1 // SPDX-License-Identifier: GPL-2.0-only
3 * mm_init.c - Memory initialisation verification and debugging
5 * Copyright 2008 IBM Corporation, 2008
6 * Author Mel Gorman <mel@csn.ul.ie>
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>
33 #include <asm/setup.h>
35 #ifdef CONFIG_DEBUG_MEMORY_INIT
36 int __meminitdata mminit_loglevel;
38 /* The zonelists are simply reported, validation is manual. */
39 void __init mminit_verify_zonelist(void)
43 if (mminit_loglevel < MMINIT_VERIFY)
46 for_each_online_node(nid) {
47 pg_data_t *pgdat = NODE_DATA(nid);
50 struct zonelist *zonelist;
51 int i, listid, zoneid;
53 BUILD_BUG_ON(MAX_ZONELISTS > 2);
54 for (i = 0; i < MAX_ZONELISTS * MAX_NR_ZONES; i++) {
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))
64 /* Print information about the zonelist */
65 printk(KERN_DEBUG "mminit::zonelist %s %d:%s = ",
66 listid > 0 ? "thisnode" : "general", nid,
69 /* Iterate the zonelist */
70 for_each_zone_zonelist(zone, z, zonelist, zoneid)
71 pr_cont("%d:%s ", zone_to_nid(zone), zone->name);
77 void __init mminit_verify_pageflags_layout(void)
80 unsigned long or_mask, add_mask;
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",
95 mminit_dprintk(MMINIT_TRACE, "pageflags_layout_shifts",
96 "Section %d Node %d Zone %d Lastcpupid %d Kasantag %d\n",
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");
120 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
121 mminit_dprintk(MMINIT_TRACE, "pageflags_layout_nodeflags",
122 "Last cpupid not in page flags");
125 if (SECTIONS_WIDTH) {
126 shift -= SECTIONS_WIDTH;
127 BUG_ON(shift != SECTIONS_PGSHIFT);
130 shift -= NODES_WIDTH;
131 BUG_ON(shift != NODES_PGSHIFT);
134 shift -= ZONES_WIDTH;
135 BUG_ON(shift != ZONES_PGSHIFT);
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);
148 static __init int set_mminit_loglevel(char *str)
150 get_option(&str, &mminit_loglevel);
153 early_param("mminit_loglevel", set_mminit_loglevel);
154 #endif /* CONFIG_DEBUG_MEMORY_INIT */
156 struct kobject *mm_kobj;
157 EXPORT_SYMBOL_GPL(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;
263 * kernelcore=size sets the amount of memory for use for allocations that
264 * cannot be reclaimed or migrated.
266 static int __init cmdline_parse_kernelcore(char *p)
268 /* parse kernelcore=mirror */
269 if (parse_option_str(p, "mirror")) {
270 mirrored_kernelcore = true;
274 return cmdline_parse_core(p, &required_kernelcore,
275 &required_kernelcore_percent);
277 early_param("kernelcore", cmdline_parse_kernelcore);
280 * movablecore=size sets the amount of memory for use for allocations that
281 * can be reclaimed or migrated.
283 static int __init cmdline_parse_movablecore(char *p)
285 return cmdline_parse_core(p, &required_movablecore,
286 &required_movablecore_percent);
288 early_param("movablecore", cmdline_parse_movablecore);
291 * early_calculate_totalpages()
292 * Sum pages in active regions for movable zone.
293 * Populate N_MEMORY for calculating usable_nodes.
295 static unsigned long __init early_calculate_totalpages(void)
297 unsigned long totalpages = 0;
298 unsigned long start_pfn, end_pfn;
301 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
302 unsigned long pages = end_pfn - start_pfn;
306 node_set_state(nid, N_MEMORY);
312 * This finds a zone that can be used for ZONE_MOVABLE pages. The
313 * assumption is made that zones within a node are ordered in monotonic
314 * increasing memory addresses so that the "highest" populated zone is used
316 static void __init find_usable_zone_for_movable(void)
319 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
320 if (zone_index == ZONE_MOVABLE)
323 if (arch_zone_highest_possible_pfn[zone_index] >
324 arch_zone_lowest_possible_pfn[zone_index])
328 VM_BUG_ON(zone_index == -1);
329 movable_zone = zone_index;
333 * Find the PFN the Movable zone begins in each node. Kernel memory
334 * is spread evenly between nodes as long as the nodes have enough
335 * memory. When they don't, some nodes will have more kernelcore than
338 static void __init find_zone_movable_pfns_for_nodes(void)
341 unsigned long usable_startpfn;
342 unsigned long kernelcore_node, kernelcore_remaining;
343 /* save the state before borrow the nodemask */
344 nodemask_t saved_node_state = node_states[N_MEMORY];
345 unsigned long totalpages = early_calculate_totalpages();
346 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
347 struct memblock_region *r;
349 /* Need to find movable_zone earlier when movable_node is specified. */
350 find_usable_zone_for_movable();
353 * If movable_node is specified, ignore kernelcore and movablecore
356 if (movable_node_is_enabled()) {
357 for_each_mem_region(r) {
358 if (!memblock_is_hotpluggable(r))
361 nid = memblock_get_region_node(r);
363 usable_startpfn = PFN_DOWN(r->base);
364 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
365 min(usable_startpfn, zone_movable_pfn[nid]) :
373 * If kernelcore=mirror is specified, ignore movablecore option
375 if (mirrored_kernelcore) {
376 bool mem_below_4gb_not_mirrored = false;
378 for_each_mem_region(r) {
379 if (memblock_is_mirror(r))
382 nid = memblock_get_region_node(r);
384 usable_startpfn = memblock_region_memory_base_pfn(r);
386 if (usable_startpfn < PHYS_PFN(SZ_4G)) {
387 mem_below_4gb_not_mirrored = true;
391 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
392 min(usable_startpfn, zone_movable_pfn[nid]) :
396 if (mem_below_4gb_not_mirrored)
397 pr_warn("This configuration results in unmirrored kernel memory.\n");
403 * If kernelcore=nn% or movablecore=nn% was specified, calculate the
404 * amount of necessary memory.
406 if (required_kernelcore_percent)
407 required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
409 if (required_movablecore_percent)
410 required_movablecore = (totalpages * 100 * required_movablecore_percent) /
414 * If movablecore= was specified, calculate what size of
415 * kernelcore that corresponds so that memory usable for
416 * any allocation type is evenly spread. If both kernelcore
417 * and movablecore are specified, then the value of kernelcore
418 * will be used for required_kernelcore if it's greater than
419 * what movablecore would have allowed.
421 if (required_movablecore) {
422 unsigned long corepages;
425 * Round-up so that ZONE_MOVABLE is at least as large as what
426 * was requested by the user
428 required_movablecore =
429 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
430 required_movablecore = min(totalpages, required_movablecore);
431 corepages = totalpages - required_movablecore;
433 required_kernelcore = max(required_kernelcore, corepages);
437 * If kernelcore was not specified or kernelcore size is larger
438 * than totalpages, there is no ZONE_MOVABLE.
440 if (!required_kernelcore || required_kernelcore >= totalpages)
443 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
444 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
447 /* Spread kernelcore memory as evenly as possible throughout nodes */
448 kernelcore_node = required_kernelcore / usable_nodes;
449 for_each_node_state(nid, N_MEMORY) {
450 unsigned long start_pfn, end_pfn;
453 * Recalculate kernelcore_node if the division per node
454 * now exceeds what is necessary to satisfy the requested
455 * amount of memory for the kernel
457 if (required_kernelcore < kernelcore_node)
458 kernelcore_node = required_kernelcore / usable_nodes;
461 * As the map is walked, we track how much memory is usable
462 * by the kernel using kernelcore_remaining. When it is
463 * 0, the rest of the node is usable by ZONE_MOVABLE
465 kernelcore_remaining = kernelcore_node;
467 /* Go through each range of PFNs within this node */
468 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
469 unsigned long size_pages;
471 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
472 if (start_pfn >= end_pfn)
475 /* Account for what is only usable for kernelcore */
476 if (start_pfn < usable_startpfn) {
477 unsigned long kernel_pages;
478 kernel_pages = min(end_pfn, usable_startpfn)
481 kernelcore_remaining -= min(kernel_pages,
482 kernelcore_remaining);
483 required_kernelcore -= min(kernel_pages,
484 required_kernelcore);
486 /* Continue if range is now fully accounted */
487 if (end_pfn <= usable_startpfn) {
490 * Push zone_movable_pfn to the end so
491 * that if we have to rebalance
492 * kernelcore across nodes, we will
493 * not double account here
495 zone_movable_pfn[nid] = end_pfn;
498 start_pfn = usable_startpfn;
502 * The usable PFN range for ZONE_MOVABLE is from
503 * start_pfn->end_pfn. Calculate size_pages as the
504 * number of pages used as kernelcore
506 size_pages = end_pfn - start_pfn;
507 if (size_pages > kernelcore_remaining)
508 size_pages = kernelcore_remaining;
509 zone_movable_pfn[nid] = start_pfn + size_pages;
512 * Some kernelcore has been met, update counts and
513 * break if the kernelcore for this node has been
516 required_kernelcore -= min(required_kernelcore,
518 kernelcore_remaining -= size_pages;
519 if (!kernelcore_remaining)
525 * If there is still required_kernelcore, we do another pass with one
526 * less node in the count. This will push zone_movable_pfn[nid] further
527 * along on the nodes that still have memory until kernelcore is
531 if (usable_nodes && required_kernelcore > usable_nodes)
535 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
536 for (nid = 0; nid < MAX_NUMNODES; nid++) {
537 unsigned long start_pfn, end_pfn;
539 zone_movable_pfn[nid] =
540 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
542 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
543 if (zone_movable_pfn[nid] >= end_pfn)
544 zone_movable_pfn[nid] = 0;
548 /* restore the node_state */
549 node_states[N_MEMORY] = saved_node_state;
552 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
553 unsigned long zone, int nid)
555 mm_zero_struct_page(page);
556 set_page_links(page, zone, nid, pfn);
557 init_page_count(page);
558 page_mapcount_reset(page);
559 page_cpupid_reset_last(page);
560 page_kasan_tag_reset(page);
562 INIT_LIST_HEAD(&page->lru);
563 #ifdef WANT_PAGE_VIRTUAL
564 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
565 if (!is_highmem_idx(zone))
566 set_page_address(page, __va(pfn << PAGE_SHIFT));
572 * During memory init memblocks map pfns to nids. The search is expensive and
573 * this caches recent lookups. The implementation of __early_pfn_to_nid
574 * treats start/end as pfns.
576 struct mminit_pfnnid_cache {
577 unsigned long last_start;
578 unsigned long last_end;
582 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
585 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
587 static int __meminit __early_pfn_to_nid(unsigned long pfn,
588 struct mminit_pfnnid_cache *state)
590 unsigned long start_pfn, end_pfn;
593 if (state->last_start <= pfn && pfn < state->last_end)
594 return state->last_nid;
596 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
597 if (nid != NUMA_NO_NODE) {
598 state->last_start = start_pfn;
599 state->last_end = end_pfn;
600 state->last_nid = nid;
606 int __meminit early_pfn_to_nid(unsigned long pfn)
608 static DEFINE_SPINLOCK(early_pfn_lock);
611 spin_lock(&early_pfn_lock);
612 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
614 nid = first_online_node;
615 spin_unlock(&early_pfn_lock);
620 int hashdist = HASHDIST_DEFAULT;
622 static int __init set_hashdist(char *str)
626 hashdist = simple_strtoul(str, &str, 0);
629 __setup("hashdist=", set_hashdist);
631 static inline void fixup_hashdist(void)
633 if (num_node_state(N_MEMORY) == 1)
637 static inline void fixup_hashdist(void) {}
638 #endif /* CONFIG_NUMA */
640 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
641 static inline void pgdat_set_deferred_range(pg_data_t *pgdat)
643 pgdat->first_deferred_pfn = ULONG_MAX;
646 /* Returns true if the struct page for the pfn is initialised */
647 static inline bool __meminit early_page_initialised(unsigned long pfn)
649 int nid = early_pfn_to_nid(pfn);
651 if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
658 * Returns true when the remaining initialisation should be deferred until
659 * later in the boot cycle when it can be parallelised.
661 static bool __meminit
662 defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
664 static unsigned long prev_end_pfn, nr_initialised;
666 if (early_page_ext_enabled())
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.
672 if (prev_end_pfn != end_pfn) {
673 prev_end_pfn = end_pfn;
677 /* Always populate low zones for address-constrained allocations */
678 if (end_pfn < pgdat_end_pfn(NODE_DATA(nid)))
681 if (NODE_DATA(nid)->first_deferred_pfn != ULONG_MAX)
684 * We start only with one section of pages, more pages are added as
685 * needed until the rest of deferred pages are initialized.
688 if ((nr_initialised > PAGES_PER_SECTION) &&
689 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
690 NODE_DATA(nid)->first_deferred_pfn = pfn;
696 static void __meminit init_reserved_page(unsigned long pfn)
701 if (early_page_initialised(pfn))
704 nid = early_pfn_to_nid(pfn);
705 pgdat = NODE_DATA(nid);
707 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
708 struct zone *zone = &pgdat->node_zones[zid];
710 if (zone_spans_pfn(zone, pfn))
713 __init_single_page(pfn_to_page(pfn), pfn, zid, nid);
716 static inline void pgdat_set_deferred_range(pg_data_t *pgdat) {}
718 static inline bool early_page_initialised(unsigned long pfn)
723 static inline bool defer_init(int nid, unsigned long pfn, unsigned long end_pfn)
728 static inline void init_reserved_page(unsigned long pfn)
731 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
734 * Initialised pages do not have PageReserved set. This function is
735 * called for each range allocated by the bootmem allocator and
736 * marks the pages PageReserved. The remaining valid pages are later
737 * sent to the buddy page allocator.
739 void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
741 unsigned long start_pfn = PFN_DOWN(start);
742 unsigned long end_pfn = PFN_UP(end);
744 for (; start_pfn < end_pfn; start_pfn++) {
745 if (pfn_valid(start_pfn)) {
746 struct page *page = pfn_to_page(start_pfn);
748 init_reserved_page(start_pfn);
750 /* Avoid false-positive PageTail() */
751 INIT_LIST_HEAD(&page->lru);
754 * no need for atomic set_bit because the struct
755 * page is not visible yet so nobody should
758 __SetPageReserved(page);
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)
767 static struct memblock_region *r;
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))
776 if (*pfn >= memblock_region_memory_base_pfn(r) &&
777 memblock_is_mirror(r)) {
778 *pfn = memblock_region_memory_end_pfn(r);
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().
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
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.
807 static void __init init_unavailable_range(unsigned long spfn,
814 for (pfn = spfn; pfn < epfn; pfn++) {
815 if (!pfn_valid(pageblock_start_pfn(pfn))) {
816 pfn = pageblock_end_pfn(pfn) - 1;
819 __init_single_page(pfn_to_page(pfn), pfn, zone, node);
820 __SetPageReserved(pfn_to_page(pfn));
825 pr_info("On node %d, zone %s: %lld pages in unavailable ranges",
826 node, zone_names[zone], pgcnt);
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.
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.
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)
843 unsigned long pfn, end_pfn = start_pfn + size;
846 if (highest_memmap_pfn < end_pfn - 1)
847 highest_memmap_pfn = end_pfn - 1;
849 #ifdef CONFIG_ZONE_DEVICE
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
857 if (zone == ZONE_DEVICE) {
861 if (start_pfn == altmap->base_pfn)
862 start_pfn += altmap->reserve;
863 end_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
867 for (pfn = start_pfn; pfn < end_pfn; ) {
869 * There can be holes in boot-time mem_map[]s handed to this
870 * function. They do not exist on hotplugged memory.
872 if (context == MEMINIT_EARLY) {
873 if (overlap_memmap_init(zone, &pfn))
875 if (defer_init(nid, pfn, zone_end_pfn)) {
876 deferred_struct_pages = true;
881 page = pfn_to_page(pfn);
882 __init_single_page(page, pfn, zone, nid);
883 if (context == MEMINIT_HOTPLUG)
884 __SetPageReserved(page);
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.
891 if (pageblock_aligned(pfn)) {
892 set_pageblock_migratetype(page, migratetype);
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)
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);
908 start_pfn = clamp(start_pfn, zone_start_pfn, zone_end_pfn);
909 end_pfn = clamp(end_pfn, zone_start_pfn, zone_end_pfn);
911 if (start_pfn >= end_pfn)
914 memmap_init_range(end_pfn - start_pfn, nid, zone_id, start_pfn,
915 zone_end_pfn, MEMINIT_EARLY, NULL, MIGRATE_MOVABLE);
917 if (*hole_pfn < start_pfn)
918 init_unavailable_range(*hole_pfn, start_pfn, zone_id, nid);
923 static void __init memmap_init(void)
925 unsigned long start_pfn, end_pfn;
926 unsigned long hole_pfn = 0;
927 int i, j, zone_id = 0, nid;
929 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
930 struct pglist_data *node = NODE_DATA(nid);
932 for (j = 0; j < MAX_NR_ZONES; j++) {
933 struct zone *zone = node->node_zones + j;
935 if (!populated_zone(zone))
938 memmap_init_zone_range(zone, start_pfn, end_pfn,
944 #ifdef CONFIG_SPARSEMEM
946 * Initialize the memory map for hole in the range [memory_end,
948 * Append the pages in this hole to the highest zone in the last
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
954 end_pfn = round_up(end_pfn, PAGES_PER_SECTION);
955 if (hole_pfn < end_pfn)
957 init_unavailable_range(hole_pfn, end_pfn, zone_id, nid);
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)
966 __init_single_page(page, pfn, zone_idx, nid);
969 * Mark page reserved as it will need to wait for onlining
970 * phase for it to be fully associated with a zone.
972 * We can use the non-atomic __set_bit operation for setting
973 * the flag as we are still initializing the pages.
975 __SetPageReserved(page);
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.
983 page->zone_device_data = NULL;
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.
992 * Please note that MEMINIT_HOTPLUG path doesn't clear memmap
993 * because this is done early in section_activate()
995 if (pageblock_aligned(pfn)) {
996 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
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.
1004 if (pgmap->type == MEMORY_DEVICE_PRIVATE ||
1005 pgmap->type == MEMORY_DEVICE_COHERENT)
1006 set_page_count(page, 0);
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().
1017 static inline unsigned long compound_nr_pages(struct vmem_altmap *altmap,
1018 struct dev_pagemap *pgmap)
1020 if (!vmemmap_can_optimize(altmap, pgmap))
1021 return pgmap_vmemmap_nr(pgmap);
1023 return 2 * (PAGE_SIZE / sizeof(struct page));
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)
1032 unsigned long pfn, end_pfn = head_pfn + nr_pages;
1033 unsigned int order = pgmap->vmemmap_shift;
1035 __SetPageHead(head);
1036 for (pfn = head_pfn + 1; pfn < end_pfn; pfn++) {
1037 struct page *page = pfn_to_page(pfn);
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);
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.
1048 if (pfn == head_pfn + 1)
1049 prep_compound_head(head, order);
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)
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;
1066 if (WARN_ON_ONCE(!pgmap || zone_idx != ZONE_DEVICE))
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.
1075 start_pfn = altmap->base_pfn + vmem_altmap_offset(altmap);
1076 nr_pages = end_pfn - start_pfn;
1079 for (pfn = start_pfn; pfn < end_pfn; pfn += pfns_per_compound) {
1080 struct page *page = pfn_to_page(pfn);
1082 __init_zone_device_page(page, pfn, zone_idx, nid, pgmap);
1084 if (pfns_per_compound == 1)
1087 memmap_init_compound(page, pfn, zone_idx, nid, pgmap,
1088 compound_nr_pages(altmap, pgmap));
1091 pr_debug("%s initialised %lu pages in %ums\n", __func__,
1092 nr_pages, jiffies_to_msecs(jiffies - start));
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
1106 static void __init adjust_zone_range_for_zone_movable(int nid,
1107 unsigned long zone_type,
1108 unsigned long node_start_pfn,
1109 unsigned long node_end_pfn,
1110 unsigned long *zone_start_pfn,
1111 unsigned long *zone_end_pfn)
1113 /* Only adjust if ZONE_MOVABLE is on this node */
1114 if (zone_movable_pfn[nid]) {
1115 /* Size ZONE_MOVABLE */
1116 if (zone_type == ZONE_MOVABLE) {
1117 *zone_start_pfn = zone_movable_pfn[nid];
1118 *zone_end_pfn = min(node_end_pfn,
1119 arch_zone_highest_possible_pfn[movable_zone]);
1121 /* Adjust for ZONE_MOVABLE starting within this range */
1122 } else if (!mirrored_kernelcore &&
1123 *zone_start_pfn < zone_movable_pfn[nid] &&
1124 *zone_end_pfn > zone_movable_pfn[nid]) {
1125 *zone_end_pfn = zone_movable_pfn[nid];
1127 /* Check if this whole range is within ZONE_MOVABLE */
1128 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
1129 *zone_start_pfn = *zone_end_pfn;
1134 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
1135 * then all holes in the requested range will be accounted for.
1137 unsigned long __init __absent_pages_in_range(int nid,
1138 unsigned long range_start_pfn,
1139 unsigned long range_end_pfn)
1141 unsigned long nr_absent = range_end_pfn - range_start_pfn;
1142 unsigned long start_pfn, end_pfn;
1145 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
1146 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
1147 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
1148 nr_absent -= end_pfn - start_pfn;
1154 * absent_pages_in_range - Return number of page frames in holes within a range
1155 * @start_pfn: The start PFN to start searching for holes
1156 * @end_pfn: The end PFN to stop searching for holes
1158 * Return: the number of pages frames in memory holes within a range.
1160 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
1161 unsigned long end_pfn)
1163 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
1166 /* Return the number of page frames in holes in a zone on a node */
1167 static unsigned long __init zone_absent_pages_in_node(int nid,
1168 unsigned long zone_type,
1169 unsigned long node_start_pfn,
1170 unsigned long node_end_pfn)
1172 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
1173 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
1174 unsigned long zone_start_pfn, zone_end_pfn;
1175 unsigned long nr_absent;
1177 /* When hotadd a new node from cpu_up(), the node should be empty */
1178 if (!node_start_pfn && !node_end_pfn)
1181 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
1182 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
1184 adjust_zone_range_for_zone_movable(nid, zone_type,
1185 node_start_pfn, node_end_pfn,
1186 &zone_start_pfn, &zone_end_pfn);
1187 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
1190 * ZONE_MOVABLE handling.
1191 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
1194 if (mirrored_kernelcore && zone_movable_pfn[nid]) {
1195 unsigned long start_pfn, end_pfn;
1196 struct memblock_region *r;
1198 for_each_mem_region(r) {
1199 start_pfn = clamp(memblock_region_memory_base_pfn(r),
1200 zone_start_pfn, zone_end_pfn);
1201 end_pfn = clamp(memblock_region_memory_end_pfn(r),
1202 zone_start_pfn, zone_end_pfn);
1204 if (zone_type == ZONE_MOVABLE &&
1205 memblock_is_mirror(r))
1206 nr_absent += end_pfn - start_pfn;
1208 if (zone_type == ZONE_NORMAL &&
1209 !memblock_is_mirror(r))
1210 nr_absent += end_pfn - start_pfn;
1218 * Return the number of pages a zone spans in a node, including holes
1219 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
1221 static unsigned long __init zone_spanned_pages_in_node(int nid,
1222 unsigned long zone_type,
1223 unsigned long node_start_pfn,
1224 unsigned long node_end_pfn,
1225 unsigned long *zone_start_pfn,
1226 unsigned long *zone_end_pfn)
1228 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
1229 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
1230 /* When hotadd a new node from cpu_up(), the node should be empty */
1231 if (!node_start_pfn && !node_end_pfn)
1234 /* Get the start and end of the zone */
1235 *zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
1236 *zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
1237 adjust_zone_range_for_zone_movable(nid, zone_type,
1238 node_start_pfn, node_end_pfn,
1239 zone_start_pfn, zone_end_pfn);
1241 /* Check that this node has pages within the zone's required range */
1242 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
1245 /* Move the zone boundaries inside the node if necessary */
1246 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
1247 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
1249 /* Return the spanned pages */
1250 return *zone_end_pfn - *zone_start_pfn;
1253 static void __init calculate_node_totalpages(struct pglist_data *pgdat,
1254 unsigned long node_start_pfn,
1255 unsigned long node_end_pfn)
1257 unsigned long realtotalpages = 0, totalpages = 0;
1260 for (i = 0; i < MAX_NR_ZONES; i++) {
1261 struct zone *zone = pgdat->node_zones + i;
1262 unsigned long zone_start_pfn, zone_end_pfn;
1263 unsigned long spanned, absent;
1264 unsigned long size, real_size;
1266 spanned = zone_spanned_pages_in_node(pgdat->node_id, i,
1271 absent = zone_absent_pages_in_node(pgdat->node_id, i,
1276 real_size = size - absent;
1279 zone->zone_start_pfn = zone_start_pfn;
1281 zone->zone_start_pfn = 0;
1282 zone->spanned_pages = size;
1283 zone->present_pages = real_size;
1284 #if defined(CONFIG_MEMORY_HOTPLUG)
1285 zone->present_early_pages = real_size;
1289 realtotalpages += real_size;
1292 pgdat->node_spanned_pages = totalpages;
1293 pgdat->node_present_pages = realtotalpages;
1294 pr_debug("On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages);
1297 static unsigned long __init calc_memmap_size(unsigned long spanned_pages,
1298 unsigned long present_pages)
1300 unsigned long pages = spanned_pages;
1303 * Provide a more accurate estimation if there are holes within
1304 * the zone and SPARSEMEM is in use. If there are holes within the
1305 * zone, each populated memory region may cost us one or two extra
1306 * memmap pages due to alignment because memmap pages for each
1307 * populated regions may not be naturally aligned on page boundary.
1308 * So the (present_pages >> 4) heuristic is a tradeoff for that.
1310 if (spanned_pages > present_pages + (present_pages >> 4) &&
1311 IS_ENABLED(CONFIG_SPARSEMEM))
1312 pages = present_pages;
1314 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
1317 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1318 static void pgdat_init_split_queue(struct pglist_data *pgdat)
1320 struct deferred_split *ds_queue = &pgdat->deferred_split_queue;
1322 spin_lock_init(&ds_queue->split_queue_lock);
1323 INIT_LIST_HEAD(&ds_queue->split_queue);
1324 ds_queue->split_queue_len = 0;
1327 static void pgdat_init_split_queue(struct pglist_data *pgdat) {}
1330 #ifdef CONFIG_COMPACTION
1331 static void pgdat_init_kcompactd(struct pglist_data *pgdat)
1333 init_waitqueue_head(&pgdat->kcompactd_wait);
1336 static void pgdat_init_kcompactd(struct pglist_data *pgdat) {}
1339 static void __meminit pgdat_init_internals(struct pglist_data *pgdat)
1343 pgdat_resize_init(pgdat);
1344 pgdat_kswapd_lock_init(pgdat);
1346 pgdat_init_split_queue(pgdat);
1347 pgdat_init_kcompactd(pgdat);
1349 init_waitqueue_head(&pgdat->kswapd_wait);
1350 init_waitqueue_head(&pgdat->pfmemalloc_wait);
1352 for (i = 0; i < NR_VMSCAN_THROTTLE; i++)
1353 init_waitqueue_head(&pgdat->reclaim_wait[i]);
1355 pgdat_page_ext_init(pgdat);
1356 lruvec_init(&pgdat->__lruvec);
1359 static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, int nid,
1360 unsigned long remaining_pages)
1362 atomic_long_set(&zone->managed_pages, remaining_pages);
1363 zone_set_nid(zone, nid);
1364 zone->name = zone_names[idx];
1365 zone->zone_pgdat = NODE_DATA(nid);
1366 spin_lock_init(&zone->lock);
1367 zone_seqlock_init(zone);
1368 zone_pcp_init(zone);
1371 static void __meminit zone_init_free_lists(struct zone *zone)
1373 unsigned int order, t;
1374 for_each_migratetype_order(order, t) {
1375 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
1376 zone->free_area[order].nr_free = 0;
1380 void __meminit init_currently_empty_zone(struct zone *zone,
1381 unsigned long zone_start_pfn,
1384 struct pglist_data *pgdat = zone->zone_pgdat;
1385 int zone_idx = zone_idx(zone) + 1;
1387 if (zone_idx > pgdat->nr_zones)
1388 pgdat->nr_zones = zone_idx;
1390 zone->zone_start_pfn = zone_start_pfn;
1392 mminit_dprintk(MMINIT_TRACE, "memmap_init",
1393 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
1395 (unsigned long)zone_idx(zone),
1396 zone_start_pfn, (zone_start_pfn + size));
1398 zone_init_free_lists(zone);
1399 zone->initialized = 1;
1402 #ifndef CONFIG_SPARSEMEM
1404 * Calculate the size of the zone->blockflags rounded to an unsigned long
1405 * Start by making sure zonesize is a multiple of pageblock_order by rounding
1406 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
1407 * round what is now in bits to nearest long in bits, then return it in
1410 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
1412 unsigned long usemapsize;
1414 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
1415 usemapsize = roundup(zonesize, pageblock_nr_pages);
1416 usemapsize = usemapsize >> pageblock_order;
1417 usemapsize *= NR_PAGEBLOCK_BITS;
1418 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
1420 return usemapsize / 8;
1423 static void __ref setup_usemap(struct zone *zone)
1425 unsigned long usemapsize = usemap_size(zone->zone_start_pfn,
1426 zone->spanned_pages);
1427 zone->pageblock_flags = NULL;
1429 zone->pageblock_flags =
1430 memblock_alloc_node(usemapsize, SMP_CACHE_BYTES,
1432 if (!zone->pageblock_flags)
1433 panic("Failed to allocate %ld bytes for zone %s pageblock flags on node %d\n",
1434 usemapsize, zone->name, zone_to_nid(zone));
1438 static inline void setup_usemap(struct zone *zone) {}
1439 #endif /* CONFIG_SPARSEMEM */
1441 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
1443 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
1444 void __init set_pageblock_order(void)
1446 unsigned int order = MAX_ORDER;
1448 /* Check that pageblock_nr_pages has not already been setup */
1449 if (pageblock_order)
1452 /* Don't let pageblocks exceed the maximum allocation granularity. */
1453 if (HPAGE_SHIFT > PAGE_SHIFT && HUGETLB_PAGE_ORDER < order)
1454 order = HUGETLB_PAGE_ORDER;
1457 * Assume the largest contiguous order of interest is a huge page.
1458 * This value may be variable depending on boot parameters on IA64 and
1461 pageblock_order = order;
1463 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
1466 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
1467 * is unused as pageblock_order is set at compile-time. See
1468 * include/linux/pageblock-flags.h for the values of pageblock_order based on
1471 void __init set_pageblock_order(void)
1475 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
1478 * Set up the zone data structures
1479 * - init pgdat internals
1480 * - init all zones belonging to this node
1482 * NOTE: this function is only called during memory hotplug
1484 #ifdef CONFIG_MEMORY_HOTPLUG
1485 void __ref free_area_init_core_hotplug(struct pglist_data *pgdat)
1487 int nid = pgdat->node_id;
1491 pgdat_init_internals(pgdat);
1493 if (pgdat->per_cpu_nodestats == &boot_nodestats)
1494 pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat);
1497 * Reset the nr_zones, order and highest_zoneidx before reuse.
1498 * Note that kswapd will init kswapd_highest_zoneidx properly
1499 * when it starts in the near future.
1501 pgdat->nr_zones = 0;
1502 pgdat->kswapd_order = 0;
1503 pgdat->kswapd_highest_zoneidx = 0;
1504 pgdat->node_start_pfn = 0;
1505 for_each_online_cpu(cpu) {
1506 struct per_cpu_nodestat *p;
1508 p = per_cpu_ptr(pgdat->per_cpu_nodestats, cpu);
1509 memset(p, 0, sizeof(*p));
1512 for (z = 0; z < MAX_NR_ZONES; z++)
1513 zone_init_internals(&pgdat->node_zones[z], z, nid, 0);
1518 * Set up the zone data structures:
1519 * - mark all pages reserved
1520 * - mark all memory queues empty
1521 * - clear the memory bitmaps
1523 * NOTE: pgdat should get zeroed by caller.
1524 * NOTE: this function is only called during early init.
1526 static void __init free_area_init_core(struct pglist_data *pgdat)
1529 int nid = pgdat->node_id;
1531 pgdat_init_internals(pgdat);
1532 pgdat->per_cpu_nodestats = &boot_nodestats;
1534 for (j = 0; j < MAX_NR_ZONES; j++) {
1535 struct zone *zone = pgdat->node_zones + j;
1536 unsigned long size, freesize, memmap_pages;
1538 size = zone->spanned_pages;
1539 freesize = zone->present_pages;
1542 * Adjust freesize so that it accounts for how much memory
1543 * is used by this zone for memmap. This affects the watermark
1544 * and per-cpu initialisations
1546 memmap_pages = calc_memmap_size(size, freesize);
1547 if (!is_highmem_idx(j)) {
1548 if (freesize >= memmap_pages) {
1549 freesize -= memmap_pages;
1551 pr_debug(" %s zone: %lu pages used for memmap\n",
1552 zone_names[j], memmap_pages);
1554 pr_warn(" %s zone: %lu memmap pages exceeds freesize %lu\n",
1555 zone_names[j], memmap_pages, freesize);
1558 /* Account for reserved pages */
1559 if (j == 0 && freesize > dma_reserve) {
1560 freesize -= dma_reserve;
1561 pr_debug(" %s zone: %lu pages reserved\n", zone_names[0], dma_reserve);
1564 if (!is_highmem_idx(j))
1565 nr_kernel_pages += freesize;
1566 /* Charge for highmem memmap if there are enough kernel pages */
1567 else if (nr_kernel_pages > memmap_pages * 2)
1568 nr_kernel_pages -= memmap_pages;
1569 nr_all_pages += freesize;
1572 * Set an approximate value for lowmem here, it will be adjusted
1573 * when the bootmem allocator frees pages into the buddy system.
1574 * And all highmem pages will be managed by the buddy system.
1576 zone_init_internals(zone, j, nid, freesize);
1581 set_pageblock_order();
1583 init_currently_empty_zone(zone, zone->zone_start_pfn, size);
1587 void __init *memmap_alloc(phys_addr_t size, phys_addr_t align,
1588 phys_addr_t min_addr, int nid, bool exact_nid)
1593 ptr = memblock_alloc_exact_nid_raw(size, align, min_addr,
1594 MEMBLOCK_ALLOC_ACCESSIBLE,
1597 ptr = memblock_alloc_try_nid_raw(size, align, min_addr,
1598 MEMBLOCK_ALLOC_ACCESSIBLE,
1601 if (ptr && size > 0)
1602 page_init_poison(ptr, size);
1607 #ifdef CONFIG_FLATMEM
1608 static void __init alloc_node_mem_map(struct pglist_data *pgdat)
1610 unsigned long __maybe_unused start = 0;
1611 unsigned long __maybe_unused offset = 0;
1613 /* Skip empty nodes */
1614 if (!pgdat->node_spanned_pages)
1617 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
1618 offset = pgdat->node_start_pfn - start;
1619 /* ia64 gets its own node_mem_map, before this, without bootmem */
1620 if (!pgdat->node_mem_map) {
1621 unsigned long size, end;
1625 * The zone's endpoints aren't required to be MAX_ORDER
1626 * aligned but the node_mem_map endpoints must be in order
1627 * for the buddy allocator to function correctly.
1629 end = pgdat_end_pfn(pgdat);
1630 end = ALIGN(end, MAX_ORDER_NR_PAGES);
1631 size = (end - start) * sizeof(struct page);
1632 map = memmap_alloc(size, SMP_CACHE_BYTES, MEMBLOCK_LOW_LIMIT,
1633 pgdat->node_id, false);
1635 panic("Failed to allocate %ld bytes for node %d memory map\n",
1636 size, pgdat->node_id);
1637 pgdat->node_mem_map = map + offset;
1639 pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
1640 __func__, pgdat->node_id, (unsigned long)pgdat,
1641 (unsigned long)pgdat->node_mem_map);
1644 * With no DISCONTIG, the global mem_map is just set as node 0's
1646 if (pgdat == NODE_DATA(0)) {
1647 mem_map = NODE_DATA(0)->node_mem_map;
1648 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
1654 static inline void alloc_node_mem_map(struct pglist_data *pgdat) { }
1655 #endif /* CONFIG_FLATMEM */
1658 * get_pfn_range_for_nid - Return the start and end page frames for a node
1659 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
1660 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
1661 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
1663 * It returns the start and end page frame of a node based on information
1664 * provided by memblock_set_node(). If called for a node
1665 * with no available memory, a warning is printed and the start and end
1668 void __init get_pfn_range_for_nid(unsigned int nid,
1669 unsigned long *start_pfn, unsigned long *end_pfn)
1671 unsigned long this_start_pfn, this_end_pfn;
1677 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
1678 *start_pfn = min(*start_pfn, this_start_pfn);
1679 *end_pfn = max(*end_pfn, this_end_pfn);
1682 if (*start_pfn == -1UL)
1686 static void __init free_area_init_node(int nid)
1688 pg_data_t *pgdat = NODE_DATA(nid);
1689 unsigned long start_pfn = 0;
1690 unsigned long end_pfn = 0;
1692 /* pg_data_t should be reset to zero when it's allocated */
1693 WARN_ON(pgdat->nr_zones || pgdat->kswapd_highest_zoneidx);
1695 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
1697 pgdat->node_id = nid;
1698 pgdat->node_start_pfn = start_pfn;
1699 pgdat->per_cpu_nodestats = NULL;
1701 if (start_pfn != end_pfn) {
1702 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
1703 (u64)start_pfn << PAGE_SHIFT,
1704 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
1706 pr_info("Initmem setup node %d as memoryless\n", nid);
1709 calculate_node_totalpages(pgdat, start_pfn, end_pfn);
1711 alloc_node_mem_map(pgdat);
1712 pgdat_set_deferred_range(pgdat);
1714 free_area_init_core(pgdat);
1715 lru_gen_init_pgdat(pgdat);
1718 /* Any regular or high memory on that node ? */
1719 static void check_for_memory(pg_data_t *pgdat, int nid)
1721 enum zone_type zone_type;
1723 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
1724 struct zone *zone = &pgdat->node_zones[zone_type];
1725 if (populated_zone(zone)) {
1726 if (IS_ENABLED(CONFIG_HIGHMEM))
1727 node_set_state(nid, N_HIGH_MEMORY);
1728 if (zone_type <= ZONE_NORMAL)
1729 node_set_state(nid, N_NORMAL_MEMORY);
1735 #if MAX_NUMNODES > 1
1737 * Figure out the number of possible node ids.
1739 void __init setup_nr_node_ids(void)
1741 unsigned int highest;
1743 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
1744 nr_node_ids = highest + 1;
1748 static void __init free_area_init_memoryless_node(int nid)
1750 free_area_init_node(nid);
1754 * Some architectures, e.g. ARC may have ZONE_HIGHMEM below ZONE_NORMAL. For
1755 * such cases we allow max_zone_pfn sorted in the descending order
1757 static bool arch_has_descending_max_zone_pfns(void)
1759 return IS_ENABLED(CONFIG_ARC) && !IS_ENABLED(CONFIG_ARC_HAS_PAE40);
1763 * free_area_init - Initialise all pg_data_t and zone data
1764 * @max_zone_pfn: an array of max PFNs for each zone
1766 * This will call free_area_init_node() for each active node in the system.
1767 * Using the page ranges provided by memblock_set_node(), the size of each
1768 * zone in each node and their holes is calculated. If the maximum PFN
1769 * between two adjacent zones match, it is assumed that the zone is empty.
1770 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
1771 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
1772 * starts where the previous one ended. For example, ZONE_DMA32 starts
1773 * at arch_max_dma_pfn.
1775 void __init free_area_init(unsigned long *max_zone_pfn)
1777 unsigned long start_pfn, end_pfn;
1781 /* Record where the zone boundaries are */
1782 memset(arch_zone_lowest_possible_pfn, 0,
1783 sizeof(arch_zone_lowest_possible_pfn));
1784 memset(arch_zone_highest_possible_pfn, 0,
1785 sizeof(arch_zone_highest_possible_pfn));
1787 start_pfn = PHYS_PFN(memblock_start_of_DRAM());
1788 descending = arch_has_descending_max_zone_pfns();
1790 for (i = 0; i < MAX_NR_ZONES; i++) {
1792 zone = MAX_NR_ZONES - i - 1;
1796 if (zone == ZONE_MOVABLE)
1799 end_pfn = max(max_zone_pfn[zone], start_pfn);
1800 arch_zone_lowest_possible_pfn[zone] = start_pfn;
1801 arch_zone_highest_possible_pfn[zone] = end_pfn;
1803 start_pfn = end_pfn;
1806 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
1807 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
1808 find_zone_movable_pfns_for_nodes();
1810 /* Print out the zone ranges */
1811 pr_info("Zone ranges:\n");
1812 for (i = 0; i < MAX_NR_ZONES; i++) {
1813 if (i == ZONE_MOVABLE)
1815 pr_info(" %-8s ", zone_names[i]);
1816 if (arch_zone_lowest_possible_pfn[i] ==
1817 arch_zone_highest_possible_pfn[i])
1820 pr_cont("[mem %#018Lx-%#018Lx]\n",
1821 (u64)arch_zone_lowest_possible_pfn[i]
1823 ((u64)arch_zone_highest_possible_pfn[i]
1824 << PAGE_SHIFT) - 1);
1827 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
1828 pr_info("Movable zone start for each node\n");
1829 for (i = 0; i < MAX_NUMNODES; i++) {
1830 if (zone_movable_pfn[i])
1831 pr_info(" Node %d: %#018Lx\n", i,
1832 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
1836 * Print out the early node map, and initialize the
1837 * subsection-map relative to active online memory ranges to
1838 * enable future "sub-section" extensions of the memory map.
1840 pr_info("Early memory node ranges\n");
1841 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
1842 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
1843 (u64)start_pfn << PAGE_SHIFT,
1844 ((u64)end_pfn << PAGE_SHIFT) - 1);
1845 subsection_map_init(start_pfn, end_pfn - start_pfn);
1848 /* Initialise every node */
1849 mminit_verify_pageflags_layout();
1850 setup_nr_node_ids();
1851 for_each_node(nid) {
1854 if (!node_online(nid)) {
1855 pr_info("Initializing node %d as memoryless\n", nid);
1857 /* Allocator not initialized yet */
1858 pgdat = arch_alloc_nodedata(nid);
1860 panic("Cannot allocate %zuB for node %d.\n",
1861 sizeof(*pgdat), nid);
1862 arch_refresh_nodedata(nid, pgdat);
1863 free_area_init_memoryless_node(nid);
1866 * We do not want to confuse userspace by sysfs
1867 * files/directories for node without any memory
1868 * attached to it, so this node is not marked as
1869 * N_MEMORY and not marked online so that no sysfs
1870 * hierarchy will be created via register_one_node for
1871 * it. The pgdat will get fully initialized by
1872 * hotadd_init_pgdat() when memory is hotplugged into
1878 pgdat = NODE_DATA(nid);
1879 free_area_init_node(nid);
1881 /* Any memory on that node */
1882 if (pgdat->node_present_pages)
1883 node_set_state(nid, N_MEMORY);
1884 check_for_memory(pgdat, nid);
1889 /* disable hash distribution for systems with a single node */
1894 * node_map_pfn_alignment - determine the maximum internode alignment
1896 * This function should be called after node map is populated and sorted.
1897 * It calculates the maximum power of two alignment which can distinguish
1900 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
1901 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
1902 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
1903 * shifted, 1GiB is enough and this function will indicate so.
1905 * This is used to test whether pfn -> nid mapping of the chosen memory
1906 * model has fine enough granularity to avoid incorrect mapping for the
1907 * populated node map.
1909 * Return: the determined alignment in pfn's. 0 if there is no alignment
1910 * requirement (single node).
1912 unsigned long __init node_map_pfn_alignment(void)
1914 unsigned long accl_mask = 0, last_end = 0;
1915 unsigned long start, end, mask;
1916 int last_nid = NUMA_NO_NODE;
1919 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
1920 if (!start || last_nid < 0 || last_nid == nid) {
1927 * Start with a mask granular enough to pin-point to the
1928 * start pfn and tick off bits one-by-one until it becomes
1929 * too coarse to separate the current node from the last.
1931 mask = ~((1 << __ffs(start)) - 1);
1932 while (mask && last_end <= (start & (mask << 1)))
1935 /* accumulate all internode masks */
1939 /* convert mask to number of pages */
1940 return ~accl_mask + 1;
1943 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1944 static void __init deferred_free_range(unsigned long pfn,
1945 unsigned long nr_pages)
1953 page = pfn_to_page(pfn);
1955 /* Free a large naturally-aligned chunk if possible */
1956 if (nr_pages == MAX_ORDER_NR_PAGES && IS_MAX_ORDER_ALIGNED(pfn)) {
1957 for (i = 0; i < nr_pages; i += pageblock_nr_pages)
1958 set_pageblock_migratetype(page + i, MIGRATE_MOVABLE);
1959 __free_pages_core(page, MAX_ORDER);
1963 for (i = 0; i < nr_pages; i++, page++, pfn++) {
1964 if (pageblock_aligned(pfn))
1965 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1966 __free_pages_core(page, 0);
1970 /* Completion tracking for deferred_init_memmap() threads */
1971 static atomic_t pgdat_init_n_undone __initdata;
1972 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1974 static inline void __init pgdat_init_report_one_done(void)
1976 if (atomic_dec_and_test(&pgdat_init_n_undone))
1977 complete(&pgdat_init_all_done_comp);
1981 * Returns true if page needs to be initialized or freed to buddy allocator.
1983 * We check if a current MAX_ORDER block is valid by only checking the validity
1986 static inline bool __init deferred_pfn_valid(unsigned long pfn)
1988 if (IS_MAX_ORDER_ALIGNED(pfn) && !pfn_valid(pfn))
1994 * Free pages to buddy allocator. Try to free aligned pages in
1995 * MAX_ORDER_NR_PAGES sizes.
1997 static void __init deferred_free_pages(unsigned long pfn,
1998 unsigned long end_pfn)
2000 unsigned long nr_free = 0;
2002 for (; pfn < end_pfn; pfn++) {
2003 if (!deferred_pfn_valid(pfn)) {
2004 deferred_free_range(pfn - nr_free, nr_free);
2006 } else if (IS_MAX_ORDER_ALIGNED(pfn)) {
2007 deferred_free_range(pfn - nr_free, nr_free);
2013 /* Free the last block of pages to allocator */
2014 deferred_free_range(pfn - nr_free, nr_free);
2018 * Initialize struct pages. We minimize pfn page lookups and scheduler checks
2019 * by performing it only once every MAX_ORDER_NR_PAGES.
2020 * Return number of pages initialized.
2022 static unsigned long __init deferred_init_pages(struct zone *zone,
2024 unsigned long end_pfn)
2026 int nid = zone_to_nid(zone);
2027 unsigned long nr_pages = 0;
2028 int zid = zone_idx(zone);
2029 struct page *page = NULL;
2031 for (; pfn < end_pfn; pfn++) {
2032 if (!deferred_pfn_valid(pfn)) {
2035 } else if (!page || IS_MAX_ORDER_ALIGNED(pfn)) {
2036 page = pfn_to_page(pfn);
2040 __init_single_page(page, pfn, zid, nid);
2047 * This function is meant to pre-load the iterator for the zone init.
2048 * Specifically it walks through the ranges until we are caught up to the
2049 * first_init_pfn value and exits there. If we never encounter the value we
2050 * return false indicating there are no valid ranges left.
2053 deferred_init_mem_pfn_range_in_zone(u64 *i, struct zone *zone,
2054 unsigned long *spfn, unsigned long *epfn,
2055 unsigned long first_init_pfn)
2060 * Start out by walking through the ranges in this zone that have
2061 * already been initialized. We don't need to do anything with them
2062 * so we just need to flush them out of the system.
2064 for_each_free_mem_pfn_range_in_zone(j, zone, spfn, epfn) {
2065 if (*epfn <= first_init_pfn)
2067 if (*spfn < first_init_pfn)
2068 *spfn = first_init_pfn;
2077 * Initialize and free pages. We do it in two loops: first we initialize
2078 * struct page, then free to buddy allocator, because while we are
2079 * freeing pages we can access pages that are ahead (computing buddy
2080 * page in __free_one_page()).
2082 * In order to try and keep some memory in the cache we have the loop
2083 * broken along max page order boundaries. This way we will not cause
2084 * any issues with the buddy page computation.
2086 static unsigned long __init
2087 deferred_init_maxorder(u64 *i, struct zone *zone, unsigned long *start_pfn,
2088 unsigned long *end_pfn)
2090 unsigned long mo_pfn = ALIGN(*start_pfn + 1, MAX_ORDER_NR_PAGES);
2091 unsigned long spfn = *start_pfn, epfn = *end_pfn;
2092 unsigned long nr_pages = 0;
2095 /* First we loop through and initialize the page values */
2096 for_each_free_mem_pfn_range_in_zone_from(j, zone, start_pfn, end_pfn) {
2099 if (mo_pfn <= *start_pfn)
2102 t = min(mo_pfn, *end_pfn);
2103 nr_pages += deferred_init_pages(zone, *start_pfn, t);
2105 if (mo_pfn < *end_pfn) {
2106 *start_pfn = mo_pfn;
2111 /* Reset values and now loop through freeing pages as needed */
2114 for_each_free_mem_pfn_range_in_zone_from(j, zone, &spfn, &epfn) {
2120 t = min(mo_pfn, epfn);
2121 deferred_free_pages(spfn, t);
2131 deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
2134 unsigned long spfn, epfn;
2135 struct zone *zone = arg;
2138 deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn, start_pfn);
2141 * Initialize and free pages in MAX_ORDER sized increments so that we
2142 * can avoid introducing any issues with the buddy allocator.
2144 while (spfn < end_pfn) {
2145 deferred_init_maxorder(&i, zone, &spfn, &epfn);
2150 /* An arch may override for more concurrency. */
2152 deferred_page_init_max_threads(const struct cpumask *node_cpumask)
2157 /* Initialise remaining memory on a node */
2158 static int __init deferred_init_memmap(void *data)
2160 pg_data_t *pgdat = data;
2161 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
2162 unsigned long spfn = 0, epfn = 0;
2163 unsigned long first_init_pfn, flags;
2164 unsigned long start = jiffies;
2166 int zid, max_threads;
2169 /* Bind memory initialisation thread to a local node if possible */
2170 if (!cpumask_empty(cpumask))
2171 set_cpus_allowed_ptr(current, cpumask);
2173 pgdat_resize_lock(pgdat, &flags);
2174 first_init_pfn = pgdat->first_deferred_pfn;
2175 if (first_init_pfn == ULONG_MAX) {
2176 pgdat_resize_unlock(pgdat, &flags);
2177 pgdat_init_report_one_done();
2181 /* Sanity check boundaries */
2182 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
2183 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
2184 pgdat->first_deferred_pfn = ULONG_MAX;
2187 * Once we unlock here, the zone cannot be grown anymore, thus if an
2188 * interrupt thread must allocate this early in boot, zone must be
2189 * pre-grown prior to start of deferred page initialization.
2191 pgdat_resize_unlock(pgdat, &flags);
2193 /* Only the highest zone is deferred so find it */
2194 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
2195 zone = pgdat->node_zones + zid;
2196 if (first_init_pfn < zone_end_pfn(zone))
2200 /* If the zone is empty somebody else may have cleared out the zone */
2201 if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2205 max_threads = deferred_page_init_max_threads(cpumask);
2207 while (spfn < epfn) {
2208 unsigned long epfn_align = ALIGN(epfn, PAGES_PER_SECTION);
2209 struct padata_mt_job job = {
2210 .thread_fn = deferred_init_memmap_chunk,
2213 .size = epfn_align - spfn,
2214 .align = PAGES_PER_SECTION,
2215 .min_chunk = PAGES_PER_SECTION,
2216 .max_threads = max_threads,
2219 padata_do_multithreaded(&job);
2220 deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2224 /* Sanity check that the next zone really is unpopulated */
2225 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
2227 pr_info("node %d deferred pages initialised in %ums\n",
2228 pgdat->node_id, jiffies_to_msecs(jiffies - start));
2230 pgdat_init_report_one_done();
2235 * If this zone has deferred pages, try to grow it by initializing enough
2236 * deferred pages to satisfy the allocation specified by order, rounded up to
2237 * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments
2238 * of SECTION_SIZE bytes by initializing struct pages in increments of
2239 * PAGES_PER_SECTION * sizeof(struct page) bytes.
2241 * Return true when zone was grown, otherwise return false. We return true even
2242 * when we grow less than requested, to let the caller decide if there are
2243 * enough pages to satisfy the allocation.
2245 * Note: We use noinline because this function is needed only during boot, and
2246 * it is called from a __ref function _deferred_grow_zone. This way we are
2247 * making sure that it is not inlined into permanent text section.
2249 bool __init deferred_grow_zone(struct zone *zone, unsigned int order)
2251 unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
2252 pg_data_t *pgdat = zone->zone_pgdat;
2253 unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
2254 unsigned long spfn, epfn, flags;
2255 unsigned long nr_pages = 0;
2258 /* Only the last zone may have deferred pages */
2259 if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
2262 pgdat_resize_lock(pgdat, &flags);
2265 * If someone grew this zone while we were waiting for spinlock, return
2266 * true, as there might be enough pages already.
2268 if (first_deferred_pfn != pgdat->first_deferred_pfn) {
2269 pgdat_resize_unlock(pgdat, &flags);
2273 /* If the zone is empty somebody else may have cleared out the zone */
2274 if (!deferred_init_mem_pfn_range_in_zone(&i, zone, &spfn, &epfn,
2275 first_deferred_pfn)) {
2276 pgdat->first_deferred_pfn = ULONG_MAX;
2277 pgdat_resize_unlock(pgdat, &flags);
2278 /* Retry only once. */
2279 return first_deferred_pfn != ULONG_MAX;
2283 * Initialize and free pages in MAX_ORDER sized increments so
2284 * that we can avoid introducing any issues with the buddy
2287 while (spfn < epfn) {
2288 /* update our first deferred PFN for this section */
2289 first_deferred_pfn = spfn;
2291 nr_pages += deferred_init_maxorder(&i, zone, &spfn, &epfn);
2292 touch_nmi_watchdog();
2294 /* We should only stop along section boundaries */
2295 if ((first_deferred_pfn ^ spfn) < PAGES_PER_SECTION)
2298 /* If our quota has been met we can stop here */
2299 if (nr_pages >= nr_pages_needed)
2303 pgdat->first_deferred_pfn = spfn;
2304 pgdat_resize_unlock(pgdat, &flags);
2306 return nr_pages > 0;
2309 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
2312 void __init init_cma_reserved_pageblock(struct page *page)
2314 unsigned i = pageblock_nr_pages;
2315 struct page *p = page;
2318 __ClearPageReserved(p);
2319 set_page_count(p, 0);
2322 set_pageblock_migratetype(page, MIGRATE_CMA);
2323 set_page_refcounted(page);
2324 __free_pages(page, pageblock_order);
2326 adjust_managed_page_count(page, pageblock_nr_pages);
2327 page_zone(page)->cma_pages += pageblock_nr_pages;
2331 void __init page_alloc_init_late(void)
2336 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
2338 /* There will be num_node_state(N_MEMORY) threads */
2339 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
2340 for_each_node_state(nid, N_MEMORY) {
2341 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
2344 /* Block until all are initialised */
2345 wait_for_completion(&pgdat_init_all_done_comp);
2348 * We initialized the rest of the deferred pages. Permanently disable
2349 * on-demand struct page initialization.
2351 static_branch_disable(&deferred_pages);
2353 /* Reinit limits that are based on free pages after the kernel is up */
2354 files_maxfiles_init();
2359 /* Discard memblock private memory */
2362 for_each_node_state(nid, N_MEMORY)
2363 shuffle_free_memory(NODE_DATA(nid));
2365 for_each_populated_zone(zone)
2366 set_zone_contiguous(zone);
2368 /* Initialize page ext after all struct pages are initialized. */
2369 if (deferred_struct_pages)
2373 #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
2375 * Returns the number of pages that arch has reserved but
2376 * is not known to alloc_large_system_hash().
2378 static unsigned long __init arch_reserved_kernel_pages(void)
2385 * Adaptive scale is meant to reduce sizes of hash tables on large memory
2386 * machines. As memory size is increased the scale is also increased but at
2387 * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory
2388 * quadruples the scale is increased by one, which means the size of hash table
2389 * only doubles, instead of quadrupling as well.
2390 * Because 32-bit systems cannot have large physical memory, where this scaling
2391 * makes sense, it is disabled on such platforms.
2393 #if __BITS_PER_LONG > 32
2394 #define ADAPT_SCALE_BASE (64ul << 30)
2395 #define ADAPT_SCALE_SHIFT 2
2396 #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT)
2400 * allocate a large system hash table from bootmem
2401 * - it is assumed that the hash table must contain an exact power-of-2
2402 * quantity of entries
2403 * - limit is the number of hash buckets, not the total allocation size
2405 void *__init alloc_large_system_hash(const char *tablename,
2406 unsigned long bucketsize,
2407 unsigned long numentries,
2410 unsigned int *_hash_shift,
2411 unsigned int *_hash_mask,
2412 unsigned long low_limit,
2413 unsigned long high_limit)
2415 unsigned long long max = high_limit;
2416 unsigned long log2qty, size;
2422 /* allow the kernel cmdline to have a say */
2424 /* round applicable memory size up to nearest megabyte */
2425 numentries = nr_kernel_pages;
2426 numentries -= arch_reserved_kernel_pages();
2428 /* It isn't necessary when PAGE_SIZE >= 1MB */
2429 if (PAGE_SIZE < SZ_1M)
2430 numentries = round_up(numentries, SZ_1M / PAGE_SIZE);
2432 #if __BITS_PER_LONG > 32
2434 unsigned long adapt;
2436 for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
2437 adapt <<= ADAPT_SCALE_SHIFT)
2442 /* limit to 1 bucket per 2^scale bytes of low memory */
2443 if (scale > PAGE_SHIFT)
2444 numentries >>= (scale - PAGE_SHIFT);
2446 numentries <<= (PAGE_SHIFT - scale);
2448 /* Make sure we've got at least a 0-order allocation.. */
2449 if (unlikely(flags & HASH_SMALL)) {
2450 /* Makes no sense without HASH_EARLY */
2451 WARN_ON(!(flags & HASH_EARLY));
2452 if (!(numentries >> *_hash_shift)) {
2453 numentries = 1UL << *_hash_shift;
2454 BUG_ON(!numentries);
2456 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
2457 numentries = PAGE_SIZE / bucketsize;
2459 numentries = roundup_pow_of_two(numentries);
2461 /* limit allocation size to 1/16 total memory by default */
2463 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
2464 do_div(max, bucketsize);
2466 max = min(max, 0x80000000ULL);
2468 if (numentries < low_limit)
2469 numentries = low_limit;
2470 if (numentries > max)
2473 log2qty = ilog2(numentries);
2475 gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
2478 size = bucketsize << log2qty;
2479 if (flags & HASH_EARLY) {
2480 if (flags & HASH_ZERO)
2481 table = memblock_alloc(size, SMP_CACHE_BYTES);
2483 table = memblock_alloc_raw(size,
2485 } else if (get_order(size) > MAX_ORDER || hashdist) {
2486 table = vmalloc_huge(size, gfp_flags);
2489 huge = is_vm_area_hugepages(table);
2492 * If bucketsize is not a power-of-two, we may free
2493 * some pages at the end of hash table which
2494 * alloc_pages_exact() automatically does
2496 table = alloc_pages_exact(size, gfp_flags);
2497 kmemleak_alloc(table, size, 1, gfp_flags);
2499 } while (!table && size > PAGE_SIZE && --log2qty);
2502 panic("Failed to allocate %s hash table\n", tablename);
2504 pr_info("%s hash table entries: %ld (order: %d, %lu bytes, %s)\n",
2505 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size,
2506 virt ? (huge ? "vmalloc hugepage" : "vmalloc") : "linear");
2509 *_hash_shift = log2qty;
2511 *_hash_mask = (1 << log2qty) - 1;
2517 * set_dma_reserve - set the specified number of pages reserved in the first zone
2518 * @new_dma_reserve: The number of pages to mark reserved
2520 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
2521 * In the DMA zone, a significant percentage may be consumed by kernel image
2522 * and other unfreeable allocations which can skew the watermarks badly. This
2523 * function may optionally be used to account for unfreeable pages in the
2524 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
2525 * smaller per-cpu batchsize.
2527 void __init set_dma_reserve(unsigned long new_dma_reserve)
2529 dma_reserve = new_dma_reserve;
2532 void __init memblock_free_pages(struct page *page, unsigned long pfn,
2535 if (!early_page_initialised(pfn))
2537 if (!kmsan_memblock_free_pages(page, order)) {
2538 /* KMSAN will take care of these pages. */
2541 __free_pages_core(page, order);
2544 static bool _init_on_alloc_enabled_early __read_mostly
2545 = IS_ENABLED(CONFIG_INIT_ON_ALLOC_DEFAULT_ON);
2546 static int __init early_init_on_alloc(char *buf)
2549 return kstrtobool(buf, &_init_on_alloc_enabled_early);
2551 early_param("init_on_alloc", early_init_on_alloc);
2553 static bool _init_on_free_enabled_early __read_mostly
2554 = IS_ENABLED(CONFIG_INIT_ON_FREE_DEFAULT_ON);
2555 static int __init early_init_on_free(char *buf)
2557 return kstrtobool(buf, &_init_on_free_enabled_early);
2559 early_param("init_on_free", early_init_on_free);
2561 DEFINE_STATIC_KEY_MAYBE(CONFIG_DEBUG_VM, check_pages_enabled);
2564 * Enable static keys related to various memory debugging and hardening options.
2565 * Some override others, and depend on early params that are evaluated in the
2566 * order of appearance. So we need to first gather the full picture of what was
2567 * enabled, and then make decisions.
2569 static void __init mem_debugging_and_hardening_init(void)
2571 bool page_poisoning_requested = false;
2572 bool want_check_pages = false;
2574 #ifdef CONFIG_PAGE_POISONING
2576 * Page poisoning is debug page alloc for some arches. If
2577 * either of those options are enabled, enable poisoning.
2579 if (page_poisoning_enabled() ||
2580 (!IS_ENABLED(CONFIG_ARCH_SUPPORTS_DEBUG_PAGEALLOC) &&
2581 debug_pagealloc_enabled())) {
2582 static_branch_enable(&_page_poisoning_enabled);
2583 page_poisoning_requested = true;
2584 want_check_pages = true;
2588 if ((_init_on_alloc_enabled_early || _init_on_free_enabled_early) &&
2589 page_poisoning_requested) {
2590 pr_info("mem auto-init: CONFIG_PAGE_POISONING is on, "
2591 "will take precedence over init_on_alloc and init_on_free\n");
2592 _init_on_alloc_enabled_early = false;
2593 _init_on_free_enabled_early = false;
2596 if (_init_on_alloc_enabled_early) {
2597 want_check_pages = true;
2598 static_branch_enable(&init_on_alloc);
2600 static_branch_disable(&init_on_alloc);
2603 if (_init_on_free_enabled_early) {
2604 want_check_pages = true;
2605 static_branch_enable(&init_on_free);
2607 static_branch_disable(&init_on_free);
2610 if (IS_ENABLED(CONFIG_KMSAN) &&
2611 (_init_on_alloc_enabled_early || _init_on_free_enabled_early))
2612 pr_info("mem auto-init: please make sure init_on_alloc and init_on_free are disabled when running KMSAN\n");
2614 #ifdef CONFIG_DEBUG_PAGEALLOC
2615 if (debug_pagealloc_enabled()) {
2616 want_check_pages = true;
2617 static_branch_enable(&_debug_pagealloc_enabled);
2619 if (debug_guardpage_minorder())
2620 static_branch_enable(&_debug_guardpage_enabled);
2625 * Any page debugging or hardening option also enables sanity checking
2626 * of struct pages being allocated or freed. With CONFIG_DEBUG_VM it's
2629 if (!IS_ENABLED(CONFIG_DEBUG_VM) && want_check_pages)
2630 static_branch_enable(&check_pages_enabled);
2633 /* Report memory auto-initialization states for this boot. */
2634 static void __init report_meminit(void)
2638 if (IS_ENABLED(CONFIG_INIT_STACK_ALL_PATTERN))
2639 stack = "all(pattern)";
2640 else if (IS_ENABLED(CONFIG_INIT_STACK_ALL_ZERO))
2641 stack = "all(zero)";
2642 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF_ALL))
2643 stack = "byref_all(zero)";
2644 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF))
2645 stack = "byref(zero)";
2646 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_USER))
2647 stack = "__user(zero)";
2651 pr_info("mem auto-init: stack:%s, heap alloc:%s, heap free:%s\n",
2652 stack, want_init_on_alloc(GFP_KERNEL) ? "on" : "off",
2653 want_init_on_free() ? "on" : "off");
2654 if (want_init_on_free())
2655 pr_info("mem auto-init: clearing system memory may take some time...\n");
2658 static void __init mem_init_print_info(void)
2660 unsigned long physpages, codesize, datasize, rosize, bss_size;
2661 unsigned long init_code_size, init_data_size;
2663 physpages = get_num_physpages();
2664 codesize = _etext - _stext;
2665 datasize = _edata - _sdata;
2666 rosize = __end_rodata - __start_rodata;
2667 bss_size = __bss_stop - __bss_start;
2668 init_data_size = __init_end - __init_begin;
2669 init_code_size = _einittext - _sinittext;
2672 * Detect special cases and adjust section sizes accordingly:
2673 * 1) .init.* may be embedded into .data sections
2674 * 2) .init.text.* may be out of [__init_begin, __init_end],
2675 * please refer to arch/tile/kernel/vmlinux.lds.S.
2676 * 3) .rodata.* may be embedded into .text or .data sections.
2678 #define adj_init_size(start, end, size, pos, adj) \
2680 if (&start[0] <= &pos[0] && &pos[0] < &end[0] && size > adj) \
2684 adj_init_size(__init_begin, __init_end, init_data_size,
2685 _sinittext, init_code_size);
2686 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
2687 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
2688 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
2689 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
2691 #undef adj_init_size
2693 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
2694 #ifdef CONFIG_HIGHMEM
2698 K(nr_free_pages()), K(physpages),
2699 codesize / SZ_1K, datasize / SZ_1K, rosize / SZ_1K,
2700 (init_data_size + init_code_size) / SZ_1K, bss_size / SZ_1K,
2701 K(physpages - totalram_pages() - totalcma_pages),
2703 #ifdef CONFIG_HIGHMEM
2704 , K(totalhigh_pages())
2710 * Set up kernel memory allocators
2712 void __init mm_core_init(void)
2714 /* Initializations relying on SMP setup */
2715 build_all_zonelists(NULL);
2716 page_alloc_init_cpuhp();
2719 * page_ext requires contiguous pages,
2720 * bigger than MAX_ORDER unless SPARSEMEM.
2722 page_ext_init_flatmem();
2723 mem_debugging_and_hardening_init();
2724 kfence_alloc_pool();
2726 kmsan_init_shadow();
2727 stack_depot_early_init();
2729 mem_init_print_info();
2732 * page_owner must be initialized after buddy is ready, and also after
2733 * slab is ready so that stack_depot_init() works properly
2735 page_ext_init_flatmem_late();
2737 ptlock_cache_init();
2738 pgtable_cache_init();
2739 debug_objects_mem_init();
2741 /* If no deferred init page_ext now, as vmap is fully initialized */
2742 if (!deferred_struct_pages)
2744 /* Should be run before the first non-init thread is created */
2746 /* Should be run after espfix64 is set up. */
2748 kmsan_init_runtime();