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)
112 #ifdef CONFIG_SPARSEMEM_EXTREME
113 unsigned long __section_nr(struct mem_section *ms)
115 unsigned long root_nr;
116 struct mem_section *root = NULL;
118 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
119 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
123 if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT)))
129 return (root_nr * SECTIONS_PER_ROOT) + (ms - root);
132 unsigned long __section_nr(struct mem_section *ms)
134 return (unsigned long)(ms - mem_section[0]);
139 * During early boot, before section_mem_map is used for an actual
140 * mem_map, we use section_mem_map to store the section's NUMA
141 * node. This keeps us from having to use another data structure. The
142 * node information is cleared just before we store the real mem_map.
144 static inline unsigned long sparse_encode_early_nid(int nid)
146 return (nid << SECTION_NID_SHIFT);
149 static inline int sparse_early_nid(struct mem_section *section)
151 return (section->section_mem_map >> SECTION_NID_SHIFT);
154 /* Validate the physical addressing limitations of the model */
155 void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
156 unsigned long *end_pfn)
158 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
161 * Sanity checks - do not allow an architecture to pass
162 * in larger pfns than the maximum scope of sparsemem:
164 if (*start_pfn > max_sparsemem_pfn) {
165 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
166 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
167 *start_pfn, *end_pfn, max_sparsemem_pfn);
169 *start_pfn = max_sparsemem_pfn;
170 *end_pfn = max_sparsemem_pfn;
171 } else if (*end_pfn > max_sparsemem_pfn) {
172 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
173 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
174 *start_pfn, *end_pfn, max_sparsemem_pfn);
176 *end_pfn = max_sparsemem_pfn;
181 * There are a number of times that we loop over NR_MEM_SECTIONS,
182 * looking for section_present() on each. But, when we have very
183 * large physical address spaces, NR_MEM_SECTIONS can also be
184 * very large which makes the loops quite long.
186 * Keeping track of this gives us an easy way to break out of
189 unsigned long __highest_present_section_nr;
190 static void __section_mark_present(struct mem_section *ms,
191 unsigned long section_nr)
193 if (section_nr > __highest_present_section_nr)
194 __highest_present_section_nr = section_nr;
196 ms->section_mem_map |= SECTION_MARKED_PRESENT;
199 #define for_each_present_section_nr(start, section_nr) \
200 for (section_nr = next_present_section_nr(start-1); \
201 ((section_nr != -1) && \
202 (section_nr <= __highest_present_section_nr)); \
203 section_nr = next_present_section_nr(section_nr))
205 static inline unsigned long first_present_section_nr(void)
207 return next_present_section_nr(-1);
210 #ifdef CONFIG_SPARSEMEM_VMEMMAP
211 static void subsection_mask_set(unsigned long *map, unsigned long pfn,
212 unsigned long nr_pages)
214 int idx = subsection_map_index(pfn);
215 int end = subsection_map_index(pfn + nr_pages - 1);
217 bitmap_set(map, idx, end - idx + 1);
220 void __init subsection_map_init(unsigned long pfn, unsigned long nr_pages)
222 int end_sec = pfn_to_section_nr(pfn + nr_pages - 1);
223 unsigned long nr, start_sec = pfn_to_section_nr(pfn);
228 for (nr = start_sec; nr <= end_sec; nr++) {
229 struct mem_section *ms;
232 pfns = min(nr_pages, PAGES_PER_SECTION
233 - (pfn & ~PAGE_SECTION_MASK));
234 ms = __nr_to_section(nr);
235 subsection_mask_set(ms->usage->subsection_map, pfn, pfns);
237 pr_debug("%s: sec: %lu pfns: %lu set(%d, %d)\n", __func__, nr,
238 pfns, subsection_map_index(pfn),
239 subsection_map_index(pfn + pfns - 1));
246 void __init subsection_map_init(unsigned long pfn, unsigned long nr_pages)
251 /* Record a memory area against a node. */
252 static void __init memory_present(int nid, unsigned long start, unsigned long end)
256 #ifdef CONFIG_SPARSEMEM_EXTREME
257 if (unlikely(!mem_section)) {
258 unsigned long size, align;
260 size = sizeof(struct mem_section *) * NR_SECTION_ROOTS;
261 align = 1 << (INTERNODE_CACHE_SHIFT);
262 mem_section = memblock_alloc(size, align);
264 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
265 __func__, size, align);
269 start &= PAGE_SECTION_MASK;
270 mminit_validate_memmodel_limits(&start, &end);
271 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
272 unsigned long section = pfn_to_section_nr(pfn);
273 struct mem_section *ms;
275 sparse_index_init(section, nid);
276 set_section_nid(section, nid);
278 ms = __nr_to_section(section);
279 if (!ms->section_mem_map) {
280 ms->section_mem_map = sparse_encode_early_nid(nid) |
282 __section_mark_present(ms, section);
288 * Mark all memblocks as present using memory_present().
289 * This is a convenience function that is useful to mark all of the systems
290 * memory as present during initialization.
292 static void __init memblocks_present(void)
294 unsigned long start, end;
297 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid)
298 memory_present(nid, start, end);
302 * Subtle, we encode the real pfn into the mem_map such that
303 * the identity pfn - section_mem_map will return the actual
304 * physical page frame number.
306 static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
308 unsigned long coded_mem_map =
309 (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
310 BUILD_BUG_ON(SECTION_MAP_LAST_BIT > (1UL<<PFN_SECTION_SHIFT));
311 BUG_ON(coded_mem_map & ~SECTION_MAP_MASK);
312 return coded_mem_map;
315 #ifdef CONFIG_MEMORY_HOTPLUG
317 * Decode mem_map from the coded memmap
319 struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
321 /* mask off the extra low bits of information */
322 coded_mem_map &= SECTION_MAP_MASK;
323 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
325 #endif /* CONFIG_MEMORY_HOTPLUG */
327 static void __meminit sparse_init_one_section(struct mem_section *ms,
328 unsigned long pnum, struct page *mem_map,
329 struct mem_section_usage *usage, unsigned long flags)
331 ms->section_mem_map &= ~SECTION_MAP_MASK;
332 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum)
333 | SECTION_HAS_MEM_MAP | flags;
337 static unsigned long usemap_size(void)
339 return BITS_TO_LONGS(SECTION_BLOCKFLAGS_BITS) * sizeof(unsigned long);
342 size_t mem_section_usage_size(void)
344 return sizeof(struct mem_section_usage) + usemap_size();
347 static inline phys_addr_t pgdat_to_phys(struct pglist_data *pgdat)
350 return __pa_symbol(pgdat);
356 #ifdef CONFIG_MEMORY_HOTREMOVE
357 static struct mem_section_usage * __init
358 sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
361 struct mem_section_usage *usage;
362 unsigned long goal, limit;
365 * A page may contain usemaps for other sections preventing the
366 * page being freed and making a section unremovable while
367 * other sections referencing the usemap remain active. Similarly,
368 * a pgdat can prevent a section being removed. If section A
369 * contains a pgdat and section B contains the usemap, both
370 * sections become inter-dependent. This allocates usemaps
371 * from the same section as the pgdat where possible to avoid
374 goal = pgdat_to_phys(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT);
375 limit = goal + (1UL << PA_SECTION_SHIFT);
376 nid = early_pfn_to_nid(goal >> PAGE_SHIFT);
378 usage = memblock_alloc_try_nid(size, SMP_CACHE_BYTES, goal, limit, nid);
379 if (!usage && limit) {
386 static void __init check_usemap_section_nr(int nid,
387 struct mem_section_usage *usage)
389 unsigned long usemap_snr, pgdat_snr;
390 static unsigned long old_usemap_snr;
391 static unsigned long old_pgdat_snr;
392 struct pglist_data *pgdat = NODE_DATA(nid);
396 if (!old_usemap_snr) {
397 old_usemap_snr = NR_MEM_SECTIONS;
398 old_pgdat_snr = NR_MEM_SECTIONS;
401 usemap_snr = pfn_to_section_nr(__pa(usage) >> PAGE_SHIFT);
402 pgdat_snr = pfn_to_section_nr(pgdat_to_phys(pgdat) >> PAGE_SHIFT);
403 if (usemap_snr == pgdat_snr)
406 if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr)
407 /* skip redundant message */
410 old_usemap_snr = usemap_snr;
411 old_pgdat_snr = pgdat_snr;
413 usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr));
414 if (usemap_nid != nid) {
415 pr_info("node %d must be removed before remove section %ld\n",
420 * There is a circular dependency.
421 * Some platforms allow un-removable section because they will just
422 * gather other removable sections for dynamic partitioning.
423 * Just notify un-removable section's number here.
425 pr_info("Section %ld and %ld (node %d) have a circular dependency on usemap and pgdat allocations\n",
426 usemap_snr, pgdat_snr, nid);
429 static struct mem_section_usage * __init
430 sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
433 return memblock_alloc_node(size, SMP_CACHE_BYTES, pgdat->node_id);
436 static void __init check_usemap_section_nr(int nid,
437 struct mem_section_usage *usage)
440 #endif /* CONFIG_MEMORY_HOTREMOVE */
442 #ifdef CONFIG_SPARSEMEM_VMEMMAP
443 static unsigned long __init section_map_size(void)
445 return ALIGN(sizeof(struct page) * PAGES_PER_SECTION, PMD_SIZE);
449 static unsigned long __init section_map_size(void)
451 return PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
454 struct page __init *__populate_section_memmap(unsigned long pfn,
455 unsigned long nr_pages, int nid, struct vmem_altmap *altmap)
457 unsigned long size = section_map_size();
458 struct page *map = sparse_buffer_alloc(size);
459 phys_addr_t addr = __pa(MAX_DMA_ADDRESS);
464 map = memblock_alloc_try_nid_raw(size, size, addr,
465 MEMBLOCK_ALLOC_ACCESSIBLE, nid);
467 panic("%s: Failed to allocate %lu bytes align=0x%lx nid=%d from=%pa\n",
468 __func__, size, PAGE_SIZE, nid, &addr);
472 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
474 static void *sparsemap_buf __meminitdata;
475 static void *sparsemap_buf_end __meminitdata;
477 static inline void __meminit sparse_buffer_free(unsigned long size)
479 WARN_ON(!sparsemap_buf || size == 0);
480 memblock_free_early(__pa(sparsemap_buf), size);
483 static void __init sparse_buffer_init(unsigned long size, int nid)
485 phys_addr_t addr = __pa(MAX_DMA_ADDRESS);
486 WARN_ON(sparsemap_buf); /* forgot to call sparse_buffer_fini()? */
488 * Pre-allocated buffer is mainly used by __populate_section_memmap
489 * and we want it to be properly aligned to the section size - this is
490 * especially the case for VMEMMAP which maps memmap to PMDs
492 sparsemap_buf = memblock_alloc_exact_nid_raw(size, section_map_size(),
493 addr, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
494 sparsemap_buf_end = sparsemap_buf + size;
497 static void __init sparse_buffer_fini(void)
499 unsigned long size = sparsemap_buf_end - sparsemap_buf;
501 if (sparsemap_buf && size > 0)
502 sparse_buffer_free(size);
503 sparsemap_buf = NULL;
506 void * __meminit sparse_buffer_alloc(unsigned long size)
511 ptr = (void *) roundup((unsigned long)sparsemap_buf, size);
512 if (ptr + size > sparsemap_buf_end)
515 /* Free redundant aligned space */
516 if ((unsigned long)(ptr - sparsemap_buf) > 0)
517 sparse_buffer_free((unsigned long)(ptr - sparsemap_buf));
518 sparsemap_buf = ptr + size;
524 void __weak __meminit vmemmap_populate_print_last(void)
529 * Initialize sparse on a specific node. The node spans [pnum_begin, pnum_end)
530 * And number of present sections in this node is map_count.
532 static void __init sparse_init_nid(int nid, unsigned long pnum_begin,
533 unsigned long pnum_end,
534 unsigned long map_count)
536 struct mem_section_usage *usage;
540 usage = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nid),
541 mem_section_usage_size() * map_count);
543 pr_err("%s: node[%d] usemap allocation failed", __func__, nid);
546 sparse_buffer_init(map_count * section_map_size(), nid);
547 for_each_present_section_nr(pnum_begin, pnum) {
548 unsigned long pfn = section_nr_to_pfn(pnum);
550 if (pnum >= pnum_end)
553 map = __populate_section_memmap(pfn, PAGES_PER_SECTION,
556 pr_err("%s: node[%d] memory map backing failed. Some memory will not be available.",
559 sparse_buffer_fini();
562 check_usemap_section_nr(nid, usage);
563 sparse_init_one_section(__nr_to_section(pnum), pnum, map, usage,
565 usage = (void *) usage + mem_section_usage_size();
567 sparse_buffer_fini();
570 /* We failed to allocate, mark all the following pnums as not present */
571 for_each_present_section_nr(pnum_begin, pnum) {
572 struct mem_section *ms;
574 if (pnum >= pnum_end)
576 ms = __nr_to_section(pnum);
577 ms->section_mem_map = 0;
582 * Allocate the accumulated non-linear sections, allocate a mem_map
583 * for each and record the physical to section mapping.
585 void __init sparse_init(void)
587 unsigned long pnum_end, pnum_begin, map_count = 1;
592 pnum_begin = first_present_section_nr();
593 nid_begin = sparse_early_nid(__nr_to_section(pnum_begin));
595 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
596 set_pageblock_order();
598 for_each_present_section_nr(pnum_begin + 1, pnum_end) {
599 int nid = sparse_early_nid(__nr_to_section(pnum_end));
601 if (nid == nid_begin) {
605 /* Init node with sections in range [pnum_begin, pnum_end) */
606 sparse_init_nid(nid_begin, pnum_begin, pnum_end, map_count);
608 pnum_begin = pnum_end;
611 /* cover the last node */
612 sparse_init_nid(nid_begin, pnum_begin, pnum_end, map_count);
613 vmemmap_populate_print_last();
616 #ifdef CONFIG_MEMORY_HOTPLUG
618 /* Mark all memory sections within the pfn range as online */
619 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
623 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
624 unsigned long section_nr = pfn_to_section_nr(pfn);
625 struct mem_section *ms;
627 /* onlining code should never touch invalid ranges */
628 if (WARN_ON(!valid_section_nr(section_nr)))
631 ms = __nr_to_section(section_nr);
632 ms->section_mem_map |= SECTION_IS_ONLINE;
636 /* Mark all memory sections within the pfn range as offline */
637 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
641 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
642 unsigned long section_nr = pfn_to_section_nr(pfn);
643 struct mem_section *ms;
646 * TODO this needs some double checking. Offlining code makes
647 * sure to check pfn_valid but those checks might be just bogus
649 if (WARN_ON(!valid_section_nr(section_nr)))
652 ms = __nr_to_section(section_nr);
653 ms->section_mem_map &= ~SECTION_IS_ONLINE;
657 #ifdef CONFIG_SPARSEMEM_VMEMMAP
658 static struct page * __meminit populate_section_memmap(unsigned long pfn,
659 unsigned long nr_pages, int nid, struct vmem_altmap *altmap)
661 return __populate_section_memmap(pfn, nr_pages, nid, altmap);
664 static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
665 struct vmem_altmap *altmap)
667 unsigned long start = (unsigned long) pfn_to_page(pfn);
668 unsigned long end = start + nr_pages * sizeof(struct page);
670 vmemmap_free(start, end, altmap);
672 static void free_map_bootmem(struct page *memmap)
674 unsigned long start = (unsigned long)memmap;
675 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
677 vmemmap_free(start, end, NULL);
680 static int clear_subsection_map(unsigned long pfn, unsigned long nr_pages)
682 DECLARE_BITMAP(map, SUBSECTIONS_PER_SECTION) = { 0 };
683 DECLARE_BITMAP(tmp, SUBSECTIONS_PER_SECTION) = { 0 };
684 struct mem_section *ms = __pfn_to_section(pfn);
685 unsigned long *subsection_map = ms->usage
686 ? &ms->usage->subsection_map[0] : NULL;
688 subsection_mask_set(map, pfn, nr_pages);
690 bitmap_and(tmp, map, subsection_map, SUBSECTIONS_PER_SECTION);
692 if (WARN(!subsection_map || !bitmap_equal(tmp, map, SUBSECTIONS_PER_SECTION),
693 "section already deactivated (%#lx + %ld)\n",
697 bitmap_xor(subsection_map, map, subsection_map, SUBSECTIONS_PER_SECTION);
701 static bool is_subsection_map_empty(struct mem_section *ms)
703 return bitmap_empty(&ms->usage->subsection_map[0],
704 SUBSECTIONS_PER_SECTION);
707 static int fill_subsection_map(unsigned long pfn, unsigned long nr_pages)
709 struct mem_section *ms = __pfn_to_section(pfn);
710 DECLARE_BITMAP(map, SUBSECTIONS_PER_SECTION) = { 0 };
711 unsigned long *subsection_map;
714 subsection_mask_set(map, pfn, nr_pages);
716 subsection_map = &ms->usage->subsection_map[0];
718 if (bitmap_empty(map, SUBSECTIONS_PER_SECTION))
720 else if (bitmap_intersects(map, subsection_map, SUBSECTIONS_PER_SECTION))
723 bitmap_or(subsection_map, map, subsection_map,
724 SUBSECTIONS_PER_SECTION);
729 struct page * __meminit populate_section_memmap(unsigned long pfn,
730 unsigned long nr_pages, int nid, struct vmem_altmap *altmap)
732 return kvmalloc_node(array_size(sizeof(struct page),
733 PAGES_PER_SECTION), GFP_KERNEL, nid);
736 static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
737 struct vmem_altmap *altmap)
739 kvfree(pfn_to_page(pfn));
742 static void free_map_bootmem(struct page *memmap)
744 unsigned long maps_section_nr, removing_section_nr, i;
745 unsigned long magic, nr_pages;
746 struct page *page = virt_to_page(memmap);
748 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
751 for (i = 0; i < nr_pages; i++, page++) {
752 magic = (unsigned long) page->freelist;
754 BUG_ON(magic == NODE_INFO);
756 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
757 removing_section_nr = page_private(page);
760 * When this function is called, the removing section is
761 * logical offlined state. This means all pages are isolated
762 * from page allocator. If removing section's memmap is placed
763 * on the same section, it must not be freed.
764 * If it is freed, page allocator may allocate it which will
765 * be removed physically soon.
767 if (maps_section_nr != removing_section_nr)
768 put_page_bootmem(page);
772 static int clear_subsection_map(unsigned long pfn, unsigned long nr_pages)
777 static bool is_subsection_map_empty(struct mem_section *ms)
782 static int fill_subsection_map(unsigned long pfn, unsigned long nr_pages)
786 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
789 * To deactivate a memory region, there are 3 cases to handle across
790 * two configurations (SPARSEMEM_VMEMMAP={y,n}):
792 * 1. deactivation of a partial hot-added section (only possible in
793 * the SPARSEMEM_VMEMMAP=y case).
794 * a) section was present at memory init.
795 * b) section was hot-added post memory init.
796 * 2. deactivation of a complete hot-added section.
797 * 3. deactivation of a complete section from memory init.
799 * For 1, when subsection_map does not empty we will not be freeing the
800 * usage map, but still need to free the vmemmap range.
802 * For 2 and 3, the SPARSEMEM_VMEMMAP={y,n} cases are unified
804 static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
805 struct vmem_altmap *altmap)
807 struct mem_section *ms = __pfn_to_section(pfn);
808 bool section_is_early = early_section(ms);
809 struct page *memmap = NULL;
812 if (clear_subsection_map(pfn, nr_pages))
815 empty = is_subsection_map_empty(ms);
817 unsigned long section_nr = pfn_to_section_nr(pfn);
820 * When removing an early section, the usage map is kept (as the
821 * usage maps of other sections fall into the same page). It
822 * will be re-used when re-adding the section - which is then no
823 * longer an early section. If the usage map is PageReserved, it
824 * was allocated during boot.
826 if (!PageReserved(virt_to_page(ms->usage))) {
830 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
832 * Mark the section invalid so that valid_section()
833 * return false. This prevents code from dereferencing
836 ms->section_mem_map &= ~SECTION_HAS_MEM_MAP;
840 * The memmap of early sections is always fully populated. See
841 * section_activate() and pfn_valid() .
843 if (!section_is_early)
844 depopulate_section_memmap(pfn, nr_pages, altmap);
846 free_map_bootmem(memmap);
849 ms->section_mem_map = (unsigned long)NULL;
852 static struct page * __meminit section_activate(int nid, unsigned long pfn,
853 unsigned long nr_pages, struct vmem_altmap *altmap)
855 struct mem_section *ms = __pfn_to_section(pfn);
856 struct mem_section_usage *usage = NULL;
861 usage = kzalloc(mem_section_usage_size(), GFP_KERNEL);
863 return ERR_PTR(-ENOMEM);
867 rc = fill_subsection_map(pfn, nr_pages);
876 * The early init code does not consider partially populated
877 * initial sections, it simply assumes that memory will never be
878 * referenced. If we hot-add memory into such a section then we
879 * do not need to populate the memmap and can simply reuse what
882 if (nr_pages < PAGES_PER_SECTION && early_section(ms))
883 return pfn_to_page(pfn);
885 memmap = populate_section_memmap(pfn, nr_pages, nid, altmap);
887 section_deactivate(pfn, nr_pages, altmap);
888 return ERR_PTR(-ENOMEM);
895 * sparse_add_section - add a memory section, or populate an existing one
896 * @nid: The node to add section on
897 * @start_pfn: start pfn of the memory range
898 * @nr_pages: number of pfns to add in the section
899 * @altmap: device page map
901 * This is only intended for hotplug.
903 * Note that only VMEMMAP supports sub-section aligned hotplug,
904 * the proper alignment and size are gated by check_pfn_span().
909 * * -EEXIST - Section has been present.
910 * * -ENOMEM - Out of memory.
912 int __meminit sparse_add_section(int nid, unsigned long start_pfn,
913 unsigned long nr_pages, struct vmem_altmap *altmap)
915 unsigned long section_nr = pfn_to_section_nr(start_pfn);
916 struct mem_section *ms;
920 ret = sparse_index_init(section_nr, nid);
924 memmap = section_activate(nid, start_pfn, nr_pages, altmap);
926 return PTR_ERR(memmap);
929 * Poison uninitialized struct pages in order to catch invalid flags
932 page_init_poison(memmap, sizeof(struct page) * nr_pages);
934 ms = __nr_to_section(section_nr);
935 set_section_nid(section_nr, nid);
936 __section_mark_present(ms, section_nr);
938 /* Align memmap to section boundary in the subsection case */
939 if (section_nr_to_pfn(section_nr) != start_pfn)
940 memmap = pfn_to_page(section_nr_to_pfn(section_nr));
941 sparse_init_one_section(ms, section_nr, memmap, ms->usage, 0);
946 #ifdef CONFIG_MEMORY_FAILURE
947 static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
952 * A further optimization is to have per section refcounted
953 * num_poisoned_pages. But that would need more space per memmap, so
954 * for now just do a quick global check to speed up this routine in the
955 * absence of bad pages.
957 if (atomic_long_read(&num_poisoned_pages) == 0)
960 for (i = 0; i < nr_pages; i++) {
961 if (PageHWPoison(&memmap[i])) {
962 num_poisoned_pages_dec();
963 ClearPageHWPoison(&memmap[i]);
968 static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
973 void sparse_remove_section(struct mem_section *ms, unsigned long pfn,
974 unsigned long nr_pages, unsigned long map_offset,
975 struct vmem_altmap *altmap)
977 clear_hwpoisoned_pages(pfn_to_page(pfn) + map_offset,
978 nr_pages - map_offset);
979 section_deactivate(pfn, nr_pages, altmap);
981 #endif /* CONFIG_MEMORY_HOTPLUG */