blob: 308d50331bc353e69f00b8ef7ea78555b1426224 [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;
Dave Hansen28ae55c2005-09-03 15:54:29 -070082
83 if (mem_section[root])
84 return -EEXIST;
85
86 section = sparse_index_alloc(nid);
WANG Congaf0cd5a2007-12-17 16:19:58 -080087 if (!section)
88 return -ENOMEM;
Dave Hansen28ae55c2005-09-03 15:54:29 -070089
90 mem_section[root] = section;
Gavin Shanc1c95182012-07-31 16:46:06 -070091
Zhang Yanfei9d1936c2013-05-17 22:10:38 +080092 return 0;
Dave Hansen28ae55c2005-09-03 15:54:29 -070093}
94#else /* !SPARSEMEM_EXTREME */
95static inline int sparse_index_init(unsigned long section_nr, int nid)
96{
97 return 0;
98}
99#endif
100
Dave Hansen4ca644d2005-10-29 18:16:51 -0700101/*
102 * Although written for the SPARSEMEM_EXTREME case, this happens
Andy Whitcroftcd881a62007-10-16 01:24:10 -0700103 * to also work for the flat array case because
Dave Hansen4ca644d2005-10-29 18:16:51 -0700104 * NR_SECTION_ROOTS==NR_MEM_SECTIONS.
105 */
106int __section_nr(struct mem_section* ms)
107{
108 unsigned long root_nr;
109 struct mem_section* root;
110
Mike Kravetz12783b02006-05-20 15:00:05 -0700111 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
112 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
Dave Hansen4ca644d2005-10-29 18:16:51 -0700113 if (!root)
114 continue;
115
116 if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT)))
117 break;
118 }
119
Gavin Shandb36a462012-07-31 16:46:04 -0700120 VM_BUG_ON(root_nr == NR_SECTION_ROOTS);
121
Dave Hansen4ca644d2005-10-29 18:16:51 -0700122 return (root_nr * SECTIONS_PER_ROOT) + (ms - root);
123}
124
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700125/*
126 * During early boot, before section_mem_map is used for an actual
127 * mem_map, we use section_mem_map to store the section's NUMA
128 * node. This keeps us from having to use another data structure. The
129 * node information is cleared just before we store the real mem_map.
130 */
131static inline unsigned long sparse_encode_early_nid(int nid)
132{
133 return (nid << SECTION_NID_SHIFT);
134}
135
136static inline int sparse_early_nid(struct mem_section *section)
137{
138 return (section->section_mem_map >> SECTION_NID_SHIFT);
139}
140
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700141/* Validate the physical addressing limitations of the model */
142void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
143 unsigned long *end_pfn)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700144{
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700145 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700146
Ingo Molnarbead9a32008-04-16 01:40:00 +0200147 /*
148 * Sanity checks - do not allow an architecture to pass
149 * in larger pfns than the maximum scope of sparsemem:
150 */
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700151 if (*start_pfn > max_sparsemem_pfn) {
152 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
153 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
154 *start_pfn, *end_pfn, max_sparsemem_pfn);
155 WARN_ON_ONCE(1);
156 *start_pfn = max_sparsemem_pfn;
157 *end_pfn = max_sparsemem_pfn;
Cyrill Gorcunovef161a92009-03-31 15:19:25 -0700158 } else if (*end_pfn > max_sparsemem_pfn) {
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700159 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
160 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
161 *start_pfn, *end_pfn, max_sparsemem_pfn);
162 WARN_ON_ONCE(1);
163 *end_pfn = max_sparsemem_pfn;
164 }
165}
166
167/* Record a memory area against a node. */
168void __init memory_present(int nid, unsigned long start, unsigned long end)
169{
170 unsigned long pfn;
Ingo Molnarbead9a32008-04-16 01:40:00 +0200171
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700172 start &= PAGE_SECTION_MASK;
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700173 mminit_validate_memmodel_limits(&start, &end);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700174 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
175 unsigned long section = pfn_to_section_nr(pfn);
Bob Picco802f1922005-09-03 15:54:26 -0700176 struct mem_section *ms;
177
178 sparse_index_init(section, nid);
Andy Whitcroft85770ff2007-08-22 14:01:03 -0700179 set_section_nid(section, nid);
Bob Picco802f1922005-09-03 15:54:26 -0700180
181 ms = __nr_to_section(section);
182 if (!ms->section_mem_map)
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700183 ms->section_mem_map = sparse_encode_early_nid(nid) |
184 SECTION_MARKED_PRESENT;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700185 }
186}
187
188/*
189 * Only used by the i386 NUMA architecures, but relatively
190 * generic code.
191 */
192unsigned long __init node_memmap_size_bytes(int nid, unsigned long start_pfn,
193 unsigned long end_pfn)
194{
195 unsigned long pfn;
196 unsigned long nr_pages = 0;
197
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700198 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700199 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
200 if (nid != early_pfn_to_nid(pfn))
201 continue;
202
Andy Whitcroft540557b2007-10-16 01:24:11 -0700203 if (pfn_present(pfn))
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700204 nr_pages += PAGES_PER_SECTION;
205 }
206
207 return nr_pages * sizeof(struct page);
208}
209
210/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700211 * Subtle, we encode the real pfn into the mem_map such that
212 * the identity pfn - section_mem_map will return the actual
213 * physical page frame number.
214 */
215static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
216{
217 return (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
218}
219
220/*
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700221 * Decode mem_map from the coded memmap
Andy Whitcroft29751f62005-06-23 00:08:00 -0700222 */
Andy Whitcroft29751f62005-06-23 00:08:00 -0700223struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
224{
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700225 /* mask off the extra low bits of information */
226 coded_mem_map &= SECTION_MAP_MASK;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700227 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
228}
229
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700230static int __meminit sparse_init_one_section(struct mem_section *ms,
Mel Gorman5c0e3062007-10-16 01:25:56 -0700231 unsigned long pnum, struct page *mem_map,
232 unsigned long *pageblock_bitmap)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700233{
Andy Whitcroft540557b2007-10-16 01:24:11 -0700234 if (!present_section(ms))
Andy Whitcroft29751f62005-06-23 00:08:00 -0700235 return -EINVAL;
236
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700237 ms->section_mem_map &= ~SECTION_MAP_MASK;
Andy Whitcroft540557b2007-10-16 01:24:11 -0700238 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) |
239 SECTION_HAS_MEM_MAP;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700240 ms->pageblock_flags = pageblock_bitmap;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700241
242 return 1;
243}
244
Yasunori Goto04753272008-04-28 02:13:31 -0700245unsigned long usemap_size(void)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700246{
247 unsigned long size_bytes;
248 size_bytes = roundup(SECTION_BLOCKFLAGS_BITS, 8) / 8;
249 size_bytes = roundup(size_bytes, sizeof(unsigned long));
250 return size_bytes;
251}
252
253#ifdef CONFIG_MEMORY_HOTPLUG
254static unsigned long *__kmalloc_section_usemap(void)
255{
256 return kmalloc(usemap_size(), GFP_KERNEL);
257}
258#endif /* CONFIG_MEMORY_HOTPLUG */
259
Yasunori Goto48c90682008-07-23 21:28:15 -0700260#ifdef CONFIG_MEMORY_HOTREMOVE
261static unsigned long * __init
Yinghai Lua4322e12010-02-10 01:20:21 -0800262sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700263 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700264{
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700265 unsigned long goal, limit;
266 unsigned long *p;
267 int nid;
Yasunori Goto48c90682008-07-23 21:28:15 -0700268 /*
269 * A page may contain usemaps for other sections preventing the
270 * page being freed and making a section unremovable while
271 * other sections referencing the usemap retmain active. Similarly,
272 * a pgdat can prevent a section being removed. If section A
273 * contains a pgdat and section B contains the usemap, both
274 * sections become inter-dependent. This allocates usemaps
275 * from the same section as the pgdat where possible to avoid
276 * this problem.
277 */
Yinghai Lu07b4e2b2012-07-11 14:02:51 -0700278 goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT);
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700279 limit = goal + (1UL << PA_SECTION_SHIFT);
280 nid = early_pfn_to_nid(goal >> PAGE_SHIFT);
281again:
282 p = ___alloc_bootmem_node_nopanic(NODE_DATA(nid), size,
283 SMP_CACHE_BYTES, goal, limit);
284 if (!p && limit) {
285 limit = 0;
286 goto again;
287 }
288 return p;
Yasunori Goto48c90682008-07-23 21:28:15 -0700289}
290
291static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
292{
293 unsigned long usemap_snr, pgdat_snr;
294 static unsigned long old_usemap_snr = NR_MEM_SECTIONS;
295 static unsigned long old_pgdat_snr = NR_MEM_SECTIONS;
296 struct pglist_data *pgdat = NODE_DATA(nid);
297 int usemap_nid;
298
299 usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT);
300 pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
301 if (usemap_snr == pgdat_snr)
302 return;
303
304 if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr)
305 /* skip redundant message */
306 return;
307
308 old_usemap_snr = usemap_snr;
309 old_pgdat_snr = pgdat_snr;
310
311 usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr));
312 if (usemap_nid != nid) {
313 printk(KERN_INFO
314 "node %d must be removed before remove section %ld\n",
315 nid, usemap_snr);
316 return;
317 }
318 /*
319 * There is a circular dependency.
320 * Some platforms allow un-removable section because they will just
321 * gather other removable sections for dynamic partitioning.
322 * Just notify un-removable section's number here.
323 */
324 printk(KERN_INFO "Section %ld and %ld (node %d)", usemap_snr,
325 pgdat_snr, nid);
326 printk(KERN_CONT
327 " have a circular dependency on usemap and pgdat allocations\n");
328}
329#else
330static unsigned long * __init
Yinghai Lua4322e12010-02-10 01:20:21 -0800331sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700332 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700333{
Johannes Weiner238305b2012-05-29 15:06:36 -0700334 return alloc_bootmem_node_nopanic(pgdat, size);
Yasunori Goto48c90682008-07-23 21:28:15 -0700335}
336
337static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
338{
339}
340#endif /* CONFIG_MEMORY_HOTREMOVE */
341
Yinghai Lua4322e12010-02-10 01:20:21 -0800342static void __init sparse_early_usemaps_alloc_node(unsigned long**usemap_map,
343 unsigned long pnum_begin,
344 unsigned long pnum_end,
345 unsigned long usemap_count, int nodeid)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700346{
Yinghai Lua4322e12010-02-10 01:20:21 -0800347 void *usemap;
348 unsigned long pnum;
349 int size = usemap_size();
Mel Gorman5c0e3062007-10-16 01:25:56 -0700350
Yinghai Lua4322e12010-02-10 01:20:21 -0800351 usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid),
Johannes Weiner238305b2012-05-29 15:06:36 -0700352 size * usemap_count);
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700353 if (!usemap) {
Johannes Weiner238305b2012-05-29 15:06:36 -0700354 printk(KERN_WARNING "%s: allocation failed\n", __func__);
355 return;
Yasunori Goto48c90682008-07-23 21:28:15 -0700356 }
357
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700358 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
359 if (!present_section_nr(pnum))
360 continue;
361 usemap_map[pnum] = usemap;
362 usemap += size;
363 check_usemap_section_nr(nodeid, usemap_map[pnum]);
Yinghai Lua4322e12010-02-10 01:20:21 -0800364 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700365}
366
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700367#ifndef CONFIG_SPARSEMEM_VMEMMAP
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700368struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700369{
370 struct page *map;
Yinghai Lue48e67e2010-05-24 14:31:57 -0700371 unsigned long size;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700372
373 map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
374 if (map)
375 return map;
376
Yinghai Lue48e67e2010-05-24 14:31:57 -0700377 size = PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
378 map = __alloc_bootmem_node_high(NODE_DATA(nid), size,
379 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700380 return map;
381}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800382void __init sparse_mem_maps_populate_node(struct page **map_map,
383 unsigned long pnum_begin,
384 unsigned long pnum_end,
385 unsigned long map_count, int nodeid)
386{
387 void *map;
388 unsigned long pnum;
389 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
390
391 map = alloc_remap(nodeid, size * map_count);
392 if (map) {
393 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
394 if (!present_section_nr(pnum))
395 continue;
396 map_map[pnum] = map;
397 map += size;
398 }
399 return;
400 }
401
402 size = PAGE_ALIGN(size);
Yinghai Lue48e67e2010-05-24 14:31:57 -0700403 map = __alloc_bootmem_node_high(NODE_DATA(nodeid), size * map_count,
404 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
Yinghai Lu9bdac912010-02-10 01:20:22 -0800405 if (map) {
406 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
407 if (!present_section_nr(pnum))
408 continue;
409 map_map[pnum] = map;
410 map += size;
411 }
412 return;
413 }
414
415 /* fallback */
416 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
417 struct mem_section *ms;
418
419 if (!present_section_nr(pnum))
420 continue;
421 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
422 if (map_map[pnum])
423 continue;
424 ms = __nr_to_section(pnum);
425 printk(KERN_ERR "%s: sparsemem memory map backing failed "
426 "some memory will not be available.\n", __func__);
427 ms->section_mem_map = 0;
428 }
429}
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700430#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
431
Yinghai Lu81d0d952010-02-27 09:29:38 -0800432#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Yinghai Lu9bdac912010-02-10 01:20:22 -0800433static void __init sparse_early_mem_maps_alloc_node(struct page **map_map,
434 unsigned long pnum_begin,
435 unsigned long pnum_end,
436 unsigned long map_count, int nodeid)
437{
438 sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end,
439 map_count, nodeid);
440}
Yinghai Lu81d0d952010-02-27 09:29:38 -0800441#else
Adrian Bunk9e5c6da2008-07-25 19:46:22 -0700442static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700443{
444 struct page *map;
445 struct mem_section *ms = __nr_to_section(pnum);
446 int nid = sparse_early_nid(ms);
447
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700448 map = sparse_mem_map_populate(pnum, nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -0700449 if (map)
450 return map;
451
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700452 printk(KERN_ERR "%s: sparsemem memory map backing failed "
Harvey Harrisond40cee22008-04-30 00:55:07 -0700453 "some memory will not be available.\n", __func__);
Bob Picco802f1922005-09-03 15:54:26 -0700454 ms->section_mem_map = 0;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700455 return NULL;
456}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800457#endif
Andy Whitcroft29751f62005-06-23 00:08:00 -0700458
Yinghai Luc2b91e22008-04-12 01:19:24 -0700459void __attribute__((weak)) __meminit vmemmap_populate_print_last(void)
460{
461}
Yinghai Lua4322e12010-02-10 01:20:21 -0800462
Stephen Rothwell193faea2007-06-08 13:46:51 -0700463/*
464 * Allocate the accumulated non-linear sections, allocate a mem_map
465 * for each and record the physical to section mapping.
466 */
467void __init sparse_init(void)
468{
469 unsigned long pnum;
470 struct page *map;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700471 unsigned long *usemap;
Yinghai Lue123dd32008-04-13 11:51:06 -0700472 unsigned long **usemap_map;
Yinghai Lu81d0d952010-02-27 09:29:38 -0800473 int size;
Yinghai Lua4322e12010-02-10 01:20:21 -0800474 int nodeid_begin = 0;
475 unsigned long pnum_begin = 0;
476 unsigned long usemap_count;
Yinghai Lu81d0d952010-02-27 09:29:38 -0800477#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Yinghai Lu9bdac912010-02-10 01:20:22 -0800478 unsigned long map_count;
Yinghai Lu81d0d952010-02-27 09:29:38 -0800479 int size2;
480 struct page **map_map;
481#endif
Yinghai Lue123dd32008-04-13 11:51:06 -0700482
Cody P Schafer55878e82013-07-03 15:04:44 -0700483 /* see include/linux/mmzone.h 'struct mem_section' definition */
484 BUILD_BUG_ON(!is_power_of_2(sizeof(struct mem_section)));
485
Xishi Qiuca57df72012-07-31 16:43:19 -0700486 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
487 set_pageblock_order();
488
Yinghai Lue123dd32008-04-13 11:51:06 -0700489 /*
490 * map is using big page (aka 2M in x86 64 bit)
491 * usemap is less one page (aka 24 bytes)
492 * so alloc 2M (with 2M align) and 24 bytes in turn will
493 * make next 2M slip to one more 2M later.
494 * then in big system, the memory will have a lot of holes...
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300495 * here try to allocate 2M pages continuously.
Yinghai Lue123dd32008-04-13 11:51:06 -0700496 *
497 * powerpc need to call sparse_init_one_section right after each
498 * sparse_early_mem_map_alloc, so allocate usemap_map at first.
499 */
500 size = sizeof(unsigned long *) * NR_MEM_SECTIONS;
501 usemap_map = alloc_bootmem(size);
502 if (!usemap_map)
503 panic("can not allocate usemap_map\n");
Stephen Rothwell193faea2007-06-08 13:46:51 -0700504
505 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
Yinghai Lua4322e12010-02-10 01:20:21 -0800506 struct mem_section *ms;
507
Andy Whitcroft540557b2007-10-16 01:24:11 -0700508 if (!present_section_nr(pnum))
Stephen Rothwell193faea2007-06-08 13:46:51 -0700509 continue;
Yinghai Lua4322e12010-02-10 01:20:21 -0800510 ms = __nr_to_section(pnum);
511 nodeid_begin = sparse_early_nid(ms);
512 pnum_begin = pnum;
513 break;
Yinghai Lue123dd32008-04-13 11:51:06 -0700514 }
Yinghai Lua4322e12010-02-10 01:20:21 -0800515 usemap_count = 1;
516 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
517 struct mem_section *ms;
518 int nodeid;
519
520 if (!present_section_nr(pnum))
521 continue;
522 ms = __nr_to_section(pnum);
523 nodeid = sparse_early_nid(ms);
524 if (nodeid == nodeid_begin) {
525 usemap_count++;
526 continue;
527 }
528 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
529 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, pnum,
530 usemap_count, nodeid_begin);
531 /* new start, update count etc*/
532 nodeid_begin = nodeid;
533 pnum_begin = pnum;
534 usemap_count = 1;
535 }
536 /* ok, last chunk */
537 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, NR_MEM_SECTIONS,
538 usemap_count, nodeid_begin);
Yinghai Lue123dd32008-04-13 11:51:06 -0700539
Yinghai Lu9bdac912010-02-10 01:20:22 -0800540#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
541 size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
542 map_map = alloc_bootmem(size2);
543 if (!map_map)
544 panic("can not allocate map_map\n");
545
546 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
547 struct mem_section *ms;
548
549 if (!present_section_nr(pnum))
550 continue;
551 ms = __nr_to_section(pnum);
552 nodeid_begin = sparse_early_nid(ms);
553 pnum_begin = pnum;
554 break;
555 }
556 map_count = 1;
557 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
558 struct mem_section *ms;
559 int nodeid;
560
561 if (!present_section_nr(pnum))
562 continue;
563 ms = __nr_to_section(pnum);
564 nodeid = sparse_early_nid(ms);
565 if (nodeid == nodeid_begin) {
566 map_count++;
567 continue;
568 }
569 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
570 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, pnum,
571 map_count, nodeid_begin);
572 /* new start, update count etc*/
573 nodeid_begin = nodeid;
574 pnum_begin = pnum;
575 map_count = 1;
576 }
577 /* ok, last chunk */
578 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, NR_MEM_SECTIONS,
579 map_count, nodeid_begin);
580#endif
581
Yinghai Lue123dd32008-04-13 11:51:06 -0700582 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
583 if (!present_section_nr(pnum))
584 continue;
585
586 usemap = usemap_map[pnum];
587 if (!usemap)
588 continue;
Stephen Rothwell193faea2007-06-08 13:46:51 -0700589
Yinghai Lu9bdac912010-02-10 01:20:22 -0800590#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
591 map = map_map[pnum];
592#else
Stephen Rothwell193faea2007-06-08 13:46:51 -0700593 map = sparse_early_mem_map_alloc(pnum);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800594#endif
Stephen Rothwell193faea2007-06-08 13:46:51 -0700595 if (!map)
596 continue;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700597
Mel Gorman5c0e3062007-10-16 01:25:56 -0700598 sparse_init_one_section(__nr_to_section(pnum), pnum, map,
599 usemap);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700600 }
Yinghai Lue123dd32008-04-13 11:51:06 -0700601
Yinghai Luc2b91e22008-04-12 01:19:24 -0700602 vmemmap_populate_print_last();
603
Yinghai Lu9bdac912010-02-10 01:20:22 -0800604#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
605 free_bootmem(__pa(map_map), size2);
606#endif
Yinghai Lue123dd32008-04-13 11:51:06 -0700607 free_bootmem(__pa(usemap_map), size);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700608}
609
610#ifdef CONFIG_MEMORY_HOTPLUG
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700611#ifdef CONFIG_SPARSEMEM_VMEMMAP
612static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
613 unsigned long nr_pages)
614{
615 /* This will make the necessary allocations eventually. */
616 return sparse_mem_map_populate(pnum, nid);
617}
618static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
619{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700620 unsigned long start = (unsigned long)memmap;
621 unsigned long end = (unsigned long)(memmap + nr_pages);
622
623 vmemmap_free(start, end);
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700624}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700625#ifdef CONFIG_MEMORY_HOTREMOVE
Jianguo Wuae64ffc2012-11-29 13:54:21 -0800626static void free_map_bootmem(struct page *memmap, unsigned long nr_pages)
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700627{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700628 unsigned long start = (unsigned long)memmap;
629 unsigned long end = (unsigned long)(memmap + nr_pages);
630
631 vmemmap_free(start, end);
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700632}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700633#endif /* CONFIG_MEMORY_HOTREMOVE */
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700634#else
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700635static struct page *__kmalloc_section_memmap(unsigned long nr_pages)
636{
637 struct page *page, *ret;
638 unsigned long memmap_size = sizeof(struct page) * nr_pages;
639
Yasunori Gotof2d0aa52006-10-28 10:38:32 -0700640 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700641 if (page)
642 goto got_map_page;
643
644 ret = vmalloc(memmap_size);
645 if (ret)
646 goto got_map_ptr;
647
648 return NULL;
649got_map_page:
650 ret = (struct page *)pfn_to_kaddr(page_to_pfn(page));
651got_map_ptr:
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700652
653 return ret;
654}
655
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700656static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
657 unsigned long nr_pages)
658{
659 return __kmalloc_section_memmap(nr_pages);
660}
661
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700662static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
663{
Christoph Lameter9e2779f2008-02-04 22:28:34 -0800664 if (is_vmalloc_addr(memmap))
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700665 vfree(memmap);
666 else
667 free_pages((unsigned long)memmap,
668 get_order(sizeof(struct page) * nr_pages));
669}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700670
David Rientjes4edd7ce2013-04-29 15:08:22 -0700671#ifdef CONFIG_MEMORY_HOTREMOVE
Jianguo Wuae64ffc2012-11-29 13:54:21 -0800672static void free_map_bootmem(struct page *memmap, unsigned long nr_pages)
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700673{
674 unsigned long maps_section_nr, removing_section_nr, i;
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800675 unsigned long magic;
Jianguo Wuae64ffc2012-11-29 13:54:21 -0800676 struct page *page = virt_to_page(memmap);
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700677
678 for (i = 0; i < nr_pages; i++, page++) {
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800679 magic = (unsigned long) page->lru.next;
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700680
681 BUG_ON(magic == NODE_INFO);
682
683 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
684 removing_section_nr = page->private;
685
686 /*
687 * When this function is called, the removing section is
688 * logical offlined state. This means all pages are isolated
689 * from page allocator. If removing section's memmap is placed
690 * on the same section, it must not be freed.
691 * If it is freed, page allocator may allocate it which will
692 * be removed physically soon.
693 */
694 if (maps_section_nr != removing_section_nr)
695 put_page_bootmem(page);
696 }
697}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700698#endif /* CONFIG_MEMORY_HOTREMOVE */
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700699#endif /* CONFIG_SPARSEMEM_VMEMMAP */
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700700
Andy Whitcroft29751f62005-06-23 00:08:00 -0700701/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700702 * returns the number of sections whose mem_maps were properly
703 * set. If this is <=0, then that means that the passed-in
704 * map was not consumed and must be freed.
705 */
Al Viro31168482008-11-22 17:33:24 +0000706int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700707 int nr_pages)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700708{
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700709 unsigned long section_nr = pfn_to_section_nr(start_pfn);
710 struct pglist_data *pgdat = zone->zone_pgdat;
711 struct mem_section *ms;
712 struct page *memmap;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700713 unsigned long *usemap;
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700714 unsigned long flags;
715 int ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700716
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700717 /*
718 * no locking for this, because it does its own
719 * plus, it does a kmalloc
720 */
WANG Congbbd06822007-12-17 16:19:59 -0800721 ret = sparse_index_init(section_nr, pgdat->node_id);
722 if (ret < 0 && ret != -EEXIST)
723 return ret;
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700724 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id, nr_pages);
WANG Congbbd06822007-12-17 16:19:59 -0800725 if (!memmap)
726 return -ENOMEM;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700727 usemap = __kmalloc_section_usemap();
WANG Congbbd06822007-12-17 16:19:59 -0800728 if (!usemap) {
729 __kfree_section_memmap(memmap, nr_pages);
730 return -ENOMEM;
731 }
Andy Whitcroft29751f62005-06-23 00:08:00 -0700732
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700733 pgdat_resize_lock(pgdat, &flags);
734
735 ms = __pfn_to_section(start_pfn);
736 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
737 ret = -EEXIST;
738 goto out;
739 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700740
Wen Congyang3ac19f82012-12-11 16:00:59 -0800741 memset(memmap, 0, sizeof(struct page) * nr_pages);
742
Andy Whitcroft29751f62005-06-23 00:08:00 -0700743 ms->section_mem_map |= SECTION_MARKED_PRESENT;
744
Mel Gorman5c0e3062007-10-16 01:25:56 -0700745 ret = sparse_init_one_section(ms, section_nr, memmap, usemap);
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700746
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700747out:
748 pgdat_resize_unlock(pgdat, &flags);
WANG Congbbd06822007-12-17 16:19:59 -0800749 if (ret <= 0) {
750 kfree(usemap);
Mike Kravetz46a66ee2006-05-01 12:16:09 -0700751 __kfree_section_memmap(memmap, nr_pages);
WANG Congbbd06822007-12-17 16:19:59 -0800752 }
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700753 return ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700754}
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700755
Zhang Yanfeif3deb682013-07-08 16:00:10 -0700756#ifdef CONFIG_MEMORY_HOTREMOVE
Wen Congyang95a47742012-12-11 16:00:47 -0800757#ifdef CONFIG_MEMORY_FAILURE
758static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
759{
760 int i;
761
762 if (!memmap)
763 return;
764
765 for (i = 0; i < PAGES_PER_SECTION; i++) {
766 if (PageHWPoison(&memmap[i])) {
Xishi Qiu293c07e2013-02-22 16:34:02 -0800767 atomic_long_sub(1, &num_poisoned_pages);
Wen Congyang95a47742012-12-11 16:00:47 -0800768 ClearPageHWPoison(&memmap[i]);
769 }
770 }
771}
772#else
773static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
774{
775}
776#endif
777
David Rientjes4edd7ce2013-04-29 15:08:22 -0700778static void free_section_usemap(struct page *memmap, unsigned long *usemap)
779{
780 struct page *usemap_page;
781 unsigned long nr_pages;
782
783 if (!usemap)
784 return;
785
786 usemap_page = virt_to_page(usemap);
787 /*
788 * Check to see if allocation came from hot-plug-add
789 */
790 if (PageSlab(usemap_page) || PageCompound(usemap_page)) {
791 kfree(usemap);
792 if (memmap)
793 __kfree_section_memmap(memmap, PAGES_PER_SECTION);
794 return;
795 }
796
797 /*
798 * The usemap came from bootmem. This is packed with other usemaps
799 * on the section which has pgdat at boot time. Just keep it as is now.
800 */
801
802 if (memmap) {
803 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
804 >> PAGE_SHIFT;
805
806 free_map_bootmem(memmap, nr_pages);
807 }
808}
809
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700810void sparse_remove_one_section(struct zone *zone, struct mem_section *ms)
811{
812 struct page *memmap = NULL;
Tang Chencd099682013-02-22 16:33:02 -0800813 unsigned long *usemap = NULL, flags;
814 struct pglist_data *pgdat = zone->zone_pgdat;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700815
Tang Chencd099682013-02-22 16:33:02 -0800816 pgdat_resize_lock(pgdat, &flags);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700817 if (ms->section_mem_map) {
818 usemap = ms->pageblock_flags;
819 memmap = sparse_decode_mem_map(ms->section_mem_map,
820 __section_nr(ms));
821 ms->section_mem_map = 0;
822 ms->pageblock_flags = NULL;
823 }
Tang Chencd099682013-02-22 16:33:02 -0800824 pgdat_resize_unlock(pgdat, &flags);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700825
Wen Congyang95a47742012-12-11 16:00:47 -0800826 clear_hwpoisoned_pages(memmap, PAGES_PER_SECTION);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700827 free_section_usemap(memmap, usemap);
828}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700829#endif /* CONFIG_MEMORY_HOTREMOVE */
830#endif /* CONFIG_MEMORY_HOTPLUG */