blob: 6a4bf9160e855ae1e2d61fefb4922918f710bb24 [file] [log] [blame]
Andy Whitcroftd41dee32005-06-23 00:07:54 -07001/*
2 * sparse memory mappings.
3 */
Andy Whitcroftd41dee32005-06-23 00:07:54 -07004#include <linux/mm.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09005#include <linux/slab.h>
Andy Whitcroftd41dee32005-06-23 00:07:54 -07006#include <linux/mmzone.h>
7#include <linux/bootmem.h>
Dave Hansen0b0acbe2005-10-29 18:16:55 -07008#include <linux/highmem.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -04009#include <linux/export.h>
Dave Hansen28ae55c2005-09-03 15:54:29 -070010#include <linux/spinlock.h>
Dave Hansen0b0acbe2005-10-29 18:16:55 -070011#include <linux/vmalloc.h>
Yasunori Goto0c0a4a52008-04-28 02:13:34 -070012#include "internal.h"
Andy Whitcroftd41dee32005-06-23 00:07:54 -070013#include <asm/dma.h>
Christoph Lameter8f6aac42007-10-16 01:24:13 -070014#include <asm/pgalloc.h>
15#include <asm/pgtable.h>
Andy Whitcroftd41dee32005-06-23 00:07:54 -070016
17/*
18 * Permanent SPARSEMEM data:
19 *
20 * 1) mem_section - memory sections, mem_map's for valid memory
21 */
Bob Picco3e347262005-09-03 15:54:28 -070022#ifdef CONFIG_SPARSEMEM_EXTREME
Bob Picco802f1922005-09-03 15:54:26 -070023struct mem_section *mem_section[NR_SECTION_ROOTS]
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -080024 ____cacheline_internodealigned_in_smp;
Bob Picco3e347262005-09-03 15:54:28 -070025#else
26struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -080027 ____cacheline_internodealigned_in_smp;
Bob Picco3e347262005-09-03 15:54:28 -070028#endif
29EXPORT_SYMBOL(mem_section);
30
Christoph Lameter89689ae2006-12-06 20:31:45 -080031#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
Ian Campbell33dd4e02011-07-25 17:11:51 -070043int page_to_nid(const struct page *page)
Christoph Lameter89689ae2006-12-06 20:31:45 -080044{
45 return section_to_node_table[page_to_section(page)];
46}
47EXPORT_SYMBOL(page_to_nid);
Andy Whitcroft85770ff2007-08-22 14:01:03 -070048
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}
Christoph Lameter89689ae2006-12-06 20:31:45 -080057#endif
58
Bob Picco3e347262005-09-03 15:54:28 -070059#ifdef CONFIG_SPARSEMEM_EXTREME
Sam Ravnborg577a32f2007-05-17 23:29:25 +020060static struct mem_section noinline __init_refok *sparse_index_alloc(int nid)
Bob Picco802f1922005-09-03 15:54:26 -070061{
Dave Hansen28ae55c2005-09-03 15:54:29 -070062 struct mem_section *section = NULL;
63 unsigned long array_size = SECTIONS_PER_ROOT *
64 sizeof(struct mem_section);
Bob Picco802f1922005-09-03 15:54:26 -070065
Shaohua Lif52407c2009-09-21 17:01:19 -070066 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
Mike Kravetz46a66ee2006-05-01 12:16:09 -070072 section = alloc_bootmem_node(NODE_DATA(nid), array_size);
Bob Picco3e347262005-09-03 15:54:28 -070073
Dave Hansen28ae55c2005-09-03 15:54:29 -070074 if (section)
75 memset(section, 0, array_size);
Bob Picco3e347262005-09-03 15:54:28 -070076
Dave Hansen28ae55c2005-09-03 15:54:29 -070077 return section;
Bob Picco802f1922005-09-03 15:54:26 -070078}
Dave Hansen28ae55c2005-09-03 15:54:29 -070079
Yasunori Gotoa3142c82007-05-08 00:23:07 -070080static int __meminit sparse_index_init(unsigned long section_nr, int nid)
Dave Hansen28ae55c2005-09-03 15:54:29 -070081{
Ingo Molnar34af9462006-06-27 02:53:55 -070082 static DEFINE_SPINLOCK(index_init_lock);
Dave Hansen28ae55c2005-09-03 15:54:29 -070083 unsigned long root = SECTION_NR_TO_ROOT(section_nr);
84 struct mem_section *section;
85 int ret = 0;
86
87 if (mem_section[root])
88 return -EEXIST;
89
90 section = sparse_index_alloc(nid);
WANG Congaf0cd5a2007-12-17 16:19:58 -080091 if (!section)
92 return -ENOMEM;
Dave Hansen28ae55c2005-09-03 15:54:29 -070093 /*
94 * This lock keeps two different sections from
95 * reallocating for the same index
96 */
97 spin_lock(&index_init_lock);
98
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;
113}
114#endif
115
Dave Hansen4ca644d2005-10-29 18:16:51 -0700116/*
117 * Although written for the SPARSEMEM_EXTREME case, this happens
Andy Whitcroftcd881a62007-10-16 01:24:10 -0700118 * to also work for the flat array case because
Dave Hansen4ca644d2005-10-29 18:16:51 -0700119 * 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
Mike Kravetz12783b02006-05-20 15:00:05 -0700126 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
127 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
Dave Hansen4ca644d2005-10-29 18:16:51 -0700128 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
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700138/*
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
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700154/* Validate the physical addressing limitations of the model */
155void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
156 unsigned long *end_pfn)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700157{
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700158 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700159
Ingo Molnarbead9a32008-04-16 01:40:00 +0200160 /*
161 * Sanity checks - do not allow an architecture to pass
162 * in larger pfns than the maximum scope of sparsemem:
163 */
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700164 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;
Cyrill Gorcunovef161a92009-03-31 15:19:25 -0700171 } else if (*end_pfn > max_sparsemem_pfn) {
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700172 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;
Ingo Molnarbead9a32008-04-16 01:40:00 +0200184
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700185 start &= PAGE_SECTION_MASK;
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700186 mminit_validate_memmodel_limits(&start, &end);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700187 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
188 unsigned long section = pfn_to_section_nr(pfn);
Bob Picco802f1922005-09-03 15:54:26 -0700189 struct mem_section *ms;
190
191 sparse_index_init(section, nid);
Andy Whitcroft85770ff2007-08-22 14:01:03 -0700192 set_section_nid(section, nid);
Bob Picco802f1922005-09-03 15:54:26 -0700193
194 ms = __nr_to_section(section);
195 if (!ms->section_mem_map)
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700196 ms->section_mem_map = sparse_encode_early_nid(nid) |
197 SECTION_MARKED_PRESENT;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700198 }
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
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700211 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700212 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
213 if (nid != early_pfn_to_nid(pfn))
214 continue;
215
Andy Whitcroft540557b2007-10-16 01:24:11 -0700216 if (pfn_present(pfn))
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700217 nr_pages += PAGES_PER_SECTION;
218 }
219
220 return nr_pages * sizeof(struct page);
221}
222
223/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700224 * 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/*
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700234 * Decode mem_map from the coded memmap
Andy Whitcroft29751f62005-06-23 00:08:00 -0700235 */
Andy Whitcroft29751f62005-06-23 00:08:00 -0700236struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
237{
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700238 /* mask off the extra low bits of information */
239 coded_mem_map &= SECTION_MAP_MASK;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700240 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
241}
242
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700243static int __meminit sparse_init_one_section(struct mem_section *ms,
Mel Gorman5c0e3062007-10-16 01:25:56 -0700244 unsigned long pnum, struct page *mem_map,
245 unsigned long *pageblock_bitmap)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700246{
Andy Whitcroft540557b2007-10-16 01:24:11 -0700247 if (!present_section(ms))
Andy Whitcroft29751f62005-06-23 00:08:00 -0700248 return -EINVAL;
249
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700250 ms->section_mem_map &= ~SECTION_MAP_MASK;
Andy Whitcroft540557b2007-10-16 01:24:11 -0700251 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) |
252 SECTION_HAS_MEM_MAP;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700253 ms->pageblock_flags = pageblock_bitmap;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700254
255 return 1;
256}
257
Yasunori Goto04753272008-04-28 02:13:31 -0700258unsigned long usemap_size(void)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700259{
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
Yasunori Goto48c90682008-07-23 21:28:15 -0700273#ifdef CONFIG_MEMORY_HOTREMOVE
274static unsigned long * __init
Yinghai Lua4322e12010-02-10 01:20:21 -0800275sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700276 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700277{
Johannes Weiner238305b2012-05-29 15:06:36 -0700278 pg_data_t *host_pgdat;
279 unsigned long goal;
Yasunori Goto48c90682008-07-23 21:28:15 -0700280 /*
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 */
Johannes Weiner238305b2012-05-29 15:06:36 -0700290 goal = __pa(pgdat) & PAGE_SECTION_MASK;
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);
Yasunori Goto48c90682008-07-23 21:28:15 -0700294}
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
Yinghai Lua4322e12010-02-10 01:20:21 -0800336sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700337 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700338{
Johannes Weiner238305b2012-05-29 15:06:36 -0700339 return alloc_bootmem_node_nopanic(pgdat, size);
Yasunori Goto48c90682008-07-23 21:28:15 -0700340}
341
342static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
343{
344}
345#endif /* CONFIG_MEMORY_HOTREMOVE */
346
Yinghai Lua4322e12010-02-10 01:20:21 -0800347static 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)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700351{
Yinghai Lua4322e12010-02-10 01:20:21 -0800352 void *usemap;
353 unsigned long pnum;
354 int size = usemap_size();
Mel Gorman5c0e3062007-10-16 01:25:56 -0700355
Yinghai Lua4322e12010-02-10 01:20:21 -0800356 usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid),
Johannes Weiner238305b2012-05-29 15:06:36 -0700357 size * usemap_count);
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700358 if (!usemap) {
Johannes Weiner238305b2012-05-29 15:06:36 -0700359 printk(KERN_WARNING "%s: allocation failed\n", __func__);
360 return;
Yasunori Goto48c90682008-07-23 21:28:15 -0700361 }
362
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700363 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]);
Yinghai Lua4322e12010-02-10 01:20:21 -0800369 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700370}
371
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700372#ifndef CONFIG_SPARSEMEM_VMEMMAP
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700373struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700374{
375 struct page *map;
Yinghai Lue48e67e2010-05-24 14:31:57 -0700376 unsigned long size;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700377
378 map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
379 if (map)
380 return map;
381
Yinghai Lue48e67e2010-05-24 14:31:57 -0700382 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));
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700385 return map;
386}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800387void __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);
Yinghai Lue48e67e2010-05-24 14:31:57 -0700408 map = __alloc_bootmem_node_high(NODE_DATA(nodeid), size * map_count,
409 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
Yinghai Lu9bdac912010-02-10 01:20:22 -0800410 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}
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700435#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
436
Yinghai Lu81d0d952010-02-27 09:29:38 -0800437#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Yinghai Lu9bdac912010-02-10 01:20:22 -0800438static 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}
Yinghai Lu81d0d952010-02-27 09:29:38 -0800446#else
Adrian Bunk9e5c6da2008-07-25 19:46:22 -0700447static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700448{
449 struct page *map;
450 struct mem_section *ms = __nr_to_section(pnum);
451 int nid = sparse_early_nid(ms);
452
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700453 map = sparse_mem_map_populate(pnum, nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -0700454 if (map)
455 return map;
456
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700457 printk(KERN_ERR "%s: sparsemem memory map backing failed "
Harvey Harrisond40cee22008-04-30 00:55:07 -0700458 "some memory will not be available.\n", __func__);
Bob Picco802f1922005-09-03 15:54:26 -0700459 ms->section_mem_map = 0;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700460 return NULL;
461}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800462#endif
Andy Whitcroft29751f62005-06-23 00:08:00 -0700463
Yinghai Luc2b91e22008-04-12 01:19:24 -0700464void __attribute__((weak)) __meminit vmemmap_populate_print_last(void)
465{
466}
Yinghai Lua4322e12010-02-10 01:20:21 -0800467
Stephen Rothwell193faea2007-06-08 13:46:51 -0700468/*
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;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700476 unsigned long *usemap;
Yinghai Lue123dd32008-04-13 11:51:06 -0700477 unsigned long **usemap_map;
Yinghai Lu81d0d952010-02-27 09:29:38 -0800478 int size;
Yinghai Lua4322e12010-02-10 01:20:21 -0800479 int nodeid_begin = 0;
480 unsigned long pnum_begin = 0;
481 unsigned long usemap_count;
Yinghai Lu81d0d952010-02-27 09:29:38 -0800482#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Yinghai Lu9bdac912010-02-10 01:20:22 -0800483 unsigned long map_count;
Yinghai Lu81d0d952010-02-27 09:29:38 -0800484 int size2;
485 struct page **map_map;
486#endif
Yinghai Lue123dd32008-04-13 11:51:06 -0700487
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...
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300494 * here try to allocate 2M pages continuously.
Yinghai Lue123dd32008-04-13 11:51:06 -0700495 *
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");
Stephen Rothwell193faea2007-06-08 13:46:51 -0700503
504 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
Yinghai Lua4322e12010-02-10 01:20:21 -0800505 struct mem_section *ms;
506
Andy Whitcroft540557b2007-10-16 01:24:11 -0700507 if (!present_section_nr(pnum))
Stephen Rothwell193faea2007-06-08 13:46:51 -0700508 continue;
Yinghai Lua4322e12010-02-10 01:20:21 -0800509 ms = __nr_to_section(pnum);
510 nodeid_begin = sparse_early_nid(ms);
511 pnum_begin = pnum;
512 break;
Yinghai Lue123dd32008-04-13 11:51:06 -0700513 }
Yinghai Lua4322e12010-02-10 01:20:21 -0800514 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;
534 }
535 /* ok, last chunk */
536 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, NR_MEM_SECTIONS,
537 usemap_count, nodeid_begin);
Yinghai Lue123dd32008-04-13 11:51:06 -0700538
Yinghai Lu9bdac912010-02-10 01:20:22 -0800539#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
Yinghai Lue123dd32008-04-13 11:51:06 -0700581 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
582 if (!present_section_nr(pnum))
583 continue;
584
585 usemap = usemap_map[pnum];
586 if (!usemap)
587 continue;
Stephen Rothwell193faea2007-06-08 13:46:51 -0700588
Yinghai Lu9bdac912010-02-10 01:20:22 -0800589#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
590 map = map_map[pnum];
591#else
Stephen Rothwell193faea2007-06-08 13:46:51 -0700592 map = sparse_early_mem_map_alloc(pnum);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800593#endif
Stephen Rothwell193faea2007-06-08 13:46:51 -0700594 if (!map)
595 continue;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700596
Mel Gorman5c0e3062007-10-16 01:25:56 -0700597 sparse_init_one_section(__nr_to_section(pnum), pnum, map,
598 usemap);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700599 }
Yinghai Lue123dd32008-04-13 11:51:06 -0700600
Yinghai Luc2b91e22008-04-12 01:19:24 -0700601 vmemmap_populate_print_last();
602
Yinghai Lu9bdac912010-02-10 01:20:22 -0800603#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
604 free_bootmem(__pa(map_map), size2);
605#endif
Yinghai Lue123dd32008-04-13 11:51:06 -0700606 free_bootmem(__pa(usemap_map), size);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700607}
608
609#ifdef CONFIG_MEMORY_HOTPLUG
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700610#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}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700621static void free_map_bootmem(struct page *page, unsigned long nr_pages)
622{
623}
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700624#else
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700625static 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
Yasunori Gotof2d0aa52006-10-28 10:38:32 -0700630 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700631 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
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700647static 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
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700653static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
654{
Christoph Lameter9e2779f2008-02-04 22:28:34 -0800655 if (is_vmalloc_addr(memmap))
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700656 vfree(memmap);
657 else
658 free_pages((unsigned long)memmap,
659 get_order(sizeof(struct page) * nr_pages));
660}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700661
662static void free_map_bootmem(struct page *page, unsigned long nr_pages)
663{
664 unsigned long maps_section_nr, removing_section_nr, i;
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800665 unsigned long magic;
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700666
667 for (i = 0; i < nr_pages; i++, page++) {
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800668 magic = (unsigned long) page->lru.next;
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700669
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}
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700687#endif /* CONFIG_SPARSEMEM_VMEMMAP */
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700688
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700689static void free_section_usemap(struct page *memmap, unsigned long *usemap)
690{
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700691 struct page *usemap_page;
692 unsigned long nr_pages;
693
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700694 if (!usemap)
695 return;
696
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700697 usemap_page = virt_to_page(usemap);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700698 /*
699 * Check to see if allocation came from hot-plug-add
700 */
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700701 if (PageSlab(usemap_page)) {
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700702 kfree(usemap);
703 if (memmap)
704 __kfree_section_memmap(memmap, PAGES_PER_SECTION);
705 return;
706 }
707
708 /*
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700709 * 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.
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700711 */
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700712
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 }
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700722}
723
Andy Whitcroft29751f62005-06-23 00:08:00 -0700724/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700725 * 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 */
Al Viro31168482008-11-22 17:33:24 +0000729int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700730 int nr_pages)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700731{
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700732 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;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700736 unsigned long *usemap;
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700737 unsigned long flags;
738 int ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700739
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700740 /*
741 * no locking for this, because it does its own
742 * plus, it does a kmalloc
743 */
WANG Congbbd06822007-12-17 16:19:59 -0800744 ret = sparse_index_init(section_nr, pgdat->node_id);
745 if (ret < 0 && ret != -EEXIST)
746 return ret;
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700747 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id, nr_pages);
WANG Congbbd06822007-12-17 16:19:59 -0800748 if (!memmap)
749 return -ENOMEM;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700750 usemap = __kmalloc_section_usemap();
WANG Congbbd06822007-12-17 16:19:59 -0800751 if (!usemap) {
752 __kfree_section_memmap(memmap, nr_pages);
753 return -ENOMEM;
754 }
Andy Whitcroft29751f62005-06-23 00:08:00 -0700755
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700756 pgdat_resize_lock(pgdat, &flags);
757
758 ms = __pfn_to_section(start_pfn);
759 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
760 ret = -EEXIST;
761 goto out;
762 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700763
Andy Whitcroft29751f62005-06-23 00:08:00 -0700764 ms->section_mem_map |= SECTION_MARKED_PRESENT;
765
Mel Gorman5c0e3062007-10-16 01:25:56 -0700766 ret = sparse_init_one_section(ms, section_nr, memmap, usemap);
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700767
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700768out:
769 pgdat_resize_unlock(pgdat, &flags);
WANG Congbbd06822007-12-17 16:19:59 -0800770 if (ret <= 0) {
771 kfree(usemap);
Mike Kravetz46a66ee2006-05-01 12:16:09 -0700772 __kfree_section_memmap(memmap, nr_pages);
WANG Congbbd06822007-12-17 16:19:59 -0800773 }
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700774 return ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700775}
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700776
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}
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700792#endif