1 // SPDX-License-Identifier: GPL-2.0
3 * sparse memory mappings.
6 #include <linux/slab.h>
7 #include <linux/mmzone.h>
8 #include <linux/memblock.h>
9 #include <linux/compiler.h>
10 #include <linux/highmem.h>
11 #include <linux/export.h>
12 #include <linux/spinlock.h>
13 #include <linux/vmalloc.h>
14 #include <linux/swap.h>
15 #include <linux/swapops.h>
16 #include <linux/bootmem_info.h>
22 * Permanent SPARSEMEM data:
24 * 1) mem_section - memory sections, mem_map's for valid memory
26 #ifdef CONFIG_SPARSEMEM_EXTREME
27 struct mem_section **mem_section;
29 struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]
30 ____cacheline_internodealigned_in_smp;
32 EXPORT_SYMBOL(mem_section);
34 #ifdef NODE_NOT_IN_PAGE_FLAGS
36 * If we did not store the node number in the page then we have to
37 * do a lookup in the section_to_node_table in order to find which
38 * node the page belongs to.
40 #if MAX_NUMNODES <= 256
41 static u8 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
43 static u16 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
46 int page_to_nid(const struct page *page)
48 return section_to_node_table[page_to_section(page)];
50 EXPORT_SYMBOL(page_to_nid);
52 static void set_section_nid(unsigned long section_nr, int nid)
54 section_to_node_table[section_nr] = nid;
56 #else /* !NODE_NOT_IN_PAGE_FLAGS */
57 static inline void set_section_nid(unsigned long section_nr, int nid)
62 #ifdef CONFIG_SPARSEMEM_EXTREME
63 static noinline struct mem_section __ref *sparse_index_alloc(int nid)
65 struct mem_section *section = NULL;
66 unsigned long array_size = SECTIONS_PER_ROOT *
67 sizeof(struct mem_section);
69 if (slab_is_available()) {
70 section = kzalloc_node(array_size, GFP_KERNEL, nid);
72 section = memblock_alloc_node(array_size, SMP_CACHE_BYTES,
75 panic("%s: Failed to allocate %lu bytes nid=%d\n",
76 __func__, array_size, nid);
82 static int __meminit sparse_index_init(unsigned long section_nr, int nid)
84 unsigned long root = SECTION_NR_TO_ROOT(section_nr);
85 struct mem_section *section;
88 * An existing section is possible in the sub-section hotplug
89 * case. First hot-add instantiates, follow-on hot-add reuses
90 * the existing section.
92 * The mem_hotplug_lock resolves the apparent race below.
94 if (mem_section[root])
97 section = sparse_index_alloc(nid);
101 mem_section[root] = section;
105 #else /* !SPARSEMEM_EXTREME */
106 static inline int sparse_index_init(unsigned long section_nr, int nid)
113 * During early boot, before section_mem_map is used for an actual
114 * mem_map, we use section_mem_map to store the section's NUMA
115 * node. This keeps us from having to use another data structure. The
116 * node information is cleared just before we store the real mem_map.
118 static inline unsigned long sparse_encode_early_nid(int nid)
120 return ((unsigned long)nid << SECTION_NID_SHIFT);
123 static inline int sparse_early_nid(struct mem_section *section)
125 return (section->section_mem_map >> SECTION_NID_SHIFT);
128 /* Validate the physical addressing limitations of the model */
129 static void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
130 unsigned long *end_pfn)
132 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
135 * Sanity checks - do not allow an architecture to pass
136 * in larger pfns than the maximum scope of sparsemem:
138 if (*start_pfn > max_sparsemem_pfn) {
139 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
140 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
141 *start_pfn, *end_pfn, max_sparsemem_pfn);
143 *start_pfn = max_sparsemem_pfn;
144 *end_pfn = max_sparsemem_pfn;
145 } else if (*end_pfn > max_sparsemem_pfn) {
146 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
147 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
148 *start_pfn, *end_pfn, max_sparsemem_pfn);
150 *end_pfn = max_sparsemem_pfn;
155 * There are a number of times that we loop over NR_MEM_SECTIONS,
156 * looking for section_present() on each. But, when we have very
157 * large physical address spaces, NR_MEM_SECTIONS can also be
158 * very large which makes the loops quite long.
160 * Keeping track of this gives us an easy way to break out of
163 unsigned long __highest_present_section_nr;
164 static void __section_mark_present(struct mem_section *ms,
165 unsigned long section_nr)
167 if (section_nr > __highest_present_section_nr)
168 __highest_present_section_nr = section_nr;
170 ms->section_mem_map |= SECTION_MARKED_PRESENT;
173 #define for_each_present_section_nr(start, section_nr) \
174 for (section_nr = next_present_section_nr(start-1); \
176 section_nr = next_present_section_nr(section_nr))
178 static inline unsigned long first_present_section_nr(void)
180 return next_present_section_nr(-1);
183 #ifdef CONFIG_SPARSEMEM_VMEMMAP
184 static void subsection_mask_set(unsigned long *map, unsigned long pfn,
185 unsigned long nr_pages)
187 int idx = subsection_map_index(pfn);
188 int end = subsection_map_index(pfn + nr_pages - 1);
190 bitmap_set(map, idx, end - idx + 1);
193 void __init subsection_map_init(unsigned long pfn, unsigned long nr_pages)
195 int end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
196 unsigned long nr, start_sec = pfn_to_section_nr(pfn);
201 for (nr = start_sec; nr <= end_sec; nr++) {
202 struct mem_section *ms;
205 pfns = min(nr_pages, PAGES_PER_SECTION
206 - (pfn & ~PAGE_SECTION_MASK));
207 ms = __nr_to_section(nr);
208 subsection_mask_set(ms->usage->subsection_map, pfn, pfns);
210 pr_debug("%s: sec: %lu pfns: %lu set(%d, %d)\n", __func__, nr,
211 pfns, subsection_map_index(pfn),
212 subsection_map_index(pfn + pfns - 1));
219 void __init subsection_map_init(unsigned long pfn, unsigned long nr_pages)
224 /* Record a memory area against a node. */
225 static void __init memory_present(int nid, unsigned long start, unsigned long end)
229 start &= PAGE_SECTION_MASK;
230 mminit_validate_memmodel_limits(&start, &end);
231 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
232 unsigned long section = pfn_to_section_nr(pfn);
233 struct mem_section *ms;
235 sparse_index_init(section, nid);
236 set_section_nid(section, nid);
238 ms = __nr_to_section(section);
239 if (!ms->section_mem_map) {
240 ms->section_mem_map = sparse_encode_early_nid(nid) |
242 __section_mark_present(ms, section);
248 * Mark all memblocks as present using memory_present().
249 * This is a convenience function that is useful to mark all of the systems
250 * memory as present during initialization.
252 static void __init memblocks_present(void)
254 unsigned long start, end;
257 #ifdef CONFIG_SPARSEMEM_EXTREME
258 if (unlikely(!mem_section)) {
259 unsigned long size, align;
261 size = sizeof(struct mem_section *) * NR_SECTION_ROOTS;
262 align = 1 << (INTERNODE_CACHE_SHIFT);
263 mem_section = memblock_alloc(size, align);
265 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
266 __func__, size, align);
270 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid)
271 memory_present(nid, start, end);
275 * Subtle, we encode the real pfn into the mem_map such that
276 * the identity pfn - section_mem_map will return the actual
277 * physical page frame number.
279 static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
281 unsigned long coded_mem_map =
282 (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
283 BUILD_BUG_ON(SECTION_MAP_LAST_BIT > PFN_SECTION_SHIFT);
284 BUG_ON(coded_mem_map & ~SECTION_MAP_MASK);
285 return coded_mem_map;
288 #ifdef CONFIG_MEMORY_HOTPLUG
290 * Decode mem_map from the coded memmap
292 struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
294 /* mask off the extra low bits of information */
295 coded_mem_map &= SECTION_MAP_MASK;
296 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
298 #endif /* CONFIG_MEMORY_HOTPLUG */
300 static void __meminit sparse_init_one_section(struct mem_section *ms,
301 unsigned long pnum, struct page *mem_map,
302 struct mem_section_usage *usage, unsigned long flags)
304 ms->section_mem_map &= ~SECTION_MAP_MASK;
305 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum)
306 | SECTION_HAS_MEM_MAP | flags;
310 static unsigned long usemap_size(void)
312 return BITS_TO_LONGS(SECTION_BLOCKFLAGS_BITS) * sizeof(unsigned long);
315 size_t mem_section_usage_size(void)
317 return sizeof(struct mem_section_usage) + usemap_size();
320 #ifdef CONFIG_MEMORY_HOTREMOVE
321 static inline phys_addr_t pgdat_to_phys(struct pglist_data *pgdat)
324 VM_BUG_ON(pgdat != &contig_page_data);
325 return __pa_symbol(&contig_page_data);
331 static struct mem_section_usage * __init
332 sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
335 struct mem_section_usage *usage;
336 unsigned long goal, limit;
339 * A page may contain usemaps for other sections preventing the
340 * page being freed and making a section unremovable while
341 * other sections referencing the usemap remain active. Similarly,
342 * a pgdat can prevent a section being removed. If section A
343 * contains a pgdat and section B contains the usemap, both
344 * sections become inter-dependent. This allocates usemaps
345 * from the same section as the pgdat where possible to avoid
348 goal = pgdat_to_phys(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT);
349 limit = goal + (1UL << PA_SECTION_SHIFT);
350 nid = early_pfn_to_nid(goal >> PAGE_SHIFT);
352 usage = memblock_alloc_try_nid(size, SMP_CACHE_BYTES, goal, limit, nid);
353 if (!usage && limit) {
360 static void __init check_usemap_section_nr(int nid,
361 struct mem_section_usage *usage)
363 unsigned long usemap_snr, pgdat_snr;
364 static unsigned long old_usemap_snr;
365 static unsigned long old_pgdat_snr;
366 struct pglist_data *pgdat = NODE_DATA(nid);
370 if (!old_usemap_snr) {
371 old_usemap_snr = NR_MEM_SECTIONS;
372 old_pgdat_snr = NR_MEM_SECTIONS;
375 usemap_snr = pfn_to_section_nr(__pa(usage) >> PAGE_SHIFT);
376 pgdat_snr = pfn_to_section_nr(pgdat_to_phys(pgdat) >> PAGE_SHIFT);
377 if (usemap_snr == pgdat_snr)
380 if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr)
381 /* skip redundant message */
384 old_usemap_snr = usemap_snr;
385 old_pgdat_snr = pgdat_snr;
387 usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr));
388 if (usemap_nid != nid) {
389 pr_info("node %d must be removed before remove section %ld\n",
394 * There is a circular dependency.
395 * Some platforms allow un-removable section because they will just
396 * gather other removable sections for dynamic partitioning.
397 * Just notify un-removable section's number here.
399 pr_info("Section %ld and %ld (node %d) have a circular dependency on usemap and pgdat allocations\n",
400 usemap_snr, pgdat_snr, nid);
403 static struct mem_section_usage * __init
404 sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
407 return memblock_alloc_node(size, SMP_CACHE_BYTES, pgdat->node_id);
410 static void __init check_usemap_section_nr(int nid,
411 struct mem_section_usage *usage)
414 #endif /* CONFIG_MEMORY_HOTREMOVE */
416 #ifdef CONFIG_SPARSEMEM_VMEMMAP
417 static unsigned long __init section_map_size(void)
419 return ALIGN(sizeof(struct page) * PAGES_PER_SECTION, PMD_SIZE);
423 static unsigned long __init section_map_size(void)
425 return PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
428 struct page __init *__populate_section_memmap(unsigned long pfn,
429 unsigned long nr_pages, int nid, struct vmem_altmap *altmap,
430 struct dev_pagemap *pgmap)
432 unsigned long size = section_map_size();
433 struct page *map = sparse_buffer_alloc(size);
434 phys_addr_t addr = __pa(MAX_DMA_ADDRESS);
439 map = memmap_alloc(size, size, addr, nid, false);
441 panic("%s: Failed to allocate %lu bytes align=0x%lx nid=%d from=%pa\n",
442 __func__, size, PAGE_SIZE, nid, &addr);
446 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
448 static void *sparsemap_buf __meminitdata;
449 static void *sparsemap_buf_end __meminitdata;
451 static inline void __meminit sparse_buffer_free(unsigned long size)
453 WARN_ON(!sparsemap_buf || size == 0);
454 memblock_free(sparsemap_buf, size);
457 static void __init sparse_buffer_init(unsigned long size, int nid)
459 phys_addr_t addr = __pa(MAX_DMA_ADDRESS);
460 WARN_ON(sparsemap_buf); /* forgot to call sparse_buffer_fini()? */
462 * Pre-allocated buffer is mainly used by __populate_section_memmap
463 * and we want it to be properly aligned to the section size - this is
464 * especially the case for VMEMMAP which maps memmap to PMDs
466 sparsemap_buf = memmap_alloc(size, section_map_size(), addr, nid, true);
467 sparsemap_buf_end = sparsemap_buf + size;
470 static void __init sparse_buffer_fini(void)
472 unsigned long size = sparsemap_buf_end - sparsemap_buf;
474 if (sparsemap_buf && size > 0)
475 sparse_buffer_free(size);
476 sparsemap_buf = NULL;
479 void * __meminit sparse_buffer_alloc(unsigned long size)
484 ptr = (void *) roundup((unsigned long)sparsemap_buf, size);
485 if (ptr + size > sparsemap_buf_end)
488 /* Free redundant aligned space */
489 if ((unsigned long)(ptr - sparsemap_buf) > 0)
490 sparse_buffer_free((unsigned long)(ptr - sparsemap_buf));
491 sparsemap_buf = ptr + size;
497 void __weak __meminit vmemmap_populate_print_last(void)
502 * Initialize sparse on a specific node. The node spans [pnum_begin, pnum_end)
503 * And number of present sections in this node is map_count.
505 static void __init sparse_init_nid(int nid, unsigned long pnum_begin,
506 unsigned long pnum_end,
507 unsigned long map_count)
509 struct mem_section_usage *usage;
513 usage = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nid),
514 mem_section_usage_size() * map_count);
516 pr_err("%s: node[%d] usemap allocation failed", __func__, nid);
519 sparse_buffer_init(map_count * section_map_size(), nid);
520 for_each_present_section_nr(pnum_begin, pnum) {
521 unsigned long pfn = section_nr_to_pfn(pnum);
523 if (pnum >= pnum_end)
526 map = __populate_section_memmap(pfn, PAGES_PER_SECTION,
529 pr_err("%s: node[%d] memory map backing failed. Some memory will not be available.",
532 sparse_buffer_fini();
535 check_usemap_section_nr(nid, usage);
536 sparse_init_one_section(__nr_to_section(pnum), pnum, map, usage,
538 usage = (void *) usage + mem_section_usage_size();
540 sparse_buffer_fini();
543 /* We failed to allocate, mark all the following pnums as not present */
544 for_each_present_section_nr(pnum_begin, pnum) {
545 struct mem_section *ms;
547 if (pnum >= pnum_end)
549 ms = __nr_to_section(pnum);
550 ms->section_mem_map = 0;
555 * Allocate the accumulated non-linear sections, allocate a mem_map
556 * for each and record the physical to section mapping.
558 void __init sparse_init(void)
560 unsigned long pnum_end, pnum_begin, map_count = 1;
563 /* see include/linux/mmzone.h 'struct mem_section' definition */
564 BUILD_BUG_ON(!is_power_of_2(sizeof(struct mem_section)));
567 pnum_begin = first_present_section_nr();
568 nid_begin = sparse_early_nid(__nr_to_section(pnum_begin));
570 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
571 set_pageblock_order();
573 for_each_present_section_nr(pnum_begin + 1, pnum_end) {
574 int nid = sparse_early_nid(__nr_to_section(pnum_end));
576 if (nid == nid_begin) {
580 /* Init node with sections in range [pnum_begin, pnum_end) */
581 sparse_init_nid(nid_begin, pnum_begin, pnum_end, map_count);
583 pnum_begin = pnum_end;
586 /* cover the last node */
587 sparse_init_nid(nid_begin, pnum_begin, pnum_end, map_count);
588 vmemmap_populate_print_last();
591 #ifdef CONFIG_MEMORY_HOTPLUG
593 /* Mark all memory sections within the pfn range as online */
594 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
598 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
599 unsigned long section_nr = pfn_to_section_nr(pfn);
600 struct mem_section *ms;
602 /* onlining code should never touch invalid ranges */
603 if (WARN_ON(!valid_section_nr(section_nr)))
606 ms = __nr_to_section(section_nr);
607 ms->section_mem_map |= SECTION_IS_ONLINE;
611 /* Mark all memory sections within the pfn range as offline */
612 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
616 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
617 unsigned long section_nr = pfn_to_section_nr(pfn);
618 struct mem_section *ms;
621 * TODO this needs some double checking. Offlining code makes
622 * sure to check pfn_valid but those checks might be just bogus
624 if (WARN_ON(!valid_section_nr(section_nr)))
627 ms = __nr_to_section(section_nr);
628 ms->section_mem_map &= ~SECTION_IS_ONLINE;
632 #ifdef CONFIG_SPARSEMEM_VMEMMAP
633 static struct page * __meminit populate_section_memmap(unsigned long pfn,
634 unsigned long nr_pages, int nid, struct vmem_altmap *altmap,
635 struct dev_pagemap *pgmap)
637 return __populate_section_memmap(pfn, nr_pages, nid, altmap, pgmap);
640 static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
641 struct vmem_altmap *altmap)
643 unsigned long start = (unsigned long) pfn_to_page(pfn);
644 unsigned long end = start + nr_pages * sizeof(struct page);
646 vmemmap_free(start, end, altmap);
648 static void free_map_bootmem(struct page *memmap)
650 unsigned long start = (unsigned long)memmap;
651 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
653 vmemmap_free(start, end, NULL);
656 static int clear_subsection_map(unsigned long pfn, unsigned long nr_pages)
658 DECLARE_BITMAP(map, SUBSECTIONS_PER_SECTION) = { 0 };
659 DECLARE_BITMAP(tmp, SUBSECTIONS_PER_SECTION) = { 0 };
660 struct mem_section *ms = __pfn_to_section(pfn);
661 unsigned long *subsection_map = ms->usage
662 ? &ms->usage->subsection_map[0] : NULL;
664 subsection_mask_set(map, pfn, nr_pages);
666 bitmap_and(tmp, map, subsection_map, SUBSECTIONS_PER_SECTION);
668 if (WARN(!subsection_map || !bitmap_equal(tmp, map, SUBSECTIONS_PER_SECTION),
669 "section already deactivated (%#lx + %ld)\n",
673 bitmap_xor(subsection_map, map, subsection_map, SUBSECTIONS_PER_SECTION);
677 static bool is_subsection_map_empty(struct mem_section *ms)
679 return bitmap_empty(&ms->usage->subsection_map[0],
680 SUBSECTIONS_PER_SECTION);
683 static int fill_subsection_map(unsigned long pfn, unsigned long nr_pages)
685 struct mem_section *ms = __pfn_to_section(pfn);
686 DECLARE_BITMAP(map, SUBSECTIONS_PER_SECTION) = { 0 };
687 unsigned long *subsection_map;
690 subsection_mask_set(map, pfn, nr_pages);
692 subsection_map = &ms->usage->subsection_map[0];
694 if (bitmap_empty(map, SUBSECTIONS_PER_SECTION))
696 else if (bitmap_intersects(map, subsection_map, SUBSECTIONS_PER_SECTION))
699 bitmap_or(subsection_map, map, subsection_map,
700 SUBSECTIONS_PER_SECTION);
705 static struct page * __meminit populate_section_memmap(unsigned long pfn,
706 unsigned long nr_pages, int nid, struct vmem_altmap *altmap,
707 struct dev_pagemap *pgmap)
709 return kvmalloc_node(array_size(sizeof(struct page),
710 PAGES_PER_SECTION), GFP_KERNEL, nid);
713 static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
714 struct vmem_altmap *altmap)
716 kvfree(pfn_to_page(pfn));
719 static void free_map_bootmem(struct page *memmap)
721 unsigned long maps_section_nr, removing_section_nr, i;
722 unsigned long magic, nr_pages;
723 struct page *page = virt_to_page(memmap);
725 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
728 for (i = 0; i < nr_pages; i++, page++) {
731 BUG_ON(magic == NODE_INFO);
733 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
734 removing_section_nr = page_private(page);
737 * When this function is called, the removing section is
738 * logical offlined state. This means all pages are isolated
739 * from page allocator. If removing section's memmap is placed
740 * on the same section, it must not be freed.
741 * If it is freed, page allocator may allocate it which will
742 * be removed physically soon.
744 if (maps_section_nr != removing_section_nr)
745 put_page_bootmem(page);
749 static int clear_subsection_map(unsigned long pfn, unsigned long nr_pages)
754 static bool is_subsection_map_empty(struct mem_section *ms)
759 static int fill_subsection_map(unsigned long pfn, unsigned long nr_pages)
763 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
766 * To deactivate a memory region, there are 3 cases to handle across
767 * two configurations (SPARSEMEM_VMEMMAP={y,n}):
769 * 1. deactivation of a partial hot-added section (only possible in
770 * the SPARSEMEM_VMEMMAP=y case).
771 * a) section was present at memory init.
772 * b) section was hot-added post memory init.
773 * 2. deactivation of a complete hot-added section.
774 * 3. deactivation of a complete section from memory init.
776 * For 1, when subsection_map does not empty we will not be freeing the
777 * usage map, but still need to free the vmemmap range.
779 * For 2 and 3, the SPARSEMEM_VMEMMAP={y,n} cases are unified
781 static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
782 struct vmem_altmap *altmap)
784 struct mem_section *ms = __pfn_to_section(pfn);
785 bool section_is_early = early_section(ms);
786 struct page *memmap = NULL;
789 if (clear_subsection_map(pfn, nr_pages))
792 empty = is_subsection_map_empty(ms);
794 unsigned long section_nr = pfn_to_section_nr(pfn);
797 * Mark the section invalid so that valid_section()
798 * return false. This prevents code from dereferencing
801 ms->section_mem_map &= ~SECTION_HAS_MEM_MAP;
804 * When removing an early section, the usage map is kept (as the
805 * usage maps of other sections fall into the same page). It
806 * will be re-used when re-adding the section - which is then no
807 * longer an early section. If the usage map is PageReserved, it
808 * was allocated during boot.
810 if (!PageReserved(virt_to_page(ms->usage))) {
811 kfree_rcu(ms->usage, rcu);
812 WRITE_ONCE(ms->usage, NULL);
814 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
818 * The memmap of early sections is always fully populated. See
819 * section_activate() and pfn_valid() .
821 if (!section_is_early)
822 depopulate_section_memmap(pfn, nr_pages, altmap);
824 free_map_bootmem(memmap);
827 ms->section_mem_map = (unsigned long)NULL;
830 static struct page * __meminit section_activate(int nid, unsigned long pfn,
831 unsigned long nr_pages, struct vmem_altmap *altmap,
832 struct dev_pagemap *pgmap)
834 struct mem_section *ms = __pfn_to_section(pfn);
835 struct mem_section_usage *usage = NULL;
840 usage = kzalloc(mem_section_usage_size(), GFP_KERNEL);
842 return ERR_PTR(-ENOMEM);
846 rc = fill_subsection_map(pfn, nr_pages);
855 * The early init code does not consider partially populated
856 * initial sections, it simply assumes that memory will never be
857 * referenced. If we hot-add memory into such a section then we
858 * do not need to populate the memmap and can simply reuse what
861 if (nr_pages < PAGES_PER_SECTION && early_section(ms))
862 return pfn_to_page(pfn);
864 memmap = populate_section_memmap(pfn, nr_pages, nid, altmap, pgmap);
866 section_deactivate(pfn, nr_pages, altmap);
867 return ERR_PTR(-ENOMEM);
874 * sparse_add_section - add a memory section, or populate an existing one
875 * @nid: The node to add section on
876 * @start_pfn: start pfn of the memory range
877 * @nr_pages: number of pfns to add in the section
878 * @altmap: alternate pfns to allocate the memmap backing store
879 * @pgmap: alternate compound page geometry for devmap mappings
881 * This is only intended for hotplug.
883 * Note that only VMEMMAP supports sub-section aligned hotplug,
884 * the proper alignment and size are gated by check_pfn_span().
889 * * -EEXIST - Section has been present.
890 * * -ENOMEM - Out of memory.
892 int __meminit sparse_add_section(int nid, unsigned long start_pfn,
893 unsigned long nr_pages, struct vmem_altmap *altmap,
894 struct dev_pagemap *pgmap)
896 unsigned long section_nr = pfn_to_section_nr(start_pfn);
897 struct mem_section *ms;
901 ret = sparse_index_init(section_nr, nid);
905 memmap = section_activate(nid, start_pfn, nr_pages, altmap, pgmap);
907 return PTR_ERR(memmap);
910 * Poison uninitialized struct pages in order to catch invalid flags
913 if (!altmap || !altmap->inaccessible)
914 page_init_poison(memmap, sizeof(struct page) * nr_pages);
916 ms = __nr_to_section(section_nr);
917 set_section_nid(section_nr, nid);
918 __section_mark_present(ms, section_nr);
920 /* Align memmap to section boundary in the subsection case */
921 if (section_nr_to_pfn(section_nr) != start_pfn)
922 memmap = pfn_to_page(section_nr_to_pfn(section_nr));
923 sparse_init_one_section(ms, section_nr, memmap, ms->usage, 0);
928 void sparse_remove_section(unsigned long pfn, unsigned long nr_pages,
929 struct vmem_altmap *altmap)
931 struct mem_section *ms = __pfn_to_section(pfn);
933 if (WARN_ON_ONCE(!valid_section(ms)))
936 section_deactivate(pfn, nr_pages, altmap);
938 #endif /* CONFIG_MEMORY_HOTPLUG */