blob: 55cda7a9dfdd83b282282ec3696782bbd38a029d [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;
Bryan Huntsman3f2bc4d2011-08-16 17:27:22 -0700124 struct mem_section *root;
Dave Hansen4ca644d2005-10-29 18:16:51 -0700125
Bryan Huntsman3f2bc4d2011-08-16 17:27:22 -0700126 if (NR_SECTION_ROOTS == 0)
127 return ms - __nr_to_section(0);
Mike Kravetz12783b02006-05-20 15:00:05 -0700128 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
129 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
Dave Hansen4ca644d2005-10-29 18:16:51 -0700130 if (!root)
131 continue;
132
133 if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT)))
134 break;
135 }
136
137 return (root_nr * SECTIONS_PER_ROOT) + (ms - root);
138}
139
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700140/*
141 * During early boot, before section_mem_map is used for an actual
142 * mem_map, we use section_mem_map to store the section's NUMA
143 * node. This keeps us from having to use another data structure. The
144 * node information is cleared just before we store the real mem_map.
145 */
146static inline unsigned long sparse_encode_early_nid(int nid)
147{
148 return (nid << SECTION_NID_SHIFT);
149}
150
151static inline int sparse_early_nid(struct mem_section *section)
152{
153 return (section->section_mem_map >> SECTION_NID_SHIFT);
154}
155
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700156/* Validate the physical addressing limitations of the model */
157void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
158 unsigned long *end_pfn)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700159{
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700160 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700161
Ingo Molnarbead9a32008-04-16 01:40:00 +0200162 /*
163 * Sanity checks - do not allow an architecture to pass
164 * in larger pfns than the maximum scope of sparsemem:
165 */
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700166 if (*start_pfn > max_sparsemem_pfn) {
167 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
168 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
169 *start_pfn, *end_pfn, max_sparsemem_pfn);
170 WARN_ON_ONCE(1);
171 *start_pfn = max_sparsemem_pfn;
172 *end_pfn = max_sparsemem_pfn;
Cyrill Gorcunovef161a92009-03-31 15:19:25 -0700173 } else if (*end_pfn > max_sparsemem_pfn) {
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700174 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
175 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
176 *start_pfn, *end_pfn, max_sparsemem_pfn);
177 WARN_ON_ONCE(1);
178 *end_pfn = max_sparsemem_pfn;
179 }
180}
181
182/* Record a memory area against a node. */
183void __init memory_present(int nid, unsigned long start, unsigned long end)
184{
185 unsigned long pfn;
Ingo Molnarbead9a32008-04-16 01:40:00 +0200186
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700187 start &= PAGE_SECTION_MASK;
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700188 mminit_validate_memmodel_limits(&start, &end);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700189 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
190 unsigned long section = pfn_to_section_nr(pfn);
Bob Picco802f1922005-09-03 15:54:26 -0700191 struct mem_section *ms;
192
193 sparse_index_init(section, nid);
Andy Whitcroft85770ff2007-08-22 14:01:03 -0700194 set_section_nid(section, nid);
Bob Picco802f1922005-09-03 15:54:26 -0700195
196 ms = __nr_to_section(section);
197 if (!ms->section_mem_map)
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700198 ms->section_mem_map = sparse_encode_early_nid(nid) |
199 SECTION_MARKED_PRESENT;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700200 }
201}
202
203/*
204 * Only used by the i386 NUMA architecures, but relatively
205 * generic code.
206 */
207unsigned long __init node_memmap_size_bytes(int nid, unsigned long start_pfn,
208 unsigned long end_pfn)
209{
210 unsigned long pfn;
211 unsigned long nr_pages = 0;
212
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700213 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700214 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
215 if (nid != early_pfn_to_nid(pfn))
216 continue;
217
Andy Whitcroft540557b2007-10-16 01:24:11 -0700218 if (pfn_present(pfn))
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700219 nr_pages += PAGES_PER_SECTION;
220 }
221
222 return nr_pages * sizeof(struct page);
223}
224
225/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700226 * Subtle, we encode the real pfn into the mem_map such that
227 * the identity pfn - section_mem_map will return the actual
228 * physical page frame number.
229 */
230static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
231{
232 return (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
233}
234
235/*
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700236 * Decode mem_map from the coded memmap
Andy Whitcroft29751f62005-06-23 00:08:00 -0700237 */
Andy Whitcroft29751f62005-06-23 00:08:00 -0700238struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
239{
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700240 /* mask off the extra low bits of information */
241 coded_mem_map &= SECTION_MAP_MASK;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700242 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
243}
244
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700245static int __meminit sparse_init_one_section(struct mem_section *ms,
Mel Gorman5c0e3062007-10-16 01:25:56 -0700246 unsigned long pnum, struct page *mem_map,
247 unsigned long *pageblock_bitmap)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700248{
Andy Whitcroft540557b2007-10-16 01:24:11 -0700249 if (!present_section(ms))
Andy Whitcroft29751f62005-06-23 00:08:00 -0700250 return -EINVAL;
251
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700252 ms->section_mem_map &= ~SECTION_MAP_MASK;
Andy Whitcroft540557b2007-10-16 01:24:11 -0700253 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) |
254 SECTION_HAS_MEM_MAP;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700255 ms->pageblock_flags = pageblock_bitmap;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700256
257 return 1;
258}
259
Yasunori Goto04753272008-04-28 02:13:31 -0700260unsigned long usemap_size(void)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700261{
262 unsigned long size_bytes;
263 size_bytes = roundup(SECTION_BLOCKFLAGS_BITS, 8) / 8;
264 size_bytes = roundup(size_bytes, sizeof(unsigned long));
265 return size_bytes;
266}
267
268#ifdef CONFIG_MEMORY_HOTPLUG
269static unsigned long *__kmalloc_section_usemap(void)
270{
271 return kmalloc(usemap_size(), GFP_KERNEL);
272}
273#endif /* CONFIG_MEMORY_HOTPLUG */
274
Yasunori Goto48c90682008-07-23 21:28:15 -0700275#ifdef CONFIG_MEMORY_HOTREMOVE
276static unsigned long * __init
Yinghai Lua4322e12010-02-10 01:20:21 -0800277sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
278 unsigned long count)
Yasunori Goto48c90682008-07-23 21:28:15 -0700279{
280 unsigned long section_nr;
281
282 /*
283 * A page may contain usemaps for other sections preventing the
284 * page being freed and making a section unremovable while
285 * other sections referencing the usemap retmain active. Similarly,
286 * a pgdat can prevent a section being removed. If section A
287 * contains a pgdat and section B contains the usemap, both
288 * sections become inter-dependent. This allocates usemaps
289 * from the same section as the pgdat where possible to avoid
290 * this problem.
291 */
292 section_nr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
Yinghai Lua4322e12010-02-10 01:20:21 -0800293 return alloc_bootmem_section(usemap_size() * count, section_nr);
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,
337 unsigned long count)
Yasunori Goto48c90682008-07-23 21:28:15 -0700338{
339 return NULL;
340}
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),
357 usemap_count);
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700358 if (!usemap) {
359 usemap = alloc_bootmem_node(NODE_DATA(nodeid), size * usemap_count);
360 if (!usemap) {
361 printk(KERN_WARNING "%s: allocation failed\n", __func__);
362 return;
Yinghai Lua4322e12010-02-10 01:20:21 -0800363 }
Yasunori Goto48c90682008-07-23 21:28:15 -0700364 }
365
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700366 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
367 if (!present_section_nr(pnum))
368 continue;
369 usemap_map[pnum] = usemap;
370 usemap += size;
371 check_usemap_section_nr(nodeid, usemap_map[pnum]);
Yinghai Lua4322e12010-02-10 01:20:21 -0800372 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700373}
374
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700375#ifndef CONFIG_SPARSEMEM_VMEMMAP
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700376struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700377{
378 struct page *map;
Yinghai Lue48e67e2010-05-24 14:31:57 -0700379 unsigned long size;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700380
381 map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
382 if (map)
383 return map;
384
Yinghai Lue48e67e2010-05-24 14:31:57 -0700385 size = PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
386 map = __alloc_bootmem_node_high(NODE_DATA(nid), size,
387 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700388 return map;
389}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800390void __init sparse_mem_maps_populate_node(struct page **map_map,
391 unsigned long pnum_begin,
392 unsigned long pnum_end,
393 unsigned long map_count, int nodeid)
394{
395 void *map;
396 unsigned long pnum;
397 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
398
399 map = alloc_remap(nodeid, size * map_count);
400 if (map) {
401 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
402 if (!present_section_nr(pnum))
403 continue;
404 map_map[pnum] = map;
405 map += size;
406 }
407 return;
408 }
409
410 size = PAGE_ALIGN(size);
Yinghai Lue48e67e2010-05-24 14:31:57 -0700411 map = __alloc_bootmem_node_high(NODE_DATA(nodeid), size * map_count,
412 PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
Yinghai Lu9bdac912010-02-10 01:20:22 -0800413 if (map) {
414 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
415 if (!present_section_nr(pnum))
416 continue;
417 map_map[pnum] = map;
418 map += size;
419 }
420 return;
421 }
422
423 /* fallback */
424 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
425 struct mem_section *ms;
426
427 if (!present_section_nr(pnum))
428 continue;
429 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
430 if (map_map[pnum])
431 continue;
432 ms = __nr_to_section(pnum);
433 printk(KERN_ERR "%s: sparsemem memory map backing failed "
434 "some memory will not be available.\n", __func__);
435 ms->section_mem_map = 0;
436 }
437}
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700438#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
439
Yinghai Lu81d0d952010-02-27 09:29:38 -0800440#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Yinghai Lu9bdac912010-02-10 01:20:22 -0800441static void __init sparse_early_mem_maps_alloc_node(struct page **map_map,
442 unsigned long pnum_begin,
443 unsigned long pnum_end,
444 unsigned long map_count, int nodeid)
445{
446 sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end,
447 map_count, nodeid);
448}
Yinghai Lu81d0d952010-02-27 09:29:38 -0800449#else
Adrian Bunk9e5c6da2008-07-25 19:46:22 -0700450static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700451{
452 struct page *map;
453 struct mem_section *ms = __nr_to_section(pnum);
454 int nid = sparse_early_nid(ms);
455
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700456 map = sparse_mem_map_populate(pnum, nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -0700457 if (map)
458 return map;
459
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700460 printk(KERN_ERR "%s: sparsemem memory map backing failed "
Harvey Harrisond40cee22008-04-30 00:55:07 -0700461 "some memory will not be available.\n", __func__);
Bob Picco802f1922005-09-03 15:54:26 -0700462 ms->section_mem_map = 0;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700463 return NULL;
464}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800465#endif
Andy Whitcroft29751f62005-06-23 00:08:00 -0700466
Yinghai Luc2b91e22008-04-12 01:19:24 -0700467void __attribute__((weak)) __meminit vmemmap_populate_print_last(void)
468{
469}
Yinghai Lua4322e12010-02-10 01:20:21 -0800470
Stephen Rothwell193faea2007-06-08 13:46:51 -0700471/*
472 * Allocate the accumulated non-linear sections, allocate a mem_map
473 * for each and record the physical to section mapping.
474 */
475void __init sparse_init(void)
476{
477 unsigned long pnum;
478 struct page *map;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700479 unsigned long *usemap;
Yinghai Lue123dd32008-04-13 11:51:06 -0700480 unsigned long **usemap_map;
Yinghai Lu81d0d952010-02-27 09:29:38 -0800481 int size;
Yinghai Lua4322e12010-02-10 01:20:21 -0800482 int nodeid_begin = 0;
483 unsigned long pnum_begin = 0;
484 unsigned long usemap_count;
Yinghai Lu81d0d952010-02-27 09:29:38 -0800485#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Yinghai Lu9bdac912010-02-10 01:20:22 -0800486 unsigned long map_count;
Yinghai Lu81d0d952010-02-27 09:29:38 -0800487 int size2;
488 struct page **map_map;
489#endif
Yinghai Lue123dd32008-04-13 11:51:06 -0700490
491 /*
492 * map is using big page (aka 2M in x86 64 bit)
493 * usemap is less one page (aka 24 bytes)
494 * so alloc 2M (with 2M align) and 24 bytes in turn will
495 * make next 2M slip to one more 2M later.
496 * then in big system, the memory will have a lot of holes...
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300497 * here try to allocate 2M pages continuously.
Yinghai Lue123dd32008-04-13 11:51:06 -0700498 *
499 * powerpc need to call sparse_init_one_section right after each
500 * sparse_early_mem_map_alloc, so allocate usemap_map at first.
501 */
502 size = sizeof(unsigned long *) * NR_MEM_SECTIONS;
503 usemap_map = alloc_bootmem(size);
504 if (!usemap_map)
505 panic("can not allocate usemap_map\n");
Stephen Rothwell193faea2007-06-08 13:46:51 -0700506
507 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
Yinghai Lua4322e12010-02-10 01:20:21 -0800508 struct mem_section *ms;
509
Andy Whitcroft540557b2007-10-16 01:24:11 -0700510 if (!present_section_nr(pnum))
Stephen Rothwell193faea2007-06-08 13:46:51 -0700511 continue;
Yinghai Lua4322e12010-02-10 01:20:21 -0800512 ms = __nr_to_section(pnum);
513 nodeid_begin = sparse_early_nid(ms);
514 pnum_begin = pnum;
515 break;
Yinghai Lue123dd32008-04-13 11:51:06 -0700516 }
Yinghai Lua4322e12010-02-10 01:20:21 -0800517 usemap_count = 1;
518 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
519 struct mem_section *ms;
520 int nodeid;
521
522 if (!present_section_nr(pnum))
523 continue;
524 ms = __nr_to_section(pnum);
525 nodeid = sparse_early_nid(ms);
526 if (nodeid == nodeid_begin) {
527 usemap_count++;
528 continue;
529 }
530 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
531 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, pnum,
532 usemap_count, nodeid_begin);
533 /* new start, update count etc*/
534 nodeid_begin = nodeid;
535 pnum_begin = pnum;
536 usemap_count = 1;
537 }
538 /* ok, last chunk */
539 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, NR_MEM_SECTIONS,
540 usemap_count, nodeid_begin);
Yinghai Lue123dd32008-04-13 11:51:06 -0700541
Yinghai Lu9bdac912010-02-10 01:20:22 -0800542#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
543 size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
544 map_map = alloc_bootmem(size2);
545 if (!map_map)
546 panic("can not allocate map_map\n");
547
548 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
549 struct mem_section *ms;
550
551 if (!present_section_nr(pnum))
552 continue;
553 ms = __nr_to_section(pnum);
554 nodeid_begin = sparse_early_nid(ms);
555 pnum_begin = pnum;
556 break;
557 }
558 map_count = 1;
559 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
560 struct mem_section *ms;
561 int nodeid;
562
563 if (!present_section_nr(pnum))
564 continue;
565 ms = __nr_to_section(pnum);
566 nodeid = sparse_early_nid(ms);
567 if (nodeid == nodeid_begin) {
568 map_count++;
569 continue;
570 }
571 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
572 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, pnum,
573 map_count, nodeid_begin);
574 /* new start, update count etc*/
575 nodeid_begin = nodeid;
576 pnum_begin = pnum;
577 map_count = 1;
578 }
579 /* ok, last chunk */
580 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, NR_MEM_SECTIONS,
581 map_count, nodeid_begin);
582#endif
583
Yinghai Lue123dd32008-04-13 11:51:06 -0700584 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
585 if (!present_section_nr(pnum))
586 continue;
587
588 usemap = usemap_map[pnum];
589 if (!usemap)
590 continue;
Stephen Rothwell193faea2007-06-08 13:46:51 -0700591
Yinghai Lu9bdac912010-02-10 01:20:22 -0800592#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
593 map = map_map[pnum];
594#else
Stephen Rothwell193faea2007-06-08 13:46:51 -0700595 map = sparse_early_mem_map_alloc(pnum);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800596#endif
Stephen Rothwell193faea2007-06-08 13:46:51 -0700597 if (!map)
598 continue;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700599
Mel Gorman5c0e3062007-10-16 01:25:56 -0700600 sparse_init_one_section(__nr_to_section(pnum), pnum, map,
601 usemap);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700602 }
Yinghai Lue123dd32008-04-13 11:51:06 -0700603
Yinghai Luc2b91e22008-04-12 01:19:24 -0700604 vmemmap_populate_print_last();
605
Yinghai Lu9bdac912010-02-10 01:20:22 -0800606#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
607 free_bootmem(__pa(map_map), size2);
608#endif
Yinghai Lue123dd32008-04-13 11:51:06 -0700609 free_bootmem(__pa(usemap_map), size);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700610}
611
612#ifdef CONFIG_MEMORY_HOTPLUG
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700613#ifdef CONFIG_SPARSEMEM_VMEMMAP
614static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
615 unsigned long nr_pages)
616{
617 /* This will make the necessary allocations eventually. */
618 return sparse_mem_map_populate(pnum, nid);
619}
620static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
621{
622 return; /* XXX: Not implemented yet */
623}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700624static void free_map_bootmem(struct page *page, unsigned long nr_pages)
625{
626}
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700627#else
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700628static struct page *__kmalloc_section_memmap(unsigned long nr_pages)
629{
630 struct page *page, *ret;
631 unsigned long memmap_size = sizeof(struct page) * nr_pages;
632
Yasunori Gotof2d0aa52006-10-28 10:38:32 -0700633 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700634 if (page)
635 goto got_map_page;
636
637 ret = vmalloc(memmap_size);
638 if (ret)
639 goto got_map_ptr;
640
641 return NULL;
642got_map_page:
643 ret = (struct page *)pfn_to_kaddr(page_to_pfn(page));
644got_map_ptr:
645 memset(ret, 0, memmap_size);
646
647 return ret;
648}
649
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700650static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
651 unsigned long nr_pages)
652{
653 return __kmalloc_section_memmap(nr_pages);
654}
655
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700656static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages)
657{
Christoph Lameter9e2779f2008-02-04 22:28:34 -0800658 if (is_vmalloc_addr(memmap))
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700659 vfree(memmap);
660 else
661 free_pages((unsigned long)memmap,
662 get_order(sizeof(struct page) * nr_pages));
663}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700664
665static void free_map_bootmem(struct page *page, unsigned long nr_pages)
666{
667 unsigned long maps_section_nr, removing_section_nr, i;
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800668 unsigned long magic;
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700669
670 for (i = 0; i < nr_pages; i++, page++) {
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800671 magic = (unsigned long) page->lru.next;
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700672
673 BUG_ON(magic == NODE_INFO);
674
675 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
676 removing_section_nr = page->private;
677
678 /*
679 * When this function is called, the removing section is
680 * logical offlined state. This means all pages are isolated
681 * from page allocator. If removing section's memmap is placed
682 * on the same section, it must not be freed.
683 * If it is freed, page allocator may allocate it which will
684 * be removed physically soon.
685 */
686 if (maps_section_nr != removing_section_nr)
687 put_page_bootmem(page);
688 }
689}
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700690#endif /* CONFIG_SPARSEMEM_VMEMMAP */
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700691
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700692static void free_section_usemap(struct page *memmap, unsigned long *usemap)
693{
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700694 struct page *usemap_page;
695 unsigned long nr_pages;
696
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700697 if (!usemap)
698 return;
699
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700700 usemap_page = virt_to_page(usemap);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700701 /*
702 * Check to see if allocation came from hot-plug-add
703 */
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700704 if (PageSlab(usemap_page)) {
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700705 kfree(usemap);
706 if (memmap)
707 __kfree_section_memmap(memmap, PAGES_PER_SECTION);
708 return;
709 }
710
711 /*
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700712 * The usemap came from bootmem. This is packed with other usemaps
713 * on the section which has pgdat at boot time. Just keep it as is now.
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700714 */
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700715
716 if (memmap) {
717 struct page *memmap_page;
718 memmap_page = virt_to_page(memmap);
719
720 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
721 >> PAGE_SHIFT;
722
723 free_map_bootmem(memmap_page, nr_pages);
724 }
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700725}
726
Andy Whitcroft29751f62005-06-23 00:08:00 -0700727/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700728 * returns the number of sections whose mem_maps were properly
729 * set. If this is <=0, then that means that the passed-in
730 * map was not consumed and must be freed.
731 */
Al Viro31168482008-11-22 17:33:24 +0000732int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn,
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700733 int nr_pages)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700734{
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700735 unsigned long section_nr = pfn_to_section_nr(start_pfn);
736 struct pglist_data *pgdat = zone->zone_pgdat;
737 struct mem_section *ms;
738 struct page *memmap;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700739 unsigned long *usemap;
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700740 unsigned long flags;
741 int ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700742
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700743 /*
744 * no locking for this, because it does its own
745 * plus, it does a kmalloc
746 */
WANG Congbbd06822007-12-17 16:19:59 -0800747 ret = sparse_index_init(section_nr, pgdat->node_id);
748 if (ret < 0 && ret != -EEXIST)
749 return ret;
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700750 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id, nr_pages);
WANG Congbbd06822007-12-17 16:19:59 -0800751 if (!memmap)
752 return -ENOMEM;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700753 usemap = __kmalloc_section_usemap();
WANG Congbbd06822007-12-17 16:19:59 -0800754 if (!usemap) {
755 __kfree_section_memmap(memmap, nr_pages);
756 return -ENOMEM;
757 }
Andy Whitcroft29751f62005-06-23 00:08:00 -0700758
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700759 pgdat_resize_lock(pgdat, &flags);
760
761 ms = __pfn_to_section(start_pfn);
762 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
763 ret = -EEXIST;
764 goto out;
765 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700766
Andy Whitcroft29751f62005-06-23 00:08:00 -0700767 ms->section_mem_map |= SECTION_MARKED_PRESENT;
768
Mel Gorman5c0e3062007-10-16 01:25:56 -0700769 ret = sparse_init_one_section(ms, section_nr, memmap, usemap);
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700770
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700771out:
772 pgdat_resize_unlock(pgdat, &flags);
WANG Congbbd06822007-12-17 16:19:59 -0800773 if (ret <= 0) {
774 kfree(usemap);
Mike Kravetz46a66ee2006-05-01 12:16:09 -0700775 __kfree_section_memmap(memmap, nr_pages);
WANG Congbbd06822007-12-17 16:19:59 -0800776 }
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700777 return ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700778}
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700779
780void sparse_remove_one_section(struct zone *zone, struct mem_section *ms)
781{
782 struct page *memmap = NULL;
783 unsigned long *usemap = NULL;
784
785 if (ms->section_mem_map) {
786 usemap = ms->pageblock_flags;
787 memmap = sparse_decode_mem_map(ms->section_mem_map,
788 __section_nr(ms));
789 ms->section_mem_map = 0;
790 ms->pageblock_flags = NULL;
791 }
792
793 free_section_usemap(memmap, usemap);
794}
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700795#endif