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 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); \
175 ((section_nr != -1) && \
176 (section_nr <= __highest_present_section_nr)); \
177 section_nr = next_present_section_nr(section_nr))
179 static inline unsigned long first_present_section_nr(void)
181 return next_present_section_nr(-1);
184 #ifdef CONFIG_SPARSEMEM_VMEMMAP
185 static void subsection_mask_set(unsigned long *map, unsigned long pfn,
186 unsigned long nr_pages)
188 int idx = subsection_map_index(pfn);
189 int end = subsection_map_index(pfn + nr_pages - 1);
191 bitmap_set(map, idx, end - idx + 1);
194 void __init subsection_map_init(unsigned long pfn, unsigned long nr_pages)
196 int end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
197 unsigned long nr, start_sec = pfn_to_section_nr(pfn);
202 for (nr = start_sec; nr <= end_sec; nr++) {
203 struct mem_section *ms;
206 pfns = min(nr_pages, PAGES_PER_SECTION
207 - (pfn & ~PAGE_SECTION_MASK));
208 ms = __nr_to_section(nr);
209 subsection_mask_set(ms->usage->subsection_map, pfn, pfns);
211 pr_debug("%s: sec: %lu pfns: %lu set(%d, %d)\n", __func__, nr,
212 pfns, subsection_map_index(pfn),
213 subsection_map_index(pfn + pfns - 1));
220 void __init subsection_map_init(unsigned long pfn, unsigned long nr_pages)
225 /* Record a memory area against a node. */
226 static void __init memory_present(int nid, unsigned long start, unsigned long end)
230 #ifdef CONFIG_SPARSEMEM_EXTREME
231 if (unlikely(!mem_section)) {
232 unsigned long size, align;
234 size = sizeof(struct mem_section *) * NR_SECTION_ROOTS;
235 align = 1 << (INTERNODE_CACHE_SHIFT);
236 mem_section = memblock_alloc(size, align);
238 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
239 __func__, size, align);
243 start &= PAGE_SECTION_MASK;
244 mminit_validate_memmodel_limits(&start, &end);
245 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
246 unsigned long section = pfn_to_section_nr(pfn);
247 struct mem_section *ms;
249 sparse_index_init(section, nid);
250 set_section_nid(section, nid);
252 ms = __nr_to_section(section);
253 if (!ms->section_mem_map) {
254 ms->section_mem_map = sparse_encode_early_nid(nid) |
256 __section_mark_present(ms, section);
262 * Mark all memblocks as present using memory_present().
263 * This is a convenience function that is useful to mark all of the systems
264 * memory as present during initialization.
266 static void __init memblocks_present(void)
268 unsigned long start, end;
271 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid)
272 memory_present(nid, start, end);
276 * Subtle, we encode the real pfn into the mem_map such that
277 * the identity pfn - section_mem_map will return the actual
278 * physical page frame number.
280 static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
282 unsigned long coded_mem_map =
283 (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
284 BUILD_BUG_ON(SECTION_MAP_LAST_BIT > (1UL<<PFN_SECTION_SHIFT));
285 BUG_ON(coded_mem_map & ~SECTION_MAP_MASK);
286 return coded_mem_map;
289 #ifdef CONFIG_MEMORY_HOTPLUG
291 * Decode mem_map from the coded memmap
293 struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
295 /* mask off the extra low bits of information */
296 coded_mem_map &= SECTION_MAP_MASK;
297 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
299 #endif /* CONFIG_MEMORY_HOTPLUG */
301 static void __meminit sparse_init_one_section(struct mem_section *ms,
302 unsigned long pnum, struct page *mem_map,
303 struct mem_section_usage *usage, unsigned long flags)
305 ms->section_mem_map &= ~SECTION_MAP_MASK;
306 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum)
307 | SECTION_HAS_MEM_MAP | flags;
311 static unsigned long usemap_size(void)
313 return BITS_TO_LONGS(SECTION_BLOCKFLAGS_BITS) * sizeof(unsigned long);
316 size_t mem_section_usage_size(void)
318 return sizeof(struct mem_section_usage) + usemap_size();
321 static inline phys_addr_t pgdat_to_phys(struct pglist_data *pgdat)
324 return __pa_symbol(pgdat);
330 #ifdef CONFIG_MEMORY_HOTREMOVE
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)
431 unsigned long size = section_map_size();
432 struct page *map = sparse_buffer_alloc(size);
433 phys_addr_t addr = __pa(MAX_DMA_ADDRESS);
438 map = memblock_alloc_try_nid_raw(size, size, addr,
439 MEMBLOCK_ALLOC_ACCESSIBLE, nid);
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_early(__pa(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 = memblock_alloc_exact_nid_raw(size, section_map_size(),
467 addr, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
468 sparsemap_buf_end = sparsemap_buf + size;
471 static void __init sparse_buffer_fini(void)
473 unsigned long size = sparsemap_buf_end - sparsemap_buf;
475 if (sparsemap_buf && size > 0)
476 sparse_buffer_free(size);
477 sparsemap_buf = NULL;
480 void * __meminit sparse_buffer_alloc(unsigned long size)
485 ptr = (void *) roundup((unsigned long)sparsemap_buf, size);
486 if (ptr + size > sparsemap_buf_end)
489 /* Free redundant aligned space */
490 if ((unsigned long)(ptr - sparsemap_buf) > 0)
491 sparse_buffer_free((unsigned long)(ptr - sparsemap_buf));
492 sparsemap_buf = ptr + size;
498 void __weak __meminit vmemmap_populate_print_last(void)
503 * Initialize sparse on a specific node. The node spans [pnum_begin, pnum_end)
504 * And number of present sections in this node is map_count.
506 static void __init sparse_init_nid(int nid, unsigned long pnum_begin,
507 unsigned long pnum_end,
508 unsigned long map_count)
510 struct mem_section_usage *usage;
514 usage = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nid),
515 mem_section_usage_size() * map_count);
517 pr_err("%s: node[%d] usemap allocation failed", __func__, nid);
520 sparse_buffer_init(map_count * section_map_size(), nid);
521 for_each_present_section_nr(pnum_begin, pnum) {
522 unsigned long pfn = section_nr_to_pfn(pnum);
524 if (pnum >= pnum_end)
527 map = __populate_section_memmap(pfn, PAGES_PER_SECTION,
530 pr_err("%s: node[%d] memory map backing failed. Some memory will not be available.",
533 sparse_buffer_fini();
536 check_usemap_section_nr(nid, usage);
537 sparse_init_one_section(__nr_to_section(pnum), pnum, map, usage,
539 usage = (void *) usage + mem_section_usage_size();
541 sparse_buffer_fini();
544 /* We failed to allocate, mark all the following pnums as not present */
545 for_each_present_section_nr(pnum_begin, pnum) {
546 struct mem_section *ms;
548 if (pnum >= pnum_end)
550 ms = __nr_to_section(pnum);
551 ms->section_mem_map = 0;
556 * Allocate the accumulated non-linear sections, allocate a mem_map
557 * for each and record the physical to section mapping.
559 void __init sparse_init(void)
561 unsigned long pnum_end, pnum_begin, map_count = 1;
566 pnum_begin = first_present_section_nr();
567 nid_begin = sparse_early_nid(__nr_to_section(pnum_begin));
569 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
570 set_pageblock_order();
572 for_each_present_section_nr(pnum_begin + 1, pnum_end) {
573 int nid = sparse_early_nid(__nr_to_section(pnum_end));
575 if (nid == nid_begin) {
579 /* Init node with sections in range [pnum_begin, pnum_end) */
580 sparse_init_nid(nid_begin, pnum_begin, pnum_end, map_count);
582 pnum_begin = pnum_end;
585 /* cover the last node */
586 sparse_init_nid(nid_begin, pnum_begin, pnum_end, map_count);
587 vmemmap_populate_print_last();
590 #ifdef CONFIG_MEMORY_HOTPLUG
592 /* Mark all memory sections within the pfn range as online */
593 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
597 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
598 unsigned long section_nr = pfn_to_section_nr(pfn);
599 struct mem_section *ms;
601 /* onlining code should never touch invalid ranges */
602 if (WARN_ON(!valid_section_nr(section_nr)))
605 ms = __nr_to_section(section_nr);
606 ms->section_mem_map |= SECTION_IS_ONLINE;
610 /* Mark all memory sections within the pfn range as offline */
611 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
615 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
616 unsigned long section_nr = pfn_to_section_nr(pfn);
617 struct mem_section *ms;
620 * TODO this needs some double checking. Offlining code makes
621 * sure to check pfn_valid but those checks might be just bogus
623 if (WARN_ON(!valid_section_nr(section_nr)))
626 ms = __nr_to_section(section_nr);
627 ms->section_mem_map &= ~SECTION_IS_ONLINE;
631 #ifdef CONFIG_SPARSEMEM_VMEMMAP
632 static struct page * __meminit populate_section_memmap(unsigned long pfn,
633 unsigned long nr_pages, int nid, struct vmem_altmap *altmap)
635 return __populate_section_memmap(pfn, nr_pages, nid, altmap);
638 static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
639 struct vmem_altmap *altmap)
641 unsigned long start = (unsigned long) pfn_to_page(pfn);
642 unsigned long end = start + nr_pages * sizeof(struct page);
644 vmemmap_free(start, end, altmap);
646 static void free_map_bootmem(struct page *memmap)
648 unsigned long start = (unsigned long)memmap;
649 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
651 vmemmap_free(start, end, NULL);
654 static int clear_subsection_map(unsigned long pfn, unsigned long nr_pages)
656 DECLARE_BITMAP(map, SUBSECTIONS_PER_SECTION) = { 0 };
657 DECLARE_BITMAP(tmp, SUBSECTIONS_PER_SECTION) = { 0 };
658 struct mem_section *ms = __pfn_to_section(pfn);
659 unsigned long *subsection_map = ms->usage
660 ? &ms->usage->subsection_map[0] : NULL;
662 subsection_mask_set(map, pfn, nr_pages);
664 bitmap_and(tmp, map, subsection_map, SUBSECTIONS_PER_SECTION);
666 if (WARN(!subsection_map || !bitmap_equal(tmp, map, SUBSECTIONS_PER_SECTION),
667 "section already deactivated (%#lx + %ld)\n",
671 bitmap_xor(subsection_map, map, subsection_map, SUBSECTIONS_PER_SECTION);
675 static bool is_subsection_map_empty(struct mem_section *ms)
677 return bitmap_empty(&ms->usage->subsection_map[0],
678 SUBSECTIONS_PER_SECTION);
681 static int fill_subsection_map(unsigned long pfn, unsigned long nr_pages)
683 struct mem_section *ms = __pfn_to_section(pfn);
684 DECLARE_BITMAP(map, SUBSECTIONS_PER_SECTION) = { 0 };
685 unsigned long *subsection_map;
688 subsection_mask_set(map, pfn, nr_pages);
690 subsection_map = &ms->usage->subsection_map[0];
692 if (bitmap_empty(map, SUBSECTIONS_PER_SECTION))
694 else if (bitmap_intersects(map, subsection_map, SUBSECTIONS_PER_SECTION))
697 bitmap_or(subsection_map, map, subsection_map,
698 SUBSECTIONS_PER_SECTION);
703 struct page * __meminit populate_section_memmap(unsigned long pfn,
704 unsigned long nr_pages, int nid, struct vmem_altmap *altmap)
706 return kvmalloc_node(array_size(sizeof(struct page),
707 PAGES_PER_SECTION), GFP_KERNEL, nid);
710 static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
711 struct vmem_altmap *altmap)
713 kvfree(pfn_to_page(pfn));
716 static void free_map_bootmem(struct page *memmap)
718 unsigned long maps_section_nr, removing_section_nr, i;
719 unsigned long magic, nr_pages;
720 struct page *page = virt_to_page(memmap);
722 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
725 for (i = 0; i < nr_pages; i++, page++) {
726 magic = (unsigned long) page->freelist;
728 BUG_ON(magic == NODE_INFO);
730 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
731 removing_section_nr = page_private(page);
734 * When this function is called, the removing section is
735 * logical offlined state. This means all pages are isolated
736 * from page allocator. If removing section's memmap is placed
737 * on the same section, it must not be freed.
738 * If it is freed, page allocator may allocate it which will
739 * be removed physically soon.
741 if (maps_section_nr != removing_section_nr)
742 put_page_bootmem(page);
746 static int clear_subsection_map(unsigned long pfn, unsigned long nr_pages)
751 static bool is_subsection_map_empty(struct mem_section *ms)
756 static int fill_subsection_map(unsigned long pfn, unsigned long nr_pages)
760 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
763 * To deactivate a memory region, there are 3 cases to handle across
764 * two configurations (SPARSEMEM_VMEMMAP={y,n}):
766 * 1. deactivation of a partial hot-added section (only possible in
767 * the SPARSEMEM_VMEMMAP=y case).
768 * a) section was present at memory init.
769 * b) section was hot-added post memory init.
770 * 2. deactivation of a complete hot-added section.
771 * 3. deactivation of a complete section from memory init.
773 * For 1, when subsection_map does not empty we will not be freeing the
774 * usage map, but still need to free the vmemmap range.
776 * For 2 and 3, the SPARSEMEM_VMEMMAP={y,n} cases are unified
778 static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
779 struct vmem_altmap *altmap)
781 struct mem_section *ms = __pfn_to_section(pfn);
782 bool section_is_early = early_section(ms);
783 struct page *memmap = NULL;
786 if (clear_subsection_map(pfn, nr_pages))
789 empty = is_subsection_map_empty(ms);
791 unsigned long section_nr = pfn_to_section_nr(pfn);
794 * When removing an early section, the usage map is kept (as the
795 * usage maps of other sections fall into the same page). It
796 * will be re-used when re-adding the section - which is then no
797 * longer an early section. If the usage map is PageReserved, it
798 * was allocated during boot.
800 if (!PageReserved(virt_to_page(ms->usage))) {
804 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
806 * Mark the section invalid so that valid_section()
807 * return false. This prevents code from dereferencing
810 ms->section_mem_map &= ~SECTION_HAS_MEM_MAP;
814 * The memmap of early sections is always fully populated. See
815 * section_activate() and pfn_valid() .
817 if (!section_is_early)
818 depopulate_section_memmap(pfn, nr_pages, altmap);
820 free_map_bootmem(memmap);
823 ms->section_mem_map = (unsigned long)NULL;
826 static struct page * __meminit section_activate(int nid, unsigned long pfn,
827 unsigned long nr_pages, struct vmem_altmap *altmap)
829 struct mem_section *ms = __pfn_to_section(pfn);
830 struct mem_section_usage *usage = NULL;
835 usage = kzalloc(mem_section_usage_size(), GFP_KERNEL);
837 return ERR_PTR(-ENOMEM);
841 rc = fill_subsection_map(pfn, nr_pages);
850 * The early init code does not consider partially populated
851 * initial sections, it simply assumes that memory will never be
852 * referenced. If we hot-add memory into such a section then we
853 * do not need to populate the memmap and can simply reuse what
856 if (nr_pages < PAGES_PER_SECTION && early_section(ms))
857 return pfn_to_page(pfn);
859 memmap = populate_section_memmap(pfn, nr_pages, nid, altmap);
861 section_deactivate(pfn, nr_pages, altmap);
862 return ERR_PTR(-ENOMEM);
869 * sparse_add_section - add a memory section, or populate an existing one
870 * @nid: The node to add section on
871 * @start_pfn: start pfn of the memory range
872 * @nr_pages: number of pfns to add in the section
873 * @altmap: device page map
875 * This is only intended for hotplug.
877 * Note that only VMEMMAP supports sub-section aligned hotplug,
878 * the proper alignment and size are gated by check_pfn_span().
883 * * -EEXIST - Section has been present.
884 * * -ENOMEM - Out of memory.
886 int __meminit sparse_add_section(int nid, unsigned long start_pfn,
887 unsigned long nr_pages, struct vmem_altmap *altmap)
889 unsigned long section_nr = pfn_to_section_nr(start_pfn);
890 struct mem_section *ms;
894 ret = sparse_index_init(section_nr, nid);
898 memmap = section_activate(nid, start_pfn, nr_pages, altmap);
900 return PTR_ERR(memmap);
903 * Poison uninitialized struct pages in order to catch invalid flags
906 page_init_poison(memmap, sizeof(struct page) * nr_pages);
908 ms = __nr_to_section(section_nr);
909 set_section_nid(section_nr, nid);
910 __section_mark_present(ms, section_nr);
912 /* Align memmap to section boundary in the subsection case */
913 if (section_nr_to_pfn(section_nr) != start_pfn)
914 memmap = pfn_to_page(section_nr_to_pfn(section_nr));
915 sparse_init_one_section(ms, section_nr, memmap, ms->usage, 0);
920 #ifdef CONFIG_MEMORY_FAILURE
921 static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
926 * A further optimization is to have per section refcounted
927 * num_poisoned_pages. But that would need more space per memmap, so
928 * for now just do a quick global check to speed up this routine in the
929 * absence of bad pages.
931 if (atomic_long_read(&num_poisoned_pages) == 0)
934 for (i = 0; i < nr_pages; i++) {
935 if (PageHWPoison(&memmap[i])) {
936 num_poisoned_pages_dec();
937 ClearPageHWPoison(&memmap[i]);
942 static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
947 void sparse_remove_section(struct mem_section *ms, unsigned long pfn,
948 unsigned long nr_pages, unsigned long map_offset,
949 struct vmem_altmap *altmap)
951 clear_hwpoisoned_pages(pfn_to_page(pfn) + map_offset,
952 nr_pages - map_offset);
953 section_deactivate(pfn, nr_pages, altmap);
955 #endif /* CONFIG_MEMORY_HOTPLUG */