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>
17 #include <asm/pgalloc.h>
18 #include <asm/pgtable.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
;
86 if (mem_section
[root
])
89 section
= sparse_index_alloc(nid
);
93 mem_section
[root
] = section
;
97 #else /* !SPARSEMEM_EXTREME */
98 static inline int sparse_index_init(unsigned long section_nr
, int nid
)
104 #ifdef CONFIG_SPARSEMEM_EXTREME
105 unsigned long __section_nr(struct mem_section
*ms
)
107 unsigned long root_nr
;
108 struct mem_section
*root
= NULL
;
110 for (root_nr
= 0; root_nr
< NR_SECTION_ROOTS
; root_nr
++) {
111 root
= __nr_to_section(root_nr
* SECTIONS_PER_ROOT
);
115 if ((ms
>= root
) && (ms
< (root
+ SECTIONS_PER_ROOT
)))
121 return (root_nr
* SECTIONS_PER_ROOT
) + (ms
- root
);
124 unsigned long __section_nr(struct mem_section
*ms
)
126 return (unsigned long)(ms
- mem_section
[0]);
131 * During early boot, before section_mem_map is used for an actual
132 * mem_map, we use section_mem_map to store the section's NUMA
133 * node. This keeps us from having to use another data structure. The
134 * node information is cleared just before we store the real mem_map.
136 static inline unsigned long sparse_encode_early_nid(int nid
)
138 return (nid
<< SECTION_NID_SHIFT
);
141 static inline int sparse_early_nid(struct mem_section
*section
)
143 return (section
->section_mem_map
>> SECTION_NID_SHIFT
);
146 /* Validate the physical addressing limitations of the model */
147 void __meminit
mminit_validate_memmodel_limits(unsigned long *start_pfn
,
148 unsigned long *end_pfn
)
150 unsigned long max_sparsemem_pfn
= 1UL << (MAX_PHYSMEM_BITS
-PAGE_SHIFT
);
153 * Sanity checks - do not allow an architecture to pass
154 * in larger pfns than the maximum scope of sparsemem:
156 if (*start_pfn
> max_sparsemem_pfn
) {
157 mminit_dprintk(MMINIT_WARNING
, "pfnvalidation",
158 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
159 *start_pfn
, *end_pfn
, max_sparsemem_pfn
);
161 *start_pfn
= max_sparsemem_pfn
;
162 *end_pfn
= max_sparsemem_pfn
;
163 } else if (*end_pfn
> max_sparsemem_pfn
) {
164 mminit_dprintk(MMINIT_WARNING
, "pfnvalidation",
165 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
166 *start_pfn
, *end_pfn
, max_sparsemem_pfn
);
168 *end_pfn
= max_sparsemem_pfn
;
173 * There are a number of times that we loop over NR_MEM_SECTIONS,
174 * looking for section_present() on each. But, when we have very
175 * large physical address spaces, NR_MEM_SECTIONS can also be
176 * very large which makes the loops quite long.
178 * Keeping track of this gives us an easy way to break out of
181 unsigned long __highest_present_section_nr
;
182 static void section_mark_present(struct mem_section
*ms
)
184 unsigned long section_nr
= __section_nr(ms
);
186 if (section_nr
> __highest_present_section_nr
)
187 __highest_present_section_nr
= section_nr
;
189 ms
->section_mem_map
|= SECTION_MARKED_PRESENT
;
192 static inline unsigned long next_present_section_nr(unsigned long section_nr
)
196 if (present_section_nr(section_nr
))
198 } while ((section_nr
<= __highest_present_section_nr
));
202 #define for_each_present_section_nr(start, section_nr) \
203 for (section_nr = next_present_section_nr(start-1); \
204 ((section_nr != -1) && \
205 (section_nr <= __highest_present_section_nr)); \
206 section_nr = next_present_section_nr(section_nr))
208 static inline unsigned long first_present_section_nr(void)
210 return next_present_section_nr(-1);
213 void subsection_mask_set(unsigned long *map
, unsigned long pfn
,
214 unsigned long nr_pages
)
216 int idx
= subsection_map_index(pfn
);
217 int end
= subsection_map_index(pfn
+ nr_pages
- 1);
219 bitmap_set(map
, idx
, end
- idx
+ 1);
222 void __init
subsection_map_init(unsigned long pfn
, unsigned long nr_pages
)
224 int end_sec
= pfn_to_section_nr(pfn
+ nr_pages
- 1);
225 int i
, start_sec
= pfn_to_section_nr(pfn
);
230 for (i
= start_sec
; i
<= end_sec
; i
++) {
231 struct mem_section
*ms
;
234 pfns
= min(nr_pages
, PAGES_PER_SECTION
235 - (pfn
& ~PAGE_SECTION_MASK
));
236 ms
= __nr_to_section(i
);
237 subsection_mask_set(ms
->usage
->subsection_map
, pfn
, pfns
);
239 pr_debug("%s: sec: %d pfns: %ld set(%d, %d)\n", __func__
, i
,
240 pfns
, subsection_map_index(pfn
),
241 subsection_map_index(pfn
+ pfns
- 1));
248 /* Record a memory area against a node. */
249 void __init
memory_present(int nid
, unsigned long start
, unsigned long end
)
253 #ifdef CONFIG_SPARSEMEM_EXTREME
254 if (unlikely(!mem_section
)) {
255 unsigned long size
, align
;
257 size
= sizeof(struct mem_section
*) * NR_SECTION_ROOTS
;
258 align
= 1 << (INTERNODE_CACHE_SHIFT
);
259 mem_section
= memblock_alloc(size
, align
);
261 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
262 __func__
, size
, align
);
266 start
&= PAGE_SECTION_MASK
;
267 mminit_validate_memmodel_limits(&start
, &end
);
268 for (pfn
= start
; pfn
< end
; pfn
+= PAGES_PER_SECTION
) {
269 unsigned long section
= pfn_to_section_nr(pfn
);
270 struct mem_section
*ms
;
272 sparse_index_init(section
, nid
);
273 set_section_nid(section
, nid
);
275 ms
= __nr_to_section(section
);
276 if (!ms
->section_mem_map
) {
277 ms
->section_mem_map
= sparse_encode_early_nid(nid
) |
279 section_mark_present(ms
);
285 * Mark all memblocks as present using memory_present(). This is a
286 * convienence function that is useful for a number of arches
287 * to mark all of the systems memory as present during initialization.
289 void __init
memblocks_present(void)
291 struct memblock_region
*reg
;
293 for_each_memblock(memory
, reg
) {
294 memory_present(memblock_get_region_node(reg
),
295 memblock_region_memory_base_pfn(reg
),
296 memblock_region_memory_end_pfn(reg
));
301 * Subtle, we encode the real pfn into the mem_map such that
302 * the identity pfn - section_mem_map will return the actual
303 * physical page frame number.
305 static unsigned long sparse_encode_mem_map(struct page
*mem_map
, unsigned long pnum
)
307 unsigned long coded_mem_map
=
308 (unsigned long)(mem_map
- (section_nr_to_pfn(pnum
)));
309 BUILD_BUG_ON(SECTION_MAP_LAST_BIT
> (1UL<<PFN_SECTION_SHIFT
));
310 BUG_ON(coded_mem_map
& ~SECTION_MAP_MASK
);
311 return coded_mem_map
;
315 * Decode mem_map from the coded memmap
317 struct page
*sparse_decode_mem_map(unsigned long coded_mem_map
, unsigned long pnum
)
319 /* mask off the extra low bits of information */
320 coded_mem_map
&= SECTION_MAP_MASK
;
321 return ((struct page
*)coded_mem_map
) + section_nr_to_pfn(pnum
);
324 static void __meminit
sparse_init_one_section(struct mem_section
*ms
,
325 unsigned long pnum
, struct page
*mem_map
,
326 struct mem_section_usage
*usage
, unsigned long flags
)
328 ms
->section_mem_map
&= ~SECTION_MAP_MASK
;
329 ms
->section_mem_map
|= sparse_encode_mem_map(mem_map
, pnum
)
330 | SECTION_HAS_MEM_MAP
| flags
;
334 static unsigned long usemap_size(void)
336 return BITS_TO_LONGS(SECTION_BLOCKFLAGS_BITS
) * sizeof(unsigned long);
339 size_t mem_section_usage_size(void)
341 return sizeof(struct mem_section_usage
) + usemap_size();
344 #ifdef CONFIG_MEMORY_HOTREMOVE
345 static struct mem_section_usage
* __init
346 sparse_early_usemaps_alloc_pgdat_section(struct pglist_data
*pgdat
,
349 struct mem_section_usage
*usage
;
350 unsigned long goal
, limit
;
353 * A page may contain usemaps for other sections preventing the
354 * page being freed and making a section unremovable while
355 * other sections referencing the usemap remain active. Similarly,
356 * a pgdat can prevent a section being removed. If section A
357 * contains a pgdat and section B contains the usemap, both
358 * sections become inter-dependent. This allocates usemaps
359 * from the same section as the pgdat where possible to avoid
362 goal
= __pa(pgdat
) & (PAGE_SECTION_MASK
<< PAGE_SHIFT
);
363 limit
= goal
+ (1UL << PA_SECTION_SHIFT
);
364 nid
= early_pfn_to_nid(goal
>> PAGE_SHIFT
);
366 usage
= memblock_alloc_try_nid(size
, SMP_CACHE_BYTES
, goal
, limit
, nid
);
367 if (!usage
&& limit
) {
374 static void __init
check_usemap_section_nr(int nid
,
375 struct mem_section_usage
*usage
)
377 unsigned long usemap_snr
, pgdat_snr
;
378 static unsigned long old_usemap_snr
;
379 static unsigned long old_pgdat_snr
;
380 struct pglist_data
*pgdat
= NODE_DATA(nid
);
384 if (!old_usemap_snr
) {
385 old_usemap_snr
= NR_MEM_SECTIONS
;
386 old_pgdat_snr
= NR_MEM_SECTIONS
;
389 usemap_snr
= pfn_to_section_nr(__pa(usage
) >> PAGE_SHIFT
);
390 pgdat_snr
= pfn_to_section_nr(__pa(pgdat
) >> PAGE_SHIFT
);
391 if (usemap_snr
== pgdat_snr
)
394 if (old_usemap_snr
== usemap_snr
&& old_pgdat_snr
== pgdat_snr
)
395 /* skip redundant message */
398 old_usemap_snr
= usemap_snr
;
399 old_pgdat_snr
= pgdat_snr
;
401 usemap_nid
= sparse_early_nid(__nr_to_section(usemap_snr
));
402 if (usemap_nid
!= nid
) {
403 pr_info("node %d must be removed before remove section %ld\n",
408 * There is a circular dependency.
409 * Some platforms allow un-removable section because they will just
410 * gather other removable sections for dynamic partitioning.
411 * Just notify un-removable section's number here.
413 pr_info("Section %ld and %ld (node %d) have a circular dependency on usemap and pgdat allocations\n",
414 usemap_snr
, pgdat_snr
, nid
);
417 static struct mem_section_usage
* __init
418 sparse_early_usemaps_alloc_pgdat_section(struct pglist_data
*pgdat
,
421 return memblock_alloc_node(size
, SMP_CACHE_BYTES
, pgdat
->node_id
);
424 static void __init
check_usemap_section_nr(int nid
,
425 struct mem_section_usage
*usage
)
428 #endif /* CONFIG_MEMORY_HOTREMOVE */
430 #ifdef CONFIG_SPARSEMEM_VMEMMAP
431 static unsigned long __init
section_map_size(void)
433 return ALIGN(sizeof(struct page
) * PAGES_PER_SECTION
, PMD_SIZE
);
437 static unsigned long __init
section_map_size(void)
439 return PAGE_ALIGN(sizeof(struct page
) * PAGES_PER_SECTION
);
442 struct page __init
*__populate_section_memmap(unsigned long pfn
,
443 unsigned long nr_pages
, int nid
, struct vmem_altmap
*altmap
)
445 unsigned long size
= section_map_size();
446 struct page
*map
= sparse_buffer_alloc(size
);
447 phys_addr_t addr
= __pa(MAX_DMA_ADDRESS
);
452 map
= memblock_alloc_try_nid(size
,
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 void __init
sparse_buffer_init(unsigned long size
, int nid
)
468 phys_addr_t addr
= __pa(MAX_DMA_ADDRESS
);
469 WARN_ON(sparsemap_buf
); /* forgot to call sparse_buffer_fini()? */
471 memblock_alloc_try_nid_raw(size
, PAGE_SIZE
,
473 MEMBLOCK_ALLOC_ACCESSIBLE
, nid
);
474 sparsemap_buf_end
= sparsemap_buf
+ size
;
477 static void __init
sparse_buffer_fini(void)
479 unsigned long size
= sparsemap_buf_end
- sparsemap_buf
;
481 if (sparsemap_buf
&& size
> 0)
482 memblock_free_early(__pa(sparsemap_buf
), size
);
483 sparsemap_buf
= NULL
;
486 void * __meminit
sparse_buffer_alloc(unsigned long size
)
491 ptr
= PTR_ALIGN(sparsemap_buf
, size
);
492 if (ptr
+ size
> sparsemap_buf_end
)
495 sparsemap_buf
= ptr
+ size
;
500 void __weak __meminit
vmemmap_populate_print_last(void)
505 * Initialize sparse on a specific node. The node spans [pnum_begin, pnum_end)
506 * And number of present sections in this node is map_count.
508 static void __init
sparse_init_nid(int nid
, unsigned long pnum_begin
,
509 unsigned long pnum_end
,
510 unsigned long map_count
)
512 struct mem_section_usage
*usage
;
516 usage
= sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nid
),
517 mem_section_usage_size() * map_count
);
519 pr_err("%s: node[%d] usemap allocation failed", __func__
, nid
);
522 sparse_buffer_init(map_count
* section_map_size(), nid
);
523 for_each_present_section_nr(pnum_begin
, pnum
) {
524 unsigned long pfn
= section_nr_to_pfn(pnum
);
526 if (pnum
>= pnum_end
)
529 map
= __populate_section_memmap(pfn
, PAGES_PER_SECTION
,
532 pr_err("%s: node[%d] memory map backing failed. Some memory will not be available.",
537 check_usemap_section_nr(nid
, usage
);
538 sparse_init_one_section(__nr_to_section(pnum
), pnum
, map
, usage
,
540 usage
= (void *) usage
+ mem_section_usage_size();
542 sparse_buffer_fini();
545 /* We failed to allocate, mark all the following pnums as not present */
546 for_each_present_section_nr(pnum_begin
, pnum
) {
547 struct mem_section
*ms
;
549 if (pnum
>= pnum_end
)
551 ms
= __nr_to_section(pnum
);
552 ms
->section_mem_map
= 0;
557 * Allocate the accumulated non-linear sections, allocate a mem_map
558 * for each and record the physical to section mapping.
560 void __init
sparse_init(void)
562 unsigned long pnum_begin
= first_present_section_nr();
563 int nid_begin
= sparse_early_nid(__nr_to_section(pnum_begin
));
564 unsigned long pnum_end
, map_count
= 1;
566 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
567 set_pageblock_order();
569 for_each_present_section_nr(pnum_begin
+ 1, pnum_end
) {
570 int nid
= sparse_early_nid(__nr_to_section(pnum_end
));
572 if (nid
== nid_begin
) {
576 /* Init node with sections in range [pnum_begin, pnum_end) */
577 sparse_init_nid(nid_begin
, pnum_begin
, pnum_end
, map_count
);
579 pnum_begin
= pnum_end
;
582 /* cover the last node */
583 sparse_init_nid(nid_begin
, pnum_begin
, pnum_end
, map_count
);
584 vmemmap_populate_print_last();
587 #ifdef CONFIG_MEMORY_HOTPLUG
589 /* Mark all memory sections within the pfn range as online */
590 void online_mem_sections(unsigned long start_pfn
, unsigned long end_pfn
)
594 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
595 unsigned long section_nr
= pfn_to_section_nr(pfn
);
596 struct mem_section
*ms
;
598 /* onlining code should never touch invalid ranges */
599 if (WARN_ON(!valid_section_nr(section_nr
)))
602 ms
= __nr_to_section(section_nr
);
603 ms
->section_mem_map
|= SECTION_IS_ONLINE
;
607 #ifdef CONFIG_MEMORY_HOTREMOVE
608 /* Mark all memory sections within the pfn range as offline */
609 void offline_mem_sections(unsigned long start_pfn
, unsigned long end_pfn
)
613 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= PAGES_PER_SECTION
) {
614 unsigned long section_nr
= pfn_to_section_nr(pfn
);
615 struct mem_section
*ms
;
618 * TODO this needs some double checking. Offlining code makes
619 * sure to check pfn_valid but those checks might be just bogus
621 if (WARN_ON(!valid_section_nr(section_nr
)))
624 ms
= __nr_to_section(section_nr
);
625 ms
->section_mem_map
&= ~SECTION_IS_ONLINE
;
630 #ifdef CONFIG_SPARSEMEM_VMEMMAP
631 static struct page
*populate_section_memmap(unsigned long pfn
,
632 unsigned long nr_pages
, int nid
, struct vmem_altmap
*altmap
)
634 return __populate_section_memmap(pfn
, nr_pages
, nid
, altmap
);
637 static void depopulate_section_memmap(unsigned long pfn
, unsigned long nr_pages
,
638 struct vmem_altmap
*altmap
)
640 unsigned long start
= (unsigned long) pfn_to_page(pfn
);
641 unsigned long end
= start
+ nr_pages
* sizeof(struct page
);
643 vmemmap_free(start
, end
, altmap
);
645 static void free_map_bootmem(struct page
*memmap
)
647 unsigned long start
= (unsigned long)memmap
;
648 unsigned long end
= (unsigned long)(memmap
+ PAGES_PER_SECTION
);
650 vmemmap_free(start
, end
, NULL
);
653 struct page
*populate_section_memmap(unsigned long pfn
,
654 unsigned long nr_pages
, int nid
, struct vmem_altmap
*altmap
)
656 struct page
*page
, *ret
;
657 unsigned long memmap_size
= sizeof(struct page
) * PAGES_PER_SECTION
;
659 page
= alloc_pages(GFP_KERNEL
|__GFP_NOWARN
, get_order(memmap_size
));
663 ret
= vmalloc(memmap_size
);
669 ret
= (struct page
*)pfn_to_kaddr(page_to_pfn(page
));
675 static void depopulate_section_memmap(unsigned long pfn
, unsigned long nr_pages
,
676 struct vmem_altmap
*altmap
)
678 struct page
*memmap
= pfn_to_page(pfn
);
680 if (is_vmalloc_addr(memmap
))
683 free_pages((unsigned long)memmap
,
684 get_order(sizeof(struct page
) * PAGES_PER_SECTION
));
687 static void free_map_bootmem(struct page
*memmap
)
689 unsigned long maps_section_nr
, removing_section_nr
, i
;
690 unsigned long magic
, nr_pages
;
691 struct page
*page
= virt_to_page(memmap
);
693 nr_pages
= PAGE_ALIGN(PAGES_PER_SECTION
* sizeof(struct page
))
696 for (i
= 0; i
< nr_pages
; i
++, page
++) {
697 magic
= (unsigned long) page
->freelist
;
699 BUG_ON(magic
== NODE_INFO
);
701 maps_section_nr
= pfn_to_section_nr(page_to_pfn(page
));
702 removing_section_nr
= page_private(page
);
705 * When this function is called, the removing section is
706 * logical offlined state. This means all pages are isolated
707 * from page allocator. If removing section's memmap is placed
708 * on the same section, it must not be freed.
709 * If it is freed, page allocator may allocate it which will
710 * be removed physically soon.
712 if (maps_section_nr
!= removing_section_nr
)
713 put_page_bootmem(page
);
716 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
719 * sparse_add_one_section - add a memory section
720 * @nid: The node to add section on
721 * @start_pfn: start pfn of the memory range
722 * @altmap: device page map
724 * This is only intended for hotplug.
728 * * -EEXIST - Section has been present.
729 * * -ENOMEM - Out of memory.
731 int __meminit
sparse_add_section(int nid
, unsigned long start_pfn
,
732 unsigned long nr_pages
, struct vmem_altmap
*altmap
)
734 unsigned long section_nr
= pfn_to_section_nr(start_pfn
);
735 struct mem_section_usage
*usage
;
736 struct mem_section
*ms
;
741 * no locking for this, because it does its own
742 * plus, it does a kmalloc
744 ret
= sparse_index_init(section_nr
, nid
);
745 if (ret
< 0 && ret
!= -EEXIST
)
748 memmap
= populate_section_memmap(start_pfn
, PAGES_PER_SECTION
, nid
,
752 usage
= kzalloc(mem_section_usage_size(), GFP_KERNEL
);
754 depopulate_section_memmap(start_pfn
, PAGES_PER_SECTION
, altmap
);
758 ms
= __pfn_to_section(start_pfn
);
759 if (ms
->section_mem_map
& SECTION_MARKED_PRESENT
) {
765 * Poison uninitialized struct pages in order to catch invalid flags
768 page_init_poison(memmap
, sizeof(struct page
) * PAGES_PER_SECTION
);
770 set_section_nid(section_nr
, nid
);
771 section_mark_present(ms
);
772 sparse_init_one_section(ms
, section_nr
, memmap
, usage
, 0);
777 depopulate_section_memmap(start_pfn
, PAGES_PER_SECTION
, altmap
);
782 #ifdef CONFIG_MEMORY_FAILURE
783 static void clear_hwpoisoned_pages(struct page
*memmap
, int nr_pages
)
791 * A further optimization is to have per section refcounted
792 * num_poisoned_pages. But that would need more space per memmap, so
793 * for now just do a quick global check to speed up this routine in the
794 * absence of bad pages.
796 if (atomic_long_read(&num_poisoned_pages
) == 0)
799 for (i
= 0; i
< nr_pages
; i
++) {
800 if (PageHWPoison(&memmap
[i
])) {
801 atomic_long_sub(1, &num_poisoned_pages
);
802 ClearPageHWPoison(&memmap
[i
]);
807 static inline void clear_hwpoisoned_pages(struct page
*memmap
, int nr_pages
)
812 static void free_section_usage(struct mem_section
*ms
, struct page
*memmap
,
813 struct mem_section_usage
*usage
, unsigned long pfn
,
814 unsigned long nr_pages
, struct vmem_altmap
*altmap
)
820 * Check to see if allocation came from hot-plug-add
822 if (!early_section(ms
)) {
825 depopulate_section_memmap(pfn
, nr_pages
, altmap
);
830 * The usemap came from bootmem. This is packed with other usemaps
831 * on the section which has pgdat at boot time. Just keep it as is now.
835 free_map_bootmem(memmap
);
838 void sparse_remove_one_section(struct mem_section
*ms
, unsigned long pfn
,
839 unsigned long nr_pages
, unsigned long map_offset
,
840 struct vmem_altmap
*altmap
)
842 struct page
*memmap
= NULL
;
843 struct mem_section_usage
*usage
= NULL
;
845 if (ms
->section_mem_map
) {
847 memmap
= sparse_decode_mem_map(ms
->section_mem_map
,
849 ms
->section_mem_map
= 0;
853 clear_hwpoisoned_pages(memmap
+ map_offset
, nr_pages
- map_offset
);
854 free_section_usage(ms
, memmap
, usage
, pfn
, nr_pages
, altmap
);
856 #endif /* CONFIG_MEMORY_HOTPLUG */