ARM: SMDK6440: Add audio devices on board
[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>
d41dee36 9#include <linux/module.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
25ba77c1 43int page_to_nid(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);
48c90682 356 if (usemap) {
a4322e1b
YL
357 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
358 if (!present_section_nr(pnum))
359 continue;
360 usemap_map[pnum] = usemap;
361 usemap += size;
362 }
363 return;
48c90682
YG
364 }
365
a4322e1b
YL
366 usemap = alloc_bootmem_node(NODE_DATA(nodeid), size * usemap_count);
367 if (usemap) {
368 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
369 if (!present_section_nr(pnum))
370 continue;
371 usemap_map[pnum] = usemap;
372 usemap += size;
373 check_usemap_section_nr(nodeid, usemap_map[pnum]);
374 }
375 return;
376 }
5c0e3066 377
d40cee24 378 printk(KERN_WARNING "%s: allocation failed\n", __func__);
5c0e3066
MG
379}
380
8f6aac41 381#ifndef CONFIG_SPARSEMEM_VMEMMAP
98f3cfc1 382struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
29751f69
AW
383{
384 struct page *map;
29751f69
AW
385
386 map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
387 if (map)
388 return map;
389
9d99217a
YG
390 map = alloc_bootmem_pages_node(NODE_DATA(nid),
391 PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION));
8f6aac41
CL
392 return map;
393}
9bdac914
YL
394void __init sparse_mem_maps_populate_node(struct page **map_map,
395 unsigned long pnum_begin,
396 unsigned long pnum_end,
397 unsigned long map_count, int nodeid)
398{
399 void *map;
400 unsigned long pnum;
401 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
402
403 map = alloc_remap(nodeid, size * map_count);
404 if (map) {
405 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
406 if (!present_section_nr(pnum))
407 continue;
408 map_map[pnum] = map;
409 map += size;
410 }
411 return;
412 }
413
414 size = PAGE_ALIGN(size);
415 map = alloc_bootmem_pages_node(NODE_DATA(nodeid), size * map_count);
416 if (map) {
417 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
418 if (!present_section_nr(pnum))
419 continue;
420 map_map[pnum] = map;
421 map += size;
422 }
423 return;
424 }
425
426 /* fallback */
427 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
428 struct mem_section *ms;
429
430 if (!present_section_nr(pnum))
431 continue;
432 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
433 if (map_map[pnum])
434 continue;
435 ms = __nr_to_section(pnum);
436 printk(KERN_ERR "%s: sparsemem memory map backing failed "
437 "some memory will not be available.\n", __func__);
438 ms->section_mem_map = 0;
439 }
440}
8f6aac41
CL
441#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
442
81d0d950 443#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
9bdac914
YL
444static void __init sparse_early_mem_maps_alloc_node(struct page **map_map,
445 unsigned long pnum_begin,
446 unsigned long pnum_end,
447 unsigned long map_count, int nodeid)
448{
449 sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end,
450 map_count, nodeid);
451}
81d0d950 452#else
9e5c6da7 453static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
8f6aac41
CL
454{
455 struct page *map;
456 struct mem_section *ms = __nr_to_section(pnum);
457 int nid = sparse_early_nid(ms);
458
98f3cfc1 459 map = sparse_mem_map_populate(pnum, nid);
29751f69
AW
460 if (map)
461 return map;
462
8f6aac41 463 printk(KERN_ERR "%s: sparsemem memory map backing failed "
d40cee24 464 "some memory will not be available.\n", __func__);
802f192e 465 ms->section_mem_map = 0;
29751f69
AW
466 return NULL;
467}
9bdac914 468#endif
29751f69 469
c2b91e2e
YL
470void __attribute__((weak)) __meminit vmemmap_populate_print_last(void)
471{
472}
a4322e1b 473
193faea9
SR
474/*
475 * Allocate the accumulated non-linear sections, allocate a mem_map
476 * for each and record the physical to section mapping.
477 */
478void __init sparse_init(void)
479{
480 unsigned long pnum;
481 struct page *map;
5c0e3066 482 unsigned long *usemap;
e123dd3f 483 unsigned long **usemap_map;
81d0d950 484 int size;
a4322e1b
YL
485 int nodeid_begin = 0;
486 unsigned long pnum_begin = 0;
487 unsigned long usemap_count;
81d0d950 488#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
9bdac914 489 unsigned long map_count;
81d0d950
YL
490 int size2;
491 struct page **map_map;
492#endif
e123dd3f
YL
493
494 /*
495 * map is using big page (aka 2M in x86 64 bit)
496 * usemap is less one page (aka 24 bytes)
497 * so alloc 2M (with 2M align) and 24 bytes in turn will
498 * make next 2M slip to one more 2M later.
499 * then in big system, the memory will have a lot of holes...
500 * here try to allocate 2M pages continously.
501 *
502 * powerpc need to call sparse_init_one_section right after each
503 * sparse_early_mem_map_alloc, so allocate usemap_map at first.
504 */
505 size = sizeof(unsigned long *) * NR_MEM_SECTIONS;
506 usemap_map = alloc_bootmem(size);
507 if (!usemap_map)
508 panic("can not allocate usemap_map\n");
193faea9
SR
509
510 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
a4322e1b
YL
511 struct mem_section *ms;
512
540557b9 513 if (!present_section_nr(pnum))
193faea9 514 continue;
a4322e1b
YL
515 ms = __nr_to_section(pnum);
516 nodeid_begin = sparse_early_nid(ms);
517 pnum_begin = pnum;
518 break;
519 }
520 usemap_count = 1;
521 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
522 struct mem_section *ms;
523 int nodeid;
524
525 if (!present_section_nr(pnum))
526 continue;
527 ms = __nr_to_section(pnum);
528 nodeid = sparse_early_nid(ms);
529 if (nodeid == nodeid_begin) {
530 usemap_count++;
531 continue;
532 }
533 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
534 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, pnum,
535 usemap_count, nodeid_begin);
536 /* new start, update count etc*/
537 nodeid_begin = nodeid;
538 pnum_begin = pnum;
539 usemap_count = 1;
e123dd3f 540 }
a4322e1b
YL
541 /* ok, last chunk */
542 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, NR_MEM_SECTIONS,
543 usemap_count, nodeid_begin);
193faea9 544
9bdac914
YL
545#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
546 size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
547 map_map = alloc_bootmem(size2);
548 if (!map_map)
549 panic("can not allocate map_map\n");
550
551 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
552 struct mem_section *ms;
553
554 if (!present_section_nr(pnum))
555 continue;
556 ms = __nr_to_section(pnum);
557 nodeid_begin = sparse_early_nid(ms);
558 pnum_begin = pnum;
559 break;
560 }
561 map_count = 1;
562 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
563 struct mem_section *ms;
564 int nodeid;
565
566 if (!present_section_nr(pnum))
567 continue;
568 ms = __nr_to_section(pnum);
569 nodeid = sparse_early_nid(ms);
570 if (nodeid == nodeid_begin) {
571 map_count++;
572 continue;
573 }
574 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
575 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, pnum,
576 map_count, nodeid_begin);
577 /* new start, update count etc*/
578 nodeid_begin = nodeid;
579 pnum_begin = pnum;
580 map_count = 1;
581 }
582 /* ok, last chunk */
583 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, NR_MEM_SECTIONS,
584 map_count, nodeid_begin);
585#endif
586
e123dd3f
YL
587 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
588 if (!present_section_nr(pnum))
193faea9 589 continue;
5c0e3066 590
e123dd3f 591 usemap = usemap_map[pnum];
5c0e3066
MG
592 if (!usemap)
593 continue;
594
9bdac914
YL
595#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
596 map = map_map[pnum];
597#else
e123dd3f 598 map = sparse_early_mem_map_alloc(pnum);
9bdac914 599#endif
e123dd3f
YL
600 if (!map)
601 continue;
602
5c0e3066
MG
603 sparse_init_one_section(__nr_to_section(pnum), pnum, map,
604 usemap);
193faea9 605 }
e123dd3f 606
c2b91e2e
YL
607 vmemmap_populate_print_last();
608
9bdac914
YL
609#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
610 free_bootmem(__pa(map_map), size2);
611#endif
e123dd3f 612 free_bootmem(__pa(usemap_map), size);
193faea9
SR
613}
614
615#ifdef CONFIG_MEMORY_HOTPLUG
98f3cfc1
YG
616#ifdef CONFIG_SPARSEMEM_VMEMMAP
617static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
618 unsigned long nr_pages)
619{
620 /* This will make the necessary allocations eventually. */
621 return sparse_mem_map_populate(pnum, nid);
622}
623static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
624{
625 return; /* XXX: Not implemented yet */
626}
0c0a4a51
YG
627static void free_map_bootmem(struct page *page, unsigned long nr_pages)
628{
629}
98f3cfc1 630#else
0b0acbec
DH
631static struct page *__kmalloc_section_memmap(unsigned long nr_pages)
632{
633 struct page *page, *ret;
634 unsigned long memmap_size = sizeof(struct page) * nr_pages;
635
f2d0aa5b 636 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
0b0acbec
DH
637 if (page)
638 goto got_map_page;
639
640 ret = vmalloc(memmap_size);
641 if (ret)
642 goto got_map_ptr;
643
644 return NULL;
645got_map_page:
646 ret = (struct page *)pfn_to_kaddr(page_to_pfn(page));
647got_map_ptr:
648 memset(ret, 0, memmap_size);
649
650 return ret;
651}
652
98f3cfc1
YG
653static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
654 unsigned long nr_pages)
655{
656 return __kmalloc_section_memmap(nr_pages);
657}
658
0b0acbec
DH
659static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
660{
9e2779fa 661 if (is_vmalloc_addr(memmap))
0b0acbec
DH
662 vfree(memmap);
663 else
664 free_pages((unsigned long)memmap,
665 get_order(sizeof(struct page) * nr_pages));
666}
0c0a4a51
YG
667
668static void free_map_bootmem(struct page *page, unsigned long nr_pages)
669{
670 unsigned long maps_section_nr, removing_section_nr, i;
671 int magic;
672
673 for (i = 0; i < nr_pages; i++, page++) {
674 magic = atomic_read(&page->_mapcount);
675
676 BUG_ON(magic == NODE_INFO);
677
678 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
679 removing_section_nr = page->private;
680
681 /*
682 * When this function is called, the removing section is
683 * logical offlined state. This means all pages are isolated
684 * from page allocator. If removing section's memmap is placed
685 * on the same section, it must not be freed.
686 * If it is freed, page allocator may allocate it which will
687 * be removed physically soon.
688 */
689 if (maps_section_nr != removing_section_nr)
690 put_page_bootmem(page);
691 }
692}
98f3cfc1 693#endif /* CONFIG_SPARSEMEM_VMEMMAP */
0b0acbec 694
ea01ea93
BP
695static void free_section_usemap(struct page *memmap, unsigned long *usemap)
696{
0c0a4a51
YG
697 struct page *usemap_page;
698 unsigned long nr_pages;
699
ea01ea93
BP
700 if (!usemap)
701 return;
702
0c0a4a51 703 usemap_page = virt_to_page(usemap);
ea01ea93
BP
704 /*
705 * Check to see if allocation came from hot-plug-add
706 */
0c0a4a51 707 if (PageSlab(usemap_page)) {
ea01ea93
BP
708 kfree(usemap);
709 if (memmap)
710 __kfree_section_memmap(memmap, PAGES_PER_SECTION);
711 return;
712 }
713
714 /*
0c0a4a51
YG
715 * The usemap came from bootmem. This is packed with other usemaps
716 * on the section which has pgdat at boot time. Just keep it as is now.
ea01ea93 717 */
0c0a4a51
YG
718
719 if (memmap) {
720 struct page *memmap_page;
721 memmap_page = virt_to_page(memmap);
722
723 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
724 >> PAGE_SHIFT;
725
726 free_map_bootmem(memmap_page, nr_pages);
727 }
ea01ea93
BP
728}
729
29751f69
AW
730/*
731 * returns the number of sections whose mem_maps were properly
732 * set. If this is <=0, then that means that the passed-in
733 * map was not consumed and must be freed.
734 */
31168481 735int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
0b0acbec 736 int nr_pages)
29751f69 737{
0b0acbec
DH
738 unsigned long section_nr = pfn_to_section_nr(start_pfn);
739 struct pglist_data *pgdat = zone->zone_pgdat;
740 struct mem_section *ms;
741 struct page *memmap;
5c0e3066 742 unsigned long *usemap;
0b0acbec
DH
743 unsigned long flags;
744 int ret;
29751f69 745
0b0acbec
DH
746 /*
747 * no locking for this, because it does its own
748 * plus, it does a kmalloc
749 */
bbd06825
WC
750 ret = sparse_index_init(section_nr, pgdat->node_id);
751 if (ret < 0 && ret != -EEXIST)
752 return ret;
98f3cfc1 753 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id, nr_pages);
bbd06825
WC
754 if (!memmap)
755 return -ENOMEM;
5c0e3066 756 usemap = __kmalloc_section_usemap();
bbd06825
WC
757 if (!usemap) {
758 __kfree_section_memmap(memmap, nr_pages);
759 return -ENOMEM;
760 }
0b0acbec
DH
761
762 pgdat_resize_lock(pgdat, &flags);
29751f69 763
0b0acbec
DH
764 ms = __pfn_to_section(start_pfn);
765 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
766 ret = -EEXIST;
767 goto out;
768 }
5c0e3066 769
29751f69
AW
770 ms->section_mem_map |= SECTION_MARKED_PRESENT;
771
5c0e3066 772 ret = sparse_init_one_section(ms, section_nr, memmap, usemap);
0b0acbec 773
0b0acbec
DH
774out:
775 pgdat_resize_unlock(pgdat, &flags);
bbd06825
WC
776 if (ret <= 0) {
777 kfree(usemap);
46a66eec 778 __kfree_section_memmap(memmap, nr_pages);
bbd06825 779 }
0b0acbec 780 return ret;
29751f69 781}
ea01ea93
BP
782
783void sparse_remove_one_section(struct zone *zone, struct mem_section *ms)
784{
785 struct page *memmap = NULL;
786 unsigned long *usemap = NULL;
787
788 if (ms->section_mem_map) {
789 usemap = ms->pageblock_flags;
790 memmap = sparse_decode_mem_map(ms->section_mem_map,
791 __section_nr(ms));
792 ms->section_mem_map = 0;
793 ms->pageblock_flags = NULL;
794 }
795
796 free_section_usemap(memmap, usemap);
797}
a3142c8e 798#endif