mm: bootmem: pass pgdat instead of pgdat->bdata down the stack
[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))
68 section = kmalloc_node(array_size, GFP_KERNEL, nid);
69 else
70 section = kmalloc(array_size, GFP_KERNEL);
71 } else
46a66eec 72 section = alloc_bootmem_node(NODE_DATA(nid), array_size);
28ae55c9
DH
73
74 if (section)
75 memset(section, 0, array_size);
76
77 return section;
3e347261 78}
802f192e 79
a3142c8e 80static int __meminit sparse_index_init(unsigned long section_nr, int nid)
802f192e 81{
34af946a 82 static DEFINE_SPINLOCK(index_init_lock);
28ae55c9
DH
83 unsigned long root = SECTION_NR_TO_ROOT(section_nr);
84 struct mem_section *section;
85 int ret = 0;
802f192e
BP
86
87 if (mem_section[root])
28ae55c9 88 return -EEXIST;
3e347261 89
28ae55c9 90 section = sparse_index_alloc(nid);
af0cd5a7
WC
91 if (!section)
92 return -ENOMEM;
28ae55c9
DH
93 /*
94 * This lock keeps two different sections from
95 * reallocating for the same index
96 */
97 spin_lock(&index_init_lock);
3e347261 98
28ae55c9
DH
99 if (mem_section[root]) {
100 ret = -EEXIST;
101 goto out;
102 }
103
104 mem_section[root] = section;
105out:
106 spin_unlock(&index_init_lock);
107 return ret;
108}
109#else /* !SPARSEMEM_EXTREME */
110static inline int sparse_index_init(unsigned long section_nr, int nid)
111{
112 return 0;
802f192e 113}
28ae55c9
DH
114#endif
115
4ca644d9
DH
116/*
117 * Although written for the SPARSEMEM_EXTREME case, this happens
cd881a6b 118 * to also work for the flat array case because
4ca644d9
DH
119 * NR_SECTION_ROOTS==NR_MEM_SECTIONS.
120 */
121int __section_nr(struct mem_section* ms)
122{
123 unsigned long root_nr;
124 struct mem_section* root;
125
12783b00
MK
126 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
127 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
4ca644d9
DH
128 if (!root)
129 continue;
130
131 if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT)))
132 break;
133 }
134
135 return (root_nr * SECTIONS_PER_ROOT) + (ms - root);
136}
137
30c253e6
AW
138/*
139 * During early boot, before section_mem_map is used for an actual
140 * mem_map, we use section_mem_map to store the section's NUMA
141 * node. This keeps us from having to use another data structure. The
142 * node information is cleared just before we store the real mem_map.
143 */
144static inline unsigned long sparse_encode_early_nid(int nid)
145{
146 return (nid << SECTION_NID_SHIFT);
147}
148
149static inline int sparse_early_nid(struct mem_section *section)
150{
151 return (section->section_mem_map >> SECTION_NID_SHIFT);
152}
153
2dbb51c4
MG
154/* Validate the physical addressing limitations of the model */
155void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
156 unsigned long *end_pfn)
d41dee36 157{
2dbb51c4 158 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
d41dee36 159
bead9a3a
IM
160 /*
161 * Sanity checks - do not allow an architecture to pass
162 * in larger pfns than the maximum scope of sparsemem:
163 */
2dbb51c4
MG
164 if (*start_pfn > max_sparsemem_pfn) {
165 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
166 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
167 *start_pfn, *end_pfn, max_sparsemem_pfn);
168 WARN_ON_ONCE(1);
169 *start_pfn = max_sparsemem_pfn;
170 *end_pfn = max_sparsemem_pfn;
ef161a98 171 } else if (*end_pfn > max_sparsemem_pfn) {
2dbb51c4
MG
172 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
173 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
174 *start_pfn, *end_pfn, max_sparsemem_pfn);
175 WARN_ON_ONCE(1);
176 *end_pfn = max_sparsemem_pfn;
177 }
178}
179
180/* Record a memory area against a node. */
181void __init memory_present(int nid, unsigned long start, unsigned long end)
182{
183 unsigned long pfn;
bead9a3a 184
d41dee36 185 start &= PAGE_SECTION_MASK;
2dbb51c4 186 mminit_validate_memmodel_limits(&start, &end);
d41dee36
AW
187 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
188 unsigned long section = pfn_to_section_nr(pfn);
802f192e
BP
189 struct mem_section *ms;
190
191 sparse_index_init(section, nid);
85770ffe 192 set_section_nid(section, nid);
802f192e
BP
193
194 ms = __nr_to_section(section);
195 if (!ms->section_mem_map)
30c253e6
AW
196 ms->section_mem_map = sparse_encode_early_nid(nid) |
197 SECTION_MARKED_PRESENT;
d41dee36
AW
198 }
199}
200
201/*
202 * Only used by the i386 NUMA architecures, but relatively
203 * generic code.
204 */
205unsigned long __init node_memmap_size_bytes(int nid, unsigned long start_pfn,
206 unsigned long end_pfn)
207{
208 unsigned long pfn;
209 unsigned long nr_pages = 0;
210
2dbb51c4 211 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
d41dee36
AW
212 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
213 if (nid != early_pfn_to_nid(pfn))
214 continue;
215
540557b9 216 if (pfn_present(pfn))
d41dee36
AW
217 nr_pages += PAGES_PER_SECTION;
218 }
219
220 return nr_pages * sizeof(struct page);
221}
222
29751f69
AW
223/*
224 * Subtle, we encode the real pfn into the mem_map such that
225 * the identity pfn - section_mem_map will return the actual
226 * physical page frame number.
227 */
228static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
229{
230 return (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
231}
232
233/*
ea01ea93 234 * Decode mem_map from the coded memmap
29751f69 235 */
29751f69
AW
236struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
237{
ea01ea93
BP
238 /* mask off the extra low bits of information */
239 coded_mem_map &= SECTION_MAP_MASK;
29751f69
AW
240 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
241}
242
a3142c8e 243static int __meminit sparse_init_one_section(struct mem_section *ms,
5c0e3066
MG
244 unsigned long pnum, struct page *mem_map,
245 unsigned long *pageblock_bitmap)
29751f69 246{
540557b9 247 if (!present_section(ms))
29751f69
AW
248 return -EINVAL;
249
30c253e6 250 ms->section_mem_map &= ~SECTION_MAP_MASK;
540557b9
AW
251 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) |
252 SECTION_HAS_MEM_MAP;
5c0e3066 253 ms->pageblock_flags = pageblock_bitmap;
29751f69
AW
254
255 return 1;
256}
257
04753278 258unsigned long usemap_size(void)
5c0e3066
MG
259{
260 unsigned long size_bytes;
261 size_bytes = roundup(SECTION_BLOCKFLAGS_BITS, 8) / 8;
262 size_bytes = roundup(size_bytes, sizeof(unsigned long));
263 return size_bytes;
264}
265
266#ifdef CONFIG_MEMORY_HOTPLUG
267static unsigned long *__kmalloc_section_usemap(void)
268{
269 return kmalloc(usemap_size(), GFP_KERNEL);
270}
271#endif /* CONFIG_MEMORY_HOTPLUG */
272
48c90682
YG
273#ifdef CONFIG_MEMORY_HOTREMOVE
274static unsigned long * __init
a4322e1b
YL
275sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
276 unsigned long count)
48c90682
YG
277{
278 unsigned long section_nr;
279
280 /*
281 * A page may contain usemaps for other sections preventing the
282 * page being freed and making a section unremovable while
283 * other sections referencing the usemap retmain active. Similarly,
284 * a pgdat can prevent a section being removed. If section A
285 * contains a pgdat and section B contains the usemap, both
286 * sections become inter-dependent. This allocates usemaps
287 * from the same section as the pgdat where possible to avoid
288 * this problem.
289 */
290 section_nr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
a4322e1b 291 return alloc_bootmem_section(usemap_size() * count, section_nr);
48c90682
YG
292}
293
294static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
295{
296 unsigned long usemap_snr, pgdat_snr;
297 static unsigned long old_usemap_snr = NR_MEM_SECTIONS;
298 static unsigned long old_pgdat_snr = NR_MEM_SECTIONS;
299 struct pglist_data *pgdat = NODE_DATA(nid);
300 int usemap_nid;
301
302 usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT);
303 pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
304 if (usemap_snr == pgdat_snr)
305 return;
306
307 if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr)
308 /* skip redundant message */
309 return;
310
311 old_usemap_snr = usemap_snr;
312 old_pgdat_snr = pgdat_snr;
313
314 usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr));
315 if (usemap_nid != nid) {
316 printk(KERN_INFO
317 "node %d must be removed before remove section %ld\n",
318 nid, usemap_snr);
319 return;
320 }
321 /*
322 * There is a circular dependency.
323 * Some platforms allow un-removable section because they will just
324 * gather other removable sections for dynamic partitioning.
325 * Just notify un-removable section's number here.
326 */
327 printk(KERN_INFO "Section %ld and %ld (node %d)", usemap_snr,
328 pgdat_snr, nid);
329 printk(KERN_CONT
330 " have a circular dependency on usemap and pgdat allocations\n");
331}
332#else
333static unsigned long * __init
a4322e1b
YL
334sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
335 unsigned long count)
48c90682
YG
336{
337 return NULL;
338}
339
340static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
341{
342}
343#endif /* CONFIG_MEMORY_HOTREMOVE */
344
a4322e1b
YL
345static void __init sparse_early_usemaps_alloc_node(unsigned long**usemap_map,
346 unsigned long pnum_begin,
347 unsigned long pnum_end,
348 unsigned long usemap_count, int nodeid)
5c0e3066 349{
a4322e1b
YL
350 void *usemap;
351 unsigned long pnum;
352 int size = usemap_size();
5c0e3066 353
a4322e1b
YL
354 usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid),
355 usemap_count);
f5bf18fa
NA
356 if (!usemap) {
357 usemap = alloc_bootmem_node(NODE_DATA(nodeid), size * usemap_count);
358 if (!usemap) {
359 printk(KERN_WARNING "%s: allocation failed\n", __func__);
360 return;
a4322e1b 361 }
48c90682
YG
362 }
363
f5bf18fa
NA
364 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
365 if (!present_section_nr(pnum))
366 continue;
367 usemap_map[pnum] = usemap;
368 usemap += size;
369 check_usemap_section_nr(nodeid, usemap_map[pnum]);
a4322e1b 370 }
5c0e3066
MG
371}
372
8f6aac41 373#ifndef CONFIG_SPARSEMEM_VMEMMAP
98f3cfc1 374struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
29751f69
AW
375{
376 struct page *map;
e48e67e0 377 unsigned long size;
29751f69
AW
378
379 map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
380 if (map)
381 return map;
382
e48e67e0
YL
383 size = PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
384 map = __alloc_bootmem_node_high(NODE_DATA(nid), size,
385 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
8f6aac41
CL
386 return map;
387}
9bdac914
YL
388void __init sparse_mem_maps_populate_node(struct page **map_map,
389 unsigned long pnum_begin,
390 unsigned long pnum_end,
391 unsigned long map_count, int nodeid)
392{
393 void *map;
394 unsigned long pnum;
395 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
396
397 map = alloc_remap(nodeid, size * map_count);
398 if (map) {
399 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
400 if (!present_section_nr(pnum))
401 continue;
402 map_map[pnum] = map;
403 map += size;
404 }
405 return;
406 }
407
408 size = PAGE_ALIGN(size);
e48e67e0
YL
409 map = __alloc_bootmem_node_high(NODE_DATA(nodeid), size * map_count,
410 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
9bdac914
YL
411 if (map) {
412 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
413 if (!present_section_nr(pnum))
414 continue;
415 map_map[pnum] = map;
416 map += size;
417 }
418 return;
419 }
420
421 /* fallback */
422 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
423 struct mem_section *ms;
424
425 if (!present_section_nr(pnum))
426 continue;
427 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
428 if (map_map[pnum])
429 continue;
430 ms = __nr_to_section(pnum);
431 printk(KERN_ERR "%s: sparsemem memory map backing failed "
432 "some memory will not be available.\n", __func__);
433 ms->section_mem_map = 0;
434 }
435}
8f6aac41
CL
436#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
437
81d0d950 438#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
9bdac914
YL
439static void __init sparse_early_mem_maps_alloc_node(struct page **map_map,
440 unsigned long pnum_begin,
441 unsigned long pnum_end,
442 unsigned long map_count, int nodeid)
443{
444 sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end,
445 map_count, nodeid);
446}
81d0d950 447#else
9e5c6da7 448static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
8f6aac41
CL
449{
450 struct page *map;
451 struct mem_section *ms = __nr_to_section(pnum);
452 int nid = sparse_early_nid(ms);
453
98f3cfc1 454 map = sparse_mem_map_populate(pnum, nid);
29751f69
AW
455 if (map)
456 return map;
457
8f6aac41 458 printk(KERN_ERR "%s: sparsemem memory map backing failed "
d40cee24 459 "some memory will not be available.\n", __func__);
802f192e 460 ms->section_mem_map = 0;
29751f69
AW
461 return NULL;
462}
9bdac914 463#endif
29751f69 464
c2b91e2e
YL
465void __attribute__((weak)) __meminit vmemmap_populate_print_last(void)
466{
467}
a4322e1b 468
193faea9
SR
469/*
470 * Allocate the accumulated non-linear sections, allocate a mem_map
471 * for each and record the physical to section mapping.
472 */
473void __init sparse_init(void)
474{
475 unsigned long pnum;
476 struct page *map;
5c0e3066 477 unsigned long *usemap;
e123dd3f 478 unsigned long **usemap_map;
81d0d950 479 int size;
a4322e1b
YL
480 int nodeid_begin = 0;
481 unsigned long pnum_begin = 0;
482 unsigned long usemap_count;
81d0d950 483#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
9bdac914 484 unsigned long map_count;
81d0d950
YL
485 int size2;
486 struct page **map_map;
487#endif
e123dd3f
YL
488
489 /*
490 * map is using big page (aka 2M in x86 64 bit)
491 * usemap is less one page (aka 24 bytes)
492 * so alloc 2M (with 2M align) and 24 bytes in turn will
493 * make next 2M slip to one more 2M later.
494 * then in big system, the memory will have a lot of holes...
25985edc 495 * here try to allocate 2M pages continuously.
e123dd3f
YL
496 *
497 * powerpc need to call sparse_init_one_section right after each
498 * sparse_early_mem_map_alloc, so allocate usemap_map at first.
499 */
500 size = sizeof(unsigned long *) * NR_MEM_SECTIONS;
501 usemap_map = alloc_bootmem(size);
502 if (!usemap_map)
503 panic("can not allocate usemap_map\n");
193faea9
SR
504
505 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
a4322e1b
YL
506 struct mem_section *ms;
507
540557b9 508 if (!present_section_nr(pnum))
193faea9 509 continue;
a4322e1b
YL
510 ms = __nr_to_section(pnum);
511 nodeid_begin = sparse_early_nid(ms);
512 pnum_begin = pnum;
513 break;
514 }
515 usemap_count = 1;
516 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
517 struct mem_section *ms;
518 int nodeid;
519
520 if (!present_section_nr(pnum))
521 continue;
522 ms = __nr_to_section(pnum);
523 nodeid = sparse_early_nid(ms);
524 if (nodeid == nodeid_begin) {
525 usemap_count++;
526 continue;
527 }
528 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
529 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, pnum,
530 usemap_count, nodeid_begin);
531 /* new start, update count etc*/
532 nodeid_begin = nodeid;
533 pnum_begin = pnum;
534 usemap_count = 1;
e123dd3f 535 }
a4322e1b
YL
536 /* ok, last chunk */
537 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, NR_MEM_SECTIONS,
538 usemap_count, nodeid_begin);
193faea9 539
9bdac914
YL
540#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
541 size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
542 map_map = alloc_bootmem(size2);
543 if (!map_map)
544 panic("can not allocate map_map\n");
545
546 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
547 struct mem_section *ms;
548
549 if (!present_section_nr(pnum))
550 continue;
551 ms = __nr_to_section(pnum);
552 nodeid_begin = sparse_early_nid(ms);
553 pnum_begin = pnum;
554 break;
555 }
556 map_count = 1;
557 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
558 struct mem_section *ms;
559 int nodeid;
560
561 if (!present_section_nr(pnum))
562 continue;
563 ms = __nr_to_section(pnum);
564 nodeid = sparse_early_nid(ms);
565 if (nodeid == nodeid_begin) {
566 map_count++;
567 continue;
568 }
569 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
570 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, pnum,
571 map_count, nodeid_begin);
572 /* new start, update count etc*/
573 nodeid_begin = nodeid;
574 pnum_begin = pnum;
575 map_count = 1;
576 }
577 /* ok, last chunk */
578 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, NR_MEM_SECTIONS,
579 map_count, nodeid_begin);
580#endif
581
e123dd3f
YL
582 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
583 if (!present_section_nr(pnum))
193faea9 584 continue;
5c0e3066 585
e123dd3f 586 usemap = usemap_map[pnum];
5c0e3066
MG
587 if (!usemap)
588 continue;
589
9bdac914
YL
590#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
591 map = map_map[pnum];
592#else
e123dd3f 593 map = sparse_early_mem_map_alloc(pnum);
9bdac914 594#endif
e123dd3f
YL
595 if (!map)
596 continue;
597
5c0e3066
MG
598 sparse_init_one_section(__nr_to_section(pnum), pnum, map,
599 usemap);
193faea9 600 }
e123dd3f 601
c2b91e2e
YL
602 vmemmap_populate_print_last();
603
9bdac914
YL
604#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
605 free_bootmem(__pa(map_map), size2);
606#endif
e123dd3f 607 free_bootmem(__pa(usemap_map), size);
193faea9
SR
608}
609
610#ifdef CONFIG_MEMORY_HOTPLUG
98f3cfc1
YG
611#ifdef CONFIG_SPARSEMEM_VMEMMAP
612static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
613 unsigned long nr_pages)
614{
615 /* This will make the necessary allocations eventually. */
616 return sparse_mem_map_populate(pnum, nid);
617}
618static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
619{
620 return; /* XXX: Not implemented yet */
621}
0c0a4a51
YG
622static void free_map_bootmem(struct page *page, unsigned long nr_pages)
623{
624}
98f3cfc1 625#else
0b0acbec
DH
626static struct page *__kmalloc_section_memmap(unsigned long nr_pages)
627{
628 struct page *page, *ret;
629 unsigned long memmap_size = sizeof(struct page) * nr_pages;
630
f2d0aa5b 631 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
0b0acbec
DH
632 if (page)
633 goto got_map_page;
634
635 ret = vmalloc(memmap_size);
636 if (ret)
637 goto got_map_ptr;
638
639 return NULL;
640got_map_page:
641 ret = (struct page *)pfn_to_kaddr(page_to_pfn(page));
642got_map_ptr:
643 memset(ret, 0, memmap_size);
644
645 return ret;
646}
647
98f3cfc1
YG
648static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
649 unsigned long nr_pages)
650{
651 return __kmalloc_section_memmap(nr_pages);
652}
653
0b0acbec
DH
654static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
655{
9e2779fa 656 if (is_vmalloc_addr(memmap))
0b0acbec
DH
657 vfree(memmap);
658 else
659 free_pages((unsigned long)memmap,
660 get_order(sizeof(struct page) * nr_pages));
661}
0c0a4a51
YG
662
663static void free_map_bootmem(struct page *page, unsigned long nr_pages)
664{
665 unsigned long maps_section_nr, removing_section_nr, i;
5f24ce5f 666 unsigned long magic;
0c0a4a51
YG
667
668 for (i = 0; i < nr_pages; i++, page++) {
5f24ce5f 669 magic = (unsigned long) page->lru.next;
0c0a4a51
YG
670
671 BUG_ON(magic == NODE_INFO);
672
673 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
674 removing_section_nr = page->private;
675
676 /*
677 * When this function is called, the removing section is
678 * logical offlined state. This means all pages are isolated
679 * from page allocator. If removing section's memmap is placed
680 * on the same section, it must not be freed.
681 * If it is freed, page allocator may allocate it which will
682 * be removed physically soon.
683 */
684 if (maps_section_nr != removing_section_nr)
685 put_page_bootmem(page);
686 }
687}
98f3cfc1 688#endif /* CONFIG_SPARSEMEM_VMEMMAP */
0b0acbec 689
ea01ea93
BP
690static void free_section_usemap(struct page *memmap, unsigned long *usemap)
691{
0c0a4a51
YG
692 struct page *usemap_page;
693 unsigned long nr_pages;
694
ea01ea93
BP
695 if (!usemap)
696 return;
697
0c0a4a51 698 usemap_page = virt_to_page(usemap);
ea01ea93
BP
699 /*
700 * Check to see if allocation came from hot-plug-add
701 */
0c0a4a51 702 if (PageSlab(usemap_page)) {
ea01ea93
BP
703 kfree(usemap);
704 if (memmap)
705 __kfree_section_memmap(memmap, PAGES_PER_SECTION);
706 return;
707 }
708
709 /*
0c0a4a51
YG
710 * The usemap came from bootmem. This is packed with other usemaps
711 * on the section which has pgdat at boot time. Just keep it as is now.
ea01ea93 712 */
0c0a4a51
YG
713
714 if (memmap) {
715 struct page *memmap_page;
716 memmap_page = virt_to_page(memmap);
717
718 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
719 >> PAGE_SHIFT;
720
721 free_map_bootmem(memmap_page, nr_pages);
722 }
ea01ea93
BP
723}
724
29751f69
AW
725/*
726 * returns the number of sections whose mem_maps were properly
727 * set. If this is <=0, then that means that the passed-in
728 * map was not consumed and must be freed.
729 */
31168481 730int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
0b0acbec 731 int nr_pages)
29751f69 732{
0b0acbec
DH
733 unsigned long section_nr = pfn_to_section_nr(start_pfn);
734 struct pglist_data *pgdat = zone->zone_pgdat;
735 struct mem_section *ms;
736 struct page *memmap;
5c0e3066 737 unsigned long *usemap;
0b0acbec
DH
738 unsigned long flags;
739 int ret;
29751f69 740
0b0acbec
DH
741 /*
742 * no locking for this, because it does its own
743 * plus, it does a kmalloc
744 */
bbd06825
WC
745 ret = sparse_index_init(section_nr, pgdat->node_id);
746 if (ret < 0 && ret != -EEXIST)
747 return ret;
98f3cfc1 748 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id, nr_pages);
bbd06825
WC
749 if (!memmap)
750 return -ENOMEM;
5c0e3066 751 usemap = __kmalloc_section_usemap();
bbd06825
WC
752 if (!usemap) {
753 __kfree_section_memmap(memmap, nr_pages);
754 return -ENOMEM;
755 }
0b0acbec
DH
756
757 pgdat_resize_lock(pgdat, &flags);
29751f69 758
0b0acbec
DH
759 ms = __pfn_to_section(start_pfn);
760 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
761 ret = -EEXIST;
762 goto out;
763 }
5c0e3066 764
29751f69
AW
765 ms->section_mem_map |= SECTION_MARKED_PRESENT;
766
5c0e3066 767 ret = sparse_init_one_section(ms, section_nr, memmap, usemap);
0b0acbec 768
0b0acbec
DH
769out:
770 pgdat_resize_unlock(pgdat, &flags);
bbd06825
WC
771 if (ret <= 0) {
772 kfree(usemap);
46a66eec 773 __kfree_section_memmap(memmap, nr_pages);
bbd06825 774 }
0b0acbec 775 return ret;
29751f69 776}
ea01ea93
BP
777
778void sparse_remove_one_section(struct zone *zone, struct mem_section *ms)
779{
780 struct page *memmap = NULL;
781 unsigned long *usemap = NULL;
782
783 if (ms->section_mem_map) {
784 usemap = ms->pageblock_flags;
785 memmap = sparse_decode_mem_map(ms->section_mem_map,
786 __section_nr(ms));
787 ms->section_mem_map = 0;
788 ms->pageblock_flags = NULL;
789 }
790
791 free_section_usemap(memmap, usemap);
792}
a3142c8e 793#endif