blob: 3717ceed4177c9183626673134116b6e0f87adda [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>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -07008#include <linux/compiler.h>
Dave Hansen0b0acbe2005-10-29 18:16:55 -07009#include <linux/highmem.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040010#include <linux/export.h>
Dave Hansen28ae55c2005-09-03 15:54:29 -070011#include <linux/spinlock.h>
Dave Hansen0b0acbe2005-10-29 18:16:55 -070012#include <linux/vmalloc.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070013
Yasunori Goto0c0a4a52008-04-28 02:13:34 -070014#include "internal.h"
Andy Whitcroftd41dee32005-06-23 00:07:54 -070015#include <asm/dma.h>
Christoph Lameter8f6aac42007-10-16 01:24:13 -070016#include <asm/pgalloc.h>
17#include <asm/pgtable.h>
Andy Whitcroftd41dee32005-06-23 00:07:54 -070018
19/*
20 * Permanent SPARSEMEM data:
21 *
22 * 1) mem_section - memory sections, mem_map's for valid memory
23 */
Bob Picco3e347262005-09-03 15:54:28 -070024#ifdef CONFIG_SPARSEMEM_EXTREME
Bob Picco802f1922005-09-03 15:54:26 -070025struct mem_section *mem_section[NR_SECTION_ROOTS]
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -080026 ____cacheline_internodealigned_in_smp;
Bob Picco3e347262005-09-03 15:54:28 -070027#else
28struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT]
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -080029 ____cacheline_internodealigned_in_smp;
Bob Picco3e347262005-09-03 15:54:28 -070030#endif
31EXPORT_SYMBOL(mem_section);
32
Christoph Lameter89689ae2006-12-06 20:31:45 -080033#ifdef NODE_NOT_IN_PAGE_FLAGS
34/*
35 * If we did not store the node number in the page then we have to
36 * do a lookup in the section_to_node_table in order to find which
37 * node the page belongs to.
38 */
39#if MAX_NUMNODES <= 256
40static u8 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
41#else
42static u16 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned;
43#endif
44
Ian Campbell33dd4e02011-07-25 17:11:51 -070045int page_to_nid(const struct page *page)
Christoph Lameter89689ae2006-12-06 20:31:45 -080046{
47 return section_to_node_table[page_to_section(page)];
48}
49EXPORT_SYMBOL(page_to_nid);
Andy Whitcroft85770ff2007-08-22 14:01:03 -070050
51static void set_section_nid(unsigned long section_nr, int nid)
52{
53 section_to_node_table[section_nr] = nid;
54}
55#else /* !NODE_NOT_IN_PAGE_FLAGS */
56static inline void set_section_nid(unsigned long section_nr, int nid)
57{
58}
Christoph Lameter89689ae2006-12-06 20:31:45 -080059#endif
60
Bob Picco3e347262005-09-03 15:54:28 -070061#ifdef CONFIG_SPARSEMEM_EXTREME
Sam Ravnborg577a32f2007-05-17 23:29:25 +020062static struct mem_section noinline __init_refok *sparse_index_alloc(int nid)
Bob Picco802f1922005-09-03 15:54:26 -070063{
Dave Hansen28ae55c2005-09-03 15:54:29 -070064 struct mem_section *section = NULL;
65 unsigned long array_size = SECTIONS_PER_ROOT *
66 sizeof(struct mem_section);
Bob Picco802f1922005-09-03 15:54:26 -070067
Shaohua Lif52407c2009-09-21 17:01:19 -070068 if (slab_is_available()) {
69 if (node_state(nid, N_HIGH_MEMORY))
Gavin Shan5b760e62012-07-31 16:46:02 -070070 section = kzalloc_node(array_size, GFP_KERNEL, nid);
Shaohua Lif52407c2009-09-21 17:01:19 -070071 else
Gavin Shan5b760e62012-07-31 16:46:02 -070072 section = kzalloc(array_size, GFP_KERNEL);
73 } else {
Santosh Shilimkarbb016b82014-01-21 15:50:34 -080074 section = memblock_virt_alloc_node(array_size, nid);
Gavin Shan5b760e62012-07-31 16:46:02 -070075 }
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{
Dave Hansen28ae55c2005-09-03 15:54:29 -070082 unsigned long root = SECTION_NR_TO_ROOT(section_nr);
83 struct mem_section *section;
Dave Hansen28ae55c2005-09-03 15:54:29 -070084
85 if (mem_section[root])
86 return -EEXIST;
87
88 section = sparse_index_alloc(nid);
WANG Congaf0cd5a2007-12-17 16:19:58 -080089 if (!section)
90 return -ENOMEM;
Dave Hansen28ae55c2005-09-03 15:54:29 -070091
92 mem_section[root] = section;
Gavin Shanc1c95182012-07-31 16:46:06 -070093
Zhang Yanfei9d1936c2013-05-17 22:10:38 +080094 return 0;
Dave Hansen28ae55c2005-09-03 15:54:29 -070095}
96#else /* !SPARSEMEM_EXTREME */
97static inline int sparse_index_init(unsigned long section_nr, int nid)
98{
99 return 0;
100}
101#endif
102
Dave Hansen4ca644d2005-10-29 18:16:51 -0700103/*
104 * Although written for the SPARSEMEM_EXTREME case, this happens
Andy Whitcroftcd881a62007-10-16 01:24:10 -0700105 * to also work for the flat array case because
Dave Hansen4ca644d2005-10-29 18:16:51 -0700106 * NR_SECTION_ROOTS==NR_MEM_SECTIONS.
107 */
108int __section_nr(struct mem_section* ms)
109{
110 unsigned long root_nr;
111 struct mem_section* root;
112
Mike Kravetz12783b02006-05-20 15:00:05 -0700113 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
114 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
Dave Hansen4ca644d2005-10-29 18:16:51 -0700115 if (!root)
116 continue;
117
118 if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT)))
119 break;
120 }
121
Gavin Shandb36a462012-07-31 16:46:04 -0700122 VM_BUG_ON(root_nr == NR_SECTION_ROOTS);
123
Dave Hansen4ca644d2005-10-29 18:16:51 -0700124 return (root_nr * SECTIONS_PER_ROOT) + (ms - root);
125}
126
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700127/*
128 * During early boot, before section_mem_map is used for an actual
129 * mem_map, we use section_mem_map to store the section's NUMA
130 * node. This keeps us from having to use another data structure. The
131 * node information is cleared just before we store the real mem_map.
132 */
133static inline unsigned long sparse_encode_early_nid(int nid)
134{
135 return (nid << SECTION_NID_SHIFT);
136}
137
138static inline int sparse_early_nid(struct mem_section *section)
139{
140 return (section->section_mem_map >> SECTION_NID_SHIFT);
141}
142
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700143/* Validate the physical addressing limitations of the model */
144void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
145 unsigned long *end_pfn)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700146{
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700147 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700148
Ingo Molnarbead9a32008-04-16 01:40:00 +0200149 /*
150 * Sanity checks - do not allow an architecture to pass
151 * in larger pfns than the maximum scope of sparsemem:
152 */
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700153 if (*start_pfn > max_sparsemem_pfn) {
154 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
155 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
156 *start_pfn, *end_pfn, max_sparsemem_pfn);
157 WARN_ON_ONCE(1);
158 *start_pfn = max_sparsemem_pfn;
159 *end_pfn = max_sparsemem_pfn;
Cyrill Gorcunovef161a92009-03-31 15:19:25 -0700160 } else if (*end_pfn > max_sparsemem_pfn) {
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700161 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
162 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
163 *start_pfn, *end_pfn, max_sparsemem_pfn);
164 WARN_ON_ONCE(1);
165 *end_pfn = max_sparsemem_pfn;
166 }
167}
168
169/* Record a memory area against a node. */
170void __init memory_present(int nid, unsigned long start, unsigned long end)
171{
172 unsigned long pfn;
Ingo Molnarbead9a32008-04-16 01:40:00 +0200173
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700174 start &= PAGE_SECTION_MASK;
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700175 mminit_validate_memmodel_limits(&start, &end);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700176 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
177 unsigned long section = pfn_to_section_nr(pfn);
Bob Picco802f1922005-09-03 15:54:26 -0700178 struct mem_section *ms;
179
180 sparse_index_init(section, nid);
Andy Whitcroft85770ff2007-08-22 14:01:03 -0700181 set_section_nid(section, nid);
Bob Picco802f1922005-09-03 15:54:26 -0700182
183 ms = __nr_to_section(section);
184 if (!ms->section_mem_map)
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700185 ms->section_mem_map = sparse_encode_early_nid(nid) |
186 SECTION_MARKED_PRESENT;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700187 }
188}
189
190/*
191 * Only used by the i386 NUMA architecures, but relatively
192 * generic code.
193 */
194unsigned long __init node_memmap_size_bytes(int nid, unsigned long start_pfn,
195 unsigned long end_pfn)
196{
197 unsigned long pfn;
198 unsigned long nr_pages = 0;
199
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700200 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700201 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
202 if (nid != early_pfn_to_nid(pfn))
203 continue;
204
Andy Whitcroft540557b2007-10-16 01:24:11 -0700205 if (pfn_present(pfn))
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700206 nr_pages += PAGES_PER_SECTION;
207 }
208
209 return nr_pages * sizeof(struct page);
210}
211
212/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700213 * Subtle, we encode the real pfn into the mem_map such that
214 * the identity pfn - section_mem_map will return the actual
215 * physical page frame number.
216 */
217static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
218{
219 return (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
220}
221
222/*
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700223 * Decode mem_map from the coded memmap
Andy Whitcroft29751f62005-06-23 00:08:00 -0700224 */
Andy Whitcroft29751f62005-06-23 00:08:00 -0700225struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
226{
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700227 /* mask off the extra low bits of information */
228 coded_mem_map &= SECTION_MAP_MASK;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700229 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
230}
231
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700232static int __meminit sparse_init_one_section(struct mem_section *ms,
Mel Gorman5c0e3062007-10-16 01:25:56 -0700233 unsigned long pnum, struct page *mem_map,
234 unsigned long *pageblock_bitmap)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700235{
Andy Whitcroft540557b2007-10-16 01:24:11 -0700236 if (!present_section(ms))
Andy Whitcroft29751f62005-06-23 00:08:00 -0700237 return -EINVAL;
238
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700239 ms->section_mem_map &= ~SECTION_MAP_MASK;
Andy Whitcroft540557b2007-10-16 01:24:11 -0700240 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) |
241 SECTION_HAS_MEM_MAP;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700242 ms->pageblock_flags = pageblock_bitmap;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700243
244 return 1;
245}
246
Yasunori Goto04753272008-04-28 02:13:31 -0700247unsigned long usemap_size(void)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700248{
249 unsigned long size_bytes;
250 size_bytes = roundup(SECTION_BLOCKFLAGS_BITS, 8) / 8;
251 size_bytes = roundup(size_bytes, sizeof(unsigned long));
252 return size_bytes;
253}
254
255#ifdef CONFIG_MEMORY_HOTPLUG
256static unsigned long *__kmalloc_section_usemap(void)
257{
258 return kmalloc(usemap_size(), GFP_KERNEL);
259}
260#endif /* CONFIG_MEMORY_HOTPLUG */
261
Yasunori Goto48c90682008-07-23 21:28:15 -0700262#ifdef CONFIG_MEMORY_HOTREMOVE
263static unsigned long * __init
Yinghai Lua4322e12010-02-10 01:20:21 -0800264sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700265 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700266{
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700267 unsigned long goal, limit;
268 unsigned long *p;
269 int nid;
Yasunori Goto48c90682008-07-23 21:28:15 -0700270 /*
271 * A page may contain usemaps for other sections preventing the
272 * page being freed and making a section unremovable while
Li Zhongc800bcd2014-03-31 16:41:58 +0800273 * other sections referencing the usemap remain active. Similarly,
Yasunori Goto48c90682008-07-23 21:28:15 -0700274 * a pgdat can prevent a section being removed. If section A
275 * contains a pgdat and section B contains the usemap, both
276 * sections become inter-dependent. This allocates usemaps
277 * from the same section as the pgdat where possible to avoid
278 * this problem.
279 */
Yinghai Lu07b4e2b2012-07-11 14:02:51 -0700280 goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT);
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700281 limit = goal + (1UL << PA_SECTION_SHIFT);
282 nid = early_pfn_to_nid(goal >> PAGE_SHIFT);
283again:
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800284 p = memblock_virt_alloc_try_nid_nopanic(size,
285 SMP_CACHE_BYTES, goal, limit,
286 nid);
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700287 if (!p && limit) {
288 limit = 0;
289 goto again;
290 }
291 return p;
Yasunori Goto48c90682008-07-23 21:28:15 -0700292}
293
294static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
295{
296 unsigned long usemap_snr, pgdat_snr;
297 static unsigned long old_usemap_snr = NR_MEM_SECTIONS;
298 static unsigned long old_pgdat_snr = NR_MEM_SECTIONS;
299 struct pglist_data *pgdat = NODE_DATA(nid);
300 int usemap_nid;
301
302 usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT);
303 pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
304 if (usemap_snr == pgdat_snr)
305 return;
306
307 if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr)
308 /* skip redundant message */
309 return;
310
311 old_usemap_snr = usemap_snr;
312 old_pgdat_snr = pgdat_snr;
313
314 usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr));
315 if (usemap_nid != nid) {
316 printk(KERN_INFO
317 "node %d must be removed before remove section %ld\n",
318 nid, usemap_snr);
319 return;
320 }
321 /*
322 * There is a circular dependency.
323 * Some platforms allow un-removable section because they will just
324 * gather other removable sections for dynamic partitioning.
325 * Just notify un-removable section's number here.
326 */
327 printk(KERN_INFO "Section %ld and %ld (node %d)", usemap_snr,
328 pgdat_snr, nid);
329 printk(KERN_CONT
330 " have a circular dependency on usemap and pgdat allocations\n");
331}
332#else
333static unsigned long * __init
Yinghai Lua4322e12010-02-10 01:20:21 -0800334sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700335 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700336{
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800337 return memblock_virt_alloc_node_nopanic(size, pgdat->node_id);
Yasunori Goto48c90682008-07-23 21:28:15 -0700338}
339
340static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
341{
342}
343#endif /* CONFIG_MEMORY_HOTREMOVE */
344
Wanpeng Li18732092013-09-11 14:22:38 -0700345static void __init sparse_early_usemaps_alloc_node(void *data,
Yinghai Lua4322e12010-02-10 01:20:21 -0800346 unsigned long pnum_begin,
347 unsigned long pnum_end,
348 unsigned long usemap_count, int nodeid)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700349{
Yinghai Lua4322e12010-02-10 01:20:21 -0800350 void *usemap;
351 unsigned long pnum;
Wanpeng Li18732092013-09-11 14:22:38 -0700352 unsigned long **usemap_map = (unsigned long **)data;
Yinghai Lua4322e12010-02-10 01:20:21 -0800353 int size = usemap_size();
Mel Gorman5c0e3062007-10-16 01:25:56 -0700354
Yinghai Lua4322e12010-02-10 01:20:21 -0800355 usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid),
Johannes Weiner238305b2012-05-29 15:06:36 -0700356 size * usemap_count);
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700357 if (!usemap) {
Johannes Weiner238305b2012-05-29 15:06:36 -0700358 printk(KERN_WARNING "%s: allocation failed\n", __func__);
359 return;
Yasunori Goto48c90682008-07-23 21:28:15 -0700360 }
361
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700362 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
363 if (!present_section_nr(pnum))
364 continue;
365 usemap_map[pnum] = usemap;
366 usemap += size;
367 check_usemap_section_nr(nodeid, usemap_map[pnum]);
Yinghai Lua4322e12010-02-10 01:20:21 -0800368 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700369}
370
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700371#ifndef CONFIG_SPARSEMEM_VMEMMAP
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700372struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700373{
374 struct page *map;
Yinghai Lue48e67e2010-05-24 14:31:57 -0700375 unsigned long size;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700376
377 map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
378 if (map)
379 return map;
380
Yinghai Lue48e67e2010-05-24 14:31:57 -0700381 size = PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800382 map = memblock_virt_alloc_try_nid(size,
383 PAGE_SIZE, __pa(MAX_DMA_ADDRESS),
384 BOOTMEM_ALLOC_ACCESSIBLE, nid);
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700385 return map;
386}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800387void __init sparse_mem_maps_populate_node(struct page **map_map,
388 unsigned long pnum_begin,
389 unsigned long pnum_end,
390 unsigned long map_count, int nodeid)
391{
392 void *map;
393 unsigned long pnum;
394 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
395
396 map = alloc_remap(nodeid, size * map_count);
397 if (map) {
398 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
399 if (!present_section_nr(pnum))
400 continue;
401 map_map[pnum] = map;
402 map += size;
403 }
404 return;
405 }
406
407 size = PAGE_ALIGN(size);
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800408 map = memblock_virt_alloc_try_nid(size * map_count,
409 PAGE_SIZE, __pa(MAX_DMA_ADDRESS),
410 BOOTMEM_ALLOC_ACCESSIBLE, nodeid);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800411 if (map) {
412 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
413 if (!present_section_nr(pnum))
414 continue;
415 map_map[pnum] = map;
416 map += size;
417 }
418 return;
419 }
420
421 /* fallback */
422 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
423 struct mem_section *ms;
424
425 if (!present_section_nr(pnum))
426 continue;
427 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
428 if (map_map[pnum])
429 continue;
430 ms = __nr_to_section(pnum);
431 printk(KERN_ERR "%s: sparsemem memory map backing failed "
432 "some memory will not be available.\n", __func__);
433 ms->section_mem_map = 0;
434 }
435}
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700436#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
437
Yinghai Lu81d0d952010-02-27 09:29:38 -0800438#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Wanpeng Li18732092013-09-11 14:22:38 -0700439static void __init sparse_early_mem_maps_alloc_node(void *data,
Yinghai Lu9bdac912010-02-10 01:20:22 -0800440 unsigned long pnum_begin,
441 unsigned long pnum_end,
442 unsigned long map_count, int nodeid)
443{
Wanpeng Li18732092013-09-11 14:22:38 -0700444 struct page **map_map = (struct page **)data;
Yinghai Lu9bdac912010-02-10 01:20:22 -0800445 sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end,
446 map_count, nodeid);
447}
Yinghai Lu81d0d952010-02-27 09:29:38 -0800448#else
Adrian Bunk9e5c6da2008-07-25 19:46:22 -0700449static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700450{
451 struct page *map;
452 struct mem_section *ms = __nr_to_section(pnum);
453 int nid = sparse_early_nid(ms);
454
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700455 map = sparse_mem_map_populate(pnum, nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -0700456 if (map)
457 return map;
458
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700459 printk(KERN_ERR "%s: sparsemem memory map backing failed "
Harvey Harrisond40cee22008-04-30 00:55:07 -0700460 "some memory will not be available.\n", __func__);
Bob Picco802f1922005-09-03 15:54:26 -0700461 ms->section_mem_map = 0;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700462 return NULL;
463}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800464#endif
Andy Whitcroft29751f62005-06-23 00:08:00 -0700465
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700466void __weak __meminit vmemmap_populate_print_last(void)
Yinghai Luc2b91e22008-04-12 01:19:24 -0700467{
468}
Yinghai Lua4322e12010-02-10 01:20:21 -0800469
Wanpeng Li18732092013-09-11 14:22:38 -0700470/**
471 * alloc_usemap_and_memmap - memory alloction for pageblock flags and vmemmap
472 * @map: usemap_map for pageblock flags or mmap_map for vmemmap
Stephen Rothwell193faea2007-06-08 13:46:51 -0700473 */
Wanpeng Li18732092013-09-11 14:22:38 -0700474static void __init alloc_usemap_and_memmap(void (*alloc_func)
475 (void *, unsigned long, unsigned long,
476 unsigned long, int), void *data)
Stephen Rothwell193faea2007-06-08 13:46:51 -0700477{
478 unsigned long pnum;
Wanpeng Li18732092013-09-11 14:22:38 -0700479 unsigned long map_count;
Yinghai Lua4322e12010-02-10 01:20:21 -0800480 int nodeid_begin = 0;
481 unsigned long pnum_begin = 0;
Yinghai Lu9bdac912010-02-10 01:20:22 -0800482
483 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
484 struct mem_section *ms;
485
486 if (!present_section_nr(pnum))
487 continue;
488 ms = __nr_to_section(pnum);
489 nodeid_begin = sparse_early_nid(ms);
490 pnum_begin = pnum;
491 break;
492 }
493 map_count = 1;
494 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
495 struct mem_section *ms;
496 int nodeid;
497
498 if (!present_section_nr(pnum))
499 continue;
500 ms = __nr_to_section(pnum);
501 nodeid = sparse_early_nid(ms);
502 if (nodeid == nodeid_begin) {
503 map_count++;
504 continue;
505 }
506 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
Wanpeng Li18732092013-09-11 14:22:38 -0700507 alloc_func(data, pnum_begin, pnum,
508 map_count, nodeid_begin);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800509 /* new start, update count etc*/
510 nodeid_begin = nodeid;
511 pnum_begin = pnum;
512 map_count = 1;
513 }
514 /* ok, last chunk */
Wanpeng Li18732092013-09-11 14:22:38 -0700515 alloc_func(data, pnum_begin, NR_MEM_SECTIONS,
516 map_count, nodeid_begin);
517}
518
519/*
520 * Allocate the accumulated non-linear sections, allocate a mem_map
521 * for each and record the physical to section mapping.
522 */
523void __init sparse_init(void)
524{
525 unsigned long pnum;
526 struct page *map;
527 unsigned long *usemap;
528 unsigned long **usemap_map;
529 int size;
530#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
531 int size2;
532 struct page **map_map;
533#endif
534
535 /* see include/linux/mmzone.h 'struct mem_section' definition */
536 BUILD_BUG_ON(!is_power_of_2(sizeof(struct mem_section)));
537
538 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
539 set_pageblock_order();
540
541 /*
542 * map is using big page (aka 2M in x86 64 bit)
543 * usemap is less one page (aka 24 bytes)
544 * so alloc 2M (with 2M align) and 24 bytes in turn will
545 * make next 2M slip to one more 2M later.
546 * then in big system, the memory will have a lot of holes...
547 * here try to allocate 2M pages continuously.
548 *
549 * powerpc need to call sparse_init_one_section right after each
550 * sparse_early_mem_map_alloc, so allocate usemap_map at first.
551 */
552 size = sizeof(unsigned long *) * NR_MEM_SECTIONS;
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800553 usemap_map = memblock_virt_alloc(size, 0);
Wanpeng Li18732092013-09-11 14:22:38 -0700554 if (!usemap_map)
555 panic("can not allocate usemap_map\n");
556 alloc_usemap_and_memmap(sparse_early_usemaps_alloc_node,
557 (void *)usemap_map);
558
559#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
560 size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800561 map_map = memblock_virt_alloc(size2, 0);
Wanpeng Li18732092013-09-11 14:22:38 -0700562 if (!map_map)
563 panic("can not allocate map_map\n");
564 alloc_usemap_and_memmap(sparse_early_mem_maps_alloc_node,
565 (void *)map_map);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800566#endif
567
Yinghai Lue123dd32008-04-13 11:51:06 -0700568 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
569 if (!present_section_nr(pnum))
570 continue;
571
572 usemap = usemap_map[pnum];
573 if (!usemap)
574 continue;
Stephen Rothwell193faea2007-06-08 13:46:51 -0700575
Yinghai Lu9bdac912010-02-10 01:20:22 -0800576#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
577 map = map_map[pnum];
578#else
Stephen Rothwell193faea2007-06-08 13:46:51 -0700579 map = sparse_early_mem_map_alloc(pnum);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800580#endif
Stephen Rothwell193faea2007-06-08 13:46:51 -0700581 if (!map)
582 continue;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700583
Mel Gorman5c0e3062007-10-16 01:25:56 -0700584 sparse_init_one_section(__nr_to_section(pnum), pnum, map,
585 usemap);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700586 }
Yinghai Lue123dd32008-04-13 11:51:06 -0700587
Yinghai Luc2b91e22008-04-12 01:19:24 -0700588 vmemmap_populate_print_last();
589
Yinghai Lu9bdac912010-02-10 01:20:22 -0800590#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800591 memblock_free_early(__pa(map_map), size2);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800592#endif
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800593 memblock_free_early(__pa(usemap_map), size);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700594}
595
596#ifdef CONFIG_MEMORY_HOTPLUG
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700597#ifdef CONFIG_SPARSEMEM_VMEMMAP
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800598static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid)
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700599{
600 /* This will make the necessary allocations eventually. */
601 return sparse_mem_map_populate(pnum, nid);
602}
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800603static void __kfree_section_memmap(struct page *memmap)
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700604{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700605 unsigned long start = (unsigned long)memmap;
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800606 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
Johannes Weiner0aad8182013-04-29 15:07:50 -0700607
608 vmemmap_free(start, end);
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700609}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700610#ifdef CONFIG_MEMORY_HOTREMOVE
Zhang Yanfei81556b02013-11-12 15:07:43 -0800611static void free_map_bootmem(struct page *memmap)
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700612{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700613 unsigned long start = (unsigned long)memmap;
Zhang Yanfei81556b02013-11-12 15:07:43 -0800614 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
Johannes Weiner0aad8182013-04-29 15:07:50 -0700615
616 vmemmap_free(start, end);
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700617}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700618#endif /* CONFIG_MEMORY_HOTREMOVE */
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700619#else
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800620static struct page *__kmalloc_section_memmap(void)
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700621{
622 struct page *page, *ret;
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800623 unsigned long memmap_size = sizeof(struct page) * PAGES_PER_SECTION;
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700624
Yasunori Gotof2d0aa52006-10-28 10:38:32 -0700625 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700626 if (page)
627 goto got_map_page;
628
629 ret = vmalloc(memmap_size);
630 if (ret)
631 goto got_map_ptr;
632
633 return NULL;
634got_map_page:
635 ret = (struct page *)pfn_to_kaddr(page_to_pfn(page));
636got_map_ptr:
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700637
638 return ret;
639}
640
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800641static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid)
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700642{
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800643 return __kmalloc_section_memmap();
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700644}
645
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800646static void __kfree_section_memmap(struct page *memmap)
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700647{
Christoph Lameter9e2779f2008-02-04 22:28:34 -0800648 if (is_vmalloc_addr(memmap))
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700649 vfree(memmap);
650 else
651 free_pages((unsigned long)memmap,
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800652 get_order(sizeof(struct page) * PAGES_PER_SECTION));
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700653}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700654
David Rientjes4edd7ce2013-04-29 15:08:22 -0700655#ifdef CONFIG_MEMORY_HOTREMOVE
Zhang Yanfei81556b02013-11-12 15:07:43 -0800656static void free_map_bootmem(struct page *memmap)
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700657{
658 unsigned long maps_section_nr, removing_section_nr, i;
Zhang Yanfei81556b02013-11-12 15:07:43 -0800659 unsigned long magic, nr_pages;
Jianguo Wuae64ffc2012-11-29 13:54:21 -0800660 struct page *page = virt_to_page(memmap);
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700661
Zhang Yanfei81556b02013-11-12 15:07:43 -0800662 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
663 >> PAGE_SHIFT;
664
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700665 for (i = 0; i < nr_pages; i++, page++) {
Andrea Arcangeli5f24ce52011-01-13 15:47:00 -0800666 magic = (unsigned long) page->lru.next;
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700667
668 BUG_ON(magic == NODE_INFO);
669
670 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
671 removing_section_nr = page->private;
672
673 /*
674 * When this function is called, the removing section is
675 * logical offlined state. This means all pages are isolated
676 * from page allocator. If removing section's memmap is placed
677 * on the same section, it must not be freed.
678 * If it is freed, page allocator may allocate it which will
679 * be removed physically soon.
680 */
681 if (maps_section_nr != removing_section_nr)
682 put_page_bootmem(page);
683 }
684}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700685#endif /* CONFIG_MEMORY_HOTREMOVE */
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700686#endif /* CONFIG_SPARSEMEM_VMEMMAP */
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700687
Andy Whitcroft29751f62005-06-23 00:08:00 -0700688/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700689 * returns the number of sections whose mem_maps were properly
690 * set. If this is <=0, then that means that the passed-in
691 * map was not consumed and must be freed.
692 */
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800693int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700694{
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700695 unsigned long section_nr = pfn_to_section_nr(start_pfn);
696 struct pglist_data *pgdat = zone->zone_pgdat;
697 struct mem_section *ms;
698 struct page *memmap;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700699 unsigned long *usemap;
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700700 unsigned long flags;
701 int ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700702
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700703 /*
704 * no locking for this, because it does its own
705 * plus, it does a kmalloc
706 */
WANG Congbbd06822007-12-17 16:19:59 -0800707 ret = sparse_index_init(section_nr, pgdat->node_id);
708 if (ret < 0 && ret != -EEXIST)
709 return ret;
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800710 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id);
WANG Congbbd06822007-12-17 16:19:59 -0800711 if (!memmap)
712 return -ENOMEM;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700713 usemap = __kmalloc_section_usemap();
WANG Congbbd06822007-12-17 16:19:59 -0800714 if (!usemap) {
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800715 __kfree_section_memmap(memmap);
WANG Congbbd06822007-12-17 16:19:59 -0800716 return -ENOMEM;
717 }
Andy Whitcroft29751f62005-06-23 00:08:00 -0700718
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700719 pgdat_resize_lock(pgdat, &flags);
720
721 ms = __pfn_to_section(start_pfn);
722 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
723 ret = -EEXIST;
724 goto out;
725 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700726
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800727 memset(memmap, 0, sizeof(struct page) * PAGES_PER_SECTION);
Wen Congyang3ac19f82012-12-11 16:00:59 -0800728
Andy Whitcroft29751f62005-06-23 00:08:00 -0700729 ms->section_mem_map |= SECTION_MARKED_PRESENT;
730
Mel Gorman5c0e3062007-10-16 01:25:56 -0700731 ret = sparse_init_one_section(ms, section_nr, memmap, usemap);
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700732
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700733out:
734 pgdat_resize_unlock(pgdat, &flags);
WANG Congbbd06822007-12-17 16:19:59 -0800735 if (ret <= 0) {
736 kfree(usemap);
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800737 __kfree_section_memmap(memmap);
WANG Congbbd06822007-12-17 16:19:59 -0800738 }
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700739 return ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700740}
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700741
Zhang Yanfeif3deb682013-07-08 16:00:10 -0700742#ifdef CONFIG_MEMORY_HOTREMOVE
Wen Congyang95a47742012-12-11 16:00:47 -0800743#ifdef CONFIG_MEMORY_FAILURE
744static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
745{
746 int i;
747
748 if (!memmap)
749 return;
750
Dan Williams4b94ffd2016-01-15 16:56:22 -0800751 for (i = 0; i < nr_pages; i++) {
Wen Congyang95a47742012-12-11 16:00:47 -0800752 if (PageHWPoison(&memmap[i])) {
Xishi Qiu293c07e2013-02-22 16:34:02 -0800753 atomic_long_sub(1, &num_poisoned_pages);
Wen Congyang95a47742012-12-11 16:00:47 -0800754 ClearPageHWPoison(&memmap[i]);
755 }
756 }
757}
758#else
759static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
760{
761}
762#endif
763
David Rientjes4edd7ce2013-04-29 15:08:22 -0700764static void free_section_usemap(struct page *memmap, unsigned long *usemap)
765{
766 struct page *usemap_page;
David Rientjes4edd7ce2013-04-29 15:08:22 -0700767
768 if (!usemap)
769 return;
770
771 usemap_page = virt_to_page(usemap);
772 /*
773 * Check to see if allocation came from hot-plug-add
774 */
775 if (PageSlab(usemap_page) || PageCompound(usemap_page)) {
776 kfree(usemap);
777 if (memmap)
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800778 __kfree_section_memmap(memmap);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700779 return;
780 }
781
782 /*
783 * The usemap came from bootmem. This is packed with other usemaps
784 * on the section which has pgdat at boot time. Just keep it as is now.
785 */
786
Zhang Yanfei81556b02013-11-12 15:07:43 -0800787 if (memmap)
788 free_map_bootmem(memmap);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700789}
790
Dan Williams4b94ffd2016-01-15 16:56:22 -0800791void sparse_remove_one_section(struct zone *zone, struct mem_section *ms,
792 unsigned long map_offset)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700793{
794 struct page *memmap = NULL;
Tang Chencd099682013-02-22 16:33:02 -0800795 unsigned long *usemap = NULL, flags;
796 struct pglist_data *pgdat = zone->zone_pgdat;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700797
Tang Chencd099682013-02-22 16:33:02 -0800798 pgdat_resize_lock(pgdat, &flags);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700799 if (ms->section_mem_map) {
800 usemap = ms->pageblock_flags;
801 memmap = sparse_decode_mem_map(ms->section_mem_map,
802 __section_nr(ms));
803 ms->section_mem_map = 0;
804 ms->pageblock_flags = NULL;
805 }
Tang Chencd099682013-02-22 16:33:02 -0800806 pgdat_resize_unlock(pgdat, &flags);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700807
Dan Williams4b94ffd2016-01-15 16:56:22 -0800808 clear_hwpoisoned_pages(memmap + map_offset,
809 PAGES_PER_SECTION - map_offset);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700810 free_section_usemap(memmap, usemap);
811}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700812#endif /* CONFIG_MEMORY_HOTREMOVE */
813#endif /* CONFIG_MEMORY_HOTPLUG */