blob: 6903c8fc308502eab4ca3570075cd6a881efa724 [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 Hansen0b0acbec2005-10-29 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 Hansen0b0acbec2005-10-29 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
Fabian Frederickbd721ea2016-08-02 14:03:33 -070062static noinline struct mem_section __ref *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
Zhou Chengming91fd8b92016-07-28 15:48:35 -0700103#ifdef CONFIG_SPARSEMEM_EXTREME
Dave Hansen4ca644d2005-10-29 18:16:51 -0700104int __section_nr(struct mem_section* ms)
105{
106 unsigned long root_nr;
107 struct mem_section* root;
108
Mike Kravetz12783b02006-05-20 15:00:05 -0700109 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
110 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
Dave Hansen4ca644d2005-10-29 18:16:51 -0700111 if (!root)
112 continue;
113
114 if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT)))
115 break;
116 }
117
Gavin Shandb36a462012-07-31 16:46:04 -0700118 VM_BUG_ON(root_nr == NR_SECTION_ROOTS);
119
Dave Hansen4ca644d2005-10-29 18:16:51 -0700120 return (root_nr * SECTIONS_PER_ROOT) + (ms - root);
121}
Zhou Chengming91fd8b92016-07-28 15:48:35 -0700122#else
123int __section_nr(struct mem_section* ms)
124{
125 return (int)(ms - mem_section[0]);
126}
127#endif
Dave Hansen4ca644d2005-10-29 18:16:51 -0700128
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700129/*
130 * During early boot, before section_mem_map is used for an actual
131 * mem_map, we use section_mem_map to store the section's NUMA
132 * node. This keeps us from having to use another data structure. The
133 * node information is cleared just before we store the real mem_map.
134 */
135static inline unsigned long sparse_encode_early_nid(int nid)
136{
137 return (nid << SECTION_NID_SHIFT);
138}
139
140static inline int sparse_early_nid(struct mem_section *section)
141{
142 return (section->section_mem_map >> SECTION_NID_SHIFT);
143}
144
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700145/* Validate the physical addressing limitations of the model */
146void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
147 unsigned long *end_pfn)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700148{
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700149 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700150
Ingo Molnarbead9a32008-04-16 01:40:00 +0200151 /*
152 * Sanity checks - do not allow an architecture to pass
153 * in larger pfns than the maximum scope of sparsemem:
154 */
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700155 if (*start_pfn > max_sparsemem_pfn) {
156 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
157 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
158 *start_pfn, *end_pfn, max_sparsemem_pfn);
159 WARN_ON_ONCE(1);
160 *start_pfn = max_sparsemem_pfn;
161 *end_pfn = max_sparsemem_pfn;
Cyrill Gorcunovef161a92009-03-31 15:19:25 -0700162 } else if (*end_pfn > max_sparsemem_pfn) {
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700163 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
164 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
165 *start_pfn, *end_pfn, max_sparsemem_pfn);
166 WARN_ON_ONCE(1);
167 *end_pfn = max_sparsemem_pfn;
168 }
169}
170
171/* Record a memory area against a node. */
172void __init memory_present(int nid, unsigned long start, unsigned long end)
173{
174 unsigned long pfn;
Ingo Molnarbead9a32008-04-16 01:40:00 +0200175
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700176 start &= PAGE_SECTION_MASK;
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700177 mminit_validate_memmodel_limits(&start, &end);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700178 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
179 unsigned long section = pfn_to_section_nr(pfn);
Bob Picco802f1922005-09-03 15:54:26 -0700180 struct mem_section *ms;
181
182 sparse_index_init(section, nid);
Andy Whitcroft85770ff2007-08-22 14:01:03 -0700183 set_section_nid(section, nid);
Bob Picco802f1922005-09-03 15:54:26 -0700184
185 ms = __nr_to_section(section);
186 if (!ms->section_mem_map)
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700187 ms->section_mem_map = sparse_encode_early_nid(nid) |
188 SECTION_MARKED_PRESENT;
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700189 }
190}
191
192/*
193 * Only used by the i386 NUMA architecures, but relatively
194 * generic code.
195 */
196unsigned long __init node_memmap_size_bytes(int nid, unsigned long start_pfn,
197 unsigned long end_pfn)
198{
199 unsigned long pfn;
200 unsigned long nr_pages = 0;
201
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700202 mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700203 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
204 if (nid != early_pfn_to_nid(pfn))
205 continue;
206
Andy Whitcroft540557b2007-10-16 01:24:11 -0700207 if (pfn_present(pfn))
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700208 nr_pages += PAGES_PER_SECTION;
209 }
210
211 return nr_pages * sizeof(struct page);
212}
213
214/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700215 * Subtle, we encode the real pfn into the mem_map such that
216 * the identity pfn - section_mem_map will return the actual
217 * physical page frame number.
218 */
219static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
220{
221 return (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
222}
223
224/*
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700225 * Decode mem_map from the coded memmap
Andy Whitcroft29751f62005-06-23 00:08:00 -0700226 */
Andy Whitcroft29751f62005-06-23 00:08:00 -0700227struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
228{
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700229 /* mask off the extra low bits of information */
230 coded_mem_map &= SECTION_MAP_MASK;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700231 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
232}
233
Yasunori Gotoa3142c82007-05-08 00:23:07 -0700234static int __meminit sparse_init_one_section(struct mem_section *ms,
Mel Gorman5c0e3062007-10-16 01:25:56 -0700235 unsigned long pnum, struct page *mem_map,
236 unsigned long *pageblock_bitmap)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700237{
Andy Whitcroft540557b2007-10-16 01:24:11 -0700238 if (!present_section(ms))
Andy Whitcroft29751f62005-06-23 00:08:00 -0700239 return -EINVAL;
240
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700241 ms->section_mem_map &= ~SECTION_MAP_MASK;
Andy Whitcroft540557b2007-10-16 01:24:11 -0700242 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) |
243 SECTION_HAS_MEM_MAP;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700244 ms->pageblock_flags = pageblock_bitmap;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700245
246 return 1;
247}
248
Yasunori Goto04753272008-04-28 02:13:31 -0700249unsigned long usemap_size(void)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700250{
Wei Yang60a7a882017-05-03 14:53:51 -0700251 return BITS_TO_LONGS(SECTION_BLOCKFLAGS_BITS) * sizeof(unsigned long);
Mel Gorman5c0e3062007-10-16 01:25:56 -0700252}
253
254#ifdef CONFIG_MEMORY_HOTPLUG
255static unsigned long *__kmalloc_section_usemap(void)
256{
257 return kmalloc(usemap_size(), GFP_KERNEL);
258}
259#endif /* CONFIG_MEMORY_HOTPLUG */
260
Yasunori Goto48c90682008-07-23 21:28:15 -0700261#ifdef CONFIG_MEMORY_HOTREMOVE
262static unsigned long * __init
Yinghai Lua4322e1b2010-02-10 01:20:21 -0800263sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700264 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700265{
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700266 unsigned long goal, limit;
267 unsigned long *p;
268 int nid;
Yasunori Goto48c90682008-07-23 21:28:15 -0700269 /*
270 * A page may contain usemaps for other sections preventing the
271 * page being freed and making a section unremovable while
Li Zhongc800bcd2014-03-31 16:41:58 +0800272 * other sections referencing the usemap remain active. Similarly,
Yasunori Goto48c90682008-07-23 21:28:15 -0700273 * a pgdat can prevent a section being removed. If section A
274 * contains a pgdat and section B contains the usemap, both
275 * sections become inter-dependent. This allocates usemaps
276 * from the same section as the pgdat where possible to avoid
277 * this problem.
278 */
Yinghai Lu07b4e2b2012-07-11 14:02:51 -0700279 goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT);
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700280 limit = goal + (1UL << PA_SECTION_SHIFT);
281 nid = early_pfn_to_nid(goal >> PAGE_SHIFT);
282again:
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800283 p = memblock_virt_alloc_try_nid_nopanic(size,
284 SMP_CACHE_BYTES, goal, limit,
285 nid);
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700286 if (!p && limit) {
287 limit = 0;
288 goto again;
289 }
290 return p;
Yasunori Goto48c90682008-07-23 21:28:15 -0700291}
292
293static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
294{
295 unsigned long usemap_snr, pgdat_snr;
296 static unsigned long old_usemap_snr = NR_MEM_SECTIONS;
297 static unsigned long old_pgdat_snr = NR_MEM_SECTIONS;
298 struct pglist_data *pgdat = NODE_DATA(nid);
299 int usemap_nid;
300
301 usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT);
302 pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
303 if (usemap_snr == pgdat_snr)
304 return;
305
306 if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr)
307 /* skip redundant message */
308 return;
309
310 old_usemap_snr = usemap_snr;
311 old_pgdat_snr = pgdat_snr;
312
313 usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr));
314 if (usemap_nid != nid) {
Joe Perches11705322016-03-17 14:19:50 -0700315 pr_info("node %d must be removed before remove section %ld\n",
316 nid, usemap_snr);
Yasunori Goto48c90682008-07-23 21:28:15 -0700317 return;
318 }
319 /*
320 * There is a circular dependency.
321 * Some platforms allow un-removable section because they will just
322 * gather other removable sections for dynamic partitioning.
323 * Just notify un-removable section's number here.
324 */
Joe Perches11705322016-03-17 14:19:50 -0700325 pr_info("Section %ld and %ld (node %d) have a circular dependency on usemap and pgdat allocations\n",
326 usemap_snr, pgdat_snr, nid);
Yasunori Goto48c90682008-07-23 21:28:15 -0700327}
328#else
329static unsigned long * __init
Yinghai Lua4322e1b2010-02-10 01:20:21 -0800330sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700331 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700332{
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800333 return memblock_virt_alloc_node_nopanic(size, pgdat->node_id);
Yasunori Goto48c90682008-07-23 21:28:15 -0700334}
335
336static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
337{
338}
339#endif /* CONFIG_MEMORY_HOTREMOVE */
340
Wanpeng Li18732092013-09-11 14:22:38 -0700341static void __init sparse_early_usemaps_alloc_node(void *data,
Yinghai Lua4322e1b2010-02-10 01:20:21 -0800342 unsigned long pnum_begin,
343 unsigned long pnum_end,
344 unsigned long usemap_count, int nodeid)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700345{
Yinghai Lua4322e1b2010-02-10 01:20:21 -0800346 void *usemap;
347 unsigned long pnum;
Wanpeng Li18732092013-09-11 14:22:38 -0700348 unsigned long **usemap_map = (unsigned long **)data;
Yinghai Lua4322e1b2010-02-10 01:20:21 -0800349 int size = usemap_size();
Mel Gorman5c0e3062007-10-16 01:25:56 -0700350
Yinghai Lua4322e1b2010-02-10 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) {
Joe Perches11705322016-03-17 14:19:50 -0700354 pr_warn("%s: allocation failed\n", __func__);
Johannes Weiner238305b2012-05-29 15:06:36 -0700355 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 Lua4322e1b2010-02-10 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);
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800378 map = memblock_virt_alloc_try_nid(size,
379 PAGE_SIZE, __pa(MAX_DMA_ADDRESS),
380 BOOTMEM_ALLOC_ACCESSIBLE, nid);
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700381 return map;
382}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800383void __init sparse_mem_maps_populate_node(struct page **map_map,
384 unsigned long pnum_begin,
385 unsigned long pnum_end,
386 unsigned long map_count, int nodeid)
387{
388 void *map;
389 unsigned long pnum;
390 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
391
392 map = alloc_remap(nodeid, size * map_count);
393 if (map) {
394 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
395 if (!present_section_nr(pnum))
396 continue;
397 map_map[pnum] = map;
398 map += size;
399 }
400 return;
401 }
402
403 size = PAGE_ALIGN(size);
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800404 map = memblock_virt_alloc_try_nid(size * map_count,
405 PAGE_SIZE, __pa(MAX_DMA_ADDRESS),
406 BOOTMEM_ALLOC_ACCESSIBLE, nodeid);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800407 if (map) {
408 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
409 if (!present_section_nr(pnum))
410 continue;
411 map_map[pnum] = map;
412 map += size;
413 }
414 return;
415 }
416
417 /* fallback */
418 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
419 struct mem_section *ms;
420
421 if (!present_section_nr(pnum))
422 continue;
423 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
424 if (map_map[pnum])
425 continue;
426 ms = __nr_to_section(pnum);
Joe Perches11705322016-03-17 14:19:50 -0700427 pr_err("%s: sparsemem memory map backing failed some memory will not be available\n",
Joe Perches756a0252016-03-17 14:19:47 -0700428 __func__);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800429 ms->section_mem_map = 0;
430 }
431}
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700432#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
433
Yinghai Lu81d0d952010-02-27 09:29:38 -0800434#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Wanpeng Li18732092013-09-11 14:22:38 -0700435static void __init sparse_early_mem_maps_alloc_node(void *data,
Yinghai Lu9bdac912010-02-10 01:20:22 -0800436 unsigned long pnum_begin,
437 unsigned long pnum_end,
438 unsigned long map_count, int nodeid)
439{
Wanpeng Li18732092013-09-11 14:22:38 -0700440 struct page **map_map = (struct page **)data;
Yinghai Lu9bdac912010-02-10 01:20:22 -0800441 sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end,
442 map_count, nodeid);
443}
Yinghai Lu81d0d952010-02-27 09:29:38 -0800444#else
Adrian Bunk9e5c6da2008-07-25 19:46:22 -0700445static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700446{
447 struct page *map;
448 struct mem_section *ms = __nr_to_section(pnum);
449 int nid = sparse_early_nid(ms);
450
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700451 map = sparse_mem_map_populate(pnum, nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -0700452 if (map)
453 return map;
454
Joe Perches11705322016-03-17 14:19:50 -0700455 pr_err("%s: sparsemem memory map backing failed some memory will not be available\n",
Joe Perches756a0252016-03-17 14:19:47 -0700456 __func__);
Bob Picco802f1922005-09-03 15:54:26 -0700457 ms->section_mem_map = 0;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700458 return NULL;
459}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800460#endif
Andy Whitcroft29751f62005-06-23 00:08:00 -0700461
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700462void __weak __meminit vmemmap_populate_print_last(void)
Yinghai Luc2b91e22008-04-12 01:19:24 -0700463{
464}
Yinghai Lua4322e1b2010-02-10 01:20:21 -0800465
Wanpeng Li18732092013-09-11 14:22:38 -0700466/**
467 * alloc_usemap_and_memmap - memory alloction for pageblock flags and vmemmap
468 * @map: usemap_map for pageblock flags or mmap_map for vmemmap
Stephen Rothwell193faea2007-06-08 13:46:51 -0700469 */
Wanpeng Li18732092013-09-11 14:22:38 -0700470static void __init alloc_usemap_and_memmap(void (*alloc_func)
471 (void *, unsigned long, unsigned long,
472 unsigned long, int), void *data)
Stephen Rothwell193faea2007-06-08 13:46:51 -0700473{
474 unsigned long pnum;
Wanpeng Li18732092013-09-11 14:22:38 -0700475 unsigned long map_count;
Yinghai Lua4322e1b2010-02-10 01:20:21 -0800476 int nodeid_begin = 0;
477 unsigned long pnum_begin = 0;
Yinghai Lu9bdac912010-02-10 01:20:22 -0800478
479 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
480 struct mem_section *ms;
481
482 if (!present_section_nr(pnum))
483 continue;
484 ms = __nr_to_section(pnum);
485 nodeid_begin = sparse_early_nid(ms);
486 pnum_begin = pnum;
487 break;
488 }
489 map_count = 1;
490 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
491 struct mem_section *ms;
492 int nodeid;
493
494 if (!present_section_nr(pnum))
495 continue;
496 ms = __nr_to_section(pnum);
497 nodeid = sparse_early_nid(ms);
498 if (nodeid == nodeid_begin) {
499 map_count++;
500 continue;
501 }
502 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
Wanpeng Li18732092013-09-11 14:22:38 -0700503 alloc_func(data, pnum_begin, pnum,
504 map_count, nodeid_begin);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800505 /* new start, update count etc*/
506 nodeid_begin = nodeid;
507 pnum_begin = pnum;
508 map_count = 1;
509 }
510 /* ok, last chunk */
Wanpeng Li18732092013-09-11 14:22:38 -0700511 alloc_func(data, pnum_begin, NR_MEM_SECTIONS,
512 map_count, nodeid_begin);
513}
514
515/*
516 * Allocate the accumulated non-linear sections, allocate a mem_map
517 * for each and record the physical to section mapping.
518 */
519void __init sparse_init(void)
520{
521 unsigned long pnum;
522 struct page *map;
523 unsigned long *usemap;
524 unsigned long **usemap_map;
525 int size;
526#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
527 int size2;
528 struct page **map_map;
529#endif
530
531 /* see include/linux/mmzone.h 'struct mem_section' definition */
532 BUILD_BUG_ON(!is_power_of_2(sizeof(struct mem_section)));
533
534 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
535 set_pageblock_order();
536
537 /*
538 * map is using big page (aka 2M in x86 64 bit)
539 * usemap is less one page (aka 24 bytes)
540 * so alloc 2M (with 2M align) and 24 bytes in turn will
541 * make next 2M slip to one more 2M later.
542 * then in big system, the memory will have a lot of holes...
543 * here try to allocate 2M pages continuously.
544 *
545 * powerpc need to call sparse_init_one_section right after each
546 * sparse_early_mem_map_alloc, so allocate usemap_map at first.
547 */
548 size = sizeof(unsigned long *) * NR_MEM_SECTIONS;
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800549 usemap_map = memblock_virt_alloc(size, 0);
Wanpeng Li18732092013-09-11 14:22:38 -0700550 if (!usemap_map)
551 panic("can not allocate usemap_map\n");
552 alloc_usemap_and_memmap(sparse_early_usemaps_alloc_node,
553 (void *)usemap_map);
554
555#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
556 size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800557 map_map = memblock_virt_alloc(size2, 0);
Wanpeng Li18732092013-09-11 14:22:38 -0700558 if (!map_map)
559 panic("can not allocate map_map\n");
560 alloc_usemap_and_memmap(sparse_early_mem_maps_alloc_node,
561 (void *)map_map);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800562#endif
563
Yinghai Lue123dd32008-04-13 11:51:06 -0700564 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
565 if (!present_section_nr(pnum))
566 continue;
567
568 usemap = usemap_map[pnum];
569 if (!usemap)
570 continue;
Stephen Rothwell193faea2007-06-08 13:46:51 -0700571
Yinghai Lu9bdac912010-02-10 01:20:22 -0800572#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
573 map = map_map[pnum];
574#else
Stephen Rothwell193faea2007-06-08 13:46:51 -0700575 map = sparse_early_mem_map_alloc(pnum);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800576#endif
Stephen Rothwell193faea2007-06-08 13:46:51 -0700577 if (!map)
578 continue;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700579
Mel Gorman5c0e3062007-10-16 01:25:56 -0700580 sparse_init_one_section(__nr_to_section(pnum), pnum, map,
581 usemap);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700582 }
Yinghai Lue123dd32008-04-13 11:51:06 -0700583
Yinghai Luc2b91e22008-04-12 01:19:24 -0700584 vmemmap_populate_print_last();
585
Yinghai Lu9bdac912010-02-10 01:20:22 -0800586#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800587 memblock_free_early(__pa(map_map), size2);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800588#endif
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800589 memblock_free_early(__pa(usemap_map), size);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700590}
591
592#ifdef CONFIG_MEMORY_HOTPLUG
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700593#ifdef CONFIG_SPARSEMEM_VMEMMAP
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800594static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid)
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700595{
596 /* This will make the necessary allocations eventually. */
597 return sparse_mem_map_populate(pnum, nid);
598}
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800599static void __kfree_section_memmap(struct page *memmap)
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700600{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700601 unsigned long start = (unsigned long)memmap;
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800602 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
Johannes Weiner0aad8182013-04-29 15:07:50 -0700603
604 vmemmap_free(start, end);
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700605}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700606#ifdef CONFIG_MEMORY_HOTREMOVE
Zhang Yanfei81556b02013-11-12 15:07:43 -0800607static void free_map_bootmem(struct page *memmap)
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700608{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700609 unsigned long start = (unsigned long)memmap;
Zhang Yanfei81556b02013-11-12 15:07:43 -0800610 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
Johannes Weiner0aad8182013-04-29 15:07:50 -0700611
612 vmemmap_free(start, end);
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700613}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700614#endif /* CONFIG_MEMORY_HOTREMOVE */
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700615#else
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800616static struct page *__kmalloc_section_memmap(void)
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700617{
618 struct page *page, *ret;
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800619 unsigned long memmap_size = sizeof(struct page) * PAGES_PER_SECTION;
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700620
Yasunori Gotof2d0aa52006-10-28 10:38:32 -0700621 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700622 if (page)
623 goto got_map_page;
624
625 ret = vmalloc(memmap_size);
626 if (ret)
627 goto got_map_ptr;
628
629 return NULL;
630got_map_page:
631 ret = (struct page *)pfn_to_kaddr(page_to_pfn(page));
632got_map_ptr:
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700633
634 return ret;
635}
636
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800637static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid)
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700638{
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800639 return __kmalloc_section_memmap();
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700640}
641
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800642static void __kfree_section_memmap(struct page *memmap)
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700643{
Christoph Lameter9e2779f2008-02-04 22:28:34 -0800644 if (is_vmalloc_addr(memmap))
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700645 vfree(memmap);
646 else
647 free_pages((unsigned long)memmap,
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800648 get_order(sizeof(struct page) * PAGES_PER_SECTION));
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700649}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700650
David Rientjes4edd7ce2013-04-29 15:08:22 -0700651#ifdef CONFIG_MEMORY_HOTREMOVE
Zhang Yanfei81556b02013-11-12 15:07:43 -0800652static void free_map_bootmem(struct page *memmap)
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700653{
654 unsigned long maps_section_nr, removing_section_nr, i;
Zhang Yanfei81556b02013-11-12 15:07:43 -0800655 unsigned long magic, nr_pages;
Jianguo Wuae64ffc2012-11-29 13:54:21 -0800656 struct page *page = virt_to_page(memmap);
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700657
Zhang Yanfei81556b02013-11-12 15:07:43 -0800658 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
659 >> PAGE_SHIFT;
660
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700661 for (i = 0; i < nr_pages; i++, page++) {
Yasuaki Ishimatsuddffe982017-02-22 15:45:13 -0800662 magic = (unsigned long) page->freelist;
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700663
664 BUG_ON(magic == NODE_INFO);
665
666 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
Yasuaki Ishimatsu857e5222017-02-22 15:45:10 -0800667 removing_section_nr = page_private(page);
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700668
669 /*
670 * When this function is called, the removing section is
671 * logical offlined state. This means all pages are isolated
672 * from page allocator. If removing section's memmap is placed
673 * on the same section, it must not be freed.
674 * If it is freed, page allocator may allocate it which will
675 * be removed physically soon.
676 */
677 if (maps_section_nr != removing_section_nr)
678 put_page_bootmem(page);
679 }
680}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700681#endif /* CONFIG_MEMORY_HOTREMOVE */
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700682#endif /* CONFIG_SPARSEMEM_VMEMMAP */
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700683
Andy Whitcroft29751f62005-06-23 00:08:00 -0700684/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700685 * returns the number of sections whose mem_maps were properly
686 * set. If this is <=0, then that means that the passed-in
687 * map was not consumed and must be freed.
688 */
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800689int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700690{
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700691 unsigned long section_nr = pfn_to_section_nr(start_pfn);
692 struct pglist_data *pgdat = zone->zone_pgdat;
693 struct mem_section *ms;
694 struct page *memmap;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700695 unsigned long *usemap;
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700696 unsigned long flags;
697 int ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700698
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700699 /*
700 * no locking for this, because it does its own
701 * plus, it does a kmalloc
702 */
WANG Congbbd06822007-12-17 16:19:59 -0800703 ret = sparse_index_init(section_nr, pgdat->node_id);
704 if (ret < 0 && ret != -EEXIST)
705 return ret;
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800706 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id);
WANG Congbbd06822007-12-17 16:19:59 -0800707 if (!memmap)
708 return -ENOMEM;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700709 usemap = __kmalloc_section_usemap();
WANG Congbbd06822007-12-17 16:19:59 -0800710 if (!usemap) {
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800711 __kfree_section_memmap(memmap);
WANG Congbbd06822007-12-17 16:19:59 -0800712 return -ENOMEM;
713 }
Andy Whitcroft29751f62005-06-23 00:08:00 -0700714
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700715 pgdat_resize_lock(pgdat, &flags);
716
717 ms = __pfn_to_section(start_pfn);
718 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
719 ret = -EEXIST;
720 goto out;
721 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700722
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800723 memset(memmap, 0, sizeof(struct page) * PAGES_PER_SECTION);
Wen Congyang3ac19f82012-12-11 16:00:59 -0800724
Andy Whitcroft29751f62005-06-23 00:08:00 -0700725 ms->section_mem_map |= SECTION_MARKED_PRESENT;
726
Mel Gorman5c0e3062007-10-16 01:25:56 -0700727 ret = sparse_init_one_section(ms, section_nr, memmap, usemap);
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700728
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700729out:
730 pgdat_resize_unlock(pgdat, &flags);
WANG Congbbd06822007-12-17 16:19:59 -0800731 if (ret <= 0) {
732 kfree(usemap);
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800733 __kfree_section_memmap(memmap);
WANG Congbbd06822007-12-17 16:19:59 -0800734 }
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700735 return ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700736}
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700737
Zhang Yanfeif3deb682013-07-08 16:00:10 -0700738#ifdef CONFIG_MEMORY_HOTREMOVE
Wen Congyang95a47742012-12-11 16:00:47 -0800739#ifdef CONFIG_MEMORY_FAILURE
740static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
741{
742 int i;
743
744 if (!memmap)
745 return;
746
Dan Williams4b94ffd2016-01-15 16:56:22 -0800747 for (i = 0; i < nr_pages; i++) {
Wen Congyang95a47742012-12-11 16:00:47 -0800748 if (PageHWPoison(&memmap[i])) {
Xishi Qiu293c07e2013-02-22 16:34:02 -0800749 atomic_long_sub(1, &num_poisoned_pages);
Wen Congyang95a47742012-12-11 16:00:47 -0800750 ClearPageHWPoison(&memmap[i]);
751 }
752 }
753}
754#else
755static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
756{
757}
758#endif
759
David Rientjes4edd7ce2013-04-29 15:08:22 -0700760static void free_section_usemap(struct page *memmap, unsigned long *usemap)
761{
762 struct page *usemap_page;
David Rientjes4edd7ce2013-04-29 15:08:22 -0700763
764 if (!usemap)
765 return;
766
767 usemap_page = virt_to_page(usemap);
768 /*
769 * Check to see if allocation came from hot-plug-add
770 */
771 if (PageSlab(usemap_page) || PageCompound(usemap_page)) {
772 kfree(usemap);
773 if (memmap)
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800774 __kfree_section_memmap(memmap);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700775 return;
776 }
777
778 /*
779 * The usemap came from bootmem. This is packed with other usemaps
780 * on the section which has pgdat at boot time. Just keep it as is now.
781 */
782
Zhang Yanfei81556b02013-11-12 15:07:43 -0800783 if (memmap)
784 free_map_bootmem(memmap);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700785}
786
Dan Williams4b94ffd2016-01-15 16:56:22 -0800787void sparse_remove_one_section(struct zone *zone, struct mem_section *ms,
788 unsigned long map_offset)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700789{
790 struct page *memmap = NULL;
Tang Chencd099682013-02-22 16:33:02 -0800791 unsigned long *usemap = NULL, flags;
792 struct pglist_data *pgdat = zone->zone_pgdat;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700793
Tang Chencd099682013-02-22 16:33:02 -0800794 pgdat_resize_lock(pgdat, &flags);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700795 if (ms->section_mem_map) {
796 usemap = ms->pageblock_flags;
797 memmap = sparse_decode_mem_map(ms->section_mem_map,
798 __section_nr(ms));
799 ms->section_mem_map = 0;
800 ms->pageblock_flags = NULL;
801 }
Tang Chencd099682013-02-22 16:33:02 -0800802 pgdat_resize_unlock(pgdat, &flags);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700803
Dan Williams4b94ffd2016-01-15 16:56:22 -0800804 clear_hwpoisoned_pages(memmap + map_offset,
805 PAGES_PER_SECTION - map_offset);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700806 free_section_usemap(memmap, usemap);
807}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700808#endif /* CONFIG_MEMORY_HOTREMOVE */
809#endif /* CONFIG_MEMORY_HOTPLUG */