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>
21 * Permanent SPARSEMEM data:
23 * 1) mem_section - memory sections, mem_map's for valid memory
25 #ifdef CONFIG_SPARSEMEM_EXTREME
26 struct mem_section **mem_section;
28 struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]
29 ____cacheline_internodealigned_in_smp;
31 EXPORT_SYMBOL(mem_section);
33 #ifdef NODE_NOT_IN_PAGE_FLAGS
35 * If we did not store the node number in the page then we have to
36 * do a lookup in the section_to_node_table in order to find which
37 * node the page belongs to.
39 #if MAX_NUMNODES <= 256
40 static u8 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
42 static u16 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
45 int page_to_nid(const struct page *page)
47 return section_to_node_table[page_to_section(page)];
49 EXPORT_SYMBOL(page_to_nid);
51 static void set_section_nid(unsigned long section_nr, int nid)
53 section_to_node_table[section_nr] = nid;
55 #else /* !NODE_NOT_IN_PAGE_FLAGS */
56 static inline void set_section_nid(unsigned long section_nr, int nid)
61 #ifdef CONFIG_SPARSEMEM_EXTREME
62 static noinline struct mem_section __ref *sparse_index_alloc(int nid)
64 struct mem_section *section = NULL;
65 unsigned long array_size = SECTIONS_PER_ROOT *
66 sizeof(struct mem_section);
68 if (slab_is_available()) {
69 section = kzalloc_node(array_size, GFP_KERNEL, nid);
71 section = memblock_alloc_node(array_size, SMP_CACHE_BYTES,
74 panic("%s: Failed to allocate %lu bytes nid=%d\n",
75 __func__, array_size, nid);
81 static int __meminit sparse_index_init(unsigned long section_nr, int nid)
83 unsigned long root = SECTION_NR_TO_ROOT(section_nr);
84 struct mem_section *section;
87 * An existing section is possible in the sub-section hotplug
88 * case. First hot-add instantiates, follow-on hot-add reuses
89 * the existing section.
91 * The mem_hotplug_lock resolves the apparent race below.
93 if (mem_section[root])
96 section = sparse_index_alloc(nid);
100 mem_section[root] = section;
104 #else /* !SPARSEMEM_EXTREME */
105 static inline int sparse_index_init(unsigned long section_nr, int nid)
111 #ifdef CONFIG_SPARSEMEM_EXTREME
112 unsigned long __section_nr(struct mem_section *ms)
114 unsigned long root_nr;
115 struct mem_section *root = NULL;
117 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
118 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
122 if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT)))
128 return (root_nr * SECTIONS_PER_ROOT) + (ms - root);
131 unsigned long __section_nr(struct mem_section *ms)
133 return (unsigned long)(ms - mem_section[0]);
138 * During early boot, before section_mem_map is used for an actual
139 * mem_map, we use section_mem_map to store the section's NUMA
140 * node. This keeps us from having to use another data structure. The
141 * node information is cleared just before we store the real mem_map.
143 static inline unsigned long sparse_encode_early_nid(int nid)
145 return (nid << SECTION_NID_SHIFT);
148 static inline int sparse_early_nid(struct mem_section *section)
150 return (section->section_mem_map >> SECTION_NID_SHIFT);
153 /* Validate the physical addressing limitations of the model */
154 void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
155 unsigned long *end_pfn)
157 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
160 * Sanity checks - do not allow an architecture to pass
161 * in larger pfns than the maximum scope of sparsemem:
163 if (*start_pfn > max_sparsemem_pfn) {
164 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
165 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
166 *start_pfn, *end_pfn, max_sparsemem_pfn);
168 *start_pfn = max_sparsemem_pfn;
169 *end_pfn = max_sparsemem_pfn;
170 } else if (*end_pfn > max_sparsemem_pfn) {
171 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
172 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
173 *start_pfn, *end_pfn, max_sparsemem_pfn);
175 *end_pfn = max_sparsemem_pfn;
180 * There are a number of times that we loop over NR_MEM_SECTIONS,
181 * looking for section_present() on each. But, when we have very
182 * large physical address spaces, NR_MEM_SECTIONS can also be
183 * very large which makes the loops quite long.
185 * Keeping track of this gives us an easy way to break out of
188 unsigned long __highest_present_section_nr;
189 static void section_mark_present(struct mem_section *ms)
191 unsigned long section_nr = __section_nr(ms);
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);
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;
316 * Decode mem_map from the coded memmap
318 struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
320 /* mask off the extra low bits of information */
321 coded_mem_map &= SECTION_MAP_MASK;
322 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
325 static void __meminit sparse_init_one_section(struct mem_section *ms,
326 unsigned long pnum, struct page *mem_map,
327 struct mem_section_usage *usage, unsigned long flags)
329 ms->section_mem_map &= ~SECTION_MAP_MASK;
330 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum)
331 | SECTION_HAS_MEM_MAP | flags;
335 static unsigned long usemap_size(void)
337 return BITS_TO_LONGS(SECTION_BLOCKFLAGS_BITS) * sizeof(unsigned long);
340 size_t mem_section_usage_size(void)
342 return sizeof(struct mem_section_usage) + usemap_size();
345 #ifdef CONFIG_MEMORY_HOTREMOVE
346 static struct mem_section_usage * __init
347 sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
350 struct mem_section_usage *usage;
351 unsigned long goal, limit;
354 * A page may contain usemaps for other sections preventing the
355 * page being freed and making a section unremovable while
356 * other sections referencing the usemap remain active. Similarly,
357 * a pgdat can prevent a section being removed. If section A
358 * contains a pgdat and section B contains the usemap, both
359 * sections become inter-dependent. This allocates usemaps
360 * from the same section as the pgdat where possible to avoid
363 goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT);
364 limit = goal + (1UL << PA_SECTION_SHIFT);
365 nid = early_pfn_to_nid(goal >> PAGE_SHIFT);
367 usage = memblock_alloc_try_nid(size, SMP_CACHE_BYTES, goal, limit, nid);
368 if (!usage && limit) {
375 static void __init check_usemap_section_nr(int nid,
376 struct mem_section_usage *usage)
378 unsigned long usemap_snr, pgdat_snr;
379 static unsigned long old_usemap_snr;
380 static unsigned long old_pgdat_snr;
381 struct pglist_data *pgdat = NODE_DATA(nid);
385 if (!old_usemap_snr) {
386 old_usemap_snr = NR_MEM_SECTIONS;
387 old_pgdat_snr = NR_MEM_SECTIONS;
390 usemap_snr = pfn_to_section_nr(__pa(usage) >> PAGE_SHIFT);
391 pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
392 if (usemap_snr == pgdat_snr)
395 if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr)
396 /* skip redundant message */
399 old_usemap_snr = usemap_snr;
400 old_pgdat_snr = pgdat_snr;
402 usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr));
403 if (usemap_nid != nid) {
404 pr_info("node %d must be removed before remove section %ld\n",
409 * There is a circular dependency.
410 * Some platforms allow un-removable section because they will just
411 * gather other removable sections for dynamic partitioning.
412 * Just notify un-removable section's number here.
414 pr_info("Section %ld and %ld (node %d) have a circular dependency on usemap and pgdat allocations\n",
415 usemap_snr, pgdat_snr, nid);
418 static struct mem_section_usage * __init
419 sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
422 return memblock_alloc_node(size, SMP_CACHE_BYTES, pgdat->node_id);
425 static void __init check_usemap_section_nr(int nid,
426 struct mem_section_usage *usage)
429 #endif /* CONFIG_MEMORY_HOTREMOVE */
431 #ifdef CONFIG_SPARSEMEM_VMEMMAP
432 static unsigned long __init section_map_size(void)
434 return ALIGN(sizeof(struct page) * PAGES_PER_SECTION, PMD_SIZE);
438 static unsigned long __init section_map_size(void)
440 return PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
443 struct page __init *__populate_section_memmap(unsigned long pfn,
444 unsigned long nr_pages, int nid, struct vmem_altmap *altmap)
446 unsigned long size = section_map_size();
447 struct page *map = sparse_buffer_alloc(size);
448 phys_addr_t addr = __pa(MAX_DMA_ADDRESS);
453 map = memblock_alloc_try_nid_raw(size, size, addr,
454 MEMBLOCK_ALLOC_ACCESSIBLE, nid);
456 panic("%s: Failed to allocate %lu bytes align=0x%lx nid=%d from=%pa\n",
457 __func__, size, PAGE_SIZE, nid, &addr);
461 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
463 static void *sparsemap_buf __meminitdata;
464 static void *sparsemap_buf_end __meminitdata;
466 static inline void __meminit sparse_buffer_free(unsigned long size)
468 WARN_ON(!sparsemap_buf || size == 0);
469 memblock_free_early(__pa(sparsemap_buf), size);
472 static void __init sparse_buffer_init(unsigned long size, int nid)
474 phys_addr_t addr = __pa(MAX_DMA_ADDRESS);
475 WARN_ON(sparsemap_buf); /* forgot to call sparse_buffer_fini()? */
477 * Pre-allocated buffer is mainly used by __populate_section_memmap
478 * and we want it to be properly aligned to the section size - this is
479 * especially the case for VMEMMAP which maps memmap to PMDs
481 sparsemap_buf = memblock_alloc_exact_nid_raw(size, section_map_size(),
482 addr, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
483 sparsemap_buf_end = sparsemap_buf + size;
486 static void __init sparse_buffer_fini(void)
488 unsigned long size = sparsemap_buf_end - sparsemap_buf;
490 if (sparsemap_buf && size > 0)
491 sparse_buffer_free(size);
492 sparsemap_buf = NULL;
495 void * __meminit sparse_buffer_alloc(unsigned long size)
500 ptr = (void *) roundup((unsigned long)sparsemap_buf, size);
501 if (ptr + size > sparsemap_buf_end)
504 /* Free redundant aligned space */
505 if ((unsigned long)(ptr - sparsemap_buf) > 0)
506 sparse_buffer_free((unsigned long)(ptr - sparsemap_buf));
507 sparsemap_buf = ptr + size;
513 void __weak __meminit vmemmap_populate_print_last(void)
518 * Initialize sparse on a specific node. The node spans [pnum_begin, pnum_end)
519 * And number of present sections in this node is map_count.
521 static void __init sparse_init_nid(int nid, unsigned long pnum_begin,
522 unsigned long pnum_end,
523 unsigned long map_count)
525 struct mem_section_usage *usage;
529 usage = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nid),
530 mem_section_usage_size() * map_count);
532 pr_err("%s: node[%d] usemap allocation failed", __func__, nid);
535 sparse_buffer_init(map_count * section_map_size(), nid);
536 for_each_present_section_nr(pnum_begin, pnum) {
537 unsigned long pfn = section_nr_to_pfn(pnum);
539 if (pnum >= pnum_end)
542 map = __populate_section_memmap(pfn, PAGES_PER_SECTION,
545 pr_err("%s: node[%d] memory map backing failed. Some memory will not be available.",
550 check_usemap_section_nr(nid, usage);
551 sparse_init_one_section(__nr_to_section(pnum), pnum, map, usage,
553 usage = (void *) usage + mem_section_usage_size();
555 sparse_buffer_fini();
558 /* We failed to allocate, mark all the following pnums as not present */
559 for_each_present_section_nr(pnum_begin, pnum) {
560 struct mem_section *ms;
562 if (pnum >= pnum_end)
564 ms = __nr_to_section(pnum);
565 ms->section_mem_map = 0;
570 * Allocate the accumulated non-linear sections, allocate a mem_map
571 * for each and record the physical to section mapping.
573 void __init sparse_init(void)
575 unsigned long pnum_end, pnum_begin, map_count = 1;
580 pnum_begin = first_present_section_nr();
581 nid_begin = sparse_early_nid(__nr_to_section(pnum_begin));
583 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
584 set_pageblock_order();
586 for_each_present_section_nr(pnum_begin + 1, pnum_end) {
587 int nid = sparse_early_nid(__nr_to_section(pnum_end));
589 if (nid == nid_begin) {
593 /* Init node with sections in range [pnum_begin, pnum_end) */
594 sparse_init_nid(nid_begin, pnum_begin, pnum_end, map_count);
596 pnum_begin = pnum_end;
599 /* cover the last node */
600 sparse_init_nid(nid_begin, pnum_begin, pnum_end, map_count);
601 vmemmap_populate_print_last();
604 #ifdef CONFIG_MEMORY_HOTPLUG
606 /* Mark all memory sections within the pfn range as online */
607 void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
611 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
612 unsigned long section_nr = pfn_to_section_nr(pfn);
613 struct mem_section *ms;
615 /* onlining code should never touch invalid ranges */
616 if (WARN_ON(!valid_section_nr(section_nr)))
619 ms = __nr_to_section(section_nr);
620 ms->section_mem_map |= SECTION_IS_ONLINE;
624 #ifdef CONFIG_MEMORY_HOTREMOVE
625 /* Mark all memory sections within the pfn range as offline */
626 void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
630 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
631 unsigned long section_nr = pfn_to_section_nr(pfn);
632 struct mem_section *ms;
635 * TODO this needs some double checking. Offlining code makes
636 * sure to check pfn_valid but those checks might be just bogus
638 if (WARN_ON(!valid_section_nr(section_nr)))
641 ms = __nr_to_section(section_nr);
642 ms->section_mem_map &= ~SECTION_IS_ONLINE;
647 #ifdef CONFIG_SPARSEMEM_VMEMMAP
648 static struct page * __meminit populate_section_memmap(unsigned long pfn,
649 unsigned long nr_pages, int nid, struct vmem_altmap *altmap)
651 return __populate_section_memmap(pfn, nr_pages, nid, altmap);
654 static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
655 struct vmem_altmap *altmap)
657 unsigned long start = (unsigned long) pfn_to_page(pfn);
658 unsigned long end = start + nr_pages * sizeof(struct page);
660 vmemmap_free(start, end, altmap);
662 static void free_map_bootmem(struct page *memmap)
664 unsigned long start = (unsigned long)memmap;
665 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
667 vmemmap_free(start, end, NULL);
670 static int clear_subsection_map(unsigned long pfn, unsigned long nr_pages)
672 DECLARE_BITMAP(map, SUBSECTIONS_PER_SECTION) = { 0 };
673 DECLARE_BITMAP(tmp, SUBSECTIONS_PER_SECTION) = { 0 };
674 struct mem_section *ms = __pfn_to_section(pfn);
675 unsigned long *subsection_map = ms->usage
676 ? &ms->usage->subsection_map[0] : NULL;
678 subsection_mask_set(map, pfn, nr_pages);
680 bitmap_and(tmp, map, subsection_map, SUBSECTIONS_PER_SECTION);
682 if (WARN(!subsection_map || !bitmap_equal(tmp, map, SUBSECTIONS_PER_SECTION),
683 "section already deactivated (%#lx + %ld)\n",
687 bitmap_xor(subsection_map, map, subsection_map, SUBSECTIONS_PER_SECTION);
691 static bool is_subsection_map_empty(struct mem_section *ms)
693 return bitmap_empty(&ms->usage->subsection_map[0],
694 SUBSECTIONS_PER_SECTION);
697 static int fill_subsection_map(unsigned long pfn, unsigned long nr_pages)
699 struct mem_section *ms = __pfn_to_section(pfn);
700 DECLARE_BITMAP(map, SUBSECTIONS_PER_SECTION) = { 0 };
701 unsigned long *subsection_map;
704 subsection_mask_set(map, pfn, nr_pages);
706 subsection_map = &ms->usage->subsection_map[0];
708 if (bitmap_empty(map, SUBSECTIONS_PER_SECTION))
710 else if (bitmap_intersects(map, subsection_map, SUBSECTIONS_PER_SECTION))
713 bitmap_or(subsection_map, map, subsection_map,
714 SUBSECTIONS_PER_SECTION);
719 struct page * __meminit populate_section_memmap(unsigned long pfn,
720 unsigned long nr_pages, int nid, struct vmem_altmap *altmap)
722 return kvmalloc_node(array_size(sizeof(struct page),
723 PAGES_PER_SECTION), GFP_KERNEL, nid);
726 static void depopulate_section_memmap(unsigned long pfn, unsigned long nr_pages,
727 struct vmem_altmap *altmap)
729 kvfree(pfn_to_page(pfn));
732 static void free_map_bootmem(struct page *memmap)
734 unsigned long maps_section_nr, removing_section_nr, i;
735 unsigned long magic, nr_pages;
736 struct page *page = virt_to_page(memmap);
738 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
741 for (i = 0; i < nr_pages; i++, page++) {
742 magic = (unsigned long) page->freelist;
744 BUG_ON(magic == NODE_INFO);
746 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
747 removing_section_nr = page_private(page);
750 * When this function is called, the removing section is
751 * logical offlined state. This means all pages are isolated
752 * from page allocator. If removing section's memmap is placed
753 * on the same section, it must not be freed.
754 * If it is freed, page allocator may allocate it which will
755 * be removed physically soon.
757 if (maps_section_nr != removing_section_nr)
758 put_page_bootmem(page);
762 static int clear_subsection_map(unsigned long pfn, unsigned long nr_pages)
767 static bool is_subsection_map_empty(struct mem_section *ms)
772 static int fill_subsection_map(unsigned long pfn, unsigned long nr_pages)
776 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
779 * To deactivate a memory region, there are 3 cases to handle across
780 * two configurations (SPARSEMEM_VMEMMAP={y,n}):
782 * 1. deactivation of a partial hot-added section (only possible in
783 * the SPARSEMEM_VMEMMAP=y case).
784 * a) section was present at memory init.
785 * b) section was hot-added post memory init.
786 * 2. deactivation of a complete hot-added section.
787 * 3. deactivation of a complete section from memory init.
789 * For 1, when subsection_map does not empty we will not be freeing the
790 * usage map, but still need to free the vmemmap range.
792 * For 2 and 3, the SPARSEMEM_VMEMMAP={y,n} cases are unified
794 static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
795 struct vmem_altmap *altmap)
797 struct mem_section *ms = __pfn_to_section(pfn);
798 bool section_is_early = early_section(ms);
799 struct page *memmap = NULL;
802 if (clear_subsection_map(pfn, nr_pages))
805 empty = is_subsection_map_empty(ms);
807 unsigned long section_nr = pfn_to_section_nr(pfn);
810 * When removing an early section, the usage map is kept (as the
811 * usage maps of other sections fall into the same page). It
812 * will be re-used when re-adding the section - which is then no
813 * longer an early section. If the usage map is PageReserved, it
814 * was allocated during boot.
816 if (!PageReserved(virt_to_page(ms->usage))) {
820 memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
822 * Mark the section invalid so that valid_section()
823 * return false. This prevents code from dereferencing
826 ms->section_mem_map &= ~SECTION_HAS_MEM_MAP;
830 * The memmap of early sections is always fully populated. See
831 * section_activate() and pfn_valid() .
833 if (!section_is_early)
834 depopulate_section_memmap(pfn, nr_pages, altmap);
836 free_map_bootmem(memmap);
839 ms->section_mem_map = (unsigned long)NULL;
842 static struct page * __meminit section_activate(int nid, unsigned long pfn,
843 unsigned long nr_pages, struct vmem_altmap *altmap)
845 struct mem_section *ms = __pfn_to_section(pfn);
846 struct mem_section_usage *usage = NULL;
851 usage = kzalloc(mem_section_usage_size(), GFP_KERNEL);
853 return ERR_PTR(-ENOMEM);
857 rc = fill_subsection_map(pfn, nr_pages);
866 * The early init code does not consider partially populated
867 * initial sections, it simply assumes that memory will never be
868 * referenced. If we hot-add memory into such a section then we
869 * do not need to populate the memmap and can simply reuse what
872 if (nr_pages < PAGES_PER_SECTION && early_section(ms))
873 return pfn_to_page(pfn);
875 memmap = populate_section_memmap(pfn, nr_pages, nid, altmap);
877 section_deactivate(pfn, nr_pages, altmap);
878 return ERR_PTR(-ENOMEM);
885 * sparse_add_section - add a memory section, or populate an existing one
886 * @nid: The node to add section on
887 * @start_pfn: start pfn of the memory range
888 * @nr_pages: number of pfns to add in the section
889 * @altmap: device page map
891 * This is only intended for hotplug.
893 * Note that only VMEMMAP supports sub-section aligned hotplug,
894 * the proper alignment and size are gated by check_pfn_span().
899 * * -EEXIST - Section has been present.
900 * * -ENOMEM - Out of memory.
902 int __meminit sparse_add_section(int nid, unsigned long start_pfn,
903 unsigned long nr_pages, struct vmem_altmap *altmap)
905 unsigned long section_nr = pfn_to_section_nr(start_pfn);
906 struct mem_section *ms;
910 ret = sparse_index_init(section_nr, nid);
914 memmap = section_activate(nid, start_pfn, nr_pages, altmap);
916 return PTR_ERR(memmap);
919 * Poison uninitialized struct pages in order to catch invalid flags
922 page_init_poison(memmap, sizeof(struct page) * nr_pages);
924 ms = __nr_to_section(section_nr);
925 set_section_nid(section_nr, nid);
926 section_mark_present(ms);
928 /* Align memmap to section boundary in the subsection case */
929 if (section_nr_to_pfn(section_nr) != start_pfn)
930 memmap = pfn_to_page(section_nr_to_pfn(section_nr));
931 sparse_init_one_section(ms, section_nr, memmap, ms->usage, 0);
936 #ifdef CONFIG_MEMORY_FAILURE
937 static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
942 * A further optimization is to have per section refcounted
943 * num_poisoned_pages. But that would need more space per memmap, so
944 * for now just do a quick global check to speed up this routine in the
945 * absence of bad pages.
947 if (atomic_long_read(&num_poisoned_pages) == 0)
950 for (i = 0; i < nr_pages; i++) {
951 if (PageHWPoison(&memmap[i])) {
952 num_poisoned_pages_dec();
953 ClearPageHWPoison(&memmap[i]);
958 static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
963 void sparse_remove_section(struct mem_section *ms, unsigned long pfn,
964 unsigned long nr_pages, unsigned long map_offset,
965 struct vmem_altmap *altmap)
967 clear_hwpoisoned_pages(pfn_to_page(pfn) + map_offset,
968 nr_pages - map_offset);
969 section_deactivate(pfn, nr_pages, altmap);
971 #endif /* CONFIG_MEMORY_HOTPLUG */