blob: 7ca6dc8479475cf982c607193b501371d94abf07 [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))
Gavin Shan5b760e62012-07-31 16:46:02 -070068 section = kzalloc_node(array_size, GFP_KERNEL, nid);
Shaohua Lif52407c2009-09-21 17:01:19 -070069 else
Gavin Shan5b760e62012-07-31 16:46:02 -070070 section = kzalloc(array_size, GFP_KERNEL);
71 } else {
Mike Kravetz46a66ee2006-05-01 12:16:09 -070072 section = alloc_bootmem_node(NODE_DATA(nid), array_size);
Gavin Shan5b760e62012-07-31 16:46:02 -070073 }
Bob Picco3e347262005-09-03 15:54:28 -070074
Dave Hansen28ae55c2005-09-03 15:54:29 -070075 return section;
Bob Picco802f1922005-09-03 15:54:26 -070076}
Dave Hansen28ae55c2005-09-03 15:54:29 -070077
Yasunori Gotoa3142c82007-05-08 00:23:07 -070078static int __meminit sparse_index_init(unsigned long section_nr, int nid)
Dave Hansen28ae55c2005-09-03 15:54:29 -070079{
Dave Hansen28ae55c2005-09-03 15:54:29 -070080 unsigned long root = SECTION_NR_TO_ROOT(section_nr);
81 struct mem_section *section;
82 int ret = 0;
83
84 if (mem_section[root])
85 return -EEXIST;
86
87 section = sparse_index_alloc(nid);
WANG Congaf0cd5a2007-12-17 16:19:58 -080088 if (!section)
89 return -ENOMEM;
Dave Hansen28ae55c2005-09-03 15:54:29 -070090
91 mem_section[root] = section;
Gavin Shanc1c95182012-07-31 16:46:06 -070092
Dave Hansen28ae55c2005-09-03 15:54:29 -070093 return ret;
94}
95#else /* !SPARSEMEM_EXTREME */
96static inline int sparse_index_init(unsigned long section_nr, int nid)
97{
98 return 0;
99}
100#endif
101
Dave Hansen4ca644d2005-10-29 18:16:51 -0700102/*
103 * Although written for the SPARSEMEM_EXTREME case, this happens
Andy Whitcroftcd881a62007-10-16 01:24:10 -0700104 * to also work for the flat array case because
Dave Hansen4ca644d2005-10-29 18:16:51 -0700105 * NR_SECTION_ROOTS==NR_MEM_SECTIONS.
106 */
107int __section_nr(struct mem_section* ms)
108{
109 unsigned long root_nr;
110 struct mem_section* root;
111
Mike Kravetz12783b02006-05-20 15:00:05 -0700112 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
113 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
Dave Hansen4ca644d2005-10-29 18:16:51 -0700114 if (!root)
115 continue;
116
117 if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT)))
118 break;
119 }
120
Gavin Shandb36a462012-07-31 16:46:04 -0700121 VM_BUG_ON(root_nr == NR_SECTION_ROOTS);
122
Dave Hansen4ca644d2005-10-29 18:16:51 -0700123 return (root_nr * SECTIONS_PER_ROOT) + (ms - root);
124}
125
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700126/*
127 * During early boot, before section_mem_map is used for an actual
128 * mem_map, we use section_mem_map to store the section's NUMA
129 * node. This keeps us from having to use another data structure. The
130 * node information is cleared just before we store the real mem_map.
131 */
132static inline unsigned long sparse_encode_early_nid(int nid)
133{
134 return (nid << SECTION_NID_SHIFT);
135}
136
137static inline int sparse_early_nid(struct mem_section *section)
138{
139 return (section->section_mem_map >> SECTION_NID_SHIFT);
140}
141
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700142/* Validate the physical addressing limitations of the model */
143void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
144 unsigned long *end_pfn)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700145{
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700146 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700147
Ingo Molnarbead9a32008-04-16 01:40:00 +0200148 /*
149 * Sanity checks - do not allow an architecture to pass
150 * in larger pfns than the maximum scope of sparsemem:
151 */
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700152 if (*start_pfn > max_sparsemem_pfn) {
153 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
154 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
155 *start_pfn, *end_pfn, max_sparsemem_pfn);
156 WARN_ON_ONCE(1);
157 *start_pfn = max_sparsemem_pfn;
158 *end_pfn = max_sparsemem_pfn;
Cyrill Gorcunovef161a92009-03-31 15:19:25 -0700159 } else if (*end_pfn > max_sparsemem_pfn) {
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700160 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
161 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
162 *start_pfn, *end_pfn, max_sparsemem_pfn);
163 WARN_ON_ONCE(1);
164 *end_pfn = max_sparsemem_pfn;
165 }
166}
167
168/* Record a memory area against a node. */
169void __init memory_present(int nid, unsigned long start, unsigned long end)
170{
171 unsigned long pfn;
Ingo Molnarbead9a32008-04-16 01:40:00 +0200172
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700173 start &= PAGE_SECTION_MASK;
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700174 mminit_validate_memmodel_limits(&start, &end);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700175 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
176 unsigned long section = pfn_to_section_nr(pfn);
Bob Picco802f1922005-09-03 15:54:26 -0700177 struct mem_section *ms;
178
179 sparse_index_init(section, nid);
Andy Whitcroft85770ff2007-08-22 14:01:03 -0700180 set_section_nid(section, nid);
Bob Picco802f1922005-09-03 15:54:26 -0700181
182 ms = __nr_to_section(section);
183 if (!ms->section_mem_map)
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700184 ms->section_mem_map = sparse_encode_early_nid(nid) |
185 SECTION_MARKED_PRESENT;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700186 }
187}
188
189/*
190 * Only used by the i386 NUMA architecures, but relatively
191 * generic code.
192 */
193unsigned long __init node_memmap_size_bytes(int nid, unsigned long start_pfn,
194 unsigned long end_pfn)
195{
196 unsigned long pfn;
197 unsigned long nr_pages = 0;
198
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700199 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700200 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
201 if (nid != early_pfn_to_nid(pfn))
202 continue;
203
Andy Whitcroft540557b2007-10-16 01:24:11 -0700204 if (pfn_present(pfn))
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700205 nr_pages += PAGES_PER_SECTION;
206 }
207
208 return nr_pages * sizeof(struct page);
209}
210
211/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700212 * Subtle, we encode the real pfn into the mem_map such that
213 * the identity pfn - section_mem_map will return the actual
214 * physical page frame number.
215 */
216static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
217{
218 return (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
219}
220
221/*
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700222 * Decode mem_map from the coded memmap
Andy Whitcroft29751f62005-06-23 00:08:00 -0700223 */
Andy Whitcroft29751f62005-06-23 00:08:00 -0700224struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
225{
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700226 /* mask off the extra low bits of information */
227 coded_mem_map &= SECTION_MAP_MASK;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700228 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
229}
230
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700231static int __meminit sparse_init_one_section(struct mem_section *ms,
Mel Gorman5c0e3062007-10-16 01:25:56 -0700232 unsigned long pnum, struct page *mem_map,
233 unsigned long *pageblock_bitmap)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700234{
Andy Whitcroft540557b2007-10-16 01:24:11 -0700235 if (!present_section(ms))
Andy Whitcroft29751f62005-06-23 00:08:00 -0700236 return -EINVAL;
237
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700238 ms->section_mem_map &= ~SECTION_MAP_MASK;
Andy Whitcroft540557b2007-10-16 01:24:11 -0700239 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) |
240 SECTION_HAS_MEM_MAP;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700241 ms->pageblock_flags = pageblock_bitmap;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700242
243 return 1;
244}
245
Yasunori Goto04753272008-04-28 02:13:31 -0700246unsigned long usemap_size(void)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700247{
248 unsigned long size_bytes;
249 size_bytes = roundup(SECTION_BLOCKFLAGS_BITS, 8) / 8;
250 size_bytes = roundup(size_bytes, sizeof(unsigned long));
251 return size_bytes;
252}
253
254#ifdef CONFIG_MEMORY_HOTPLUG
255static unsigned long *__kmalloc_section_usemap(void)
256{
257 return kmalloc(usemap_size(), GFP_KERNEL);
258}
259#endif /* CONFIG_MEMORY_HOTPLUG */
260
Yasunori Goto48c90682008-07-23 21:28:15 -0700261#ifdef CONFIG_MEMORY_HOTREMOVE
262static unsigned long * __init
Yinghai Lua4322e12010-02-10 01:20:21 -0800263sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700264 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700265{
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700266 unsigned long goal, limit;
267 unsigned long *p;
268 int nid;
Yasunori Goto48c90682008-07-23 21:28:15 -0700269 /*
270 * A page may contain usemaps for other sections preventing the
271 * page being freed and making a section unremovable while
272 * other sections referencing the usemap retmain active. Similarly,
273 * a pgdat can prevent a section being removed. If section A
274 * contains a pgdat and section B contains the usemap, both
275 * sections become inter-dependent. This allocates usemaps
276 * from the same section as the pgdat where possible to avoid
277 * this problem.
278 */
Yinghai Lu07b4e2b2012-07-11 14:02:51 -0700279 goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT);
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700280 limit = goal + (1UL << PA_SECTION_SHIFT);
281 nid = early_pfn_to_nid(goal >> PAGE_SHIFT);
282again:
283 p = ___alloc_bootmem_node_nopanic(NODE_DATA(nid), size,
284 SMP_CACHE_BYTES, goal, limit);
285 if (!p && limit) {
286 limit = 0;
287 goto again;
288 }
289 return p;
Yasunori Goto48c90682008-07-23 21:28:15 -0700290}
291
292static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
293{
294 unsigned long usemap_snr, pgdat_snr;
295 static unsigned long old_usemap_snr = NR_MEM_SECTIONS;
296 static unsigned long old_pgdat_snr = NR_MEM_SECTIONS;
297 struct pglist_data *pgdat = NODE_DATA(nid);
298 int usemap_nid;
299
300 usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT);
301 pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
302 if (usemap_snr == pgdat_snr)
303 return;
304
305 if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr)
306 /* skip redundant message */
307 return;
308
309 old_usemap_snr = usemap_snr;
310 old_pgdat_snr = pgdat_snr;
311
312 usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr));
313 if (usemap_nid != nid) {
314 printk(KERN_INFO
315 "node %d must be removed before remove section %ld\n",
316 nid, usemap_snr);
317 return;
318 }
319 /*
320 * There is a circular dependency.
321 * Some platforms allow un-removable section because they will just
322 * gather other removable sections for dynamic partitioning.
323 * Just notify un-removable section's number here.
324 */
325 printk(KERN_INFO "Section %ld and %ld (node %d)", usemap_snr,
326 pgdat_snr, nid);
327 printk(KERN_CONT
328 " have a circular dependency on usemap and pgdat allocations\n");
329}
330#else
331static unsigned long * __init
Yinghai Lua4322e12010-02-10 01:20:21 -0800332sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700333 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700334{
Johannes Weiner238305b2012-05-29 15:06:36 -0700335 return alloc_bootmem_node_nopanic(pgdat, size);
Yasunori Goto48c90682008-07-23 21:28:15 -0700336}
337
338static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
339{
340}
341#endif /* CONFIG_MEMORY_HOTREMOVE */
342
Yinghai Lua4322e12010-02-10 01:20:21 -0800343static void __init sparse_early_usemaps_alloc_node(unsigned long**usemap_map,
344 unsigned long pnum_begin,
345 unsigned long pnum_end,
346 unsigned long usemap_count, int nodeid)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700347{
Yinghai Lua4322e12010-02-10 01:20:21 -0800348 void *usemap;
349 unsigned long pnum;
350 int size = usemap_size();
Mel Gorman5c0e3062007-10-16 01:25:56 -0700351
Yinghai Lua4322e12010-02-10 01:20:21 -0800352 usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid),
Johannes Weiner238305b2012-05-29 15:06:36 -0700353 size * usemap_count);
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700354 if (!usemap) {
Johannes Weiner238305b2012-05-29 15:06:36 -0700355 printk(KERN_WARNING "%s: allocation failed\n", __func__);
356 return;
Yasunori Goto48c90682008-07-23 21:28:15 -0700357 }
358
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700359 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
360 if (!present_section_nr(pnum))
361 continue;
362 usemap_map[pnum] = usemap;
363 usemap += size;
364 check_usemap_section_nr(nodeid, usemap_map[pnum]);
Yinghai Lua4322e12010-02-10 01:20:21 -0800365 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700366}
367
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700368#ifndef CONFIG_SPARSEMEM_VMEMMAP
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700369struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700370{
371 struct page *map;
Yinghai Lue48e67e2010-05-24 14:31:57 -0700372 unsigned long size;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700373
374 map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
375 if (map)
376 return map;
377
Yinghai Lue48e67e2010-05-24 14:31:57 -0700378 size = PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
379 map = __alloc_bootmem_node_high(NODE_DATA(nid), size,
380 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700381 return map;
382}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800383void __init sparse_mem_maps_populate_node(struct page **map_map,
384 unsigned long pnum_begin,
385 unsigned long pnum_end,
386 unsigned long map_count, int nodeid)
387{
388 void *map;
389 unsigned long pnum;
390 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
391
392 map = alloc_remap(nodeid, size * map_count);
393 if (map) {
394 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
395 if (!present_section_nr(pnum))
396 continue;
397 map_map[pnum] = map;
398 map += size;
399 }
400 return;
401 }
402
403 size = PAGE_ALIGN(size);
Yinghai Lue48e67e2010-05-24 14:31:57 -0700404 map = __alloc_bootmem_node_high(NODE_DATA(nodeid), size * map_count,
405 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
Yinghai Lu9bdac912010-02-10 01:20:22 -0800406 if (map) {
407 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
408 if (!present_section_nr(pnum))
409 continue;
410 map_map[pnum] = map;
411 map += size;
412 }
413 return;
414 }
415
416 /* fallback */
417 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
418 struct mem_section *ms;
419
420 if (!present_section_nr(pnum))
421 continue;
422 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
423 if (map_map[pnum])
424 continue;
425 ms = __nr_to_section(pnum);
426 printk(KERN_ERR "%s: sparsemem memory map backing failed "
427 "some memory will not be available.\n", __func__);
428 ms->section_mem_map = 0;
429 }
430}
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700431#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
432
Yinghai Lu81d0d952010-02-27 09:29:38 -0800433#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Yinghai Lu9bdac912010-02-10 01:20:22 -0800434static void __init sparse_early_mem_maps_alloc_node(struct page **map_map,
435 unsigned long pnum_begin,
436 unsigned long pnum_end,
437 unsigned long map_count, int nodeid)
438{
439 sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end,
440 map_count, nodeid);
441}
Yinghai Lu81d0d952010-02-27 09:29:38 -0800442#else
Adrian Bunk9e5c6da2008-07-25 19:46:22 -0700443static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700444{
445 struct page *map;
446 struct mem_section *ms = __nr_to_section(pnum);
447 int nid = sparse_early_nid(ms);
448
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700449 map = sparse_mem_map_populate(pnum, nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -0700450 if (map)
451 return map;
452
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700453 printk(KERN_ERR "%s: sparsemem memory map backing failed "
Harvey Harrisond40cee22008-04-30 00:55:07 -0700454 "some memory will not be available.\n", __func__);
Bob Picco802f1922005-09-03 15:54:26 -0700455 ms->section_mem_map = 0;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700456 return NULL;
457}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800458#endif
Andy Whitcroft29751f62005-06-23 00:08:00 -0700459
Yinghai Luc2b91e22008-04-12 01:19:24 -0700460void __attribute__((weak)) __meminit vmemmap_populate_print_last(void)
461{
462}
Yinghai Lua4322e12010-02-10 01:20:21 -0800463
Stephen Rothwell193faea2007-06-08 13:46:51 -0700464/*
465 * Allocate the accumulated non-linear sections, allocate a mem_map
466 * for each and record the physical to section mapping.
467 */
468void __init sparse_init(void)
469{
470 unsigned long pnum;
471 struct page *map;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700472 unsigned long *usemap;
Yinghai Lue123dd32008-04-13 11:51:06 -0700473 unsigned long **usemap_map;
Yinghai Lu81d0d952010-02-27 09:29:38 -0800474 int size;
Yinghai Lua4322e12010-02-10 01:20:21 -0800475 int nodeid_begin = 0;
476 unsigned long pnum_begin = 0;
477 unsigned long usemap_count;
Yinghai Lu81d0d952010-02-27 09:29:38 -0800478#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Yinghai Lu9bdac912010-02-10 01:20:22 -0800479 unsigned long map_count;
Yinghai Lu81d0d952010-02-27 09:29:38 -0800480 int size2;
481 struct page **map_map;
482#endif
Yinghai Lue123dd32008-04-13 11:51:06 -0700483
Xishi Qiuca57df72012-07-31 16:43:19 -0700484 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
485 set_pageblock_order();
486
Yinghai Lue123dd32008-04-13 11:51:06 -0700487 /*
488 * map is using big page (aka 2M in x86 64 bit)
489 * usemap is less one page (aka 24 bytes)
490 * so alloc 2M (with 2M align) and 24 bytes in turn will
491 * make next 2M slip to one more 2M later.
492 * then in big system, the memory will have a lot of holes...
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300493 * here try to allocate 2M pages continuously.
Yinghai Lue123dd32008-04-13 11:51:06 -0700494 *
495 * powerpc need to call sparse_init_one_section right after each
496 * sparse_early_mem_map_alloc, so allocate usemap_map at first.
497 */
498 size = sizeof(unsigned long *) * NR_MEM_SECTIONS;
499 usemap_map = alloc_bootmem(size);
500 if (!usemap_map)
501 panic("can not allocate usemap_map\n");
Stephen Rothwell193faea2007-06-08 13:46:51 -0700502
503 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
Yinghai Lua4322e12010-02-10 01:20:21 -0800504 struct mem_section *ms;
505
Andy Whitcroft540557b2007-10-16 01:24:11 -0700506 if (!present_section_nr(pnum))
Stephen Rothwell193faea2007-06-08 13:46:51 -0700507 continue;
Yinghai Lua4322e12010-02-10 01:20:21 -0800508 ms = __nr_to_section(pnum);
509 nodeid_begin = sparse_early_nid(ms);
510 pnum_begin = pnum;
511 break;
Yinghai Lue123dd32008-04-13 11:51:06 -0700512 }
Yinghai Lua4322e12010-02-10 01:20:21 -0800513 usemap_count = 1;
514 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
515 struct mem_section *ms;
516 int nodeid;
517
518 if (!present_section_nr(pnum))
519 continue;
520 ms = __nr_to_section(pnum);
521 nodeid = sparse_early_nid(ms);
522 if (nodeid == nodeid_begin) {
523 usemap_count++;
524 continue;
525 }
526 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
527 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, pnum,
528 usemap_count, nodeid_begin);
529 /* new start, update count etc*/
530 nodeid_begin = nodeid;
531 pnum_begin = pnum;
532 usemap_count = 1;
533 }
534 /* ok, last chunk */
535 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, NR_MEM_SECTIONS,
536 usemap_count, nodeid_begin);
Yinghai Lue123dd32008-04-13 11:51:06 -0700537
Yinghai Lu9bdac912010-02-10 01:20:22 -0800538#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
539 size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
540 map_map = alloc_bootmem(size2);
541 if (!map_map)
542 panic("can not allocate map_map\n");
543
544 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
545 struct mem_section *ms;
546
547 if (!present_section_nr(pnum))
548 continue;
549 ms = __nr_to_section(pnum);
550 nodeid_begin = sparse_early_nid(ms);
551 pnum_begin = pnum;
552 break;
553 }
554 map_count = 1;
555 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
556 struct mem_section *ms;
557 int nodeid;
558
559 if (!present_section_nr(pnum))
560 continue;
561 ms = __nr_to_section(pnum);
562 nodeid = sparse_early_nid(ms);
563 if (nodeid == nodeid_begin) {
564 map_count++;
565 continue;
566 }
567 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
568 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, pnum,
569 map_count, nodeid_begin);
570 /* new start, update count etc*/
571 nodeid_begin = nodeid;
572 pnum_begin = pnum;
573 map_count = 1;
574 }
575 /* ok, last chunk */
576 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, NR_MEM_SECTIONS,
577 map_count, nodeid_begin);
578#endif
579
Yinghai Lue123dd32008-04-13 11:51:06 -0700580 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
581 if (!present_section_nr(pnum))
582 continue;
583
584 usemap = usemap_map[pnum];
585 if (!usemap)
586 continue;
Stephen Rothwell193faea2007-06-08 13:46:51 -0700587
Yinghai Lu9bdac912010-02-10 01:20:22 -0800588#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
589 map = map_map[pnum];
590#else
Stephen Rothwell193faea2007-06-08 13:46:51 -0700591 map = sparse_early_mem_map_alloc(pnum);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800592#endif
Stephen Rothwell193faea2007-06-08 13:46:51 -0700593 if (!map)
594 continue;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700595
Mel Gorman5c0e3062007-10-16 01:25:56 -0700596 sparse_init_one_section(__nr_to_section(pnum), pnum, map,
597 usemap);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700598 }
Yinghai Lue123dd32008-04-13 11:51:06 -0700599
Yinghai Luc2b91e22008-04-12 01:19:24 -0700600 vmemmap_populate_print_last();
601
Yinghai Lu9bdac912010-02-10 01:20:22 -0800602#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
603 free_bootmem(__pa(map_map), size2);
604#endif
Yinghai Lue123dd32008-04-13 11:51:06 -0700605 free_bootmem(__pa(usemap_map), size);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700606}
607
608#ifdef CONFIG_MEMORY_HOTPLUG
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700609#ifdef CONFIG_SPARSEMEM_VMEMMAP
610static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
611 unsigned long nr_pages)
612{
613 /* This will make the necessary allocations eventually. */
614 return sparse_mem_map_populate(pnum, nid);
615}
616static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
617{
Tang Chen01975182013-02-22 16:33:08 -0800618 vmemmap_free(memmap, nr_pages);
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700619}
Jianguo Wuae64ffc2012-11-29 13:54:21 -0800620static void free_map_bootmem(struct page *memmap, unsigned long nr_pages)
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700621{
Tang Chen01975182013-02-22 16:33:08 -0800622 vmemmap_free(memmap, nr_pages);
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700623}
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:
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700642
643 return ret;
644}
645
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700646static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
647 unsigned long nr_pages)
648{
649 return __kmalloc_section_memmap(nr_pages);
650}
651
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700652static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
653{
Christoph Lameter9e2779f2008-02-04 22:28:34 -0800654 if (is_vmalloc_addr(memmap))
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700655 vfree(memmap);
656 else
657 free_pages((unsigned long)memmap,
658 get_order(sizeof(struct page) * nr_pages));
659}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700660
Jianguo Wuae64ffc2012-11-29 13:54:21 -0800661static void free_map_bootmem(struct page *memmap, unsigned long nr_pages)
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700662{
663 unsigned long maps_section_nr, removing_section_nr, i;
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800664 unsigned long magic;
Jianguo Wuae64ffc2012-11-29 13:54:21 -0800665 struct page *page = virt_to_page(memmap);
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 */
Wen Congyang8a356ce2013-02-22 16:33:21 -0800701 if (PageSlab(usemap_page) || PageCompound(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) {
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700714 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
715 >> PAGE_SHIFT;
716
Jianguo Wuae64ffc2012-11-29 13:54:21 -0800717 free_map_bootmem(memmap, nr_pages);
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700718 }
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700719}
720
Andy Whitcroft29751f62005-06-23 00:08:00 -0700721/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700722 * returns the number of sections whose mem_maps were properly
723 * set. If this is <=0, then that means that the passed-in
724 * map was not consumed and must be freed.
725 */
Al Viro31168482008-11-22 17:33:24 +0000726int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700727 int nr_pages)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700728{
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700729 unsigned long section_nr = pfn_to_section_nr(start_pfn);
730 struct pglist_data *pgdat = zone->zone_pgdat;
731 struct mem_section *ms;
732 struct page *memmap;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700733 unsigned long *usemap;
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700734 unsigned long flags;
735 int ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700736
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700737 /*
738 * no locking for this, because it does its own
739 * plus, it does a kmalloc
740 */
WANG Congbbd06822007-12-17 16:19:59 -0800741 ret = sparse_index_init(section_nr, pgdat->node_id);
742 if (ret < 0 && ret != -EEXIST)
743 return ret;
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700744 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id, nr_pages);
WANG Congbbd06822007-12-17 16:19:59 -0800745 if (!memmap)
746 return -ENOMEM;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700747 usemap = __kmalloc_section_usemap();
WANG Congbbd06822007-12-17 16:19:59 -0800748 if (!usemap) {
749 __kfree_section_memmap(memmap, nr_pages);
750 return -ENOMEM;
751 }
Andy Whitcroft29751f62005-06-23 00:08:00 -0700752
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700753 pgdat_resize_lock(pgdat, &flags);
754
755 ms = __pfn_to_section(start_pfn);
756 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
757 ret = -EEXIST;
758 goto out;
759 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700760
Wen Congyang3ac19f82012-12-11 16:00:59 -0800761 memset(memmap, 0, sizeof(struct page) * nr_pages);
762
Andy Whitcroft29751f62005-06-23 00:08:00 -0700763 ms->section_mem_map |= SECTION_MARKED_PRESENT;
764
Mel Gorman5c0e3062007-10-16 01:25:56 -0700765 ret = sparse_init_one_section(ms, section_nr, memmap, usemap);
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700766
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700767out:
768 pgdat_resize_unlock(pgdat, &flags);
WANG Congbbd06822007-12-17 16:19:59 -0800769 if (ret <= 0) {
770 kfree(usemap);
Mike Kravetz46a66ee2006-05-01 12:16:09 -0700771 __kfree_section_memmap(memmap, nr_pages);
WANG Congbbd06822007-12-17 16:19:59 -0800772 }
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700773 return ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700774}
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700775
Wen Congyang95a47742012-12-11 16:00:47 -0800776#ifdef CONFIG_MEMORY_FAILURE
777static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
778{
779 int i;
780
781 if (!memmap)
782 return;
783
784 for (i = 0; i < PAGES_PER_SECTION; i++) {
785 if (PageHWPoison(&memmap[i])) {
Xishi Qiu293c07e2013-02-22 16:34:02 -0800786 atomic_long_sub(1, &num_poisoned_pages);
Wen Congyang95a47742012-12-11 16:00:47 -0800787 ClearPageHWPoison(&memmap[i]);
788 }
789 }
790}
791#else
792static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
793{
794}
795#endif
796
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700797void sparse_remove_one_section(struct zone *zone, struct mem_section *ms)
798{
799 struct page *memmap = NULL;
Tang Chencd099682013-02-22 16:33:02 -0800800 unsigned long *usemap = NULL, flags;
801 struct pglist_data *pgdat = zone->zone_pgdat;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700802
Tang Chencd099682013-02-22 16:33:02 -0800803 pgdat_resize_lock(pgdat, &flags);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700804 if (ms->section_mem_map) {
805 usemap = ms->pageblock_flags;
806 memmap = sparse_decode_mem_map(ms->section_mem_map,
807 __section_nr(ms));
808 ms->section_mem_map = 0;
809 ms->pageblock_flags = NULL;
810 }
Tang Chencd099682013-02-22 16:33:02 -0800811 pgdat_resize_unlock(pgdat, &flags);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700812
Wen Congyang95a47742012-12-11 16:00:47 -0800813 clear_hwpoisoned_pages(memmap, PAGES_PER_SECTION);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700814 free_section_usemap(memmap, usemap);
815}
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700816#endif