blob: fe44b2d3bd7e9f0a23110a230d8b4de32ea925cd [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Andy Whitcroftd41dee32005-06-23 00:07:54 -07002/*
3 * sparse memory mappings.
4 */
Andy Whitcroftd41dee32005-06-23 00:07:54 -07005#include <linux/mm.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09006#include <linux/slab.h>
Andy Whitcroftd41dee32005-06-23 00:07:54 -07007#include <linux/mmzone.h>
Mike Rapoport97ad1082018-10-30 15:09:44 -07008#include <linux/memblock.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -07009#include <linux/compiler.h>
Dave Hansen0b0acbec2005-10-29 18:16:55 -070010#include <linux/highmem.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040011#include <linux/export.h>
Dave Hansen28ae55c2005-09-03 15:54:29 -070012#include <linux/spinlock.h>
Dave Hansen0b0acbec2005-10-29 18:16:55 -070013#include <linux/vmalloc.h>
Gideon Israel Dsouza3b321232014-04-07 15:37:26 -070014
Yasunori Goto0c0a4a52008-04-28 02:13:34 -070015#include "internal.h"
Andy Whitcroftd41dee32005-06-23 00:07:54 -070016#include <asm/dma.h>
Christoph Lameter8f6aac42007-10-16 01:24:13 -070017#include <asm/pgalloc.h>
18#include <asm/pgtable.h>
Andy Whitcroftd41dee32005-06-23 00:07:54 -070019
20/*
21 * Permanent SPARSEMEM data:
22 *
23 * 1) mem_section - memory sections, mem_map's for valid memory
24 */
Bob Picco3e347262005-09-03 15:54:28 -070025#ifdef CONFIG_SPARSEMEM_EXTREME
Kirill A. Shutemov83e3c4872017-09-29 17:08:16 +030026struct mem_section **mem_section;
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
Mike Rapoport8a7f97b2019-03-11 23:30:31 -070068 if (slab_is_available()) {
Michal Hockob95046b2017-09-06 16:20:41 -070069 section = kzalloc_node(array_size, GFP_KERNEL, nid);
Mike Rapoport8a7f97b2019-03-11 23:30:31 -070070 } else {
Mike Rapoport7e1c4e22018-10-30 15:09:57 -070071 section = memblock_alloc_node(array_size, SMP_CACHE_BYTES,
72 nid);
Mike Rapoport8a7f97b2019-03-11 23:30:31 -070073 if (!section)
74 panic("%s: Failed to allocate %lu bytes nid=%d\n",
75 __func__, array_size, nid);
76 }
Bob Picco3e347262005-09-03 15:54:28 -070077
Dave Hansen28ae55c2005-09-03 15:54:29 -070078 return section;
Bob Picco802f1922005-09-03 15:54:26 -070079}
Dave Hansen28ae55c2005-09-03 15:54:29 -070080
Yasunori Gotoa3142c82007-05-08 00:23:07 -070081static int __meminit sparse_index_init(unsigned long section_nr, int nid)
Dave Hansen28ae55c2005-09-03 15:54:29 -070082{
Dave Hansen28ae55c2005-09-03 15:54:29 -070083 unsigned long root = SECTION_NR_TO_ROOT(section_nr);
84 struct mem_section *section;
Dave Hansen28ae55c2005-09-03 15:54:29 -070085
86 if (mem_section[root])
87 return -EEXIST;
88
89 section = sparse_index_alloc(nid);
WANG Congaf0cd5a2007-12-17 16:19:58 -080090 if (!section)
91 return -ENOMEM;
Dave Hansen28ae55c2005-09-03 15:54:29 -070092
93 mem_section[root] = section;
Gavin Shanc1c95182012-07-31 16:46:06 -070094
Zhang Yanfei9d1936c2013-05-17 22:10:38 +080095 return 0;
Dave Hansen28ae55c2005-09-03 15:54:29 -070096}
97#else /* !SPARSEMEM_EXTREME */
98static inline int sparse_index_init(unsigned long section_nr, int nid)
99{
100 return 0;
101}
102#endif
103
Zhou Chengming91fd8b92016-07-28 15:48:35 -0700104#ifdef CONFIG_SPARSEMEM_EXTREME
Dave Hansen4ca644d2005-10-29 18:16:51 -0700105int __section_nr(struct mem_section* ms)
106{
107 unsigned long root_nr;
Kirill A. Shutemov83e3c4872017-09-29 17:08:16 +0300108 struct mem_section *root = NULL;
Dave Hansen4ca644d2005-10-29 18:16:51 -0700109
Mike Kravetz12783b02006-05-20 15:00:05 -0700110 for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) {
111 root = __nr_to_section(root_nr * SECTIONS_PER_ROOT);
Dave Hansen4ca644d2005-10-29 18:16:51 -0700112 if (!root)
113 continue;
114
115 if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT)))
116 break;
117 }
118
Kirill A. Shutemov83e3c4872017-09-29 17:08:16 +0300119 VM_BUG_ON(!root);
Gavin Shandb36a462012-07-31 16:46:04 -0700120
Dave Hansen4ca644d2005-10-29 18:16:51 -0700121 return (root_nr * SECTIONS_PER_ROOT) + (ms - root);
122}
Zhou Chengming91fd8b92016-07-28 15:48:35 -0700123#else
124int __section_nr(struct mem_section* ms)
125{
126 return (int)(ms - mem_section[0]);
127}
128#endif
Dave Hansen4ca644d2005-10-29 18:16:51 -0700129
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700130/*
131 * During early boot, before section_mem_map is used for an actual
132 * mem_map, we use section_mem_map to store the section's NUMA
133 * node. This keeps us from having to use another data structure. The
134 * node information is cleared just before we store the real mem_map.
135 */
136static inline unsigned long sparse_encode_early_nid(int nid)
137{
138 return (nid << SECTION_NID_SHIFT);
139}
140
141static inline int sparse_early_nid(struct mem_section *section)
142{
143 return (section->section_mem_map >> SECTION_NID_SHIFT);
144}
145
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700146/* Validate the physical addressing limitations of the model */
147void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn,
148 unsigned long *end_pfn)
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700149{
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700150 unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700151
Ingo Molnarbead9a32008-04-16 01:40:00 +0200152 /*
153 * Sanity checks - do not allow an architecture to pass
154 * in larger pfns than the maximum scope of sparsemem:
155 */
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700156 if (*start_pfn > max_sparsemem_pfn) {
157 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
158 "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
159 *start_pfn, *end_pfn, max_sparsemem_pfn);
160 WARN_ON_ONCE(1);
161 *start_pfn = max_sparsemem_pfn;
162 *end_pfn = max_sparsemem_pfn;
Cyrill Gorcunovef161a92009-03-31 15:19:25 -0700163 } else if (*end_pfn > max_sparsemem_pfn) {
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700164 mminit_dprintk(MMINIT_WARNING, "pfnvalidation",
165 "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n",
166 *start_pfn, *end_pfn, max_sparsemem_pfn);
167 WARN_ON_ONCE(1);
168 *end_pfn = max_sparsemem_pfn;
169 }
170}
171
Dave Hansenc4e1be92017-07-06 15:36:44 -0700172/*
173 * There are a number of times that we loop over NR_MEM_SECTIONS,
174 * looking for section_present() on each. But, when we have very
175 * large physical address spaces, NR_MEM_SECTIONS can also be
176 * very large which makes the loops quite long.
177 *
178 * Keeping track of this gives us an easy way to break out of
179 * those loops early.
180 */
181int __highest_present_section_nr;
182static void section_mark_present(struct mem_section *ms)
183{
184 int section_nr = __section_nr(ms);
185
186 if (section_nr > __highest_present_section_nr)
187 __highest_present_section_nr = section_nr;
188
189 ms->section_mem_map |= SECTION_MARKED_PRESENT;
190}
191
192static inline int next_present_section_nr(int section_nr)
193{
194 do {
195 section_nr++;
196 if (present_section_nr(section_nr))
197 return section_nr;
Wei Yangd538c162018-06-07 17:06:39 -0700198 } while ((section_nr <= __highest_present_section_nr));
Dave Hansenc4e1be92017-07-06 15:36:44 -0700199
200 return -1;
201}
202#define for_each_present_section_nr(start, section_nr) \
203 for (section_nr = next_present_section_nr(start-1); \
Qian Caid7780152019-03-05 15:50:11 -0800204 ((section_nr != -1) && \
Dave Hansenc4e1be92017-07-06 15:36:44 -0700205 (section_nr <= __highest_present_section_nr)); \
206 section_nr = next_present_section_nr(section_nr))
207
Pavel Tatashin85c77f72018-08-17 15:49:33 -0700208static inline unsigned long first_present_section_nr(void)
209{
210 return next_present_section_nr(-1);
211}
212
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700213/* Record a memory area against a node. */
214void __init memory_present(int nid, unsigned long start, unsigned long end)
215{
216 unsigned long pfn;
Ingo Molnarbead9a32008-04-16 01:40:00 +0200217
Kirill A. Shutemov629a3592017-11-07 11:33:37 +0300218#ifdef CONFIG_SPARSEMEM_EXTREME
219 if (unlikely(!mem_section)) {
220 unsigned long size, align;
221
Baoquan Hed09cfbb2018-01-04 16:18:06 -0800222 size = sizeof(struct mem_section*) * NR_SECTION_ROOTS;
Kirill A. Shutemov629a3592017-11-07 11:33:37 +0300223 align = 1 << (INTERNODE_CACHE_SHIFT);
Mike Rapoporteb31d552018-10-30 15:08:04 -0700224 mem_section = memblock_alloc(size, align);
Mike Rapoport8a7f97b2019-03-11 23:30:31 -0700225 if (!mem_section)
226 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
227 __func__, size, align);
Kirill A. Shutemov629a3592017-11-07 11:33:37 +0300228 }
229#endif
230
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700231 start &= PAGE_SECTION_MASK;
Mel Gorman2dbb51c2008-07-23 21:26:52 -0700232 mminit_validate_memmodel_limits(&start, &end);
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700233 for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
234 unsigned long section = pfn_to_section_nr(pfn);
Bob Picco802f1922005-09-03 15:54:26 -0700235 struct mem_section *ms;
236
237 sparse_index_init(section, nid);
Andy Whitcroft85770ff2007-08-22 14:01:03 -0700238 set_section_nid(section, nid);
Bob Picco802f1922005-09-03 15:54:26 -0700239
240 ms = __nr_to_section(section);
Dave Hansenc4e1be92017-07-06 15:36:44 -0700241 if (!ms->section_mem_map) {
Michal Hocko2d070ea2017-07-06 15:37:56 -0700242 ms->section_mem_map = sparse_encode_early_nid(nid) |
243 SECTION_IS_ONLINE;
Dave Hansenc4e1be92017-07-06 15:36:44 -0700244 section_mark_present(ms);
245 }
Andy Whitcroftd41dee32005-06-23 00:07:54 -0700246 }
247}
248
249/*
Logan Gunthorpe9def36e2018-12-14 14:16:57 -0800250 * Mark all memblocks as present using memory_present(). This is a
251 * convienence function that is useful for a number of arches
252 * to mark all of the systems memory as present during initialization.
253 */
254void __init memblocks_present(void)
255{
256 struct memblock_region *reg;
257
258 for_each_memblock(memory, reg) {
259 memory_present(memblock_get_region_node(reg),
260 memblock_region_memory_base_pfn(reg),
261 memblock_region_memory_end_pfn(reg));
262 }
263}
264
265/*
Andy Whitcroft29751f62005-06-23 00:08:00 -0700266 * Subtle, we encode the real pfn into the mem_map such that
267 * the identity pfn - section_mem_map will return the actual
268 * physical page frame number.
269 */
270static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
271{
Petr Tesarikdef9b712018-01-31 16:20:26 -0800272 unsigned long coded_mem_map =
273 (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
274 BUILD_BUG_ON(SECTION_MAP_LAST_BIT > (1UL<<PFN_SECTION_SHIFT));
275 BUG_ON(coded_mem_map & ~SECTION_MAP_MASK);
276 return coded_mem_map;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700277}
278
279/*
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700280 * Decode mem_map from the coded memmap
Andy Whitcroft29751f62005-06-23 00:08:00 -0700281 */
Andy Whitcroft29751f62005-06-23 00:08:00 -0700282struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
283{
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700284 /* mask off the extra low bits of information */
285 coded_mem_map &= SECTION_MAP_MASK;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700286 return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
287}
288
Oscar Salvador4e409872018-08-17 15:47:14 -0700289static void __meminit sparse_init_one_section(struct mem_section *ms,
Mel Gorman5c0e3062007-10-16 01:25:56 -0700290 unsigned long pnum, struct page *mem_map,
291 unsigned long *pageblock_bitmap)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700292{
Andy Whitcroft30c253e2006-06-23 02:03:41 -0700293 ms->section_mem_map &= ~SECTION_MAP_MASK;
Andy Whitcroft540557b2007-10-16 01:24:11 -0700294 ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) |
295 SECTION_HAS_MEM_MAP;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700296 ms->pageblock_flags = pageblock_bitmap;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700297}
298
Yasunori Goto04753272008-04-28 02:13:31 -0700299unsigned long usemap_size(void)
Mel Gorman5c0e3062007-10-16 01:25:56 -0700300{
Wei Yang60a7a882017-05-03 14:53:51 -0700301 return BITS_TO_LONGS(SECTION_BLOCKFLAGS_BITS) * sizeof(unsigned long);
Mel Gorman5c0e3062007-10-16 01:25:56 -0700302}
303
304#ifdef CONFIG_MEMORY_HOTPLUG
305static unsigned long *__kmalloc_section_usemap(void)
306{
307 return kmalloc(usemap_size(), GFP_KERNEL);
308}
309#endif /* CONFIG_MEMORY_HOTPLUG */
310
Yasunori Goto48c90682008-07-23 21:28:15 -0700311#ifdef CONFIG_MEMORY_HOTREMOVE
312static unsigned long * __init
Yinghai Lua4322e1b2010-02-10 01:20:21 -0800313sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700314 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700315{
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700316 unsigned long goal, limit;
317 unsigned long *p;
318 int nid;
Yasunori Goto48c90682008-07-23 21:28:15 -0700319 /*
320 * A page may contain usemaps for other sections preventing the
321 * page being freed and making a section unremovable while
Li Zhongc800bcd2014-03-31 16:41:58 +0800322 * other sections referencing the usemap remain active. Similarly,
Yasunori Goto48c90682008-07-23 21:28:15 -0700323 * a pgdat can prevent a section being removed. If section A
324 * contains a pgdat and section B contains the usemap, both
325 * sections become inter-dependent. This allocates usemaps
326 * from the same section as the pgdat where possible to avoid
327 * this problem.
328 */
Yinghai Lu07b4e2b2012-07-11 14:02:51 -0700329 goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT);
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700330 limit = goal + (1UL << PA_SECTION_SHIFT);
331 nid = early_pfn_to_nid(goal >> PAGE_SHIFT);
332again:
Mike Rapoport26fb3da2019-03-11 23:30:42 -0700333 p = memblock_alloc_try_nid(size, SMP_CACHE_BYTES, goal, limit, nid);
Yinghai Lu99ab7b12012-07-11 14:02:53 -0700334 if (!p && limit) {
335 limit = 0;
336 goto again;
337 }
338 return p;
Yasunori Goto48c90682008-07-23 21:28:15 -0700339}
340
341static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
342{
343 unsigned long usemap_snr, pgdat_snr;
Kirill A. Shutemov83e3c4872017-09-29 17:08:16 +0300344 static unsigned long old_usemap_snr;
345 static unsigned long old_pgdat_snr;
Yasunori Goto48c90682008-07-23 21:28:15 -0700346 struct pglist_data *pgdat = NODE_DATA(nid);
347 int usemap_nid;
348
Kirill A. Shutemov83e3c4872017-09-29 17:08:16 +0300349 /* First call */
350 if (!old_usemap_snr) {
351 old_usemap_snr = NR_MEM_SECTIONS;
352 old_pgdat_snr = NR_MEM_SECTIONS;
353 }
354
Yasunori Goto48c90682008-07-23 21:28:15 -0700355 usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT);
356 pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
357 if (usemap_snr == pgdat_snr)
358 return;
359
360 if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr)
361 /* skip redundant message */
362 return;
363
364 old_usemap_snr = usemap_snr;
365 old_pgdat_snr = pgdat_snr;
366
367 usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr));
368 if (usemap_nid != nid) {
Joe Perches11705322016-03-17 14:19:50 -0700369 pr_info("node %d must be removed before remove section %ld\n",
370 nid, usemap_snr);
Yasunori Goto48c90682008-07-23 21:28:15 -0700371 return;
372 }
373 /*
374 * There is a circular dependency.
375 * Some platforms allow un-removable section because they will just
376 * gather other removable sections for dynamic partitioning.
377 * Just notify un-removable section's number here.
378 */
Joe Perches11705322016-03-17 14:19:50 -0700379 pr_info("Section %ld and %ld (node %d) have a circular dependency on usemap and pgdat allocations\n",
380 usemap_snr, pgdat_snr, nid);
Yasunori Goto48c90682008-07-23 21:28:15 -0700381}
382#else
383static unsigned long * __init
Yinghai Lua4322e1b2010-02-10 01:20:21 -0800384sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
Johannes Weiner238305b2012-05-29 15:06:36 -0700385 unsigned long size)
Yasunori Goto48c90682008-07-23 21:28:15 -0700386{
Mike Rapoport26fb3da2019-03-11 23:30:42 -0700387 return memblock_alloc_node(size, SMP_CACHE_BYTES, pgdat->node_id);
Yasunori Goto48c90682008-07-23 21:28:15 -0700388}
389
390static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
391{
392}
393#endif /* CONFIG_MEMORY_HOTREMOVE */
394
Pavel Tatashin35fd1eb2018-08-17 15:49:21 -0700395#ifdef CONFIG_SPARSEMEM_VMEMMAP
Pavel Tatashinafda57b2018-08-17 15:49:30 -0700396static unsigned long __init section_map_size(void)
Pavel Tatashin35fd1eb2018-08-17 15:49:21 -0700397{
398 return ALIGN(sizeof(struct page) * PAGES_PER_SECTION, PMD_SIZE);
399}
400
401#else
Pavel Tatashinafda57b2018-08-17 15:49:30 -0700402static unsigned long __init section_map_size(void)
Pavel Tatashine131c062018-08-17 15:49:26 -0700403{
404 return PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION);
405}
406
Christoph Hellwig7b73d972017-12-29 08:53:54 +0100407struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid,
408 struct vmem_altmap *altmap)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700409{
Pavel Tatashine131c062018-08-17 15:49:26 -0700410 unsigned long size = section_map_size();
411 struct page *map = sparse_buffer_alloc(size);
Mike Rapoport8a7f97b2019-03-11 23:30:31 -0700412 phys_addr_t addr = __pa(MAX_DMA_ADDRESS);
Andy Whitcroft29751f62005-06-23 00:08:00 -0700413
Pavel Tatashine131c062018-08-17 15:49:26 -0700414 if (map)
415 return map;
416
Mike Rapoporteb31d552018-10-30 15:08:04 -0700417 map = memblock_alloc_try_nid(size,
Mike Rapoport8a7f97b2019-03-11 23:30:31 -0700418 PAGE_SIZE, addr,
Mike Rapoport97ad1082018-10-30 15:09:44 -0700419 MEMBLOCK_ALLOC_ACCESSIBLE, nid);
Mike Rapoport8a7f97b2019-03-11 23:30:31 -0700420 if (!map)
421 panic("%s: Failed to allocate %lu bytes align=0x%lx nid=%d from=%pa\n",
422 __func__, size, PAGE_SIZE, nid, &addr);
423
Christoph Lameter8f6aac42007-10-16 01:24:13 -0700424 return map;
425}
426#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
427
Pavel Tatashin35fd1eb2018-08-17 15:49:21 -0700428static void *sparsemap_buf __meminitdata;
429static void *sparsemap_buf_end __meminitdata;
430
Pavel Tatashinafda57b2018-08-17 15:49:30 -0700431static void __init sparse_buffer_init(unsigned long size, int nid)
Pavel Tatashin35fd1eb2018-08-17 15:49:21 -0700432{
Mike Rapoport8a7f97b2019-03-11 23:30:31 -0700433 phys_addr_t addr = __pa(MAX_DMA_ADDRESS);
Pavel Tatashin35fd1eb2018-08-17 15:49:21 -0700434 WARN_ON(sparsemap_buf); /* forgot to call sparse_buffer_fini()? */
435 sparsemap_buf =
Mike Rapoporteb31d552018-10-30 15:08:04 -0700436 memblock_alloc_try_nid_raw(size, PAGE_SIZE,
Mike Rapoport8a7f97b2019-03-11 23:30:31 -0700437 addr,
Mike Rapoport97ad1082018-10-30 15:09:44 -0700438 MEMBLOCK_ALLOC_ACCESSIBLE, nid);
Pavel Tatashin35fd1eb2018-08-17 15:49:21 -0700439 sparsemap_buf_end = sparsemap_buf + size;
440}
441
Pavel Tatashinafda57b2018-08-17 15:49:30 -0700442static void __init sparse_buffer_fini(void)
Pavel Tatashin35fd1eb2018-08-17 15:49:21 -0700443{
444 unsigned long size = sparsemap_buf_end - sparsemap_buf;
445
446 if (sparsemap_buf && size > 0)
447 memblock_free_early(__pa(sparsemap_buf), size);
448 sparsemap_buf = NULL;
449}
450
451void * __meminit sparse_buffer_alloc(unsigned long size)
452{
453 void *ptr = NULL;
454
455 if (sparsemap_buf) {
456 ptr = PTR_ALIGN(sparsemap_buf, size);
457 if (ptr + size > sparsemap_buf_end)
458 ptr = NULL;
459 else
460 sparsemap_buf = ptr + size;
461 }
462 return ptr;
463}
464
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 Lua4322e1b2010-02-10 01:20:21 -0800468
Pavel Tatashin85c77f72018-08-17 15:49:33 -0700469/*
470 * Initialize sparse on a specific node. The node spans [pnum_begin, pnum_end)
471 * And number of present sections in this node is map_count.
472 */
473static void __init sparse_init_nid(int nid, unsigned long pnum_begin,
474 unsigned long pnum_end,
475 unsigned long map_count)
476{
477 unsigned long pnum, usemap_longs, *usemap;
478 struct page *map;
479
480 usemap_longs = BITS_TO_LONGS(SECTION_BLOCKFLAGS_BITS);
481 usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nid),
482 usemap_size() *
483 map_count);
484 if (!usemap) {
485 pr_err("%s: node[%d] usemap allocation failed", __func__, nid);
486 goto failed;
487 }
488 sparse_buffer_init(map_count * section_map_size(), nid);
489 for_each_present_section_nr(pnum_begin, pnum) {
490 if (pnum >= pnum_end)
491 break;
492
493 map = sparse_mem_map_populate(pnum, nid, NULL);
494 if (!map) {
495 pr_err("%s: node[%d] memory map backing failed. Some memory will not be available.",
496 __func__, nid);
497 pnum_begin = pnum;
498 goto failed;
499 }
500 check_usemap_section_nr(nid, usemap);
501 sparse_init_one_section(__nr_to_section(pnum), pnum, map, usemap);
502 usemap += usemap_longs;
503 }
504 sparse_buffer_fini();
505 return;
506failed:
507 /* We failed to allocate, mark all the following pnums as not present */
508 for_each_present_section_nr(pnum_begin, pnum) {
509 struct mem_section *ms;
510
511 if (pnum >= pnum_end)
512 break;
513 ms = __nr_to_section(pnum);
514 ms->section_mem_map = 0;
515 }
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 */
Pavel Tatashin2a3cb8b2018-08-17 15:49:37 -0700522void __init sparse_init(void)
Pavel Tatashin85c77f72018-08-17 15:49:33 -0700523{
524 unsigned long pnum_begin = first_present_section_nr();
525 int nid_begin = sparse_early_nid(__nr_to_section(pnum_begin));
526 unsigned long pnum_end, map_count = 1;
527
528 /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */
529 set_pageblock_order();
530
531 for_each_present_section_nr(pnum_begin + 1, pnum_end) {
532 int nid = sparse_early_nid(__nr_to_section(pnum_end));
533
534 if (nid == nid_begin) {
535 map_count++;
536 continue;
537 }
538 /* Init node with sections in range [pnum_begin, pnum_end) */
539 sparse_init_nid(nid_begin, pnum_begin, pnum_end, map_count);
540 nid_begin = nid;
541 pnum_begin = pnum_end;
542 map_count = 1;
543 }
544 /* cover the last node */
545 sparse_init_nid(nid_begin, pnum_begin, pnum_end, map_count);
546 vmemmap_populate_print_last();
547}
548
Stephen Rothwell193faea2007-06-08 13:46:51 -0700549#ifdef CONFIG_MEMORY_HOTPLUG
Michal Hocko2d070ea2017-07-06 15:37:56 -0700550
551/* Mark all memory sections within the pfn range as online */
552void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
553{
554 unsigned long pfn;
555
556 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
Michal Hockob4ccec42017-09-08 16:13:15 -0700557 unsigned long section_nr = pfn_to_section_nr(pfn);
Michal Hocko2d070ea2017-07-06 15:37:56 -0700558 struct mem_section *ms;
559
560 /* onlining code should never touch invalid ranges */
561 if (WARN_ON(!valid_section_nr(section_nr)))
562 continue;
563
564 ms = __nr_to_section(section_nr);
565 ms->section_mem_map |= SECTION_IS_ONLINE;
566 }
567}
568
569#ifdef CONFIG_MEMORY_HOTREMOVE
Qian Cai9b7ea462019-03-28 20:43:34 -0700570/* Mark all memory sections within the pfn range as offline */
Michal Hocko2d070ea2017-07-06 15:37:56 -0700571void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn)
572{
573 unsigned long pfn;
574
575 for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
Pavel Tatashin27227c72018-05-11 16:01:50 -0700576 unsigned long section_nr = pfn_to_section_nr(pfn);
Michal Hocko2d070ea2017-07-06 15:37:56 -0700577 struct mem_section *ms;
578
579 /*
580 * TODO this needs some double checking. Offlining code makes
581 * sure to check pfn_valid but those checks might be just bogus
582 */
583 if (WARN_ON(!valid_section_nr(section_nr)))
584 continue;
585
586 ms = __nr_to_section(section_nr);
587 ms->section_mem_map &= ~SECTION_IS_ONLINE;
588 }
589}
590#endif
591
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700592#ifdef CONFIG_SPARSEMEM_VMEMMAP
Christoph Hellwig7b73d972017-12-29 08:53:54 +0100593static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
594 struct vmem_altmap *altmap)
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700595{
596 /* This will make the necessary allocations eventually. */
Christoph Hellwig7b73d972017-12-29 08:53:54 +0100597 return sparse_mem_map_populate(pnum, nid, altmap);
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700598}
Christoph Hellwig24b6d412017-12-29 08:53:56 +0100599static void __kfree_section_memmap(struct page *memmap,
600 struct vmem_altmap *altmap)
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700601{
Johannes Weiner0aad8182013-04-29 15:07:50 -0700602 unsigned long start = (unsigned long)memmap;
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800603 unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION);
Johannes Weiner0aad8182013-04-29 15:07:50 -0700604
Christoph Hellwig24b6d412017-12-29 08:53:56 +0100605 vmemmap_free(start, end, altmap);
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700606}
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
Christoph Hellwig24b6d412017-12-29 08:53:56 +0100612 vmemmap_free(start, end, NULL);
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700613}
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700614#else
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800615static struct page *__kmalloc_section_memmap(void)
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700616{
617 struct page *page, *ret;
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800618 unsigned long memmap_size = sizeof(struct page) * PAGES_PER_SECTION;
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700619
Yasunori Gotof2d0aa52006-10-28 10:38:32 -0700620 page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size));
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700621 if (page)
622 goto got_map_page;
623
624 ret = vmalloc(memmap_size);
625 if (ret)
626 goto got_map_ptr;
627
628 return NULL;
629got_map_page:
630 ret = (struct page *)pfn_to_kaddr(page_to_pfn(page));
631got_map_ptr:
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700632
633 return ret;
634}
635
Christoph Hellwig7b73d972017-12-29 08:53:54 +0100636static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid,
637 struct vmem_altmap *altmap)
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
Christoph Hellwig24b6d412017-12-29 08:53:56 +0100642static void __kfree_section_memmap(struct page *memmap,
643 struct vmem_altmap *altmap)
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700644{
Christoph Lameter9e2779f2008-02-04 22:28:34 -0800645 if (is_vmalloc_addr(memmap))
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700646 vfree(memmap);
647 else
648 free_pages((unsigned long)memmap,
Zhang Yanfei85b35fe2013-11-12 15:07:42 -0800649 get_order(sizeof(struct page) * PAGES_PER_SECTION));
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700650}
Yasunori Goto0c0a4a52008-04-28 02:13:34 -0700651
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}
Yasunori Goto98f3cfc2007-10-16 01:26:14 -0700681#endif /* CONFIG_SPARSEMEM_VMEMMAP */
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700682
Baoquan He7567cfc2019-05-13 17:19:32 -0700683/**
684 * sparse_add_one_section - add a memory section
685 * @nid: The node to add section on
686 * @start_pfn: start pfn of the memory range
687 * @altmap: device page map
688 *
689 * This is only intended for hotplug.
690 *
691 * Return:
692 * * 0 - On success.
693 * * -EEXIST - Section has been present.
694 * * -ENOMEM - Out of memory.
Andy Whitcroft29751f62005-06-23 00:08:00 -0700695 */
Wei Yang4e0d2e72018-12-28 00:37:06 -0800696int __meminit sparse_add_one_section(int nid, unsigned long start_pfn,
697 struct vmem_altmap *altmap)
Andy Whitcroft29751f62005-06-23 00:08:00 -0700698{
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700699 unsigned long section_nr = pfn_to_section_nr(start_pfn);
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700700 struct mem_section *ms;
701 struct page *memmap;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700702 unsigned long *usemap;
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700703 int ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700704
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700705 /*
706 * no locking for this, because it does its own
707 * plus, it does a kmalloc
708 */
Wei Yang4e0d2e72018-12-28 00:37:06 -0800709 ret = sparse_index_init(section_nr, nid);
WANG Congbbd06822007-12-17 16:19:59 -0800710 if (ret < 0 && ret != -EEXIST)
711 return ret;
Oscar Salvador4e409872018-08-17 15:47:14 -0700712 ret = 0;
Wei Yang4e0d2e72018-12-28 00:37:06 -0800713 memmap = kmalloc_section_memmap(section_nr, nid, altmap);
WANG Congbbd06822007-12-17 16:19:59 -0800714 if (!memmap)
715 return -ENOMEM;
Mel Gorman5c0e3062007-10-16 01:25:56 -0700716 usemap = __kmalloc_section_usemap();
WANG Congbbd06822007-12-17 16:19:59 -0800717 if (!usemap) {
Christoph Hellwig24b6d412017-12-29 08:53:56 +0100718 __kfree_section_memmap(memmap, altmap);
WANG Congbbd06822007-12-17 16:19:59 -0800719 return -ENOMEM;
720 }
Andy Whitcroft29751f62005-06-23 00:08:00 -0700721
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700722 ms = __pfn_to_section(start_pfn);
723 if (ms->section_mem_map & SECTION_MARKED_PRESENT) {
724 ret = -EEXIST;
725 goto out;
726 }
Mel Gorman5c0e3062007-10-16 01:25:56 -0700727
Pavel Tatashind0dc12e2018-04-05 16:23:00 -0700728 /*
729 * Poison uninitialized struct pages in order to catch invalid flags
730 * combinations.
731 */
Alexander Duyckf682a972018-10-26 15:07:45 -0700732 page_init_poison(memmap, sizeof(struct page) * PAGES_PER_SECTION);
Wen Congyang3ac19f82012-12-11 16:00:59 -0800733
Wei Yang26f26be2019-07-18 15:57:21 -0700734 set_section_nid(section_nr, nid);
Dave Hansenc4e1be92017-07-06 15:36:44 -0700735 section_mark_present(ms);
Oscar Salvador4e409872018-08-17 15:47:14 -0700736 sparse_init_one_section(ms, section_nr, memmap, usemap);
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700737
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700738out:
Oscar Salvador4e409872018-08-17 15:47:14 -0700739 if (ret < 0) {
WANG Congbbd06822007-12-17 16:19:59 -0800740 kfree(usemap);
Christoph Hellwig24b6d412017-12-29 08:53:56 +0100741 __kfree_section_memmap(memmap, altmap);
WANG Congbbd06822007-12-17 16:19:59 -0800742 }
Dave Hansen0b0acbec2005-10-29 18:16:55 -0700743 return ret;
Andy Whitcroft29751f62005-06-23 00:08:00 -0700744}
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700745
Wen Congyang95a47742012-12-11 16:00:47 -0800746#ifdef CONFIG_MEMORY_FAILURE
747static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
748{
749 int i;
750
751 if (!memmap)
752 return;
753
Balbir Singh5eb570a2018-12-28 00:33:24 -0800754 /*
755 * A further optimization is to have per section refcounted
756 * num_poisoned_pages. But that would need more space per memmap, so
757 * for now just do a quick global check to speed up this routine in the
758 * absence of bad pages.
759 */
760 if (atomic_long_read(&num_poisoned_pages) == 0)
761 return;
762
Dan Williams4b94ffd2016-01-15 16:56:22 -0800763 for (i = 0; i < nr_pages; i++) {
Wen Congyang95a47742012-12-11 16:00:47 -0800764 if (PageHWPoison(&memmap[i])) {
Xishi Qiu293c07e2013-02-22 16:34:02 -0800765 atomic_long_sub(1, &num_poisoned_pages);
Wen Congyang95a47742012-12-11 16:00:47 -0800766 ClearPageHWPoison(&memmap[i]);
767 }
768 }
769}
770#else
771static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages)
772{
773}
774#endif
775
Christoph Hellwig24b6d412017-12-29 08:53:56 +0100776static void free_section_usemap(struct page *memmap, unsigned long *usemap,
777 struct vmem_altmap *altmap)
David Rientjes4edd7ce2013-04-29 15:08:22 -0700778{
779 struct page *usemap_page;
David Rientjes4edd7ce2013-04-29 15:08:22 -0700780
781 if (!usemap)
782 return;
783
784 usemap_page = virt_to_page(usemap);
785 /*
786 * Check to see if allocation came from hot-plug-add
787 */
788 if (PageSlab(usemap_page) || PageCompound(usemap_page)) {
789 kfree(usemap);
790 if (memmap)
Christoph Hellwig24b6d412017-12-29 08:53:56 +0100791 __kfree_section_memmap(memmap, altmap);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700792 return;
793 }
794
795 /*
796 * The usemap came from bootmem. This is packed with other usemaps
797 * on the section which has pgdat at boot time. Just keep it as is now.
798 */
799
Zhang Yanfei81556b02013-11-12 15:07:43 -0800800 if (memmap)
801 free_map_bootmem(memmap);
David Rientjes4edd7ce2013-04-29 15:08:22 -0700802}
803
David Hildenbrandb9bf8d32019-07-18 15:57:17 -0700804void sparse_remove_one_section(struct mem_section *ms, unsigned long map_offset,
805 struct vmem_altmap *altmap)
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700806{
807 struct page *memmap = NULL;
Wei Yang83af6582018-12-28 00:37:02 -0800808 unsigned long *usemap = NULL;
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700809
810 if (ms->section_mem_map) {
811 usemap = ms->pageblock_flags;
812 memmap = sparse_decode_mem_map(ms->section_mem_map,
813 __section_nr(ms));
814 ms->section_mem_map = 0;
815 ms->pageblock_flags = NULL;
816 }
817
Dan Williams4b94ffd2016-01-15 16:56:22 -0800818 clear_hwpoisoned_pages(memmap + map_offset,
819 PAGES_PER_SECTION - map_offset);
Christoph Hellwig24b6d412017-12-29 08:53:56 +0100820 free_section_usemap(memmap, usemap, altmap);
Badari Pulavartyea01ea92008-04-28 02:12:01 -0700821}
David Rientjes4edd7ce2013-04-29 15:08:22 -0700822#endif /* CONFIG_MEMORY_HOTPLUG */