Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
d41dee36 AW |
2 | /* |
3 | * sparse memory mappings. | |
4 | */ | |
d41dee36 | 5 | #include <linux/mm.h> |
5a0e3ad6 | 6 | #include <linux/slab.h> |
d41dee36 AW |
7 | #include <linux/mmzone.h> |
8 | #include <linux/bootmem.h> | |
3b32123d | 9 | #include <linux/compiler.h> |
0b0acbec | 10 | #include <linux/highmem.h> |
b95f1b31 | 11 | #include <linux/export.h> |
28ae55c9 | 12 | #include <linux/spinlock.h> |
0b0acbec | 13 | #include <linux/vmalloc.h> |
3b32123d | 14 | |
0c0a4a51 | 15 | #include "internal.h" |
d41dee36 | 16 | #include <asm/dma.h> |
8f6aac41 CL |
17 | #include <asm/pgalloc.h> |
18 | #include <asm/pgtable.h> | |
d41dee36 AW |
19 | |
20 | /* | |
21 | * Permanent SPARSEMEM data: | |
22 | * | |
23 | * 1) mem_section - memory sections, mem_map's for valid memory | |
24 | */ | |
3e347261 | 25 | #ifdef CONFIG_SPARSEMEM_EXTREME |
83e3c487 | 26 | struct mem_section **mem_section; |
3e347261 BP |
27 | #else |
28 | struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT] | |
22fc6ecc | 29 | ____cacheline_internodealigned_in_smp; |
3e347261 BP |
30 | #endif |
31 | EXPORT_SYMBOL(mem_section); | |
32 | ||
89689ae7 CL |
33 | #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 | |
40 | static u8 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned; | |
41 | #else | |
42 | static u16 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned; | |
43 | #endif | |
44 | ||
33dd4e0e | 45 | int page_to_nid(const struct page *page) |
89689ae7 CL |
46 | { |
47 | return section_to_node_table[page_to_section(page)]; | |
48 | } | |
49 | EXPORT_SYMBOL(page_to_nid); | |
85770ffe AW |
50 | |
51 | static 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 */ | |
56 | static inline void set_section_nid(unsigned long section_nr, int nid) | |
57 | { | |
58 | } | |
89689ae7 CL |
59 | #endif |
60 | ||
3e347261 | 61 | #ifdef CONFIG_SPARSEMEM_EXTREME |
bd721ea7 | 62 | static noinline struct mem_section __ref *sparse_index_alloc(int nid) |
28ae55c9 DH |
63 | { |
64 | struct mem_section *section = NULL; | |
65 | unsigned long array_size = SECTIONS_PER_ROOT * | |
66 | sizeof(struct mem_section); | |
67 | ||
b95046b0 MH |
68 | if (slab_is_available()) |
69 | section = kzalloc_node(array_size, GFP_KERNEL, nid); | |
70 | else | |
bb016b84 | 71 | section = memblock_virt_alloc_node(array_size, nid); |
28ae55c9 DH |
72 | |
73 | return section; | |
3e347261 | 74 | } |
802f192e | 75 | |
a3142c8e | 76 | static int __meminit sparse_index_init(unsigned long section_nr, int nid) |
802f192e | 77 | { |
28ae55c9 DH |
78 | unsigned long root = SECTION_NR_TO_ROOT(section_nr); |
79 | struct mem_section *section; | |
802f192e BP |
80 | |
81 | if (mem_section[root]) | |
28ae55c9 | 82 | return -EEXIST; |
3e347261 | 83 | |
28ae55c9 | 84 | section = sparse_index_alloc(nid); |
af0cd5a7 WC |
85 | if (!section) |
86 | return -ENOMEM; | |
28ae55c9 DH |
87 | |
88 | mem_section[root] = section; | |
c1c95183 | 89 | |
9d1936cf | 90 | return 0; |
28ae55c9 DH |
91 | } |
92 | #else /* !SPARSEMEM_EXTREME */ | |
93 | static inline int sparse_index_init(unsigned long section_nr, int nid) | |
94 | { | |
95 | return 0; | |
802f192e | 96 | } |
28ae55c9 DH |
97 | #endif |
98 | ||
91fd8b95 | 99 | #ifdef CONFIG_SPARSEMEM_EXTREME |
4ca644d9 DH |
100 | int __section_nr(struct mem_section* ms) |
101 | { | |
102 | unsigned long root_nr; | |
83e3c487 | 103 | struct mem_section *root = NULL; |
4ca644d9 | 104 | |
12783b00 MK |
105 | for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) { |
106 | root = __nr_to_section(root_nr * SECTIONS_PER_ROOT); | |
4ca644d9 DH |
107 | if (!root) |
108 | continue; | |
109 | ||
110 | if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT))) | |
111 | break; | |
112 | } | |
113 | ||
83e3c487 | 114 | VM_BUG_ON(!root); |
db36a461 | 115 | |
4ca644d9 DH |
116 | return (root_nr * SECTIONS_PER_ROOT) + (ms - root); |
117 | } | |
91fd8b95 ZC |
118 | #else |
119 | int __section_nr(struct mem_section* ms) | |
120 | { | |
121 | return (int)(ms - mem_section[0]); | |
122 | } | |
123 | #endif | |
4ca644d9 | 124 | |
30c253e6 AW |
125 | /* |
126 | * During early boot, before section_mem_map is used for an actual | |
127 | * mem_map, we use section_mem_map to store the section's NUMA | |
128 | * node. This keeps us from having to use another data structure. The | |
129 | * node information is cleared just before we store the real mem_map. | |
130 | */ | |
131 | static inline unsigned long sparse_encode_early_nid(int nid) | |
132 | { | |
133 | return (nid << SECTION_NID_SHIFT); | |
134 | } | |
135 | ||
136 | static inline int sparse_early_nid(struct mem_section *section) | |
137 | { | |
138 | return (section->section_mem_map >> SECTION_NID_SHIFT); | |
139 | } | |
140 | ||
2dbb51c4 MG |
141 | /* Validate the physical addressing limitations of the model */ |
142 | void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn, | |
143 | unsigned long *end_pfn) | |
d41dee36 | 144 | { |
2dbb51c4 | 145 | unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT); |
d41dee36 | 146 | |
bead9a3a IM |
147 | /* |
148 | * Sanity checks - do not allow an architecture to pass | |
149 | * in larger pfns than the maximum scope of sparsemem: | |
150 | */ | |
2dbb51c4 MG |
151 | if (*start_pfn > max_sparsemem_pfn) { |
152 | mminit_dprintk(MMINIT_WARNING, "pfnvalidation", | |
153 | "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n", | |
154 | *start_pfn, *end_pfn, max_sparsemem_pfn); | |
155 | WARN_ON_ONCE(1); | |
156 | *start_pfn = max_sparsemem_pfn; | |
157 | *end_pfn = max_sparsemem_pfn; | |
ef161a98 | 158 | } else if (*end_pfn > max_sparsemem_pfn) { |
2dbb51c4 MG |
159 | mminit_dprintk(MMINIT_WARNING, "pfnvalidation", |
160 | "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n", | |
161 | *start_pfn, *end_pfn, max_sparsemem_pfn); | |
162 | WARN_ON_ONCE(1); | |
163 | *end_pfn = max_sparsemem_pfn; | |
164 | } | |
165 | } | |
166 | ||
c4e1be9e DH |
167 | /* |
168 | * There are a number of times that we loop over NR_MEM_SECTIONS, | |
169 | * looking for section_present() on each. But, when we have very | |
170 | * large physical address spaces, NR_MEM_SECTIONS can also be | |
171 | * very large which makes the loops quite long. | |
172 | * | |
173 | * Keeping track of this gives us an easy way to break out of | |
174 | * those loops early. | |
175 | */ | |
176 | int __highest_present_section_nr; | |
177 | static void section_mark_present(struct mem_section *ms) | |
178 | { | |
179 | int section_nr = __section_nr(ms); | |
180 | ||
181 | if (section_nr > __highest_present_section_nr) | |
182 | __highest_present_section_nr = section_nr; | |
183 | ||
184 | ms->section_mem_map |= SECTION_MARKED_PRESENT; | |
185 | } | |
186 | ||
187 | static inline int next_present_section_nr(int section_nr) | |
188 | { | |
189 | do { | |
190 | section_nr++; | |
191 | if (present_section_nr(section_nr)) | |
192 | return section_nr; | |
193 | } while ((section_nr < NR_MEM_SECTIONS) && | |
194 | (section_nr <= __highest_present_section_nr)); | |
195 | ||
196 | return -1; | |
197 | } | |
198 | #define for_each_present_section_nr(start, section_nr) \ | |
199 | for (section_nr = next_present_section_nr(start-1); \ | |
200 | ((section_nr >= 0) && \ | |
201 | (section_nr < NR_MEM_SECTIONS) && \ | |
202 | (section_nr <= __highest_present_section_nr)); \ | |
203 | section_nr = next_present_section_nr(section_nr)) | |
204 | ||
2dbb51c4 MG |
205 | /* Record a memory area against a node. */ |
206 | void __init memory_present(int nid, unsigned long start, unsigned long end) | |
207 | { | |
208 | unsigned long pfn; | |
bead9a3a | 209 | |
629a359b KS |
210 | #ifdef CONFIG_SPARSEMEM_EXTREME |
211 | if (unlikely(!mem_section)) { | |
212 | unsigned long size, align; | |
213 | ||
d09cfbbf | 214 | size = sizeof(struct mem_section*) * NR_SECTION_ROOTS; |
629a359b KS |
215 | align = 1 << (INTERNODE_CACHE_SHIFT); |
216 | mem_section = memblock_virt_alloc(size, align); | |
217 | } | |
218 | #endif | |
219 | ||
d41dee36 | 220 | start &= PAGE_SECTION_MASK; |
2dbb51c4 | 221 | mminit_validate_memmodel_limits(&start, &end); |
d41dee36 AW |
222 | for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) { |
223 | unsigned long section = pfn_to_section_nr(pfn); | |
802f192e BP |
224 | struct mem_section *ms; |
225 | ||
226 | sparse_index_init(section, nid); | |
85770ffe | 227 | set_section_nid(section, nid); |
802f192e BP |
228 | |
229 | ms = __nr_to_section(section); | |
c4e1be9e | 230 | if (!ms->section_mem_map) { |
2d070eab MH |
231 | ms->section_mem_map = sparse_encode_early_nid(nid) | |
232 | SECTION_IS_ONLINE; | |
c4e1be9e DH |
233 | section_mark_present(ms); |
234 | } | |
d41dee36 AW |
235 | } |
236 | } | |
237 | ||
238 | /* | |
239 | * Only used by the i386 NUMA architecures, but relatively | |
240 | * generic code. | |
241 | */ | |
242 | unsigned long __init node_memmap_size_bytes(int nid, unsigned long start_pfn, | |
243 | unsigned long end_pfn) | |
244 | { | |
245 | unsigned long pfn; | |
246 | unsigned long nr_pages = 0; | |
247 | ||
2dbb51c4 | 248 | mminit_validate_memmodel_limits(&start_pfn, &end_pfn); |
d41dee36 AW |
249 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
250 | if (nid != early_pfn_to_nid(pfn)) | |
251 | continue; | |
252 | ||
540557b9 | 253 | if (pfn_present(pfn)) |
d41dee36 AW |
254 | nr_pages += PAGES_PER_SECTION; |
255 | } | |
256 | ||
257 | return nr_pages * sizeof(struct page); | |
258 | } | |
259 | ||
29751f69 AW |
260 | /* |
261 | * Subtle, we encode the real pfn into the mem_map such that | |
262 | * the identity pfn - section_mem_map will return the actual | |
263 | * physical page frame number. | |
264 | */ | |
265 | static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum) | |
266 | { | |
def9b71e PT |
267 | unsigned long coded_mem_map = |
268 | (unsigned long)(mem_map - (section_nr_to_pfn(pnum))); | |
269 | BUILD_BUG_ON(SECTION_MAP_LAST_BIT > (1UL<<PFN_SECTION_SHIFT)); | |
270 | BUG_ON(coded_mem_map & ~SECTION_MAP_MASK); | |
271 | return coded_mem_map; | |
29751f69 AW |
272 | } |
273 | ||
274 | /* | |
ea01ea93 | 275 | * Decode mem_map from the coded memmap |
29751f69 | 276 | */ |
29751f69 AW |
277 | struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum) |
278 | { | |
ea01ea93 BP |
279 | /* mask off the extra low bits of information */ |
280 | coded_mem_map &= SECTION_MAP_MASK; | |
29751f69 AW |
281 | return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum); |
282 | } | |
283 | ||
a3142c8e | 284 | static int __meminit sparse_init_one_section(struct mem_section *ms, |
5c0e3066 MG |
285 | unsigned long pnum, struct page *mem_map, |
286 | unsigned long *pageblock_bitmap) | |
29751f69 | 287 | { |
540557b9 | 288 | if (!present_section(ms)) |
29751f69 AW |
289 | return -EINVAL; |
290 | ||
30c253e6 | 291 | ms->section_mem_map &= ~SECTION_MAP_MASK; |
540557b9 AW |
292 | ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) | |
293 | SECTION_HAS_MEM_MAP; | |
5c0e3066 | 294 | ms->pageblock_flags = pageblock_bitmap; |
29751f69 AW |
295 | |
296 | return 1; | |
297 | } | |
298 | ||
04753278 | 299 | unsigned long usemap_size(void) |
5c0e3066 | 300 | { |
60a7a88d | 301 | return BITS_TO_LONGS(SECTION_BLOCKFLAGS_BITS) * sizeof(unsigned long); |
5c0e3066 MG |
302 | } |
303 | ||
304 | #ifdef CONFIG_MEMORY_HOTPLUG | |
305 | static unsigned long *__kmalloc_section_usemap(void) | |
306 | { | |
307 | return kmalloc(usemap_size(), GFP_KERNEL); | |
308 | } | |
309 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
310 | ||
48c90682 YG |
311 | #ifdef CONFIG_MEMORY_HOTREMOVE |
312 | static unsigned long * __init | |
a4322e1b | 313 | sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat, |
238305bb | 314 | unsigned long size) |
48c90682 | 315 | { |
99ab7b19 YL |
316 | unsigned long goal, limit; |
317 | unsigned long *p; | |
318 | int nid; | |
48c90682 YG |
319 | /* |
320 | * A page may contain usemaps for other sections preventing the | |
321 | * page being freed and making a section unremovable while | |
c800bcd5 | 322 | * other sections referencing the usemap remain active. Similarly, |
48c90682 YG |
323 | * 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 | */ | |
07b4e2bc | 329 | goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT); |
99ab7b19 YL |
330 | limit = goal + (1UL << PA_SECTION_SHIFT); |
331 | nid = early_pfn_to_nid(goal >> PAGE_SHIFT); | |
332 | again: | |
bb016b84 SS |
333 | p = memblock_virt_alloc_try_nid_nopanic(size, |
334 | SMP_CACHE_BYTES, goal, limit, | |
335 | nid); | |
99ab7b19 YL |
336 | if (!p && limit) { |
337 | limit = 0; | |
338 | goto again; | |
339 | } | |
340 | return p; | |
48c90682 YG |
341 | } |
342 | ||
343 | static void __init check_usemap_section_nr(int nid, unsigned long *usemap) | |
344 | { | |
345 | unsigned long usemap_snr, pgdat_snr; | |
83e3c487 KS |
346 | static unsigned long old_usemap_snr; |
347 | static unsigned long old_pgdat_snr; | |
48c90682 YG |
348 | struct pglist_data *pgdat = NODE_DATA(nid); |
349 | int usemap_nid; | |
350 | ||
83e3c487 KS |
351 | /* First call */ |
352 | if (!old_usemap_snr) { | |
353 | old_usemap_snr = NR_MEM_SECTIONS; | |
354 | old_pgdat_snr = NR_MEM_SECTIONS; | |
355 | } | |
356 | ||
48c90682 YG |
357 | usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT); |
358 | pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT); | |
359 | if (usemap_snr == pgdat_snr) | |
360 | return; | |
361 | ||
362 | if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr) | |
363 | /* skip redundant message */ | |
364 | return; | |
365 | ||
366 | old_usemap_snr = usemap_snr; | |
367 | old_pgdat_snr = pgdat_snr; | |
368 | ||
369 | usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr)); | |
370 | if (usemap_nid != nid) { | |
1170532b JP |
371 | pr_info("node %d must be removed before remove section %ld\n", |
372 | nid, usemap_snr); | |
48c90682 YG |
373 | return; |
374 | } | |
375 | /* | |
376 | * There is a circular dependency. | |
377 | * Some platforms allow un-removable section because they will just | |
378 | * gather other removable sections for dynamic partitioning. | |
379 | * Just notify un-removable section's number here. | |
380 | */ | |
1170532b JP |
381 | pr_info("Section %ld and %ld (node %d) have a circular dependency on usemap and pgdat allocations\n", |
382 | usemap_snr, pgdat_snr, nid); | |
48c90682 YG |
383 | } |
384 | #else | |
385 | static unsigned long * __init | |
a4322e1b | 386 | sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat, |
238305bb | 387 | unsigned long size) |
48c90682 | 388 | { |
bb016b84 | 389 | return memblock_virt_alloc_node_nopanic(size, pgdat->node_id); |
48c90682 YG |
390 | } |
391 | ||
392 | static void __init check_usemap_section_nr(int nid, unsigned long *usemap) | |
393 | { | |
394 | } | |
395 | #endif /* CONFIG_MEMORY_HOTREMOVE */ | |
396 | ||
18732093 | 397 | static void __init sparse_early_usemaps_alloc_node(void *data, |
a4322e1b YL |
398 | unsigned long pnum_begin, |
399 | unsigned long pnum_end, | |
400 | unsigned long usemap_count, int nodeid) | |
5c0e3066 | 401 | { |
a4322e1b YL |
402 | void *usemap; |
403 | unsigned long pnum; | |
18732093 | 404 | unsigned long **usemap_map = (unsigned long **)data; |
a4322e1b | 405 | int size = usemap_size(); |
5c0e3066 | 406 | |
a4322e1b | 407 | usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid), |
238305bb | 408 | size * usemap_count); |
f5bf18fa | 409 | if (!usemap) { |
1170532b | 410 | pr_warn("%s: allocation failed\n", __func__); |
238305bb | 411 | return; |
48c90682 YG |
412 | } |
413 | ||
f5bf18fa NA |
414 | for (pnum = pnum_begin; pnum < pnum_end; pnum++) { |
415 | if (!present_section_nr(pnum)) | |
416 | continue; | |
417 | usemap_map[pnum] = usemap; | |
418 | usemap += size; | |
419 | check_usemap_section_nr(nodeid, usemap_map[pnum]); | |
a4322e1b | 420 | } |
5c0e3066 MG |
421 | } |
422 | ||
8f6aac41 | 423 | #ifndef CONFIG_SPARSEMEM_VMEMMAP |
7b73d978 CH |
424 | struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid, |
425 | struct vmem_altmap *altmap) | |
29751f69 AW |
426 | { |
427 | struct page *map; | |
e48e67e0 | 428 | unsigned long size; |
29751f69 AW |
429 | |
430 | map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION); | |
431 | if (map) | |
432 | return map; | |
433 | ||
e48e67e0 | 434 | size = PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION); |
bb016b84 SS |
435 | map = memblock_virt_alloc_try_nid(size, |
436 | PAGE_SIZE, __pa(MAX_DMA_ADDRESS), | |
437 | BOOTMEM_ALLOC_ACCESSIBLE, nid); | |
8f6aac41 CL |
438 | return map; |
439 | } | |
9bdac914 YL |
440 | void __init sparse_mem_maps_populate_node(struct page **map_map, |
441 | unsigned long pnum_begin, | |
442 | unsigned long pnum_end, | |
443 | unsigned long map_count, int nodeid) | |
444 | { | |
445 | void *map; | |
446 | unsigned long pnum; | |
447 | unsigned long size = sizeof(struct page) * PAGES_PER_SECTION; | |
448 | ||
449 | map = alloc_remap(nodeid, size * map_count); | |
450 | if (map) { | |
451 | for (pnum = pnum_begin; pnum < pnum_end; pnum++) { | |
452 | if (!present_section_nr(pnum)) | |
453 | continue; | |
454 | map_map[pnum] = map; | |
455 | map += size; | |
456 | } | |
457 | return; | |
458 | } | |
459 | ||
460 | size = PAGE_ALIGN(size); | |
f7f99100 PT |
461 | map = memblock_virt_alloc_try_nid_raw(size * map_count, |
462 | PAGE_SIZE, __pa(MAX_DMA_ADDRESS), | |
463 | BOOTMEM_ALLOC_ACCESSIBLE, nodeid); | |
9bdac914 YL |
464 | if (map) { |
465 | for (pnum = pnum_begin; pnum < pnum_end; pnum++) { | |
466 | if (!present_section_nr(pnum)) | |
467 | continue; | |
468 | map_map[pnum] = map; | |
469 | map += size; | |
470 | } | |
471 | return; | |
472 | } | |
473 | ||
474 | /* fallback */ | |
475 | for (pnum = pnum_begin; pnum < pnum_end; pnum++) { | |
476 | struct mem_section *ms; | |
477 | ||
478 | if (!present_section_nr(pnum)) | |
479 | continue; | |
7b73d978 | 480 | map_map[pnum] = sparse_mem_map_populate(pnum, nodeid, NULL); |
9bdac914 YL |
481 | if (map_map[pnum]) |
482 | continue; | |
483 | ms = __nr_to_section(pnum); | |
1170532b | 484 | pr_err("%s: sparsemem memory map backing failed some memory will not be available\n", |
756a025f | 485 | __func__); |
9bdac914 YL |
486 | ms->section_mem_map = 0; |
487 | } | |
488 | } | |
8f6aac41 CL |
489 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ |
490 | ||
81d0d950 | 491 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
18732093 | 492 | static void __init sparse_early_mem_maps_alloc_node(void *data, |
9bdac914 YL |
493 | unsigned long pnum_begin, |
494 | unsigned long pnum_end, | |
495 | unsigned long map_count, int nodeid) | |
496 | { | |
18732093 | 497 | struct page **map_map = (struct page **)data; |
9bdac914 YL |
498 | sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end, |
499 | map_count, nodeid); | |
500 | } | |
81d0d950 | 501 | #else |
9e5c6da7 | 502 | static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum) |
8f6aac41 CL |
503 | { |
504 | struct page *map; | |
505 | struct mem_section *ms = __nr_to_section(pnum); | |
506 | int nid = sparse_early_nid(ms); | |
507 | ||
7b73d978 | 508 | map = sparse_mem_map_populate(pnum, nid, NULL); |
29751f69 AW |
509 | if (map) |
510 | return map; | |
511 | ||
1170532b | 512 | pr_err("%s: sparsemem memory map backing failed some memory will not be available\n", |
756a025f | 513 | __func__); |
802f192e | 514 | ms->section_mem_map = 0; |
29751f69 AW |
515 | return NULL; |
516 | } | |
9bdac914 | 517 | #endif |
29751f69 | 518 | |
3b32123d | 519 | void __weak __meminit vmemmap_populate_print_last(void) |
c2b91e2e YL |
520 | { |
521 | } | |
a4322e1b | 522 | |
18732093 WL |
523 | /** |
524 | * alloc_usemap_and_memmap - memory alloction for pageblock flags and vmemmap | |
525 | * @map: usemap_map for pageblock flags or mmap_map for vmemmap | |
526 | */ | |
527 | static void __init alloc_usemap_and_memmap(void (*alloc_func) | |
528 | (void *, unsigned long, unsigned long, | |
529 | unsigned long, int), void *data) | |
530 | { | |
531 | unsigned long pnum; | |
532 | unsigned long map_count; | |
533 | int nodeid_begin = 0; | |
534 | unsigned long pnum_begin = 0; | |
535 | ||
c4e1be9e | 536 | for_each_present_section_nr(0, pnum) { |
18732093 WL |
537 | struct mem_section *ms; |
538 | ||
18732093 WL |
539 | ms = __nr_to_section(pnum); |
540 | nodeid_begin = sparse_early_nid(ms); | |
541 | pnum_begin = pnum; | |
542 | break; | |
543 | } | |
544 | map_count = 1; | |
c4e1be9e | 545 | for_each_present_section_nr(pnum_begin + 1, pnum) { |
18732093 WL |
546 | struct mem_section *ms; |
547 | int nodeid; | |
548 | ||
18732093 WL |
549 | ms = __nr_to_section(pnum); |
550 | nodeid = sparse_early_nid(ms); | |
551 | if (nodeid == nodeid_begin) { | |
552 | map_count++; | |
553 | continue; | |
554 | } | |
555 | /* ok, we need to take cake of from pnum_begin to pnum - 1*/ | |
556 | alloc_func(data, pnum_begin, pnum, | |
557 | map_count, nodeid_begin); | |
558 | /* new start, update count etc*/ | |
559 | nodeid_begin = nodeid; | |
560 | pnum_begin = pnum; | |
561 | map_count = 1; | |
562 | } | |
563 | /* ok, last chunk */ | |
564 | alloc_func(data, pnum_begin, NR_MEM_SECTIONS, | |
565 | map_count, nodeid_begin); | |
566 | } | |
567 | ||
193faea9 SR |
568 | /* |
569 | * Allocate the accumulated non-linear sections, allocate a mem_map | |
570 | * for each and record the physical to section mapping. | |
571 | */ | |
572 | void __init sparse_init(void) | |
573 | { | |
574 | unsigned long pnum; | |
575 | struct page *map; | |
5c0e3066 | 576 | unsigned long *usemap; |
e123dd3f | 577 | unsigned long **usemap_map; |
81d0d950 | 578 | int size; |
81d0d950 | 579 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
81d0d950 YL |
580 | int size2; |
581 | struct page **map_map; | |
582 | #endif | |
e123dd3f | 583 | |
55878e88 CS |
584 | /* see include/linux/mmzone.h 'struct mem_section' definition */ |
585 | BUILD_BUG_ON(!is_power_of_2(sizeof(struct mem_section))); | |
586 | ||
ca57df79 XQ |
587 | /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */ |
588 | set_pageblock_order(); | |
589 | ||
e123dd3f YL |
590 | /* |
591 | * map is using big page (aka 2M in x86 64 bit) | |
592 | * usemap is less one page (aka 24 bytes) | |
593 | * so alloc 2M (with 2M align) and 24 bytes in turn will | |
594 | * make next 2M slip to one more 2M later. | |
595 | * then in big system, the memory will have a lot of holes... | |
25985edc | 596 | * here try to allocate 2M pages continuously. |
e123dd3f YL |
597 | * |
598 | * powerpc need to call sparse_init_one_section right after each | |
599 | * sparse_early_mem_map_alloc, so allocate usemap_map at first. | |
600 | */ | |
601 | size = sizeof(unsigned long *) * NR_MEM_SECTIONS; | |
bb016b84 | 602 | usemap_map = memblock_virt_alloc(size, 0); |
e123dd3f YL |
603 | if (!usemap_map) |
604 | panic("can not allocate usemap_map\n"); | |
18732093 WL |
605 | alloc_usemap_and_memmap(sparse_early_usemaps_alloc_node, |
606 | (void *)usemap_map); | |
193faea9 | 607 | |
9bdac914 YL |
608 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
609 | size2 = sizeof(struct page *) * NR_MEM_SECTIONS; | |
bb016b84 | 610 | map_map = memblock_virt_alloc(size2, 0); |
9bdac914 YL |
611 | if (!map_map) |
612 | panic("can not allocate map_map\n"); | |
18732093 WL |
613 | alloc_usemap_and_memmap(sparse_early_mem_maps_alloc_node, |
614 | (void *)map_map); | |
9bdac914 YL |
615 | #endif |
616 | ||
c4e1be9e | 617 | for_each_present_section_nr(0, pnum) { |
e123dd3f | 618 | usemap = usemap_map[pnum]; |
5c0e3066 MG |
619 | if (!usemap) |
620 | continue; | |
621 | ||
9bdac914 YL |
622 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
623 | map = map_map[pnum]; | |
624 | #else | |
e123dd3f | 625 | map = sparse_early_mem_map_alloc(pnum); |
9bdac914 | 626 | #endif |
e123dd3f YL |
627 | if (!map) |
628 | continue; | |
629 | ||
5c0e3066 MG |
630 | sparse_init_one_section(__nr_to_section(pnum), pnum, map, |
631 | usemap); | |
193faea9 | 632 | } |
e123dd3f | 633 | |
c2b91e2e YL |
634 | vmemmap_populate_print_last(); |
635 | ||
9bdac914 | 636 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
bb016b84 | 637 | memblock_free_early(__pa(map_map), size2); |
9bdac914 | 638 | #endif |
bb016b84 | 639 | memblock_free_early(__pa(usemap_map), size); |
193faea9 SR |
640 | } |
641 | ||
642 | #ifdef CONFIG_MEMORY_HOTPLUG | |
2d070eab MH |
643 | |
644 | /* Mark all memory sections within the pfn range as online */ | |
645 | void online_mem_sections(unsigned long start_pfn, unsigned long end_pfn) | |
646 | { | |
647 | unsigned long pfn; | |
648 | ||
649 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
b4ccec41 | 650 | unsigned long section_nr = pfn_to_section_nr(pfn); |
2d070eab MH |
651 | struct mem_section *ms; |
652 | ||
653 | /* onlining code should never touch invalid ranges */ | |
654 | if (WARN_ON(!valid_section_nr(section_nr))) | |
655 | continue; | |
656 | ||
657 | ms = __nr_to_section(section_nr); | |
658 | ms->section_mem_map |= SECTION_IS_ONLINE; | |
659 | } | |
660 | } | |
661 | ||
662 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
663 | /* Mark all memory sections within the pfn range as online */ | |
664 | void offline_mem_sections(unsigned long start_pfn, unsigned long end_pfn) | |
665 | { | |
666 | unsigned long pfn; | |
667 | ||
668 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { | |
669 | unsigned long section_nr = pfn_to_section_nr(start_pfn); | |
670 | struct mem_section *ms; | |
671 | ||
672 | /* | |
673 | * TODO this needs some double checking. Offlining code makes | |
674 | * sure to check pfn_valid but those checks might be just bogus | |
675 | */ | |
676 | if (WARN_ON(!valid_section_nr(section_nr))) | |
677 | continue; | |
678 | ||
679 | ms = __nr_to_section(section_nr); | |
680 | ms->section_mem_map &= ~SECTION_IS_ONLINE; | |
681 | } | |
682 | } | |
683 | #endif | |
684 | ||
98f3cfc1 | 685 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
7b73d978 CH |
686 | static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid, |
687 | struct vmem_altmap *altmap) | |
98f3cfc1 YG |
688 | { |
689 | /* This will make the necessary allocations eventually. */ | |
7b73d978 | 690 | return sparse_mem_map_populate(pnum, nid, altmap); |
98f3cfc1 | 691 | } |
24b6d416 CH |
692 | static void __kfree_section_memmap(struct page *memmap, |
693 | struct vmem_altmap *altmap) | |
98f3cfc1 | 694 | { |
0aad818b | 695 | unsigned long start = (unsigned long)memmap; |
85b35fea | 696 | unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION); |
0aad818b | 697 | |
24b6d416 | 698 | vmemmap_free(start, end, altmap); |
98f3cfc1 | 699 | } |
4edd7cef | 700 | #ifdef CONFIG_MEMORY_HOTREMOVE |
81556b02 | 701 | static void free_map_bootmem(struct page *memmap) |
0c0a4a51 | 702 | { |
0aad818b | 703 | unsigned long start = (unsigned long)memmap; |
81556b02 | 704 | unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION); |
0aad818b | 705 | |
24b6d416 | 706 | vmemmap_free(start, end, NULL); |
0c0a4a51 | 707 | } |
4edd7cef | 708 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
98f3cfc1 | 709 | #else |
85b35fea | 710 | static struct page *__kmalloc_section_memmap(void) |
0b0acbec DH |
711 | { |
712 | struct page *page, *ret; | |
85b35fea | 713 | unsigned long memmap_size = sizeof(struct page) * PAGES_PER_SECTION; |
0b0acbec | 714 | |
f2d0aa5b | 715 | page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size)); |
0b0acbec DH |
716 | if (page) |
717 | goto got_map_page; | |
718 | ||
719 | ret = vmalloc(memmap_size); | |
720 | if (ret) | |
721 | goto got_map_ptr; | |
722 | ||
723 | return NULL; | |
724 | got_map_page: | |
725 | ret = (struct page *)pfn_to_kaddr(page_to_pfn(page)); | |
726 | got_map_ptr: | |
0b0acbec DH |
727 | |
728 | return ret; | |
729 | } | |
730 | ||
7b73d978 CH |
731 | static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid, |
732 | struct vmem_altmap *altmap) | |
98f3cfc1 | 733 | { |
85b35fea | 734 | return __kmalloc_section_memmap(); |
98f3cfc1 YG |
735 | } |
736 | ||
24b6d416 CH |
737 | static void __kfree_section_memmap(struct page *memmap, |
738 | struct vmem_altmap *altmap) | |
0b0acbec | 739 | { |
9e2779fa | 740 | if (is_vmalloc_addr(memmap)) |
0b0acbec DH |
741 | vfree(memmap); |
742 | else | |
743 | free_pages((unsigned long)memmap, | |
85b35fea | 744 | get_order(sizeof(struct page) * PAGES_PER_SECTION)); |
0b0acbec | 745 | } |
0c0a4a51 | 746 | |
4edd7cef | 747 | #ifdef CONFIG_MEMORY_HOTREMOVE |
81556b02 | 748 | static void free_map_bootmem(struct page *memmap) |
0c0a4a51 YG |
749 | { |
750 | unsigned long maps_section_nr, removing_section_nr, i; | |
81556b02 | 751 | unsigned long magic, nr_pages; |
ae64ffca | 752 | struct page *page = virt_to_page(memmap); |
0c0a4a51 | 753 | |
81556b02 ZY |
754 | nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page)) |
755 | >> PAGE_SHIFT; | |
756 | ||
0c0a4a51 | 757 | for (i = 0; i < nr_pages; i++, page++) { |
ddffe98d | 758 | magic = (unsigned long) page->freelist; |
0c0a4a51 YG |
759 | |
760 | BUG_ON(magic == NODE_INFO); | |
761 | ||
762 | maps_section_nr = pfn_to_section_nr(page_to_pfn(page)); | |
857e522a | 763 | removing_section_nr = page_private(page); |
0c0a4a51 YG |
764 | |
765 | /* | |
766 | * When this function is called, the removing section is | |
767 | * logical offlined state. This means all pages are isolated | |
768 | * from page allocator. If removing section's memmap is placed | |
769 | * on the same section, it must not be freed. | |
770 | * If it is freed, page allocator may allocate it which will | |
771 | * be removed physically soon. | |
772 | */ | |
773 | if (maps_section_nr != removing_section_nr) | |
774 | put_page_bootmem(page); | |
775 | } | |
776 | } | |
4edd7cef | 777 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
98f3cfc1 | 778 | #endif /* CONFIG_SPARSEMEM_VMEMMAP */ |
0b0acbec | 779 | |
29751f69 AW |
780 | /* |
781 | * returns the number of sections whose mem_maps were properly | |
782 | * set. If this is <=0, then that means that the passed-in | |
783 | * map was not consumed and must be freed. | |
784 | */ | |
7b73d978 CH |
785 | int __meminit sparse_add_one_section(struct pglist_data *pgdat, |
786 | unsigned long start_pfn, struct vmem_altmap *altmap) | |
29751f69 | 787 | { |
0b0acbec | 788 | unsigned long section_nr = pfn_to_section_nr(start_pfn); |
0b0acbec DH |
789 | struct mem_section *ms; |
790 | struct page *memmap; | |
5c0e3066 | 791 | unsigned long *usemap; |
0b0acbec DH |
792 | unsigned long flags; |
793 | int ret; | |
29751f69 | 794 | |
0b0acbec DH |
795 | /* |
796 | * no locking for this, because it does its own | |
797 | * plus, it does a kmalloc | |
798 | */ | |
bbd06825 WC |
799 | ret = sparse_index_init(section_nr, pgdat->node_id); |
800 | if (ret < 0 && ret != -EEXIST) | |
801 | return ret; | |
7b73d978 | 802 | memmap = kmalloc_section_memmap(section_nr, pgdat->node_id, altmap); |
bbd06825 WC |
803 | if (!memmap) |
804 | return -ENOMEM; | |
5c0e3066 | 805 | usemap = __kmalloc_section_usemap(); |
bbd06825 | 806 | if (!usemap) { |
24b6d416 | 807 | __kfree_section_memmap(memmap, altmap); |
bbd06825 WC |
808 | return -ENOMEM; |
809 | } | |
0b0acbec DH |
810 | |
811 | pgdat_resize_lock(pgdat, &flags); | |
29751f69 | 812 | |
0b0acbec DH |
813 | ms = __pfn_to_section(start_pfn); |
814 | if (ms->section_mem_map & SECTION_MARKED_PRESENT) { | |
815 | ret = -EEXIST; | |
816 | goto out; | |
817 | } | |
5c0e3066 | 818 | |
85b35fea | 819 | memset(memmap, 0, sizeof(struct page) * PAGES_PER_SECTION); |
3ac19f8e | 820 | |
c4e1be9e | 821 | section_mark_present(ms); |
29751f69 | 822 | |
5c0e3066 | 823 | ret = sparse_init_one_section(ms, section_nr, memmap, usemap); |
0b0acbec | 824 | |
0b0acbec DH |
825 | out: |
826 | pgdat_resize_unlock(pgdat, &flags); | |
bbd06825 WC |
827 | if (ret <= 0) { |
828 | kfree(usemap); | |
24b6d416 | 829 | __kfree_section_memmap(memmap, altmap); |
bbd06825 | 830 | } |
0b0acbec | 831 | return ret; |
29751f69 | 832 | } |
ea01ea93 | 833 | |
f3deb687 | 834 | #ifdef CONFIG_MEMORY_HOTREMOVE |
95a4774d WC |
835 | #ifdef CONFIG_MEMORY_FAILURE |
836 | static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages) | |
837 | { | |
838 | int i; | |
839 | ||
840 | if (!memmap) | |
841 | return; | |
842 | ||
4b94ffdc | 843 | for (i = 0; i < nr_pages; i++) { |
95a4774d | 844 | if (PageHWPoison(&memmap[i])) { |
293c07e3 | 845 | atomic_long_sub(1, &num_poisoned_pages); |
95a4774d WC |
846 | ClearPageHWPoison(&memmap[i]); |
847 | } | |
848 | } | |
849 | } | |
850 | #else | |
851 | static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages) | |
852 | { | |
853 | } | |
854 | #endif | |
855 | ||
24b6d416 CH |
856 | static void free_section_usemap(struct page *memmap, unsigned long *usemap, |
857 | struct vmem_altmap *altmap) | |
4edd7cef DR |
858 | { |
859 | struct page *usemap_page; | |
4edd7cef DR |
860 | |
861 | if (!usemap) | |
862 | return; | |
863 | ||
864 | usemap_page = virt_to_page(usemap); | |
865 | /* | |
866 | * Check to see if allocation came from hot-plug-add | |
867 | */ | |
868 | if (PageSlab(usemap_page) || PageCompound(usemap_page)) { | |
869 | kfree(usemap); | |
870 | if (memmap) | |
24b6d416 | 871 | __kfree_section_memmap(memmap, altmap); |
4edd7cef DR |
872 | return; |
873 | } | |
874 | ||
875 | /* | |
876 | * The usemap came from bootmem. This is packed with other usemaps | |
877 | * on the section which has pgdat at boot time. Just keep it as is now. | |
878 | */ | |
879 | ||
81556b02 ZY |
880 | if (memmap) |
881 | free_map_bootmem(memmap); | |
4edd7cef DR |
882 | } |
883 | ||
4b94ffdc | 884 | void sparse_remove_one_section(struct zone *zone, struct mem_section *ms, |
24b6d416 | 885 | unsigned long map_offset, struct vmem_altmap *altmap) |
ea01ea93 BP |
886 | { |
887 | struct page *memmap = NULL; | |
cd099682 TC |
888 | unsigned long *usemap = NULL, flags; |
889 | struct pglist_data *pgdat = zone->zone_pgdat; | |
ea01ea93 | 890 | |
cd099682 | 891 | pgdat_resize_lock(pgdat, &flags); |
ea01ea93 BP |
892 | if (ms->section_mem_map) { |
893 | usemap = ms->pageblock_flags; | |
894 | memmap = sparse_decode_mem_map(ms->section_mem_map, | |
895 | __section_nr(ms)); | |
896 | ms->section_mem_map = 0; | |
897 | ms->pageblock_flags = NULL; | |
898 | } | |
cd099682 | 899 | pgdat_resize_unlock(pgdat, &flags); |
ea01ea93 | 900 | |
4b94ffdc DW |
901 | clear_hwpoisoned_pages(memmap + map_offset, |
902 | PAGES_PER_SECTION - map_offset); | |
24b6d416 | 903 | free_section_usemap(memmap, usemap, altmap); |
ea01ea93 | 904 | } |
4edd7cef DR |
905 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
906 | #endif /* CONFIG_MEMORY_HOTPLUG */ |