mm: sparse: fix section usemap placement calculation
[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 275sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
238305bb 276 unsigned long size)
48c90682 277{
238305bb
JW
278 pg_data_t *host_pgdat;
279 unsigned long goal;
48c90682
YG
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 */
07b4e2bc 290 goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT);
238305bb
JW
291 host_pgdat = NODE_DATA(early_pfn_to_nid(goal >> PAGE_SHIFT));
292 return __alloc_bootmem_node_nopanic(host_pgdat, size,
293 SMP_CACHE_BYTES, goal);
48c90682
YG
294}
295
296static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
297{
298 unsigned long usemap_snr, pgdat_snr;
299 static unsigned long old_usemap_snr = NR_MEM_SECTIONS;
300 static unsigned long old_pgdat_snr = NR_MEM_SECTIONS;
301 struct pglist_data *pgdat = NODE_DATA(nid);
302 int usemap_nid;
303
304 usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT);
305 pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
306 if (usemap_snr == pgdat_snr)
307 return;
308
309 if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr)
310 /* skip redundant message */
311 return;
312
313 old_usemap_snr = usemap_snr;
314 old_pgdat_snr = pgdat_snr;
315
316 usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr));
317 if (usemap_nid != nid) {
318 printk(KERN_INFO
319 "node %d must be removed before remove section %ld\n",
320 nid, usemap_snr);
321 return;
322 }
323 /*
324 * There is a circular dependency.
325 * Some platforms allow un-removable section because they will just
326 * gather other removable sections for dynamic partitioning.
327 * Just notify un-removable section's number here.
328 */
329 printk(KERN_INFO "Section %ld and %ld (node %d)", usemap_snr,
330 pgdat_snr, nid);
331 printk(KERN_CONT
332 " have a circular dependency on usemap and pgdat allocations\n");
333}
334#else
335static unsigned long * __init
a4322e1b 336sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
238305bb 337 unsigned long size)
48c90682 338{
238305bb 339 return alloc_bootmem_node_nopanic(pgdat, size);
48c90682
YG
340}
341
342static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
343{
344}
345#endif /* CONFIG_MEMORY_HOTREMOVE */
346
a4322e1b
YL
347static void __init sparse_early_usemaps_alloc_node(unsigned long**usemap_map,
348 unsigned long pnum_begin,
349 unsigned long pnum_end,
350 unsigned long usemap_count, int nodeid)
5c0e3066 351{
a4322e1b
YL
352 void *usemap;
353 unsigned long pnum;
354 int size = usemap_size();
5c0e3066 355
a4322e1b 356 usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid),
238305bb 357 size * usemap_count);
f5bf18fa 358 if (!usemap) {
238305bb
JW
359 printk(KERN_WARNING "%s: allocation failed\n", __func__);
360 return;
48c90682
YG
361 }
362
f5bf18fa
NA
363 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
364 if (!present_section_nr(pnum))
365 continue;
366 usemap_map[pnum] = usemap;
367 usemap += size;
368 check_usemap_section_nr(nodeid, usemap_map[pnum]);
a4322e1b 369 }
5c0e3066
MG
370}
371
8f6aac41 372#ifndef CONFIG_SPARSEMEM_VMEMMAP
98f3cfc1 373struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
29751f69
AW
374{
375 struct page *map;
e48e67e0 376 unsigned long size;
29751f69
AW
377
378 map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
379 if (map)
380 return map;
381
e48e67e0
YL
382 size = PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
383 map = __alloc_bootmem_node_high(NODE_DATA(nid), size,
384 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
8f6aac41
CL
385 return map;
386}
9bdac914
YL
387void __init sparse_mem_maps_populate_node(struct page **map_map,
388 unsigned long pnum_begin,
389 unsigned long pnum_end,
390 unsigned long map_count, int nodeid)
391{
392 void *map;
393 unsigned long pnum;
394 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
395
396 map = alloc_remap(nodeid, size * map_count);
397 if (map) {
398 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
399 if (!present_section_nr(pnum))
400 continue;
401 map_map[pnum] = map;
402 map += size;
403 }
404 return;
405 }
406
407 size = PAGE_ALIGN(size);
e48e67e0
YL
408 map = __alloc_bootmem_node_high(NODE_DATA(nodeid), size * map_count,
409 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
9bdac914
YL
410 if (map) {
411 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
412 if (!present_section_nr(pnum))
413 continue;
414 map_map[pnum] = map;
415 map += size;
416 }
417 return;
418 }
419
420 /* fallback */
421 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
422 struct mem_section *ms;
423
424 if (!present_section_nr(pnum))
425 continue;
426 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
427 if (map_map[pnum])
428 continue;
429 ms = __nr_to_section(pnum);
430 printk(KERN_ERR "%s: sparsemem memory map backing failed "
431 "some memory will not be available.\n", __func__);
432 ms->section_mem_map = 0;
433 }
434}
8f6aac41
CL
435#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
436
81d0d950 437#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
9bdac914
YL
438static void __init sparse_early_mem_maps_alloc_node(struct page **map_map,
439 unsigned long pnum_begin,
440 unsigned long pnum_end,
441 unsigned long map_count, int nodeid)
442{
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
193faea9
SR
468/*
469 * Allocate the accumulated non-linear sections, allocate a mem_map
470 * for each and record the physical to section mapping.
471 */
472void __init sparse_init(void)
473{
474 unsigned long pnum;
475 struct page *map;
5c0e3066 476 unsigned long *usemap;
e123dd3f 477 unsigned long **usemap_map;
81d0d950 478 int size;
a4322e1b
YL
479 int nodeid_begin = 0;
480 unsigned long pnum_begin = 0;
481 unsigned long usemap_count;
81d0d950 482#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
9bdac914 483 unsigned long map_count;
81d0d950
YL
484 int size2;
485 struct page **map_map;
486#endif
e123dd3f
YL
487
488 /*
489 * map is using big page (aka 2M in x86 64 bit)
490 * usemap is less one page (aka 24 bytes)
491 * so alloc 2M (with 2M align) and 24 bytes in turn will
492 * make next 2M slip to one more 2M later.
493 * then in big system, the memory will have a lot of holes...
25985edc 494 * here try to allocate 2M pages continuously.
e123dd3f
YL
495 *
496 * powerpc need to call sparse_init_one_section right after each
497 * sparse_early_mem_map_alloc, so allocate usemap_map at first.
498 */
499 size = sizeof(unsigned long *) * NR_MEM_SECTIONS;
500 usemap_map = alloc_bootmem(size);
501 if (!usemap_map)
502 panic("can not allocate usemap_map\n");
193faea9
SR
503
504 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
a4322e1b
YL
505 struct mem_section *ms;
506
540557b9 507 if (!present_section_nr(pnum))
193faea9 508 continue;
a4322e1b
YL
509 ms = __nr_to_section(pnum);
510 nodeid_begin = sparse_early_nid(ms);
511 pnum_begin = pnum;
512 break;
513 }
514 usemap_count = 1;
515 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
516 struct mem_section *ms;
517 int nodeid;
518
519 if (!present_section_nr(pnum))
520 continue;
521 ms = __nr_to_section(pnum);
522 nodeid = sparse_early_nid(ms);
523 if (nodeid == nodeid_begin) {
524 usemap_count++;
525 continue;
526 }
527 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
528 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, pnum,
529 usemap_count, nodeid_begin);
530 /* new start, update count etc*/
531 nodeid_begin = nodeid;
532 pnum_begin = pnum;
533 usemap_count = 1;
e123dd3f 534 }
a4322e1b
YL
535 /* ok, last chunk */
536 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, NR_MEM_SECTIONS,
537 usemap_count, nodeid_begin);
193faea9 538
9bdac914
YL
539#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
540 size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
541 map_map = alloc_bootmem(size2);
542 if (!map_map)
543 panic("can not allocate map_map\n");
544
545 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
546 struct mem_section *ms;
547
548 if (!present_section_nr(pnum))
549 continue;
550 ms = __nr_to_section(pnum);
551 nodeid_begin = sparse_early_nid(ms);
552 pnum_begin = pnum;
553 break;
554 }
555 map_count = 1;
556 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
557 struct mem_section *ms;
558 int nodeid;
559
560 if (!present_section_nr(pnum))
561 continue;
562 ms = __nr_to_section(pnum);
563 nodeid = sparse_early_nid(ms);
564 if (nodeid == nodeid_begin) {
565 map_count++;
566 continue;
567 }
568 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
569 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, pnum,
570 map_count, nodeid_begin);
571 /* new start, update count etc*/
572 nodeid_begin = nodeid;
573 pnum_begin = pnum;
574 map_count = 1;
575 }
576 /* ok, last chunk */
577 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, NR_MEM_SECTIONS,
578 map_count, nodeid_begin);
579#endif
580
e123dd3f
YL
581 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
582 if (!present_section_nr(pnum))
193faea9 583 continue;
5c0e3066 584
e123dd3f 585 usemap = usemap_map[pnum];
5c0e3066
MG
586 if (!usemap)
587 continue;
588
9bdac914
YL
589#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
590 map = map_map[pnum];
591#else
e123dd3f 592 map = sparse_early_mem_map_alloc(pnum);
9bdac914 593#endif
e123dd3f
YL
594 if (!map)
595 continue;
596
5c0e3066
MG
597 sparse_init_one_section(__nr_to_section(pnum), pnum, map,
598 usemap);
193faea9 599 }
e123dd3f 600
c2b91e2e
YL
601 vmemmap_populate_print_last();
602
9bdac914
YL
603#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
604 free_bootmem(__pa(map_map), size2);
605#endif
e123dd3f 606 free_bootmem(__pa(usemap_map), size);
193faea9
SR
607}
608
609#ifdef CONFIG_MEMORY_HOTPLUG
98f3cfc1
YG
610#ifdef CONFIG_SPARSEMEM_VMEMMAP
611static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
612 unsigned long nr_pages)
613{
614 /* This will make the necessary allocations eventually. */
615 return sparse_mem_map_populate(pnum, nid);
616}
617static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
618{
619 return; /* XXX: Not implemented yet */
620}
0c0a4a51
YG
621static void free_map_bootmem(struct page *page, unsigned long nr_pages)
622{
623}
98f3cfc1 624#else
0b0acbec
DH
625static struct page *__kmalloc_section_memmap(unsigned long nr_pages)
626{
627 struct page *page, *ret;
628 unsigned long memmap_size = sizeof(struct page) * nr_pages;
629
f2d0aa5b 630 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
0b0acbec
DH
631 if (page)
632 goto got_map_page;
633
634 ret = vmalloc(memmap_size);
635 if (ret)
636 goto got_map_ptr;
637
638 return NULL;
639got_map_page:
640 ret = (struct page *)pfn_to_kaddr(page_to_pfn(page));
641got_map_ptr:
642 memset(ret, 0, memmap_size);
643
644 return ret;
645}
646
98f3cfc1
YG
647static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
648 unsigned long nr_pages)
649{
650 return __kmalloc_section_memmap(nr_pages);
651}
652
0b0acbec
DH
653static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
654{
9e2779fa 655 if (is_vmalloc_addr(memmap))
0b0acbec
DH
656 vfree(memmap);
657 else
658 free_pages((unsigned long)memmap,
659 get_order(sizeof(struct page) * nr_pages));
660}
0c0a4a51
YG
661
662static void free_map_bootmem(struct page *page, unsigned long nr_pages)
663{
664 unsigned long maps_section_nr, removing_section_nr, i;
5f24ce5f 665 unsigned long magic;
0c0a4a51
YG
666
667 for (i = 0; i < nr_pages; i++, page++) {
5f24ce5f 668 magic = (unsigned long) page->lru.next;
0c0a4a51
YG
669
670 BUG_ON(magic == NODE_INFO);
671
672 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
673 removing_section_nr = page->private;
674
675 /*
676 * When this function is called, the removing section is
677 * logical offlined state. This means all pages are isolated
678 * from page allocator. If removing section's memmap is placed
679 * on the same section, it must not be freed.
680 * If it is freed, page allocator may allocate it which will
681 * be removed physically soon.
682 */
683 if (maps_section_nr != removing_section_nr)
684 put_page_bootmem(page);
685 }
686}
98f3cfc1 687#endif /* CONFIG_SPARSEMEM_VMEMMAP */
0b0acbec 688
ea01ea93
BP
689static void free_section_usemap(struct page *memmap, unsigned long *usemap)
690{
0c0a4a51
YG
691 struct page *usemap_page;
692 unsigned long nr_pages;
693
ea01ea93
BP
694 if (!usemap)
695 return;
696
0c0a4a51 697 usemap_page = virt_to_page(usemap);
ea01ea93
BP
698 /*
699 * Check to see if allocation came from hot-plug-add
700 */
0c0a4a51 701 if (PageSlab(usemap_page)) {
ea01ea93
BP
702 kfree(usemap);
703 if (memmap)
704 __kfree_section_memmap(memmap, PAGES_PER_SECTION);
705 return;
706 }
707
708 /*
0c0a4a51
YG
709 * The usemap came from bootmem. This is packed with other usemaps
710 * on the section which has pgdat at boot time. Just keep it as is now.
ea01ea93 711 */
0c0a4a51
YG
712
713 if (memmap) {
714 struct page *memmap_page;
715 memmap_page = virt_to_page(memmap);
716
717 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
718 >> PAGE_SHIFT;
719
720 free_map_bootmem(memmap_page, nr_pages);
721 }
ea01ea93
BP
722}
723
29751f69
AW
724/*
725 * returns the number of sections whose mem_maps were properly
726 * set. If this is <=0, then that means that the passed-in
727 * map was not consumed and must be freed.
728 */
31168481 729int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
0b0acbec 730 int nr_pages)
29751f69 731{
0b0acbec
DH
732 unsigned long section_nr = pfn_to_section_nr(start_pfn);
733 struct pglist_data *pgdat = zone->zone_pgdat;
734 struct mem_section *ms;
735 struct page *memmap;
5c0e3066 736 unsigned long *usemap;
0b0acbec
DH
737 unsigned long flags;
738 int ret;
29751f69 739
0b0acbec
DH
740 /*
741 * no locking for this, because it does its own
742 * plus, it does a kmalloc
743 */
bbd06825
WC
744 ret = sparse_index_init(section_nr, pgdat->node_id);
745 if (ret < 0 && ret != -EEXIST)
746 return ret;
98f3cfc1 747 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id, nr_pages);
bbd06825
WC
748 if (!memmap)
749 return -ENOMEM;
5c0e3066 750 usemap = __kmalloc_section_usemap();
bbd06825
WC
751 if (!usemap) {
752 __kfree_section_memmap(memmap, nr_pages);
753 return -ENOMEM;
754 }
0b0acbec
DH
755
756 pgdat_resize_lock(pgdat, &flags);
29751f69 757
0b0acbec
DH
758 ms = __pfn_to_section(start_pfn);
759 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
760 ret = -EEXIST;
761 goto out;
762 }
5c0e3066 763
29751f69
AW
764 ms->section_mem_map |= SECTION_MARKED_PRESENT;
765
5c0e3066 766 ret = sparse_init_one_section(ms, section_nr, memmap, usemap);
0b0acbec 767
0b0acbec
DH
768out:
769 pgdat_resize_unlock(pgdat, &flags);
bbd06825
WC
770 if (ret <= 0) {
771 kfree(usemap);
46a66eec 772 __kfree_section_memmap(memmap, nr_pages);
bbd06825 773 }
0b0acbec 774 return ret;
29751f69 775}
ea01ea93
BP
776
777void sparse_remove_one_section(struct zone *zone, struct mem_section *ms)
778{
779 struct page *memmap = NULL;
780 unsigned long *usemap = NULL;
781
782 if (ms->section_mem_map) {
783 usemap = ms->pageblock_flags;
784 memmap = sparse_decode_mem_map(ms->section_mem_map,
785 __section_nr(ms));
786 ms->section_mem_map = 0;
787 ms->pageblock_flags = NULL;
788 }
789
790 free_section_usemap(memmap, usemap);
791}
a3142c8e 792#endif