blob: db6bf3c97ea2cd7e593922ab88840b84d67cdd88 [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
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{
251 unsigned long size_bytes;
252 size_bytes = roundup(SECTION_BLOCKFLAGS_BITS, 8) / 8;
253 size_bytes = roundup(size_bytes, sizeof(unsigned long));
254 return size_bytes;
255}
256
257#ifdef CONFIG_MEMORY_HOTPLUG
258static unsigned long *__kmalloc_section_usemap(void)
259{
260 return kmalloc(usemap_size(), GFP_KERNEL);
261}
262#endif /* CONFIG_MEMORY_HOTPLUG */
263
Yasunori Goto48c90682008-07-23 21:28:15 -0700264#ifdef CONFIG_MEMORY_HOTREMOVE
265static unsigned long * __init
Yinghai Lua4322e12010-02-10 01:20:21 -0800266sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700267 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700268{
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700269 unsigned long goal, limit;
270 unsigned long *p;
271 int nid;
Yasunori Goto48c90682008-07-23 21:28:15 -0700272 /*
273 * A page may contain usemaps for other sections preventing the
274 * page being freed and making a section unremovable while
Li Zhongc800bcd2014-03-31 16:41:58 +0800275 * other sections referencing the usemap remain active. Similarly,
Yasunori Goto48c90682008-07-23 21:28:15 -0700276 * a pgdat can prevent a section being removed. If section A
277 * contains a pgdat and section B contains the usemap, both
278 * sections become inter-dependent. This allocates usemaps
279 * from the same section as the pgdat where possible to avoid
280 * this problem.
281 */
Yinghai Lu07b4e2b2012-07-11 14:02:51 -0700282 goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT);
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700283 limit = goal + (1UL << PA_SECTION_SHIFT);
284 nid = early_pfn_to_nid(goal >> PAGE_SHIFT);
285again:
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800286 p = memblock_virt_alloc_try_nid_nopanic(size,
287 SMP_CACHE_BYTES, goal, limit,
288 nid);
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700289 if (!p && limit) {
290 limit = 0;
291 goto again;
292 }
293 return p;
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) {
Joe Perches11705322016-03-17 14:19:50 -0700318 pr_info("node %d must be removed before remove section %ld\n",
319 nid, usemap_snr);
Yasunori Goto48c90682008-07-23 21:28:15 -0700320 return;
321 }
322 /*
323 * There is a circular dependency.
324 * Some platforms allow un-removable section because they will just
325 * gather other removable sections for dynamic partitioning.
326 * Just notify un-removable section's number here.
327 */
Joe Perches11705322016-03-17 14:19:50 -0700328 pr_info("Section %ld and %ld (node %d) have a circular dependency on usemap and pgdat allocations\n",
329 usemap_snr, pgdat_snr, nid);
Yasunori Goto48c90682008-07-23 21:28:15 -0700330}
331#else
332static unsigned long * __init
Yinghai Lua4322e12010-02-10 01:20:21 -0800333sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700334 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700335{
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800336 return memblock_virt_alloc_node_nopanic(size, pgdat->node_id);
Yasunori Goto48c90682008-07-23 21:28:15 -0700337}
338
339static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
340{
341}
342#endif /* CONFIG_MEMORY_HOTREMOVE */
343
Wanpeng Li18732092013-09-11 14:22:38 -0700344static void __init sparse_early_usemaps_alloc_node(void *data,
Yinghai Lua4322e12010-02-10 01:20:21 -0800345 unsigned long pnum_begin,
346 unsigned long pnum_end,
347 unsigned long usemap_count, int nodeid)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700348{
Yinghai Lua4322e12010-02-10 01:20:21 -0800349 void *usemap;
350 unsigned long pnum;
Wanpeng Li18732092013-09-11 14:22:38 -0700351 unsigned long **usemap_map = (unsigned long **)data;
Yinghai Lua4322e12010-02-10 01:20:21 -0800352 int size = usemap_size();
Mel Gorman5c0e3062007-10-16 01:25:56 -0700353
Yinghai Lua4322e12010-02-10 01:20:21 -0800354 usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid),
Johannes Weiner238305b2012-05-29 15:06:36 -0700355 size * usemap_count);
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700356 if (!usemap) {
Joe Perches11705322016-03-17 14:19:50 -0700357 pr_warn("%s: allocation failed\n", __func__);
Johannes Weiner238305b2012-05-29 15:06:36 -0700358 return;
Yasunori Goto48c90682008-07-23 21:28:15 -0700359 }
360
Nishanth Aravamudanf5bf18f2012-03-21 16:34:07 -0700361 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
362 if (!present_section_nr(pnum))
363 continue;
364 usemap_map[pnum] = usemap;
365 usemap += size;
366 check_usemap_section_nr(nodeid, usemap_map[pnum]);
Yinghai Lua4322e12010-02-10 01:20:21 -0800367 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700368}
369
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700370#ifndef CONFIG_SPARSEMEM_VMEMMAP
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700371struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700372{
373 struct page *map;
Yinghai Lue48e67e2010-05-24 14:31:57 -0700374 unsigned long size;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700375
376 map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
377 if (map)
378 return map;
379
Yinghai Lue48e67e2010-05-24 14:31:57 -0700380 size = PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800381 map = memblock_virt_alloc_try_nid(size,
382 PAGE_SIZE, __pa(MAX_DMA_ADDRESS),
383 BOOTMEM_ALLOC_ACCESSIBLE, nid);
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700384 return map;
385}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800386void __init sparse_mem_maps_populate_node(struct page **map_map,
387 unsigned long pnum_begin,
388 unsigned long pnum_end,
389 unsigned long map_count, int nodeid)
390{
391 void *map;
392 unsigned long pnum;
393 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
394
395 map = alloc_remap(nodeid, size * map_count);
396 if (map) {
397 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
398 if (!present_section_nr(pnum))
399 continue;
400 map_map[pnum] = map;
401 map += size;
402 }
403 return;
404 }
405
406 size = PAGE_ALIGN(size);
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800407 map = memblock_virt_alloc_try_nid(size * map_count,
408 PAGE_SIZE, __pa(MAX_DMA_ADDRESS),
409 BOOTMEM_ALLOC_ACCESSIBLE, nodeid);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800410 if (map) {
411 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
412 if (!present_section_nr(pnum))
413 continue;
414 map_map[pnum] = map;
415 map += size;
416 }
417 return;
418 }
419
420 /* fallback */
421 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
422 struct mem_section *ms;
423
424 if (!present_section_nr(pnum))
425 continue;
426 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
427 if (map_map[pnum])
428 continue;
429 ms = __nr_to_section(pnum);
Joe Perches11705322016-03-17 14:19:50 -0700430 pr_err("%s: sparsemem memory map backing failed some memory will not be available\n",
Joe Perches756a0252016-03-17 14:19:47 -0700431 __func__);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800432 ms->section_mem_map = 0;
433 }
434}
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700435#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
436
Yinghai Lu81d0d952010-02-27 09:29:38 -0800437#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Wanpeng Li18732092013-09-11 14:22:38 -0700438static void __init sparse_early_mem_maps_alloc_node(void *data,
Yinghai Lu9bdac912010-02-10 01:20:22 -0800439 unsigned long pnum_begin,
440 unsigned long pnum_end,
441 unsigned long map_count, int nodeid)
442{
Wanpeng Li18732092013-09-11 14:22:38 -0700443 struct page **map_map = (struct page **)data;
Yinghai Lu9bdac912010-02-10 01:20:22 -0800444 sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end,
445 map_count, nodeid);
446}
Yinghai Lu81d0d952010-02-27 09:29:38 -0800447#else
Adrian Bunk9e5c6da2008-07-25 19:46:22 -0700448static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700449{
450 struct page *map;
451 struct mem_section *ms = __nr_to_section(pnum);
452 int nid = sparse_early_nid(ms);
453
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700454 map = sparse_mem_map_populate(pnum, nid);
Andy Whitcroft29751f62005-06-23 00:08:00 -0700455 if (map)
456 return map;
457
Joe Perches11705322016-03-17 14:19:50 -0700458 pr_err("%s: sparsemem memory map backing failed some memory will not be available\n",
Joe Perches756a0252016-03-17 14:19:47 -0700459 __func__);
Bob Picco802f1922005-09-03 15:54:26 -0700460 ms->section_mem_map = 0;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700461 return NULL;
462}
Yinghai Lu9bdac912010-02-10 01:20:22 -0800463#endif
Andy Whitcroft29751f62005-06-23 00:08:00 -0700464
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -0700465void __weak __meminit vmemmap_populate_print_last(void)
Yinghai Luc2b91e22008-04-12 01:19:24 -0700466{
467}
Yinghai Lua4322e12010-02-10 01:20:21 -0800468
Wanpeng Li18732092013-09-11 14:22:38 -0700469/**
470 * alloc_usemap_and_memmap - memory alloction for pageblock flags and vmemmap
471 * @map: usemap_map for pageblock flags or mmap_map for vmemmap
Stephen Rothwell193faea2007-06-08 13:46:51 -0700472 */
Wanpeng Li18732092013-09-11 14:22:38 -0700473static void __init alloc_usemap_and_memmap(void (*alloc_func)
474 (void *, unsigned long, unsigned long,
475 unsigned long, int), void *data)
Stephen Rothwell193faea2007-06-08 13:46:51 -0700476{
477 unsigned long pnum;
Wanpeng Li18732092013-09-11 14:22:38 -0700478 unsigned long map_count;
Yinghai Lua4322e12010-02-10 01:20:21 -0800479 int nodeid_begin = 0;
480 unsigned long pnum_begin = 0;
Yinghai Lu9bdac912010-02-10 01:20:22 -0800481
482 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
483 struct mem_section *ms;
484
485 if (!present_section_nr(pnum))
486 continue;
487 ms = __nr_to_section(pnum);
488 nodeid_begin = sparse_early_nid(ms);
489 pnum_begin = pnum;
490 break;
491 }
492 map_count = 1;
493 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
494 struct mem_section *ms;
495 int nodeid;
496
497 if (!present_section_nr(pnum))
498 continue;
499 ms = __nr_to_section(pnum);
500 nodeid = sparse_early_nid(ms);
501 if (nodeid == nodeid_begin) {
502 map_count++;
503 continue;
504 }
505 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
Wanpeng Li18732092013-09-11 14:22:38 -0700506 alloc_func(data, pnum_begin, pnum,
507 map_count, nodeid_begin);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800508 /* new start, update count etc*/
509 nodeid_begin = nodeid;
510 pnum_begin = pnum;
511 map_count = 1;
512 }
513 /* ok, last chunk */
Wanpeng Li18732092013-09-11 14:22:38 -0700514 alloc_func(data, pnum_begin, NR_MEM_SECTIONS,
515 map_count, nodeid_begin);
516}
517
518/*
519 * Allocate the accumulated non-linear sections, allocate a mem_map
520 * for each and record the physical to section mapping.
521 */
522void __init sparse_init(void)
523{
524 unsigned long pnum;
525 struct page *map;
526 unsigned long *usemap;
527 unsigned long **usemap_map;
528 int size;
529#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
530 int size2;
531 struct page **map_map;
532#endif
533
534 /* see include/linux/mmzone.h 'struct mem_section' definition */
535 BUILD_BUG_ON(!is_power_of_2(sizeof(struct mem_section)));
536
537 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
538 set_pageblock_order();
539
540 /*
541 * map is using big page (aka 2M in x86 64 bit)
542 * usemap is less one page (aka 24 bytes)
543 * so alloc 2M (with 2M align) and 24 bytes in turn will
544 * make next 2M slip to one more 2M later.
545 * then in big system, the memory will have a lot of holes...
546 * here try to allocate 2M pages continuously.
547 *
548 * powerpc need to call sparse_init_one_section right after each
549 * sparse_early_mem_map_alloc, so allocate usemap_map at first.
550 */
551 size = sizeof(unsigned long *) * NR_MEM_SECTIONS;
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800552 usemap_map = memblock_virt_alloc(size, 0);
Wanpeng Li18732092013-09-11 14:22:38 -0700553 if (!usemap_map)
554 panic("can not allocate usemap_map\n");
555 alloc_usemap_and_memmap(sparse_early_usemaps_alloc_node,
556 (void *)usemap_map);
557
558#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
559 size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800560 map_map = memblock_virt_alloc(size2, 0);
Wanpeng Li18732092013-09-11 14:22:38 -0700561 if (!map_map)
562 panic("can not allocate map_map\n");
563 alloc_usemap_and_memmap(sparse_early_mem_maps_alloc_node,
564 (void *)map_map);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800565#endif
566
Yinghai Lue123dd32008-04-13 11:51:06 -0700567 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
568 if (!present_section_nr(pnum))
569 continue;
570
571 usemap = usemap_map[pnum];
572 if (!usemap)
573 continue;
Stephen Rothwell193faea2007-06-08 13:46:51 -0700574
Yinghai Lu9bdac912010-02-10 01:20:22 -0800575#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
576 map = map_map[pnum];
577#else
Stephen Rothwell193faea2007-06-08 13:46:51 -0700578 map = sparse_early_mem_map_alloc(pnum);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800579#endif
Stephen Rothwell193faea2007-06-08 13:46:51 -0700580 if (!map)
581 continue;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700582
Mel Gorman5c0e3062007-10-16 01:25:56 -0700583 sparse_init_one_section(__nr_to_section(pnum), pnum, map,
584 usemap);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700585 }
Yinghai Lue123dd32008-04-13 11:51:06 -0700586
Yinghai Luc2b91e22008-04-12 01:19:24 -0700587 vmemmap_populate_print_last();
588
Yinghai Lu9bdac912010-02-10 01:20:22 -0800589#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800590 memblock_free_early(__pa(map_map), size2);
Yinghai Lu9bdac912010-02-10 01:20:22 -0800591#endif
Santosh Shilimkarbb016b82014-01-21 15:50:34 -0800592 memblock_free_early(__pa(usemap_map), size);
Stephen Rothwell193faea2007-06-08 13:46:51 -0700593}
594
595#ifdef CONFIG_MEMORY_HOTPLUG
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700596#ifdef CONFIG_SPARSEMEM_VMEMMAP
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800597static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid)
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700598{
599 /* This will make the necessary allocations eventually. */
600 return sparse_mem_map_populate(pnum, nid);
601}
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800602static void __kfree_section_memmap(struct page *memmap)
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700603{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700604 unsigned long start = (unsigned long)memmap;
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800605 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
Johannes Weiner0aad8182013-04-29 15:07:50 -0700606
607 vmemmap_free(start, end);
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700608}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700609#ifdef CONFIG_MEMORY_HOTREMOVE
Zhang Yanfei81556b02013-11-12 15:07:43 -0800610static void free_map_bootmem(struct page *memmap)
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700611{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700612 unsigned long start = (unsigned long)memmap;
Zhang Yanfei81556b02013-11-12 15:07:43 -0800613 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
Johannes Weiner0aad8182013-04-29 15:07:50 -0700614
615 vmemmap_free(start, end);
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700616}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700617#endif /* CONFIG_MEMORY_HOTREMOVE */
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700618#else
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800619static struct page *__kmalloc_section_memmap(void)
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700620{
621 struct page *page, *ret;
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800622 unsigned long memmap_size = sizeof(struct page) * PAGES_PER_SECTION;
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700623
Yasunori Gotof2d0aa52006-10-28 10:38:32 -0700624 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700625 if (page)
626 goto got_map_page;
627
628 ret = vmalloc(memmap_size);
629 if (ret)
630 goto got_map_ptr;
631
632 return NULL;
633got_map_page:
634 ret = (struct page *)pfn_to_kaddr(page_to_pfn(page));
635got_map_ptr:
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700636
637 return ret;
638}
639
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800640static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid)
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700641{
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800642 return __kmalloc_section_memmap();
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700643}
644
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800645static void __kfree_section_memmap(struct page *memmap)
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700646{
Christoph Lameter9e2779f2008-02-04 22:28:34 -0800647 if (is_vmalloc_addr(memmap))
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700648 vfree(memmap);
649 else
650 free_pages((unsigned long)memmap,
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800651 get_order(sizeof(struct page) * PAGES_PER_SECTION));
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700652}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700653
David Rientjes4edd7ce2013-04-29 15:08:22 -0700654#ifdef CONFIG_MEMORY_HOTREMOVE
Zhang Yanfei81556b02013-11-12 15:07:43 -0800655static void free_map_bootmem(struct page *memmap)
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700656{
657 unsigned long maps_section_nr, removing_section_nr, i;
Zhang Yanfei81556b02013-11-12 15:07:43 -0800658 unsigned long magic, nr_pages;
Jianguo Wuae64ffc2012-11-29 13:54:21 -0800659 struct page *page = virt_to_page(memmap);
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700660
Zhang Yanfei81556b02013-11-12 15:07:43 -0800661 nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page))
662 >> PAGE_SHIFT;
663
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700664 for (i = 0; i < nr_pages; i++, page++) {
Yasuaki Ishimatsuddffe982017-02-22 15:45:13 -0800665 magic = (unsigned long) page->freelist;
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700666
667 BUG_ON(magic == NODE_INFO);
668
669 maps_section_nr = pfn_to_section_nr(page_to_pfn(page));
Yasuaki Ishimatsu857e5222017-02-22 15:45:10 -0800670 removing_section_nr = page_private(page);
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700671
672 /*
673 * When this function is called, the removing section is
674 * logical offlined state. This means all pages are isolated
675 * from page allocator. If removing section's memmap is placed
676 * on the same section, it must not be freed.
677 * If it is freed, page allocator may allocate it which will
678 * be removed physically soon.
679 */
680 if (maps_section_nr != removing_section_nr)
681 put_page_bootmem(page);
682 }
683}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700684#endif /* CONFIG_MEMORY_HOTREMOVE */
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700685#endif /* CONFIG_SPARSEMEM_VMEMMAP */
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700686
Andy Whitcroft29751f62005-06-23 00:08:00 -0700687/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700688 * returns the number of sections whose mem_maps were properly
689 * set. If this is <=0, then that means that the passed-in
690 * map was not consumed and must be freed.
691 */
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800692int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700693{
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700694 unsigned long section_nr = pfn_to_section_nr(start_pfn);
695 struct pglist_data *pgdat = zone->zone_pgdat;
696 struct mem_section *ms;
697 struct page *memmap;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700698 unsigned long *usemap;
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700699 unsigned long flags;
700 int ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700701
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700702 /*
703 * no locking for this, because it does its own
704 * plus, it does a kmalloc
705 */
WANG Congbbd06822007-12-17 16:19:59 -0800706 ret = sparse_index_init(section_nr, pgdat->node_id);
707 if (ret < 0 && ret != -EEXIST)
708 return ret;
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800709 memmap = kmalloc_section_memmap(section_nr, pgdat->node_id);
WANG Congbbd06822007-12-17 16:19:59 -0800710 if (!memmap)
711 return -ENOMEM;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700712 usemap = __kmalloc_section_usemap();
WANG Congbbd06822007-12-17 16:19:59 -0800713 if (!usemap) {
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800714 __kfree_section_memmap(memmap);
WANG Congbbd06822007-12-17 16:19:59 -0800715 return -ENOMEM;
716 }
Andy Whitcroft29751f62005-06-23 00:08:00 -0700717
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700718 pgdat_resize_lock(pgdat, &flags);
719
720 ms = __pfn_to_section(start_pfn);
721 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
722 ret = -EEXIST;
723 goto out;
724 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700725
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800726 memset(memmap, 0, sizeof(struct page) * PAGES_PER_SECTION);
Wen Congyang3ac19f82012-12-11 16:00:59 -0800727
Andy Whitcroft29751f62005-06-23 00:08:00 -0700728 ms->section_mem_map |= SECTION_MARKED_PRESENT;
729
Mel Gorman5c0e3062007-10-16 01:25:56 -0700730 ret = sparse_init_one_section(ms, section_nr, memmap, usemap);
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700731
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700732out:
733 pgdat_resize_unlock(pgdat, &flags);
WANG Congbbd06822007-12-17 16:19:59 -0800734 if (ret <= 0) {
735 kfree(usemap);
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800736 __kfree_section_memmap(memmap);
WANG Congbbd06822007-12-17 16:19:59 -0800737 }
Dave Hansen0b0acbe2005-10-29 18:16:55 -0700738 return ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700739}
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700740
Zhang Yanfeif3deb682013-07-08 16:00:10 -0700741#ifdef CONFIG_MEMORY_HOTREMOVE
Wen Congyang95a47742012-12-11 16:00:47 -0800742#ifdef CONFIG_MEMORY_FAILURE
743static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
744{
745 int i;
746
747 if (!memmap)
748 return;
749
Dan Williams4b94ffd2016-01-15 16:56:22 -0800750 for (i = 0; i < nr_pages; i++) {
Wen Congyang95a47742012-12-11 16:00:47 -0800751 if (PageHWPoison(&memmap[i])) {
Xishi Qiu293c07e2013-02-22 16:34:02 -0800752 atomic_long_sub(1, &num_poisoned_pages);
Wen Congyang95a47742012-12-11 16:00:47 -0800753 ClearPageHWPoison(&memmap[i]);
754 }
755 }
756}
757#else
758static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
759{
760}
761#endif
762
David Rientjes4edd7ce2013-04-29 15:08:22 -0700763static void free_section_usemap(struct page *memmap, unsigned long *usemap)
764{
765 struct page *usemap_page;
David Rientjes4edd7ce2013-04-29 15:08:22 -0700766
767 if (!usemap)
768 return;
769
770 usemap_page = virt_to_page(usemap);
771 /*
772 * Check to see if allocation came from hot-plug-add
773 */
774 if (PageSlab(usemap_page) || PageCompound(usemap_page)) {
775 kfree(usemap);
776 if (memmap)
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800777 __kfree_section_memmap(memmap);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700778 return;
779 }
780
781 /*
782 * The usemap came from bootmem. This is packed with other usemaps
783 * on the section which has pgdat at boot time. Just keep it as is now.
784 */
785
Zhang Yanfei81556b02013-11-12 15:07:43 -0800786 if (memmap)
787 free_map_bootmem(memmap);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700788}
789
Dan Williams4b94ffd2016-01-15 16:56:22 -0800790void sparse_remove_one_section(struct zone *zone, struct mem_section *ms,
791 unsigned long map_offset)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700792{
793 struct page *memmap = NULL;
Tang Chencd099682013-02-22 16:33:02 -0800794 unsigned long *usemap = NULL, flags;
795 struct pglist_data *pgdat = zone->zone_pgdat;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700796
Tang Chencd099682013-02-22 16:33:02 -0800797 pgdat_resize_lock(pgdat, &flags);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700798 if (ms->section_mem_map) {
799 usemap = ms->pageblock_flags;
800 memmap = sparse_decode_mem_map(ms->section_mem_map,
801 __section_nr(ms));
802 ms->section_mem_map = 0;
803 ms->pageblock_flags = NULL;
804 }
Tang Chencd099682013-02-22 16:33:02 -0800805 pgdat_resize_unlock(pgdat, &flags);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700806
Dan Williams4b94ffd2016-01-15 16:56:22 -0800807 clear_hwpoisoned_pages(memmap + map_offset,
808 PAGES_PER_SECTION - map_offset);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700809 free_section_usemap(memmap, usemap);
810}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700811#endif /* CONFIG_MEMORY_HOTREMOVE */
812#endif /* CONFIG_MEMORY_HOTPLUG */