doc: fix double words
[linux-2.6-block.git] / mm / sparse.c
CommitLineData
d41dee36
AW
1/*
2 * sparse memory mappings.
3 */
d41dee36 4#include <linux/mm.h>
5a0e3ad6 5#include <linux/slab.h>
d41dee36
AW
6#include <linux/mmzone.h>
7#include <linux/bootmem.h>
0b0acbec 8#include <linux/highmem.h>
b95f1b31 9#include <linux/export.h>
28ae55c9 10#include <linux/spinlock.h>
0b0acbec 11#include <linux/vmalloc.h>
0c0a4a51 12#include "internal.h"
d41dee36 13#include <asm/dma.h>
8f6aac41
CL
14#include <asm/pgalloc.h>
15#include <asm/pgtable.h>
d41dee36
AW
16
17/*
18 * Permanent SPARSEMEM data:
19 *
20 * 1) mem_section - memory sections, mem_map's for valid memory
21 */
3e347261 22#ifdef CONFIG_SPARSEMEM_EXTREME
802f192e 23struct mem_section *mem_section[NR_SECTION_ROOTS]
22fc6ecc 24 ____cacheline_internodealigned_in_smp;
3e347261
BP
25#else
26struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]
22fc6ecc 27 ____cacheline_internodealigned_in_smp;
3e347261
BP
28#endif
29EXPORT_SYMBOL(mem_section);
30
89689ae7
CL
31#ifdef NODE_NOT_IN_PAGE_FLAGS
32/*
33 * If we did not store the node number in the page then we have to
34 * do a lookup in the section_to_node_table in order to find which
35 * node the page belongs to.
36 */
37#if MAX_NUMNODES <= 256
38static u8 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
39#else
40static u16 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
41#endif
42
33dd4e0e 43int page_to_nid(const struct page *page)
89689ae7
CL
44{
45 return section_to_node_table[page_to_section(page)];
46}
47EXPORT_SYMBOL(page_to_nid);
85770ffe
AW
48
49static void set_section_nid(unsigned long section_nr, int nid)
50{
51 section_to_node_table[section_nr] = nid;
52}
53#else /* !NODE_NOT_IN_PAGE_FLAGS */
54static inline void set_section_nid(unsigned long section_nr, int nid)
55{
56}
89689ae7
CL
57#endif
58
3e347261 59#ifdef CONFIG_SPARSEMEM_EXTREME
577a32f6 60static struct mem_section noinline __init_refok *sparse_index_alloc(int nid)
28ae55c9
DH
61{
62 struct mem_section *section = NULL;
63 unsigned long array_size = SECTIONS_PER_ROOT *
64 sizeof(struct mem_section);
65
f52407ce
SL
66 if (slab_is_available()) {
67 if (node_state(nid, N_HIGH_MEMORY))
5b760e64 68 section = kzalloc_node(array_size, GFP_KERNEL, nid);
f52407ce 69 else
5b760e64
GS
70 section = kzalloc(array_size, GFP_KERNEL);
71 } else {
bb016b84 72 section = memblock_virt_alloc_node(array_size, nid);
5b760e64 73 }
28ae55c9
DH
74
75 return section;
3e347261 76}
802f192e 77
a3142c8e 78static int __meminit sparse_index_init(unsigned long section_nr, int nid)
802f192e 79{
28ae55c9
DH
80 unsigned long root = SECTION_NR_TO_ROOT(section_nr);
81 struct mem_section *section;
802f192e
BP
82
83 if (mem_section[root])
28ae55c9 84 return -EEXIST;
3e347261 85
28ae55c9 86 section = sparse_index_alloc(nid);
af0cd5a7
WC
87 if (!section)
88 return -ENOMEM;
28ae55c9
DH
89
90 mem_section[root] = section;
c1c95183 91
9d1936cf 92 return 0;
28ae55c9
DH
93}
94#else /* !SPARSEMEM_EXTREME */
95static inline int sparse_index_init(unsigned long section_nr, int nid)
96{
97 return 0;
802f192e 98}
28ae55c9
DH
99#endif
100
4ca644d9
DH
101/*
102 * Although written for the SPARSEMEM_EXTREME case, this happens
cd881a6b 103 * to also work for the flat array case because
4ca644d9
DH
104 * NR_SECTION_ROOTS==NR_MEM_SECTIONS.
105 */
106int __section_nr(struct mem_section* ms)
107{
108 unsigned long root_nr;
109 struct mem_section* root;
110
12783b00
MK
111 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
112 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
4ca644d9
DH
113 if (!root)
114 continue;
115
116 if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT)))
117 break;
118 }
119
db36a461
GS
120 VM_BUG_ON(root_nr == NR_SECTION_ROOTS);
121
4ca644d9
DH
122 return (root_nr * SECTIONS_PER_ROOT) + (ms - root);
123}
124
30c253e6
AW
125/*
126 * During early boot, before section_mem_map is used for an actual
127 * mem_map, we use section_mem_map to store the section's NUMA
128 * node. This keeps us from having to use another data structure. The
129 * node information is cleared just before we store the real mem_map.
130 */
131static inline unsigned long sparse_encode_early_nid(int nid)
132{
133 return (nid << SECTION_NID_SHIFT);
134}
135
136static inline int sparse_early_nid(struct mem_section *section)
137{
138 return (section->section_mem_map >> SECTION_NID_SHIFT);
139}
140
2dbb51c4
MG
141/* Validate the physical addressing limitations of the model */
142void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
143 unsigned long *end_pfn)
d41dee36 144{
2dbb51c4 145 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
d41dee36 146
bead9a3a
IM
147 /*
148 * Sanity checks - do not allow an architecture to pass
149 * in larger pfns than the maximum scope of sparsemem:
150 */
2dbb51c4
MG
151 if (*start_pfn > max_sparsemem_pfn) {
152 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
153 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
154 *start_pfn, *end_pfn, max_sparsemem_pfn);
155 WARN_ON_ONCE(1);
156 *start_pfn = max_sparsemem_pfn;
157 *end_pfn = max_sparsemem_pfn;
ef161a98 158 } else if (*end_pfn > max_sparsemem_pfn) {
2dbb51c4
MG
159 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
160 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
161 *start_pfn, *end_pfn, max_sparsemem_pfn);
162 WARN_ON_ONCE(1);
163 *end_pfn = max_sparsemem_pfn;
164 }
165}
166
167/* Record a memory area against a node. */
168void __init memory_present(int nid, unsigned long start, unsigned long end)
169{
170 unsigned long pfn;
bead9a3a 171
d41dee36 172 start &= PAGE_SECTION_MASK;
2dbb51c4 173 mminit_validate_memmodel_limits(&start, &end);
d41dee36
AW
174 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
175 unsigned long section = pfn_to_section_nr(pfn);
802f192e
BP
176 struct mem_section *ms;
177
178 sparse_index_init(section, nid);
85770ffe 179 set_section_nid(section, nid);
802f192e
BP
180
181 ms = __nr_to_section(section);
182 if (!ms->section_mem_map)
30c253e6
AW
183 ms->section_mem_map = sparse_encode_early_nid(nid) |
184 SECTION_MARKED_PRESENT;
d41dee36
AW
185 }
186}
187
188/*
189 * Only used by the i386 NUMA architecures, but relatively
190 * generic code.
191 */
192unsigned long __init node_memmap_size_bytes(int nid, unsigned long start_pfn,
193 unsigned long end_pfn)
194{
195 unsigned long pfn;
196 unsigned long nr_pages = 0;
197
2dbb51c4 198 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
d41dee36
AW
199 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
200 if (nid != early_pfn_to_nid(pfn))
201 continue;
202
540557b9 203 if (pfn_present(pfn))
d41dee36
AW
204 nr_pages += PAGES_PER_SECTION;
205 }
206
207 return nr_pages * sizeof(struct page);
208}
209
29751f69
AW
210/*
211 * Subtle, we encode the real pfn into the mem_map such that
212 * the identity pfn - section_mem_map will return the actual
213 * physical page frame number.
214 */
215static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
216{
217 return (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
218}
219
220/*
ea01ea93 221 * Decode mem_map from the coded memmap
29751f69 222 */
29751f69
AW
223struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
224{
ea01ea93
BP
225 /* mask off the extra low bits of information */
226 coded_mem_map &= SECTION_MAP_MASK;
29751f69
AW
227 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
228}
229
a3142c8e 230static int __meminit sparse_init_one_section(struct mem_section *ms,
5c0e3066
MG
231 unsigned long pnum, struct page *mem_map,
232 unsigned long *pageblock_bitmap)
29751f69 233{
540557b9 234 if (!present_section(ms))
29751f69
AW
235 return -EINVAL;
236
30c253e6 237 ms->section_mem_map &= ~SECTION_MAP_MASK;
540557b9
AW
238 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) |
239 SECTION_HAS_MEM_MAP;
5c0e3066 240 ms->pageblock_flags = pageblock_bitmap;
29751f69
AW
241
242 return 1;
243}
244
04753278 245unsigned long usemap_size(void)
5c0e3066
MG
246{
247 unsigned long size_bytes;
248 size_bytes = roundup(SECTION_BLOCKFLAGS_BITS, 8) / 8;
249 size_bytes = roundup(size_bytes, sizeof(unsigned long));
250 return size_bytes;
251}
252
253#ifdef CONFIG_MEMORY_HOTPLUG
254static unsigned long *__kmalloc_section_usemap(void)
255{
256 return kmalloc(usemap_size(), GFP_KERNEL);
257}
258#endif /* CONFIG_MEMORY_HOTPLUG */
259
48c90682
YG
260#ifdef CONFIG_MEMORY_HOTREMOVE
261static unsigned long * __init
a4322e1b 262sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
238305bb 263 unsigned long size)
48c90682 264{
99ab7b19
YL
265 unsigned long goal, limit;
266 unsigned long *p;
267 int nid;
48c90682
YG
268 /*
269 * A page may contain usemaps for other sections preventing the
270 * page being freed and making a section unremovable while
271 * other sections referencing the usemap retmain active. Similarly,
272 * a pgdat can prevent a section being removed. If section A
273 * contains a pgdat and section B contains the usemap, both
274 * sections become inter-dependent. This allocates usemaps
275 * from the same section as the pgdat where possible to avoid
276 * this problem.
277 */
07b4e2bc 278 goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT);
99ab7b19
YL
279 limit = goal + (1UL << PA_SECTION_SHIFT);
280 nid = early_pfn_to_nid(goal >> PAGE_SHIFT);
281again:
bb016b84
SS
282 p = memblock_virt_alloc_try_nid_nopanic(size,
283 SMP_CACHE_BYTES, goal, limit,
284 nid);
99ab7b19
YL
285 if (!p && limit) {
286 limit = 0;
287 goto again;
288 }
289 return p;
48c90682
YG
290}
291
292static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
293{
294 unsigned long usemap_snr, pgdat_snr;
295 static unsigned long old_usemap_snr = NR_MEM_SECTIONS;
296 static unsigned long old_pgdat_snr = NR_MEM_SECTIONS;
297 struct pglist_data *pgdat = NODE_DATA(nid);
298 int usemap_nid;
299
300 usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT);
301 pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
302 if (usemap_snr == pgdat_snr)
303 return;
304
305 if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr)
306 /* skip redundant message */
307 return;
308
309 old_usemap_snr = usemap_snr;
310 old_pgdat_snr = pgdat_snr;
311
312 usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr));
313 if (usemap_nid != nid) {
314 printk(KERN_INFO
315 "node %d must be removed before remove section %ld\n",
316 nid, usemap_snr);
317 return;
318 }
319 /*
320 * There is a circular dependency.
321 * Some platforms allow un-removable section because they will just
322 * gather other removable sections for dynamic partitioning.
323 * Just notify un-removable section's number here.
324 */
325 printk(KERN_INFO "Section %ld and %ld (node %d)", usemap_snr,
326 pgdat_snr, nid);
327 printk(KERN_CONT
328 " have a circular dependency on usemap and pgdat allocations\n");
329}
330#else
331static unsigned long * __init
a4322e1b 332sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
238305bb 333 unsigned long size)
48c90682 334{
bb016b84 335 return memblock_virt_alloc_node_nopanic(size, pgdat->node_id);
48c90682
YG
336}
337
338static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
339{
340}
341#endif /* CONFIG_MEMORY_HOTREMOVE */
342
18732093 343static void __init sparse_early_usemaps_alloc_node(void *data,
a4322e1b
YL
344 unsigned long pnum_begin,
345 unsigned long pnum_end,
346 unsigned long usemap_count, int nodeid)
5c0e3066 347{
a4322e1b
YL
348 void *usemap;
349 unsigned long pnum;
18732093 350 unsigned long **usemap_map = (unsigned long **)data;
a4322e1b 351 int size = usemap_size();
5c0e3066 352
a4322e1b 353 usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid),
238305bb 354 size * usemap_count);
f5bf18fa 355 if (!usemap) {
238305bb
JW
356 printk(KERN_WARNING "%s: allocation failed\n", __func__);
357 return;
48c90682
YG
358 }
359
f5bf18fa
NA
360 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
361 if (!present_section_nr(pnum))
362 continue;
363 usemap_map[pnum] = usemap;
364 usemap += size;
365 check_usemap_section_nr(nodeid, usemap_map[pnum]);
a4322e1b 366 }
5c0e3066
MG
367}
368
8f6aac41 369#ifndef CONFIG_SPARSEMEM_VMEMMAP
98f3cfc1 370struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
29751f69
AW
371{
372 struct page *map;
e48e67e0 373 unsigned long size;
29751f69
AW
374
375 map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
376 if (map)
377 return map;
378
e48e67e0 379 size = PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
bb016b84
SS
380 map = memblock_virt_alloc_try_nid(size,
381 PAGE_SIZE, __pa(MAX_DMA_ADDRESS),
382 BOOTMEM_ALLOC_ACCESSIBLE, nid);
8f6aac41
CL
383 return map;
384}
9bdac914
YL
385void __init sparse_mem_maps_populate_node(struct page **map_map,
386 unsigned long pnum_begin,
387 unsigned long pnum_end,
388 unsigned long map_count, int nodeid)
389{
390 void *map;
391 unsigned long pnum;
392 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
393
394 map = alloc_remap(nodeid, size * map_count);
395 if (map) {
396 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
397 if (!present_section_nr(pnum))
398 continue;
399 map_map[pnum] = map;
400 map += size;
401 }
402 return;
403 }
404
405 size = PAGE_ALIGN(size);
bb016b84
SS
406 map = memblock_virt_alloc_try_nid(size * map_count,
407 PAGE_SIZE, __pa(MAX_DMA_ADDRESS),
408 BOOTMEM_ALLOC_ACCESSIBLE, nodeid);
9bdac914
YL
409 if (map) {
410 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
411 if (!present_section_nr(pnum))
412 continue;
413 map_map[pnum] = map;
414 map += size;
415 }
416 return;
417 }
418
419 /* fallback */
420 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
421 struct mem_section *ms;
422
423 if (!present_section_nr(pnum))
424 continue;
425 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
426 if (map_map[pnum])
427 continue;
428 ms = __nr_to_section(pnum);
429 printk(KERN_ERR "%s: sparsemem memory map backing failed "
430 "some memory will not be available.\n", __func__);
431 ms->section_mem_map = 0;
432 }
433}
8f6aac41
CL
434#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
435
81d0d950 436#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
18732093 437static void __init sparse_early_mem_maps_alloc_node(void *data,
9bdac914
YL
438 unsigned long pnum_begin,
439 unsigned long pnum_end,
440 unsigned long map_count, int nodeid)
441{
18732093 442 struct page **map_map = (struct page **)data;
9bdac914
YL
443 sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end,
444 map_count, nodeid);
445}
81d0d950 446#else
9e5c6da7 447static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
8f6aac41
CL
448{
449 struct page *map;
450 struct mem_section *ms = __nr_to_section(pnum);
451 int nid = sparse_early_nid(ms);
452
98f3cfc1 453 map = sparse_mem_map_populate(pnum, nid);
29751f69
AW
454 if (map)
455 return map;
456
8f6aac41 457 printk(KERN_ERR "%s: sparsemem memory map backing failed "
d40cee24 458 "some memory will not be available.\n", __func__);
802f192e 459 ms->section_mem_map = 0;
29751f69
AW
460 return NULL;
461}
9bdac914 462#endif
29751f69 463
c2b91e2e
YL
464void __attribute__((weak)) __meminit vmemmap_populate_print_last(void)
465{
466}
a4322e1b 467
18732093
WL
468/**
469 * alloc_usemap_and_memmap - memory alloction for pageblock flags and vmemmap
470 * @map: usemap_map for pageblock flags or mmap_map for vmemmap
471 */
472static void __init alloc_usemap_and_memmap(void (*alloc_func)
473 (void *, unsigned long, unsigned long,
474 unsigned long, int), void *data)
475{
476 unsigned long pnum;
477 unsigned long map_count;
478 int nodeid_begin = 0;
479 unsigned long pnum_begin = 0;
480
481 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
482 struct mem_section *ms;
483
484 if (!present_section_nr(pnum))
485 continue;
486 ms = __nr_to_section(pnum);
487 nodeid_begin = sparse_early_nid(ms);
488 pnum_begin = pnum;
489 break;
490 }
491 map_count = 1;
492 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
493 struct mem_section *ms;
494 int nodeid;
495
496 if (!present_section_nr(pnum))
497 continue;
498 ms = __nr_to_section(pnum);
499 nodeid = sparse_early_nid(ms);
500 if (nodeid == nodeid_begin) {
501 map_count++;
502 continue;
503 }
504 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
505 alloc_func(data, pnum_begin, pnum,
506 map_count, nodeid_begin);
507 /* new start, update count etc*/
508 nodeid_begin = nodeid;
509 pnum_begin = pnum;
510 map_count = 1;
511 }
512 /* ok, last chunk */
513 alloc_func(data, pnum_begin, NR_MEM_SECTIONS,
514 map_count, nodeid_begin);
515}
516
193faea9
SR
517/*
518 * Allocate the accumulated non-linear sections, allocate a mem_map
519 * for each and record the physical to section mapping.
520 */
521void __init sparse_init(void)
522{
523 unsigned long pnum;
524 struct page *map;
5c0e3066 525 unsigned long *usemap;
e123dd3f 526 unsigned long **usemap_map;
81d0d950 527 int size;
81d0d950 528#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
81d0d950
YL
529 int size2;
530 struct page **map_map;
531#endif
e123dd3f 532
55878e88
CS
533 /* see include/linux/mmzone.h 'struct mem_section' definition */
534 BUILD_BUG_ON(!is_power_of_2(sizeof(struct mem_section)));
535
ca57df79
XQ
536 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
537 set_pageblock_order();
538
e123dd3f
YL
539 /*
540 * map is using big page (aka 2M in x86 64 bit)
541 * usemap is less one page (aka 24 bytes)
542 * so alloc 2M (with 2M align) and 24 bytes in turn will
543 * make next 2M slip to one more 2M later.
544 * then in big system, the memory will have a lot of holes...
25985edc 545 * here try to allocate 2M pages continuously.
e123dd3f
YL
546 *
547 * powerpc need to call sparse_init_one_section right after each
548 * sparse_early_mem_map_alloc, so allocate usemap_map at first.
549 */
550 size = sizeof(unsigned long *) * NR_MEM_SECTIONS;
bb016b84 551 usemap_map = memblock_virt_alloc(size, 0);
e123dd3f
YL
552 if (!usemap_map)
553 panic("can not allocate usemap_map\n");
18732093
WL
554 alloc_usemap_and_memmap(sparse_early_usemaps_alloc_node,
555 (void *)usemap_map);
193faea9 556
9bdac914
YL
557#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
558 size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
bb016b84 559 map_map = memblock_virt_alloc(size2, 0);
9bdac914
YL
560 if (!map_map)
561 panic("can not allocate map_map\n");
18732093
WL
562 alloc_usemap_and_memmap(sparse_early_mem_maps_alloc_node,
563 (void *)map_map);
9bdac914
YL
564#endif
565
e123dd3f
YL
566 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
567 if (!present_section_nr(pnum))
193faea9 568 continue;
5c0e3066 569
e123dd3f 570 usemap = usemap_map[pnum];
5c0e3066
MG
571 if (!usemap)
572 continue;
573
9bdac914
YL
574#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
575 map = map_map[pnum];
576#else
e123dd3f 577 map = sparse_early_mem_map_alloc(pnum);
9bdac914 578#endif
e123dd3f
YL
579 if (!map)
580 continue;
581
5c0e3066
MG
582 sparse_init_one_section(__nr_to_section(pnum), pnum, map,
583 usemap);
193faea9 584 }
e123dd3f 585
c2b91e2e
YL
586 vmemmap_populate_print_last();
587
9bdac914 588#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
bb016b84 589 memblock_free_early(__pa(map_map), size2);
9bdac914 590#endif
bb016b84 591 memblock_free_early(__pa(usemap_map), size);
193faea9
SR
592}
593
594#ifdef CONFIG_MEMORY_HOTPLUG
98f3cfc1 595#ifdef CONFIG_SPARSEMEM_VMEMMAP
85b35fea 596static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid)
98f3cfc1
YG
597{
598 /* This will make the necessary allocations eventually. */
599 return sparse_mem_map_populate(pnum, nid);
600}
85b35fea 601static void __kfree_section_memmap(struct page *memmap)
98f3cfc1 602{
0aad818b 603 unsigned long start = (unsigned long)memmap;
85b35fea 604 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
0aad818b
JW
605
606 vmemmap_free(start, end);
98f3cfc1 607}
4edd7cef 608#ifdef CONFIG_MEMORY_HOTREMOVE
81556b02 609static void free_map_bootmem(struct page *memmap)
0c0a4a51 610{
0aad818b 611 unsigned long start = (unsigned long)memmap;
81556b02 612 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
0aad818b
JW
613
614 vmemmap_free(start, end);
0c0a4a51 615}
4edd7cef 616#endif /* CONFIG_MEMORY_HOTREMOVE */
98f3cfc1 617#else
85b35fea 618static struct page *__kmalloc_section_memmap(void)
0b0acbec
DH
619{
620 struct page *page, *ret;
85b35fea 621 unsigned long memmap_size = sizeof(struct page) * PAGES_PER_SECTION;
0b0acbec 622
f2d0aa5b 623 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
0b0acbec
DH
624 if (page)
625 goto got_map_page;
626
627 ret = vmalloc(memmap_size);
628 if (ret)
629 goto got_map_ptr;
630
631 return NULL;
632got_map_page:
633 ret = (struct page *)pfn_to_kaddr(page_to_pfn(page));
634got_map_ptr:
0b0acbec
DH
635
636 return ret;
637}
638
85b35fea 639static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid)
98f3cfc1 640{
85b35fea 641 return __kmalloc_section_memmap();
98f3cfc1
YG
642}
643
85b35fea 644static void __kfree_section_memmap(struct page *memmap)
0b0acbec 645{
9e2779fa 646 if (is_vmalloc_addr(memmap))
0b0acbec
DH
647 vfree(memmap);
648 else
649 free_pages((unsigned long)memmap,
85b35fea 650 get_order(sizeof(struct page) * PAGES_PER_SECTION));
0b0acbec 651}
0c0a4a51 652
4edd7cef 653#ifdef CONFIG_MEMORY_HOTREMOVE
81556b02 654static void free_map_bootmem(struct page *memmap)
0c0a4a51
YG
655{
656 unsigned long maps_section_nr, removing_section_nr, i;
81556b02 657 unsigned long magic, nr_pages;
ae64ffca 658 struct page *page = virt_to_page(memmap);
0c0a4a51 659
81556b02
ZY
660 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
661 >> PAGE_SHIFT;
662
0c0a4a51 663 for (i = 0; i < nr_pages; i++, page++) {
5f24ce5f 664 magic = (unsigned long) page->lru.next;
0c0a4a51
YG
665
666 BUG_ON(magic == NODE_INFO);
667
668 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
669 removing_section_nr = page->private;
670
671 /*
672 * When this function is called, the removing section is
673 * logical offlined state. This means all pages are isolated
674 * from page allocator. If removing section's memmap is placed
675 * on the same section, it must not be freed.
676 * If it is freed, page allocator may allocate it which will
677 * be removed physically soon.
678 */
679 if (maps_section_nr != removing_section_nr)
680 put_page_bootmem(page);
681 }
682}
4edd7cef 683#endif /* CONFIG_MEMORY_HOTREMOVE */
98f3cfc1 684#endif /* CONFIG_SPARSEMEM_VMEMMAP */
0b0acbec 685
29751f69
AW
686/*
687 * returns the number of sections whose mem_maps were properly
688 * set. If this is <=0, then that means that the passed-in
689 * map was not consumed and must be freed.
690 */
85b35fea 691int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn)
29751f69 692{
0b0acbec
DH
693 unsigned long section_nr = pfn_to_section_nr(start_pfn);
694 struct pglist_data *pgdat = zone->zone_pgdat;
695 struct mem_section *ms;
696 struct page *memmap;
5c0e3066 697 unsigned long *usemap;
0b0acbec
DH
698 unsigned long flags;
699 int ret;
29751f69 700
0b0acbec
DH
701 /*
702 * no locking for this, because it does its own
703 * plus, it does a kmalloc
704 */
bbd06825
WC
705 ret = sparse_index_init(section_nr, pgdat->node_id);
706 if (ret < 0 && ret != -EEXIST)
707 return ret;
85b35fea 708 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id);
bbd06825
WC
709 if (!memmap)
710 return -ENOMEM;
5c0e3066 711 usemap = __kmalloc_section_usemap();
bbd06825 712 if (!usemap) {
85b35fea 713 __kfree_section_memmap(memmap);
bbd06825
WC
714 return -ENOMEM;
715 }
0b0acbec
DH
716
717 pgdat_resize_lock(pgdat, &flags);
29751f69 718
0b0acbec
DH
719 ms = __pfn_to_section(start_pfn);
720 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
721 ret = -EEXIST;
722 goto out;
723 }
5c0e3066 724
85b35fea 725 memset(memmap, 0, sizeof(struct page) * PAGES_PER_SECTION);
3ac19f8e 726
29751f69
AW
727 ms->section_mem_map |= SECTION_MARKED_PRESENT;
728
5c0e3066 729 ret = sparse_init_one_section(ms, section_nr, memmap, usemap);
0b0acbec 730
0b0acbec
DH
731out:
732 pgdat_resize_unlock(pgdat, &flags);
bbd06825
WC
733 if (ret <= 0) {
734 kfree(usemap);
85b35fea 735 __kfree_section_memmap(memmap);
bbd06825 736 }
0b0acbec 737 return ret;
29751f69 738}
ea01ea93 739
f3deb687 740#ifdef CONFIG_MEMORY_HOTREMOVE
95a4774d
WC
741#ifdef CONFIG_MEMORY_FAILURE
742static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
743{
744 int i;
745
746 if (!memmap)
747 return;
748
749 for (i = 0; i < PAGES_PER_SECTION; i++) {
750 if (PageHWPoison(&memmap[i])) {
293c07e3 751 atomic_long_sub(1, &num_poisoned_pages);
95a4774d
WC
752 ClearPageHWPoison(&memmap[i]);
753 }
754 }
755}
756#else
757static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
758{
759}
760#endif
761
4edd7cef
DR
762static void free_section_usemap(struct page *memmap, unsigned long *usemap)
763{
764 struct page *usemap_page;
4edd7cef
DR
765
766 if (!usemap)
767 return;
768
769 usemap_page = virt_to_page(usemap);
770 /*
771 * Check to see if allocation came from hot-plug-add
772 */
773 if (PageSlab(usemap_page) || PageCompound(usemap_page)) {
774 kfree(usemap);
775 if (memmap)
85b35fea 776 __kfree_section_memmap(memmap);
4edd7cef
DR
777 return;
778 }
779
780 /*
781 * The usemap came from bootmem. This is packed with other usemaps
782 * on the section which has pgdat at boot time. Just keep it as is now.
783 */
784
81556b02
ZY
785 if (memmap)
786 free_map_bootmem(memmap);
4edd7cef
DR
787}
788
ea01ea93
BP
789void sparse_remove_one_section(struct zone *zone, struct mem_section *ms)
790{
791 struct page *memmap = NULL;
cd099682
TC
792 unsigned long *usemap = NULL, flags;
793 struct pglist_data *pgdat = zone->zone_pgdat;
ea01ea93 794
cd099682 795 pgdat_resize_lock(pgdat, &flags);
ea01ea93
BP
796 if (ms->section_mem_map) {
797 usemap = ms->pageblock_flags;
798 memmap = sparse_decode_mem_map(ms->section_mem_map,
799 __section_nr(ms));
800 ms->section_mem_map = 0;
801 ms->pageblock_flags = NULL;
802 }
cd099682 803 pgdat_resize_unlock(pgdat, &flags);
ea01ea93 804
95a4774d 805 clear_hwpoisoned_pages(memmap, PAGES_PER_SECTION);
ea01ea93
BP
806 free_section_usemap(memmap, usemap);
807}
4edd7cef
DR
808#endif /* CONFIG_MEMORY_HOTREMOVE */
809#endif /* CONFIG_MEMORY_HOTPLUG */