2099 * In order to try and keep some memory in the cache we have the loop
2100 * broken along max page order boundaries. This way we will not cause
2101 * any issues with the buddy page computation.
2103 static unsigned long __init
2104 deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn,
2105 unsigned long *end_pfn)
2107 unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES);
2108 unsigned long spfn = *start_pfn, epfn = *end_pfn;
2109 unsigned long nr_pages = 0;
2112 /* First we loop through and initialize the page values */
2113 for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) {
2116 if (mo_pfn <= *start_pfn)
2119 t = min(mo_pfn, *end_pfn);
2120 nr_pages += deferred_init_pages(zone, *start_pfn, t);
2122 if (mo_pfn < *end_pfn) {
2123 *start_pfn = mo_pfn;
2128 /* Reset values and now loop through freeing pages as needed */
2131 for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) {
2137 t = min(mo_pfn, epfn);
2138 deferred_free_pages(spfn, t);
2148 deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
2151 unsigned long spfn, epfn;
2152 struct zone *zone = arg;
2155 deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn);
2158 * Initialize and free pages in MAX_PAGE_ORDER sized increments so that
2159 * we can avoid introducing any issues with the buddy allocator.
2161 while (spfn < end_pfn) {
2162 deferred_init_maxorder(&i, zone, &spfn, &epfn);
2167 /* An arch may override for more concurrency. */
2169 deferred_page_init_max_threads(const struct cpumask *node_cpumask)
2174 /* Initialise remaining memory on a node */
2175 static int __init deferred_init_memmap(void *data)
2177 pg_data_t *pgdat = data;
2178 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
2179 unsigned long spfn = 0, epfn = 0;
2180 unsigned long first_init_pfn, flags;
2181 unsigned long start = jiffies;
2183 int zid, max_threads;
2186 /* Bind memory initialisation thread to a local node if possible */
2187 if (!cpumask_empty(cpumask))
2188 set_cpus_allowed_ptr(current, cpumask);
2190 pgdat_resize_lock(pgdat, &flags);
2191 first_init_pfn = pgdat->first_deferred_pfn;
2192 if (first_init_pfn == ULONG_MAX) {
2193 pgdat_resize_unlock(pgdat, &flags);
2194 pgdat_init_report_one_done();
2198 /* Sanity check boundaries */
2199 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
2200 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
2201 pgdat->first_deferred_pfn = ULONG_MAX;
2204 * Once we unlock here, the zone cannot be grown anymore, thus if an
2205 * interrupt thread must allocate this early in boot, zone must be
2206 * pre-grown prior to start of deferred page initialization.
2208 pgdat_resize_unlock(pgdat, &flags);
2210 /* Only the highest zone is deferred so find it */
2211 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2212 zone = pgdat->node_zones + zid;
2213 if (first_init_pfn < zone_end_pfn(zone))
2217 /* If the zone is empty somebody else may have cleared out the zone */
2218 if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2222 max_threads = deferred_page_init_max_threads(cpumask);
2224 while (spfn < epfn) {
2225 unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION);
2226 struct padata_mt_job job = {
2227 .thread_fn = deferred_init_memmap_chunk,
2230 .size = epfn_align - spfn,
2231 .align = PAGES_PER_SECTION,
2232 .min_chunk = PAGES_PER_SECTION,
2233 .max_threads = max_threads,
2234 .numa_aware = false,
2237 padata_do_multithreaded(&job);
2238 deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2242 /* Sanity check that the next zone really is unpopulated */
2243 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
2245 pr_info("node %d deferred pages initialised in %ums\n",
2246 pgdat->node_id, jiffies_to_msecs(jiffies - start));
2248 pgdat_init_report_one_done();
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.
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.
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.
2267 bool __init deferred_grow_zone(struct zone *zone, unsigned int order)
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;
2276 /* Only the last zone may have deferred pages */
2277 if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
2280 pgdat_resize_lock(pgdat, &flags);
2283 * If someone grew this zone while we were waiting for spinlock, return
2284 * true, as there might be enough pages already.
2286 if (first_deferred_pfn != pgdat->first_deferred_pfn) {
2287 pgdat_resize_unlock(pgdat, &flags);
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;
2301 * Initialize and free pages in MAX_PAGE_ORDER sized increments so
2302 * that we can avoid introducing any issues with the buddy
2305 while (spfn < epfn) {
2306 /* update our first deferred PFN for this section */
2307 first_deferred_pfn = spfn;
2309 nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn);
2310 touch_nmi_watchdog();
2312 /* We should only stop along section boundaries */
2313 if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION)
2316 /* If our quota has been met we can stop here */
2317 if (nr_pages >= nr_pages_needed)
2321 pgdat->first_deferred_pfn = spfn;
2322 pgdat_resize_unlock(pgdat, &flags);
2324 return nr_pages > 0;
2327 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
2330 void __init init_cma_reserved_pageblock(struct page *page)
2332 unsigned i = pageblock_nr_pages;
2333 struct page *p = page;
2336 __ClearPageReserved(p);
2337 set_page_count(p, 0);
2340 set_pageblock_migratetype(page, MIGRATE_CMA);
2341 set_page_refcounted(page);
2342 __free_pages(page, pageblock_order);
2344 adjust_managed_page_count(page, pageblock_nr_pages);
2345 page_zone(page)->cma_pages += pageblock_nr_pages;
2349 void set_zone_contiguous(struct zone *zone)
2351 unsigned long block_start_pfn = zone->zone_start_pfn;
2352 unsigned long block_end_pfn;
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) {
2359 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
2361 if (!__pageblock_pfn_to_page(block_start_pfn,
2362 block_end_pfn, zone))
2367 /* We confirm that there is no hole */
2368 zone->contiguous = true;
2371 void __init page_alloc_init_late(void)
2376 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
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);
2384 /* Block until all are initialised */
2385 wait_for_completion(&pgdat_init_all_done_comp);
2388 * We initialized the rest of the deferred pages. Permanently disable
2389 * on-demand struct page initialization.
2391 static_branch_disable(&deferred_pages);
2393 /* Reinit limits that are based on free pages after the kernel is up */
2394 files_maxfiles_init();
2399 /* Discard memblock private memory */
2402 for_each_node_state(nid, N_MEMORY)
2403 shuffle_free_memory(NODE_DATA(nid));
2405 for_each_populated_zone(zone)
2406 set_zone_contiguous(zone);
2408 /* Initialize page ext after all struct pages are initialized. */
2409 if (deferred_struct_pages)
2412 page_alloc_sysctl_init();
2415 #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
2417 * Returns the number of pages that arch has reserved but
2418 * is not known to alloc_large_system_hash().
2420 static unsigned long __init arch_reserved_kernel_pages(void)
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.
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)
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
2447 void *__init alloc_large_system_hash(const char *tablename,
2448 unsigned long bucketsize,
2449 unsigned long numentries,
2452 unsigned int *_hash_shift,
2453 unsigned int *_hash_mask,
2454 unsigned long low_limit,
2455 unsigned long high_limit)
2457 unsigned long long max = high_limit;
2458 unsigned long log2qty, size;
2464 /* allow the kernel cmdline to have a say */
2466 /* round applicable memory size up to nearest megabyte */
2467 numentries = nr_kernel_pages;
2468 numentries -= arch_reserved_kernel_pages();
2470 /* It isn't necessary when PAGE_SIZE >= 1MB */
2471 if (PAGE_SIZE < SZ_1M)
2472 numentries = round_up(numentries, SZ_1M / PAGE_SIZE);
2474 #if __BITS_PER_LONG > 32
2476 unsigned long adapt;
2478 for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
2479 adapt <<= ADAPT_SCALE_SHIFT)
2484 /* limit to 1 bucket per 2^scale bytes of low memory */
2485 if (scale > PAGE_SHIFT)
2486 numentries >>= (scale - PAGE_SHIFT);
2488 numentries <<= (PAGE_SHIFT - scale);
2490 if (unlikely((numentries * bucketsize) < PAGE_SIZE))
2491 numentries = PAGE_SIZE / bucketsize;
2493 numentries = roundup_pow_of_two(numentries);
2495 /* limit allocation size to 1/16 total memory by default */
2497 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2498 do_div(max, bucketsize);
2500 max = min(max, 0x80000000ULL);
2502 if (numentries < low_limit)
2503 numentries = low_limit;
2504 if (numentries > max)
2507 log2qty = ilog2(numentries);
2509 gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
2512 size = bucketsize << log2qty;
2513 if (flags & HASH_EARLY) {
2514 if (flags & HASH_ZERO)
2515 table = memblock_alloc(size, SMP_CACHE_BYTES);
2517 table = memblock_alloc_raw(size,
2519 } else if (get_order(size) > MAX_PAGE_ORDER || hashdist) {
2520 table = vmalloc_huge(size, gfp_flags);
2523 huge = is_vm_area_hugepages(table);
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
2530 table = alloc_pages_exact(size, gfp_flags);
2531 kmemleak_alloc(table, size, 1, gfp_flags);
2533 } while (!table && size > PAGE_SIZE && --log2qty);
2536 panic("Failed to allocate %s hash table\n", tablename);
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");
2543 *_hash_shift = log2qty;
2545 *_hash_mask = (1 << log2qty) - 1;
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
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.
2561 void __init set_dma_reserve(unsigned long new_dma_reserve)
2563 dma_reserve = new_dma_reserve;
2566 void __init memblock_free_pages(struct page *page, unsigned long pfn,
2570 if (IS_ENABLED(CONFIG_DEFERRED_STRUCT_PAGE_INIT)) {
2571 int nid = early_pfn_to_nid(pfn);
2573 if (!early_page_initialised(pfn, nid))
2577 if (!kmsan_memblock_free_pages(page, order)) {
2578 /* KMSAN will take care of these pages. */
2581 __free_pages_core(page, order);
2584 DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, init_on_alloc);
2585 EXPORT_SYMBOL(init_on_alloc);
2587 DEFINE_STATIC_KEY_MAYBE(CONFIG_INIT_ON_FREE_DEFAULT_ON, init_on_free);
2588 EXPORT_SYMBOL(init_on_free);
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)
2595 return kstrtobool(buf, &_init_on_alloc_enabled_early);
2597 early_param("init_on_alloc", early_init_on_alloc);
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)
2603 return kstrtobool(buf, &_init_on_free_enabled_early);
2605 early_param("init_on_free", early_init_on_free);
2607 DEFINE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled);
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.
2615 static void __init mem_debugging_and_hardening_init(void)
2617 bool page_poisoning_requested = false;
2618 bool want_check_pages = false;
2620 #ifdef CONFIG_PAGE_POISONING
2622 * Page poisoning is debug page alloc for some arches. If
2623 * either of those options are enabled, enable poisoning.
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;
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;
2642 if (_init_on_alloc_enabled_early) {
2643 want_check_pages = true;
2644 static_branch_enable(&init_on_alloc);
2646 static_branch_disable(&init_on_alloc);
2649 if (_init_on_free_enabled_early) {
2650 want_check_pages = true;
2651 static_branch_enable(&init_on_free);
2653 static_branch_disable(&init_on_free);
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");
2660 #ifdef CONFIG_DEBUG_PAGEALLOC
2661 if (debug_pagealloc_enabled()) {
2662 want_check_pages = true;
2663 static_branch_enable(&_debug_pagealloc_enabled);
2665 if (debug_guardpage_minorder())
2666 static_branch_enable(&_debug_guardpage_enabled);
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
2675 if (!IS_ENABLED(CONFIG_DEBUG_VM) && want_check_pages)
2676 static_branch_enable(&check_pages_enabled);
2679 /* Report memory auto-initialization states for this boot. */
2680 static void __init report_meminit(void)
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)";
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");
2704 static void __init mem_init_print_info(void)
2706 unsigned long physpages, codesize, datasize, rosize, bss_size;
2707 unsigned long init_code_size, init_data_size;
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;
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.
2724 #define adj_init_size(start, end, size, pos, adj) \
2726 if (&start[0] <= &pos[0] && &pos[0] < &end[0] && size > adj) \
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);
2737 #undef adj_init_size
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
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),
2749 #ifdef CONFIG_HIGHMEM
2750 , K(totalhigh_pages())
2756 * Set up kernel memory allocators
2758 void __init mm_core_init(void)
2760 /* Initializations relying on SMP setup */
2761 build_all_zonelists(NULL);
2762 page_alloc_init_cpuhp();
2765 * page_ext requires contiguous pages,
2766 * bigger than MAX_PAGE_ORDER unless SPARSEMEM.
2768 page_ext_init_flatmem();
2769 mem_debugging_and_hardening_init();
2770 kfence_alloc_pool_and_metadata();
2772 kmsan_init_shadow();
2773 stack_depot_early_init();
2775 mem_init_print_info();
2778 * page_owner must be initialized after buddy is ready, and also after
2779 * slab is ready so that stack_depot_init() works properly
2781 page_ext_init_flatmem_late();
2783 ptlock_cache_init();
2784 pgtable_cache_init();
2785 debug_objects_mem_init();
2787 /* If no deferred init page_ext now, as vmap is fully initialized */
2788 if (!deferred_struct_pages)
2790 /* Should be run before the first non-init thread is created */
2792 /* Should be run after espfix64 is set up. */
2794 kmsan_init_runtime();