Merge tag 'arc-4.20-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc
[linux-2.6-block.git] / mm / sparse.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
d41dee36
AW
2/*
3 * sparse memory mappings.
4 */
d41dee36 5#include <linux/mm.h>
5a0e3ad6 6#include <linux/slab.h>
d41dee36 7#include <linux/mmzone.h>
97ad1087 8#include <linux/memblock.h>
3b32123d 9#include <linux/compiler.h>
0b0acbec 10#include <linux/highmem.h>
b95f1b31 11#include <linux/export.h>
28ae55c9 12#include <linux/spinlock.h>
0b0acbec 13#include <linux/vmalloc.h>
3b32123d 14
0c0a4a51 15#include "internal.h"
d41dee36 16#include <asm/dma.h>
8f6aac41
CL
17#include <asm/pgalloc.h>
18#include <asm/pgtable.h>
d41dee36
AW
19
20/*
21 * Permanent SPARSEMEM data:
22 *
23 * 1) mem_section - memory sections, mem_map's for valid memory
24 */
3e347261 25#ifdef CONFIG_SPARSEMEM_EXTREME
83e3c487 26struct mem_section **mem_section;
3e347261
BP
27#else
28struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]
22fc6ecc 29 ____cacheline_internodealigned_in_smp;
3e347261
BP
30#endif
31EXPORT_SYMBOL(mem_section);
32
89689ae7
CL
33#ifdef NODE_NOT_IN_PAGE_FLAGS
34/*
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.
38 */
39#if MAX_NUMNODES <= 256
40static u8 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
41#else
42static u16 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
43#endif
44
33dd4e0e 45int page_to_nid(const struct page *page)
89689ae7
CL
46{
47 return section_to_node_table[page_to_section(page)];
48}
49EXPORT_SYMBOL(page_to_nid);
85770ffe
AW
50
51static void set_section_nid(unsigned long section_nr, int nid)
52{
53 section_to_node_table[section_nr] = nid;
54}
55#else /* !NODE_NOT_IN_PAGE_FLAGS */
56static inline void set_section_nid(unsigned long section_nr, int nid)
57{
58}
89689ae7
CL
59#endif
60
3e347261 61#ifdef CONFIG_SPARSEMEM_EXTREME
bd721ea7 62static noinline struct mem_section __ref *sparse_index_alloc(int nid)
28ae55c9
DH
63{
64 struct mem_section *section = NULL;
65 unsigned long array_size = SECTIONS_PER_ROOT *
66 sizeof(struct mem_section);
67
b95046b0
MH
68 if (slab_is_available())
69 section = kzalloc_node(array_size, GFP_KERNEL, nid);
70 else
7e1c4e27
MR
71 section = memblock_alloc_node(array_size, SMP_CACHE_BYTES,
72 nid);
28ae55c9
DH
73
74 return section;
3e347261 75}
802f192e 76
a3142c8e 77static int __meminit sparse_index_init(unsigned long section_nr, int nid)
802f192e 78{
28ae55c9
DH
79 unsigned long root = SECTION_NR_TO_ROOT(section_nr);
80 struct mem_section *section;
802f192e
BP
81
82 if (mem_section[root])
28ae55c9 83 return -EEXIST;
3e347261 84
28ae55c9 85 section = sparse_index_alloc(nid);
af0cd5a7
WC
86 if (!section)
87 return -ENOMEM;
28ae55c9
DH
88
89 mem_section[root] = section;
c1c95183 90
9d1936cf 91 return 0;
28ae55c9
DH
92}
93#else /* !SPARSEMEM_EXTREME */
94static inline int sparse_index_init(unsigned long section_nr, int nid)
95{
96 return 0;
802f192e 97}
28ae55c9
DH
98#endif
99
91fd8b95 100#ifdef CONFIG_SPARSEMEM_EXTREME
4ca644d9
DH
101int __section_nr(struct mem_section* ms)
102{
103 unsigned long root_nr;
83e3c487 104 struct mem_section *root = NULL;
4ca644d9 105
12783b00
MK
106 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
107 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
4ca644d9
DH
108 if (!root)
109 continue;
110
111 if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT)))
112 break;
113 }
114
83e3c487 115 VM_BUG_ON(!root);
db36a461 116
4ca644d9
DH
117 return (root_nr * SECTIONS_PER_ROOT) + (ms - root);
118}
91fd8b95
ZC
119#else
120int __section_nr(struct mem_section* ms)
121{
122 return (int)(ms - mem_section[0]);
123}
124#endif
4ca644d9 125
30c253e6
AW
126/*
127 * During early boot, before section_mem_map is used for an actual
128 * mem_map, we use section_mem_map to store the section's NUMA
129 * node. This keeps us from having to use another data structure. The
130 * node information is cleared just before we store the real mem_map.
131 */
132static inline unsigned long sparse_encode_early_nid(int nid)
133{
134 return (nid << SECTION_NID_SHIFT);
135}
136
137static inline int sparse_early_nid(struct mem_section *section)
138{
139 return (section->section_mem_map >> SECTION_NID_SHIFT);
140}
141
2dbb51c4
MG
142/* Validate the physical addressing limitations of the model */
143void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
144 unsigned long *end_pfn)
d41dee36 145{
2dbb51c4 146 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
d41dee36 147
bead9a3a
IM
148 /*
149 * Sanity checks - do not allow an architecture to pass
150 * in larger pfns than the maximum scope of sparsemem:
151 */
2dbb51c4
MG
152 if (*start_pfn > max_sparsemem_pfn) {
153 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
154 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
155 *start_pfn, *end_pfn, max_sparsemem_pfn);
156 WARN_ON_ONCE(1);
157 *start_pfn = max_sparsemem_pfn;
158 *end_pfn = max_sparsemem_pfn;
ef161a98 159 } else if (*end_pfn > max_sparsemem_pfn) {
2dbb51c4
MG
160 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
161 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
162 *start_pfn, *end_pfn, max_sparsemem_pfn);
163 WARN_ON_ONCE(1);
164 *end_pfn = max_sparsemem_pfn;
165 }
166}
167
c4e1be9e
DH
168/*
169 * There are a number of times that we loop over NR_MEM_SECTIONS,
170 * looking for section_present() on each. But, when we have very
171 * large physical address spaces, NR_MEM_SECTIONS can also be
172 * very large which makes the loops quite long.
173 *
174 * Keeping track of this gives us an easy way to break out of
175 * those loops early.
176 */
177int __highest_present_section_nr;
178static void section_mark_present(struct mem_section *ms)
179{
180 int section_nr = __section_nr(ms);
181
182 if (section_nr > __highest_present_section_nr)
183 __highest_present_section_nr = section_nr;
184
185 ms->section_mem_map |= SECTION_MARKED_PRESENT;
186}
187
188static inline int next_present_section_nr(int section_nr)
189{
190 do {
191 section_nr++;
192 if (present_section_nr(section_nr))
193 return section_nr;
d538c164 194 } while ((section_nr <= __highest_present_section_nr));
c4e1be9e
DH
195
196 return -1;
197}
198#define for_each_present_section_nr(start, section_nr) \
199 for (section_nr = next_present_section_nr(start-1); \
200 ((section_nr >= 0) && \
c4e1be9e
DH
201 (section_nr <= __highest_present_section_nr)); \
202 section_nr = next_present_section_nr(section_nr))
203
85c77f79
PT
204static inline unsigned long first_present_section_nr(void)
205{
206 return next_present_section_nr(-1);
207}
208
2dbb51c4
MG
209/* Record a memory area against a node. */
210void __init memory_present(int nid, unsigned long start, unsigned long end)
211{
212 unsigned long pfn;
bead9a3a 213
629a359b
KS
214#ifdef CONFIG_SPARSEMEM_EXTREME
215 if (unlikely(!mem_section)) {
216 unsigned long size, align;
217
d09cfbbf 218 size = sizeof(struct mem_section*) * NR_SECTION_ROOTS;
629a359b 219 align = 1 << (INTERNODE_CACHE_SHIFT);
eb31d559 220 mem_section = memblock_alloc(size, align);
629a359b
KS
221 }
222#endif
223
d41dee36 224 start &= PAGE_SECTION_MASK;
2dbb51c4 225 mminit_validate_memmodel_limits(&start, &end);
d41dee36
AW
226 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
227 unsigned long section = pfn_to_section_nr(pfn);
802f192e
BP
228 struct mem_section *ms;
229
230 sparse_index_init(section, nid);
85770ffe 231 set_section_nid(section, nid);
802f192e
BP
232
233 ms = __nr_to_section(section);
c4e1be9e 234 if (!ms->section_mem_map) {
2d070eab
MH
235 ms->section_mem_map = sparse_encode_early_nid(nid) |
236 SECTION_IS_ONLINE;
c4e1be9e
DH
237 section_mark_present(ms);
238 }
d41dee36
AW
239 }
240}
241
29751f69
AW
242/*
243 * Subtle, we encode the real pfn into the mem_map such that
244 * the identity pfn - section_mem_map will return the actual
245 * physical page frame number.
246 */
247static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
248{
def9b71e
PT
249 unsigned long coded_mem_map =
250 (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
251 BUILD_BUG_ON(SECTION_MAP_LAST_BIT > (1UL<<PFN_SECTION_SHIFT));
252 BUG_ON(coded_mem_map & ~SECTION_MAP_MASK);
253 return coded_mem_map;
29751f69
AW
254}
255
256/*
ea01ea93 257 * Decode mem_map from the coded memmap
29751f69 258 */
29751f69
AW
259struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
260{
ea01ea93
BP
261 /* mask off the extra low bits of information */
262 coded_mem_map &= SECTION_MAP_MASK;
29751f69
AW
263 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
264}
265
4e40987f 266static void __meminit sparse_init_one_section(struct mem_section *ms,
5c0e3066
MG
267 unsigned long pnum, struct page *mem_map,
268 unsigned long *pageblock_bitmap)
29751f69 269{
30c253e6 270 ms->section_mem_map &= ~SECTION_MAP_MASK;
540557b9
AW
271 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) |
272 SECTION_HAS_MEM_MAP;
5c0e3066 273 ms->pageblock_flags = pageblock_bitmap;
29751f69
AW
274}
275
04753278 276unsigned long usemap_size(void)
5c0e3066 277{
60a7a88d 278 return BITS_TO_LONGS(SECTION_BLOCKFLAGS_BITS) * sizeof(unsigned long);
5c0e3066
MG
279}
280
281#ifdef CONFIG_MEMORY_HOTPLUG
282static unsigned long *__kmalloc_section_usemap(void)
283{
284 return kmalloc(usemap_size(), GFP_KERNEL);
285}
286#endif /* CONFIG_MEMORY_HOTPLUG */
287
48c90682
YG
288#ifdef CONFIG_MEMORY_HOTREMOVE
289static unsigned long * __init
a4322e1b 290sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
238305bb 291 unsigned long size)
48c90682 292{
99ab7b19
YL
293 unsigned long goal, limit;
294 unsigned long *p;
295 int nid;
48c90682
YG
296 /*
297 * A page may contain usemaps for other sections preventing the
298 * page being freed and making a section unremovable while
c800bcd5 299 * other sections referencing the usemap remain active. Similarly,
48c90682
YG
300 * a pgdat can prevent a section being removed. If section A
301 * contains a pgdat and section B contains the usemap, both
302 * sections become inter-dependent. This allocates usemaps
303 * from the same section as the pgdat where possible to avoid
304 * this problem.
305 */
07b4e2bc 306 goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT);
99ab7b19
YL
307 limit = goal + (1UL << PA_SECTION_SHIFT);
308 nid = early_pfn_to_nid(goal >> PAGE_SHIFT);
309again:
eb31d559 310 p = memblock_alloc_try_nid_nopanic(size,
bb016b84
SS
311 SMP_CACHE_BYTES, goal, limit,
312 nid);
99ab7b19
YL
313 if (!p && limit) {
314 limit = 0;
315 goto again;
316 }
317 return p;
48c90682
YG
318}
319
320static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
321{
322 unsigned long usemap_snr, pgdat_snr;
83e3c487
KS
323 static unsigned long old_usemap_snr;
324 static unsigned long old_pgdat_snr;
48c90682
YG
325 struct pglist_data *pgdat = NODE_DATA(nid);
326 int usemap_nid;
327
83e3c487
KS
328 /* First call */
329 if (!old_usemap_snr) {
330 old_usemap_snr = NR_MEM_SECTIONS;
331 old_pgdat_snr = NR_MEM_SECTIONS;
332 }
333
48c90682
YG
334 usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT);
335 pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
336 if (usemap_snr == pgdat_snr)
337 return;
338
339 if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr)
340 /* skip redundant message */
341 return;
342
343 old_usemap_snr = usemap_snr;
344 old_pgdat_snr = pgdat_snr;
345
346 usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr));
347 if (usemap_nid != nid) {
1170532b
JP
348 pr_info("node %d must be removed before remove section %ld\n",
349 nid, usemap_snr);
48c90682
YG
350 return;
351 }
352 /*
353 * There is a circular dependency.
354 * Some platforms allow un-removable section because they will just
355 * gather other removable sections for dynamic partitioning.
356 * Just notify un-removable section's number here.
357 */
1170532b
JP
358 pr_info("Section %ld and %ld (node %d) have a circular dependency on usemap and pgdat allocations\n",
359 usemap_snr, pgdat_snr, nid);
48c90682
YG
360}
361#else
362static unsigned long * __init
a4322e1b 363sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
238305bb 364 unsigned long size)
48c90682 365{
eb31d559 366 return memblock_alloc_node_nopanic(size, pgdat->node_id);
48c90682
YG
367}
368
369static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
370{
371}
372#endif /* CONFIG_MEMORY_HOTREMOVE */
373
35fd1eb1 374#ifdef CONFIG_SPARSEMEM_VMEMMAP
afda57bc 375static unsigned long __init section_map_size(void)
35fd1eb1
PT
376{
377 return ALIGN(sizeof(struct page) * PAGES_PER_SECTION, PMD_SIZE);
378}
379
380#else
afda57bc 381static unsigned long __init section_map_size(void)
e131c06b
PT
382{
383 return PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
384}
385
7b73d978
CH
386struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid,
387 struct vmem_altmap *altmap)
29751f69 388{
e131c06b
PT
389 unsigned long size = section_map_size();
390 struct page *map = sparse_buffer_alloc(size);
391
392 if (map)
393 return map;
29751f69 394
eb31d559 395 map = memblock_alloc_try_nid(size,
bb016b84 396 PAGE_SIZE, __pa(MAX_DMA_ADDRESS),
97ad1087 397 MEMBLOCK_ALLOC_ACCESSIBLE, nid);
8f6aac41
CL
398 return map;
399}
400#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
401
35fd1eb1
PT
402static void *sparsemap_buf __meminitdata;
403static void *sparsemap_buf_end __meminitdata;
404
afda57bc 405static void __init sparse_buffer_init(unsigned long size, int nid)
35fd1eb1
PT
406{
407 WARN_ON(sparsemap_buf); /* forgot to call sparse_buffer_fini()? */
408 sparsemap_buf =
eb31d559 409 memblock_alloc_try_nid_raw(size, PAGE_SIZE,
35fd1eb1 410 __pa(MAX_DMA_ADDRESS),
97ad1087 411 MEMBLOCK_ALLOC_ACCESSIBLE, nid);
35fd1eb1
PT
412 sparsemap_buf_end = sparsemap_buf + size;
413}
414
afda57bc 415static void __init sparse_buffer_fini(void)
35fd1eb1
PT
416{
417 unsigned long size = sparsemap_buf_end - sparsemap_buf;
418
419 if (sparsemap_buf && size > 0)
420 memblock_free_early(__pa(sparsemap_buf), size);
421 sparsemap_buf = NULL;
422}
423
424void * __meminit sparse_buffer_alloc(unsigned long size)
425{
426 void *ptr = NULL;
427
428 if (sparsemap_buf) {
429 ptr = PTR_ALIGN(sparsemap_buf, size);
430 if (ptr + size > sparsemap_buf_end)
431 ptr = NULL;
432 else
433 sparsemap_buf = ptr + size;
434 }
435 return ptr;
436}
437
3b32123d 438void __weak __meminit vmemmap_populate_print_last(void)
c2b91e2e
YL
439{
440}
a4322e1b 441
85c77f79
PT
442/*
443 * Initialize sparse on a specific node. The node spans [pnum_begin, pnum_end)
444 * And number of present sections in this node is map_count.
445 */
446static void __init sparse_init_nid(int nid, unsigned long pnum_begin,
447 unsigned long pnum_end,
448 unsigned long map_count)
449{
450 unsigned long pnum, usemap_longs, *usemap;
451 struct page *map;
452
453 usemap_longs = BITS_TO_LONGS(SECTION_BLOCKFLAGS_BITS);
454 usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nid),
455 usemap_size() *
456 map_count);
457 if (!usemap) {
458 pr_err("%s: node[%d] usemap allocation failed", __func__, nid);
459 goto failed;
460 }
461 sparse_buffer_init(map_count * section_map_size(), nid);
462 for_each_present_section_nr(pnum_begin, pnum) {
463 if (pnum >= pnum_end)
464 break;
465
466 map = sparse_mem_map_populate(pnum, nid, NULL);
467 if (!map) {
468 pr_err("%s: node[%d] memory map backing failed. Some memory will not be available.",
469 __func__, nid);
470 pnum_begin = pnum;
471 goto failed;
472 }
473 check_usemap_section_nr(nid, usemap);
474 sparse_init_one_section(__nr_to_section(pnum), pnum, map, usemap);
475 usemap += usemap_longs;
476 }
477 sparse_buffer_fini();
478 return;
479failed:
480 /* We failed to allocate, mark all the following pnums as not present */
481 for_each_present_section_nr(pnum_begin, pnum) {
482 struct mem_section *ms;
483
484 if (pnum >= pnum_end)
485 break;
486 ms = __nr_to_section(pnum);
487 ms->section_mem_map = 0;
488 }
489}
490
491/*
492 * Allocate the accumulated non-linear sections, allocate a mem_map
493 * for each and record the physical to section mapping.
494 */
2a3cb8ba 495void __init sparse_init(void)
85c77f79
PT
496{
497 unsigned long pnum_begin = first_present_section_nr();
498 int nid_begin = sparse_early_nid(__nr_to_section(pnum_begin));
499 unsigned long pnum_end, map_count = 1;
500
501 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
502 set_pageblock_order();
503
504 for_each_present_section_nr(pnum_begin + 1, pnum_end) {
505 int nid = sparse_early_nid(__nr_to_section(pnum_end));
506
507 if (nid == nid_begin) {
508 map_count++;
509 continue;
510 }
511 /* Init node with sections in range [pnum_begin, pnum_end) */
512 sparse_init_nid(nid_begin, pnum_begin, pnum_end, map_count);
513 nid_begin = nid;
514 pnum_begin = pnum_end;
515 map_count = 1;
516 }
517 /* cover the last node */
518 sparse_init_nid(nid_begin, pnum_begin, pnum_end, map_count);
519 vmemmap_populate_print_last();
520}
521
193faea9 522#ifdef CONFIG_MEMORY_HOTPLUG
2d070eab
MH
523
524/* Mark all memory sections within the pfn range as online */
525void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
526{
527 unsigned long pfn;
528
529 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
b4ccec41 530 unsigned long section_nr = pfn_to_section_nr(pfn);
2d070eab
MH
531 struct mem_section *ms;
532
533 /* onlining code should never touch invalid ranges */
534 if (WARN_ON(!valid_section_nr(section_nr)))
535 continue;
536
537 ms = __nr_to_section(section_nr);
538 ms->section_mem_map |= SECTION_IS_ONLINE;
539 }
540}
541
542#ifdef CONFIG_MEMORY_HOTREMOVE
543/* Mark all memory sections within the pfn range as online */
544void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
545{
546 unsigned long pfn;
547
548 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
27227c73 549 unsigned long section_nr = pfn_to_section_nr(pfn);
2d070eab
MH
550 struct mem_section *ms;
551
552 /*
553 * TODO this needs some double checking. Offlining code makes
554 * sure to check pfn_valid but those checks might be just bogus
555 */
556 if (WARN_ON(!valid_section_nr(section_nr)))
557 continue;
558
559 ms = __nr_to_section(section_nr);
560 ms->section_mem_map &= ~SECTION_IS_ONLINE;
561 }
562}
563#endif
564
98f3cfc1 565#ifdef CONFIG_SPARSEMEM_VMEMMAP
7b73d978
CH
566static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
567 struct vmem_altmap *altmap)
98f3cfc1
YG
568{
569 /* This will make the necessary allocations eventually. */
7b73d978 570 return sparse_mem_map_populate(pnum, nid, altmap);
98f3cfc1 571}
24b6d416
CH
572static void __kfree_section_memmap(struct page *memmap,
573 struct vmem_altmap *altmap)
98f3cfc1 574{
0aad818b 575 unsigned long start = (unsigned long)memmap;
85b35fea 576 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
0aad818b 577
24b6d416 578 vmemmap_free(start, end, altmap);
98f3cfc1 579}
4edd7cef 580#ifdef CONFIG_MEMORY_HOTREMOVE
81556b02 581static void free_map_bootmem(struct page *memmap)
0c0a4a51 582{
0aad818b 583 unsigned long start = (unsigned long)memmap;
81556b02 584 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
0aad818b 585
24b6d416 586 vmemmap_free(start, end, NULL);
0c0a4a51 587}
4edd7cef 588#endif /* CONFIG_MEMORY_HOTREMOVE */
98f3cfc1 589#else
85b35fea 590static struct page *__kmalloc_section_memmap(void)
0b0acbec
DH
591{
592 struct page *page, *ret;
85b35fea 593 unsigned long memmap_size = sizeof(struct page) * PAGES_PER_SECTION;
0b0acbec 594
f2d0aa5b 595 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
0b0acbec
DH
596 if (page)
597 goto got_map_page;
598
599 ret = vmalloc(memmap_size);
600 if (ret)
601 goto got_map_ptr;
602
603 return NULL;
604got_map_page:
605 ret = (struct page *)pfn_to_kaddr(page_to_pfn(page));
606got_map_ptr:
0b0acbec
DH
607
608 return ret;
609}
610
7b73d978
CH
611static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
612 struct vmem_altmap *altmap)
98f3cfc1 613{
85b35fea 614 return __kmalloc_section_memmap();
98f3cfc1
YG
615}
616
24b6d416
CH
617static void __kfree_section_memmap(struct page *memmap,
618 struct vmem_altmap *altmap)
0b0acbec 619{
9e2779fa 620 if (is_vmalloc_addr(memmap))
0b0acbec
DH
621 vfree(memmap);
622 else
623 free_pages((unsigned long)memmap,
85b35fea 624 get_order(sizeof(struct page) * PAGES_PER_SECTION));
0b0acbec 625}
0c0a4a51 626
4edd7cef 627#ifdef CONFIG_MEMORY_HOTREMOVE
81556b02 628static void free_map_bootmem(struct page *memmap)
0c0a4a51
YG
629{
630 unsigned long maps_section_nr, removing_section_nr, i;
81556b02 631 unsigned long magic, nr_pages;
ae64ffca 632 struct page *page = virt_to_page(memmap);
0c0a4a51 633
81556b02
ZY
634 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
635 >> PAGE_SHIFT;
636
0c0a4a51 637 for (i = 0; i < nr_pages; i++, page++) {
ddffe98d 638 magic = (unsigned long) page->freelist;
0c0a4a51
YG
639
640 BUG_ON(magic == NODE_INFO);
641
642 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
857e522a 643 removing_section_nr = page_private(page);
0c0a4a51
YG
644
645 /*
646 * When this function is called, the removing section is
647 * logical offlined state. This means all pages are isolated
648 * from page allocator. If removing section's memmap is placed
649 * on the same section, it must not be freed.
650 * If it is freed, page allocator may allocate it which will
651 * be removed physically soon.
652 */
653 if (maps_section_nr != removing_section_nr)
654 put_page_bootmem(page);
655 }
656}
4edd7cef 657#endif /* CONFIG_MEMORY_HOTREMOVE */
98f3cfc1 658#endif /* CONFIG_SPARSEMEM_VMEMMAP */
0b0acbec 659
29751f69
AW
660/*
661 * returns the number of sections whose mem_maps were properly
662 * set. If this is <=0, then that means that the passed-in
663 * map was not consumed and must be freed.
664 */
7b73d978
CH
665int __meminit sparse_add_one_section(struct pglist_data *pgdat,
666 unsigned long start_pfn, struct vmem_altmap *altmap)
29751f69 667{
0b0acbec 668 unsigned long section_nr = pfn_to_section_nr(start_pfn);
0b0acbec
DH
669 struct mem_section *ms;
670 struct page *memmap;
5c0e3066 671 unsigned long *usemap;
0b0acbec
DH
672 unsigned long flags;
673 int ret;
29751f69 674
0b0acbec
DH
675 /*
676 * no locking for this, because it does its own
677 * plus, it does a kmalloc
678 */
bbd06825
WC
679 ret = sparse_index_init(section_nr, pgdat->node_id);
680 if (ret < 0 && ret != -EEXIST)
681 return ret;
4e40987f 682 ret = 0;
7b73d978 683 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id, altmap);
bbd06825
WC
684 if (!memmap)
685 return -ENOMEM;
5c0e3066 686 usemap = __kmalloc_section_usemap();
bbd06825 687 if (!usemap) {
24b6d416 688 __kfree_section_memmap(memmap, altmap);
bbd06825
WC
689 return -ENOMEM;
690 }
0b0acbec
DH
691
692 pgdat_resize_lock(pgdat, &flags);
29751f69 693
0b0acbec
DH
694 ms = __pfn_to_section(start_pfn);
695 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
696 ret = -EEXIST;
697 goto out;
698 }
5c0e3066 699
d0dc12e8
PT
700 /*
701 * Poison uninitialized struct pages in order to catch invalid flags
702 * combinations.
703 */
f682a97a 704 page_init_poison(memmap, sizeof(struct page) * PAGES_PER_SECTION);
3ac19f8e 705
c4e1be9e 706 section_mark_present(ms);
4e40987f 707 sparse_init_one_section(ms, section_nr, memmap, usemap);
0b0acbec 708
0b0acbec
DH
709out:
710 pgdat_resize_unlock(pgdat, &flags);
4e40987f 711 if (ret < 0) {
bbd06825 712 kfree(usemap);
24b6d416 713 __kfree_section_memmap(memmap, altmap);
bbd06825 714 }
0b0acbec 715 return ret;
29751f69 716}
ea01ea93 717
f3deb687 718#ifdef CONFIG_MEMORY_HOTREMOVE
95a4774d
WC
719#ifdef CONFIG_MEMORY_FAILURE
720static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
721{
722 int i;
723
724 if (!memmap)
725 return;
726
4b94ffdc 727 for (i = 0; i < nr_pages; i++) {
95a4774d 728 if (PageHWPoison(&memmap[i])) {
293c07e3 729 atomic_long_sub(1, &num_poisoned_pages);
95a4774d
WC
730 ClearPageHWPoison(&memmap[i]);
731 }
732 }
733}
734#else
735static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
736{
737}
738#endif
739
24b6d416
CH
740static void free_section_usemap(struct page *memmap, unsigned long *usemap,
741 struct vmem_altmap *altmap)
4edd7cef
DR
742{
743 struct page *usemap_page;
4edd7cef
DR
744
745 if (!usemap)
746 return;
747
748 usemap_page = virt_to_page(usemap);
749 /*
750 * Check to see if allocation came from hot-plug-add
751 */
752 if (PageSlab(usemap_page) || PageCompound(usemap_page)) {
753 kfree(usemap);
754 if (memmap)
24b6d416 755 __kfree_section_memmap(memmap, altmap);
4edd7cef
DR
756 return;
757 }
758
759 /*
760 * The usemap came from bootmem. This is packed with other usemaps
761 * on the section which has pgdat at boot time. Just keep it as is now.
762 */
763
81556b02
ZY
764 if (memmap)
765 free_map_bootmem(memmap);
4edd7cef
DR
766}
767
4b94ffdc 768void sparse_remove_one_section(struct zone *zone, struct mem_section *ms,
24b6d416 769 unsigned long map_offset, struct vmem_altmap *altmap)
ea01ea93
BP
770{
771 struct page *memmap = NULL;
cd099682
TC
772 unsigned long *usemap = NULL, flags;
773 struct pglist_data *pgdat = zone->zone_pgdat;
ea01ea93 774
cd099682 775 pgdat_resize_lock(pgdat, &flags);
ea01ea93
BP
776 if (ms->section_mem_map) {
777 usemap = ms->pageblock_flags;
778 memmap = sparse_decode_mem_map(ms->section_mem_map,
779 __section_nr(ms));
780 ms->section_mem_map = 0;
781 ms->pageblock_flags = NULL;
782 }
cd099682 783 pgdat_resize_unlock(pgdat, &flags);
ea01ea93 784
4b94ffdc
DW
785 clear_hwpoisoned_pages(memmap + map_offset,
786 PAGES_PER_SECTION - map_offset);
24b6d416 787 free_section_usemap(memmap, usemap, altmap);
ea01ea93 788}
4edd7cef
DR
789#endif /* CONFIG_MEMORY_HOTREMOVE */
790#endif /* CONFIG_MEMORY_HOTPLUG */