Commit | Line | Data |
---|---|---|
d41dee36 AW |
1 | /* |
2 | * sparse memory mappings. | |
3 | */ | |
d41dee36 | 4 | #include <linux/mm.h> |
5a0e3ad6 | 5 | #include <linux/slab.h> |
d41dee36 AW |
6 | #include <linux/mmzone.h> |
7 | #include <linux/bootmem.h> | |
3b32123d | 8 | #include <linux/compiler.h> |
0b0acbec | 9 | #include <linux/highmem.h> |
b95f1b31 | 10 | #include <linux/export.h> |
28ae55c9 | 11 | #include <linux/spinlock.h> |
0b0acbec | 12 | #include <linux/vmalloc.h> |
3b32123d | 13 | |
0c0a4a51 | 14 | #include "internal.h" |
d41dee36 | 15 | #include <asm/dma.h> |
8f6aac41 CL |
16 | #include <asm/pgalloc.h> |
17 | #include <asm/pgtable.h> | |
d41dee36 AW |
18 | |
19 | /* | |
20 | * Permanent SPARSEMEM data: | |
21 | * | |
22 | * 1) mem_section - memory sections, mem_map's for valid memory | |
23 | */ | |
3e347261 | 24 | #ifdef CONFIG_SPARSEMEM_EXTREME |
802f192e | 25 | struct mem_section *mem_section[NR_SECTION_ROOTS] |
22fc6ecc | 26 | ____cacheline_internodealigned_in_smp; |
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 |
577a32f6 | 62 | static struct mem_section noinline __init_refok *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 | ||
f52407ce SL |
68 | if (slab_is_available()) { |
69 | if (node_state(nid, N_HIGH_MEMORY)) | |
5b760e64 | 70 | section = kzalloc_node(array_size, GFP_KERNEL, nid); |
f52407ce | 71 | else |
5b760e64 GS |
72 | section = kzalloc(array_size, GFP_KERNEL); |
73 | } else { | |
bb016b84 | 74 | section = memblock_virt_alloc_node(array_size, nid); |
5b760e64 | 75 | } |
28ae55c9 DH |
76 | |
77 | return section; | |
3e347261 | 78 | } |
802f192e | 79 | |
a3142c8e | 80 | static int __meminit sparse_index_init(unsigned long section_nr, int nid) |
802f192e | 81 | { |
28ae55c9 DH |
82 | unsigned long root = SECTION_NR_TO_ROOT(section_nr); |
83 | struct mem_section *section; | |
802f192e BP |
84 | |
85 | if (mem_section[root]) | |
28ae55c9 | 86 | return -EEXIST; |
3e347261 | 87 | |
28ae55c9 | 88 | section = sparse_index_alloc(nid); |
af0cd5a7 WC |
89 | if (!section) |
90 | return -ENOMEM; | |
28ae55c9 DH |
91 | |
92 | mem_section[root] = section; | |
c1c95183 | 93 | |
9d1936cf | 94 | return 0; |
28ae55c9 DH |
95 | } |
96 | #else /* !SPARSEMEM_EXTREME */ | |
97 | static inline int sparse_index_init(unsigned long section_nr, int nid) | |
98 | { | |
99 | return 0; | |
802f192e | 100 | } |
28ae55c9 DH |
101 | #endif |
102 | ||
4ca644d9 DH |
103 | /* |
104 | * Although written for the SPARSEMEM_EXTREME case, this happens | |
cd881a6b | 105 | * to also work for the flat array case because |
4ca644d9 DH |
106 | * NR_SECTION_ROOTS==NR_MEM_SECTIONS. |
107 | */ | |
108 | int __section_nr(struct mem_section* ms) | |
109 | { | |
110 | unsigned long root_nr; | |
111 | struct mem_section* root; | |
112 | ||
12783b00 MK |
113 | for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) { |
114 | root = __nr_to_section(root_nr * SECTIONS_PER_ROOT); | |
4ca644d9 DH |
115 | if (!root) |
116 | continue; | |
117 | ||
118 | if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT))) | |
119 | break; | |
120 | } | |
121 | ||
db36a461 GS |
122 | VM_BUG_ON(root_nr == NR_SECTION_ROOTS); |
123 | ||
4ca644d9 DH |
124 | return (root_nr * SECTIONS_PER_ROOT) + (ms - root); |
125 | } | |
126 | ||
30c253e6 AW |
127 | /* |
128 | * During early boot, before section_mem_map is used for an actual | |
129 | * mem_map, we use section_mem_map to store the section's NUMA | |
130 | * node. This keeps us from having to use another data structure. The | |
131 | * node information is cleared just before we store the real mem_map. | |
132 | */ | |
133 | static inline unsigned long sparse_encode_early_nid(int nid) | |
134 | { | |
135 | return (nid << SECTION_NID_SHIFT); | |
136 | } | |
137 | ||
138 | static inline int sparse_early_nid(struct mem_section *section) | |
139 | { | |
140 | return (section->section_mem_map >> SECTION_NID_SHIFT); | |
141 | } | |
142 | ||
2dbb51c4 MG |
143 | /* Validate the physical addressing limitations of the model */ |
144 | void __meminit mminit_validate_memmodel_limits(unsigned long *start_pfn, | |
145 | unsigned long *end_pfn) | |
d41dee36 | 146 | { |
2dbb51c4 | 147 | unsigned long max_sparsemem_pfn = 1UL << (MAX_PHYSMEM_BITS-PAGE_SHIFT); |
d41dee36 | 148 | |
bead9a3a IM |
149 | /* |
150 | * Sanity checks - do not allow an architecture to pass | |
151 | * in larger pfns than the maximum scope of sparsemem: | |
152 | */ | |
2dbb51c4 MG |
153 | if (*start_pfn > max_sparsemem_pfn) { |
154 | mminit_dprintk(MMINIT_WARNING, "pfnvalidation", | |
155 | "Start of range %lu -> %lu exceeds SPARSEMEM max %lu\n", | |
156 | *start_pfn, *end_pfn, max_sparsemem_pfn); | |
157 | WARN_ON_ONCE(1); | |
158 | *start_pfn = max_sparsemem_pfn; | |
159 | *end_pfn = max_sparsemem_pfn; | |
ef161a98 | 160 | } else if (*end_pfn > max_sparsemem_pfn) { |
2dbb51c4 MG |
161 | mminit_dprintk(MMINIT_WARNING, "pfnvalidation", |
162 | "End of range %lu -> %lu exceeds SPARSEMEM max %lu\n", | |
163 | *start_pfn, *end_pfn, max_sparsemem_pfn); | |
164 | WARN_ON_ONCE(1); | |
165 | *end_pfn = max_sparsemem_pfn; | |
166 | } | |
167 | } | |
168 | ||
169 | /* Record a memory area against a node. */ | |
170 | void __init memory_present(int nid, unsigned long start, unsigned long end) | |
171 | { | |
172 | unsigned long pfn; | |
bead9a3a | 173 | |
d41dee36 | 174 | start &= PAGE_SECTION_MASK; |
2dbb51c4 | 175 | mminit_validate_memmodel_limits(&start, &end); |
d41dee36 AW |
176 | for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) { |
177 | unsigned long section = pfn_to_section_nr(pfn); | |
802f192e BP |
178 | struct mem_section *ms; |
179 | ||
180 | sparse_index_init(section, nid); | |
85770ffe | 181 | set_section_nid(section, nid); |
802f192e BP |
182 | |
183 | ms = __nr_to_section(section); | |
184 | if (!ms->section_mem_map) | |
30c253e6 AW |
185 | ms->section_mem_map = sparse_encode_early_nid(nid) | |
186 | SECTION_MARKED_PRESENT; | |
d41dee36 AW |
187 | } |
188 | } | |
189 | ||
190 | /* | |
191 | * Only used by the i386 NUMA architecures, but relatively | |
192 | * generic code. | |
193 | */ | |
194 | unsigned long __init node_memmap_size_bytes(int nid, unsigned long start_pfn, | |
195 | unsigned long end_pfn) | |
196 | { | |
197 | unsigned long pfn; | |
198 | unsigned long nr_pages = 0; | |
199 | ||
2dbb51c4 | 200 | mminit_validate_memmodel_limits(&start_pfn, &end_pfn); |
d41dee36 AW |
201 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
202 | if (nid != early_pfn_to_nid(pfn)) | |
203 | continue; | |
204 | ||
540557b9 | 205 | if (pfn_present(pfn)) |
d41dee36 AW |
206 | nr_pages += PAGES_PER_SECTION; |
207 | } | |
208 | ||
209 | return nr_pages * sizeof(struct page); | |
210 | } | |
211 | ||
29751f69 AW |
212 | /* |
213 | * Subtle, we encode the real pfn into the mem_map such that | |
214 | * the identity pfn - section_mem_map will return the actual | |
215 | * physical page frame number. | |
216 | */ | |
217 | static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum) | |
218 | { | |
219 | return (unsigned long)(mem_map - (section_nr_to_pfn(pnum))); | |
220 | } | |
221 | ||
222 | /* | |
ea01ea93 | 223 | * Decode mem_map from the coded memmap |
29751f69 | 224 | */ |
29751f69 AW |
225 | struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum) |
226 | { | |
ea01ea93 BP |
227 | /* mask off the extra low bits of information */ |
228 | coded_mem_map &= SECTION_MAP_MASK; | |
29751f69 AW |
229 | return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum); |
230 | } | |
231 | ||
a3142c8e | 232 | static int __meminit sparse_init_one_section(struct mem_section *ms, |
5c0e3066 MG |
233 | unsigned long pnum, struct page *mem_map, |
234 | unsigned long *pageblock_bitmap) | |
29751f69 | 235 | { |
540557b9 | 236 | if (!present_section(ms)) |
29751f69 AW |
237 | return -EINVAL; |
238 | ||
30c253e6 | 239 | ms->section_mem_map &= ~SECTION_MAP_MASK; |
540557b9 AW |
240 | ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) | |
241 | SECTION_HAS_MEM_MAP; | |
5c0e3066 | 242 | ms->pageblock_flags = pageblock_bitmap; |
29751f69 AW |
243 | |
244 | return 1; | |
245 | } | |
246 | ||
04753278 | 247 | unsigned long usemap_size(void) |
5c0e3066 MG |
248 | { |
249 | unsigned long size_bytes; | |
250 | size_bytes = roundup(SECTION_BLOCKFLAGS_BITS, 8) / 8; | |
251 | size_bytes = roundup(size_bytes, sizeof(unsigned long)); | |
252 | return size_bytes; | |
253 | } | |
254 | ||
255 | #ifdef CONFIG_MEMORY_HOTPLUG | |
256 | static unsigned long *__kmalloc_section_usemap(void) | |
257 | { | |
258 | return kmalloc(usemap_size(), GFP_KERNEL); | |
259 | } | |
260 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
261 | ||
48c90682 YG |
262 | #ifdef CONFIG_MEMORY_HOTREMOVE |
263 | static unsigned long * __init | |
a4322e1b | 264 | sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat, |
238305bb | 265 | unsigned long size) |
48c90682 | 266 | { |
99ab7b19 YL |
267 | unsigned long goal, limit; |
268 | unsigned long *p; | |
269 | int nid; | |
48c90682 YG |
270 | /* |
271 | * A page may contain usemaps for other sections preventing the | |
272 | * page being freed and making a section unremovable while | |
c800bcd5 | 273 | * other sections referencing the usemap remain active. Similarly, |
48c90682 YG |
274 | * a pgdat can prevent a section being removed. If section A |
275 | * contains a pgdat and section B contains the usemap, both | |
276 | * sections become inter-dependent. This allocates usemaps | |
277 | * from the same section as the pgdat where possible to avoid | |
278 | * this problem. | |
279 | */ | |
07b4e2bc | 280 | goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT); |
99ab7b19 YL |
281 | limit = goal + (1UL << PA_SECTION_SHIFT); |
282 | nid = early_pfn_to_nid(goal >> PAGE_SHIFT); | |
283 | again: | |
bb016b84 SS |
284 | p = memblock_virt_alloc_try_nid_nopanic(size, |
285 | SMP_CACHE_BYTES, goal, limit, | |
286 | nid); | |
99ab7b19 YL |
287 | if (!p && limit) { |
288 | limit = 0; | |
289 | goto again; | |
290 | } | |
291 | return p; | |
48c90682 YG |
292 | } |
293 | ||
294 | static void __init check_usemap_section_nr(int nid, unsigned long *usemap) | |
295 | { | |
296 | unsigned long usemap_snr, pgdat_snr; | |
297 | static unsigned long old_usemap_snr = NR_MEM_SECTIONS; | |
298 | static unsigned long old_pgdat_snr = NR_MEM_SECTIONS; | |
299 | struct pglist_data *pgdat = NODE_DATA(nid); | |
300 | int usemap_nid; | |
301 | ||
302 | usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT); | |
303 | pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT); | |
304 | if (usemap_snr == pgdat_snr) | |
305 | return; | |
306 | ||
307 | if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr) | |
308 | /* skip redundant message */ | |
309 | return; | |
310 | ||
311 | old_usemap_snr = usemap_snr; | |
312 | old_pgdat_snr = pgdat_snr; | |
313 | ||
314 | usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr)); | |
315 | if (usemap_nid != nid) { | |
1170532b JP |
316 | pr_info("node %d must be removed before remove section %ld\n", |
317 | nid, usemap_snr); | |
48c90682 YG |
318 | return; |
319 | } | |
320 | /* | |
321 | * There is a circular dependency. | |
322 | * Some platforms allow un-removable section because they will just | |
323 | * gather other removable sections for dynamic partitioning. | |
324 | * Just notify un-removable section's number here. | |
325 | */ | |
1170532b JP |
326 | pr_info("Section %ld and %ld (node %d) have a circular dependency on usemap and pgdat allocations\n", |
327 | usemap_snr, pgdat_snr, nid); | |
48c90682 YG |
328 | } |
329 | #else | |
330 | static unsigned long * __init | |
a4322e1b | 331 | sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat, |
238305bb | 332 | unsigned long size) |
48c90682 | 333 | { |
bb016b84 | 334 | return memblock_virt_alloc_node_nopanic(size, pgdat->node_id); |
48c90682 YG |
335 | } |
336 | ||
337 | static void __init check_usemap_section_nr(int nid, unsigned long *usemap) | |
338 | { | |
339 | } | |
340 | #endif /* CONFIG_MEMORY_HOTREMOVE */ | |
341 | ||
18732093 | 342 | static void __init sparse_early_usemaps_alloc_node(void *data, |
a4322e1b YL |
343 | unsigned long pnum_begin, |
344 | unsigned long pnum_end, | |
345 | unsigned long usemap_count, int nodeid) | |
5c0e3066 | 346 | { |
a4322e1b YL |
347 | void *usemap; |
348 | unsigned long pnum; | |
18732093 | 349 | unsigned long **usemap_map = (unsigned long **)data; |
a4322e1b | 350 | int size = usemap_size(); |
5c0e3066 | 351 | |
a4322e1b | 352 | usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid), |
238305bb | 353 | size * usemap_count); |
f5bf18fa | 354 | if (!usemap) { |
1170532b | 355 | pr_warn("%s: allocation failed\n", __func__); |
238305bb | 356 | return; |
48c90682 YG |
357 | } |
358 | ||
f5bf18fa NA |
359 | for (pnum = pnum_begin; pnum < pnum_end; pnum++) { |
360 | if (!present_section_nr(pnum)) | |
361 | continue; | |
362 | usemap_map[pnum] = usemap; | |
363 | usemap += size; | |
364 | check_usemap_section_nr(nodeid, usemap_map[pnum]); | |
a4322e1b | 365 | } |
5c0e3066 MG |
366 | } |
367 | ||
8f6aac41 | 368 | #ifndef CONFIG_SPARSEMEM_VMEMMAP |
98f3cfc1 | 369 | struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid) |
29751f69 AW |
370 | { |
371 | struct page *map; | |
e48e67e0 | 372 | unsigned long size; |
29751f69 AW |
373 | |
374 | map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION); | |
375 | if (map) | |
376 | return map; | |
377 | ||
e48e67e0 | 378 | size = PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION); |
bb016b84 SS |
379 | map = memblock_virt_alloc_try_nid(size, |
380 | PAGE_SIZE, __pa(MAX_DMA_ADDRESS), | |
381 | BOOTMEM_ALLOC_ACCESSIBLE, nid); | |
8f6aac41 CL |
382 | return map; |
383 | } | |
9bdac914 YL |
384 | void __init sparse_mem_maps_populate_node(struct page **map_map, |
385 | unsigned long pnum_begin, | |
386 | unsigned long pnum_end, | |
387 | unsigned long map_count, int nodeid) | |
388 | { | |
389 | void *map; | |
390 | unsigned long pnum; | |
391 | unsigned long size = sizeof(struct page) * PAGES_PER_SECTION; | |
392 | ||
393 | map = alloc_remap(nodeid, size * map_count); | |
394 | if (map) { | |
395 | for (pnum = pnum_begin; pnum < pnum_end; pnum++) { | |
396 | if (!present_section_nr(pnum)) | |
397 | continue; | |
398 | map_map[pnum] = map; | |
399 | map += size; | |
400 | } | |
401 | return; | |
402 | } | |
403 | ||
404 | size = PAGE_ALIGN(size); | |
bb016b84 SS |
405 | map = memblock_virt_alloc_try_nid(size * map_count, |
406 | PAGE_SIZE, __pa(MAX_DMA_ADDRESS), | |
407 | BOOTMEM_ALLOC_ACCESSIBLE, nodeid); | |
9bdac914 YL |
408 | if (map) { |
409 | for (pnum = pnum_begin; pnum < pnum_end; pnum++) { | |
410 | if (!present_section_nr(pnum)) | |
411 | continue; | |
412 | map_map[pnum] = map; | |
413 | map += size; | |
414 | } | |
415 | return; | |
416 | } | |
417 | ||
418 | /* fallback */ | |
419 | for (pnum = pnum_begin; pnum < pnum_end; pnum++) { | |
420 | struct mem_section *ms; | |
421 | ||
422 | if (!present_section_nr(pnum)) | |
423 | continue; | |
424 | map_map[pnum] = sparse_mem_map_populate(pnum, nodeid); | |
425 | if (map_map[pnum]) | |
426 | continue; | |
427 | ms = __nr_to_section(pnum); | |
1170532b | 428 | pr_err("%s: sparsemem memory map backing failed some memory will not be available\n", |
756a025f | 429 | __func__); |
9bdac914 YL |
430 | ms->section_mem_map = 0; |
431 | } | |
432 | } | |
8f6aac41 CL |
433 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ |
434 | ||
81d0d950 | 435 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
18732093 | 436 | static void __init sparse_early_mem_maps_alloc_node(void *data, |
9bdac914 YL |
437 | unsigned long pnum_begin, |
438 | unsigned long pnum_end, | |
439 | unsigned long map_count, int nodeid) | |
440 | { | |
18732093 | 441 | struct page **map_map = (struct page **)data; |
9bdac914 YL |
442 | sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end, |
443 | map_count, nodeid); | |
444 | } | |
81d0d950 | 445 | #else |
9e5c6da7 | 446 | static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum) |
8f6aac41 CL |
447 | { |
448 | struct page *map; | |
449 | struct mem_section *ms = __nr_to_section(pnum); | |
450 | int nid = sparse_early_nid(ms); | |
451 | ||
98f3cfc1 | 452 | map = sparse_mem_map_populate(pnum, nid); |
29751f69 AW |
453 | if (map) |
454 | return map; | |
455 | ||
1170532b | 456 | pr_err("%s: sparsemem memory map backing failed some memory will not be available\n", |
756a025f | 457 | __func__); |
802f192e | 458 | ms->section_mem_map = 0; |
29751f69 AW |
459 | return NULL; |
460 | } | |
9bdac914 | 461 | #endif |
29751f69 | 462 | |
3b32123d | 463 | void __weak __meminit vmemmap_populate_print_last(void) |
c2b91e2e YL |
464 | { |
465 | } | |
a4322e1b | 466 | |
18732093 WL |
467 | /** |
468 | * alloc_usemap_and_memmap - memory alloction for pageblock flags and vmemmap | |
469 | * @map: usemap_map for pageblock flags or mmap_map for vmemmap | |
470 | */ | |
471 | static void __init alloc_usemap_and_memmap(void (*alloc_func) | |
472 | (void *, unsigned long, unsigned long, | |
473 | unsigned long, int), void *data) | |
474 | { | |
475 | unsigned long pnum; | |
476 | unsigned long map_count; | |
477 | int nodeid_begin = 0; | |
478 | unsigned long pnum_begin = 0; | |
479 | ||
480 | for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) { | |
481 | struct mem_section *ms; | |
482 | ||
483 | if (!present_section_nr(pnum)) | |
484 | continue; | |
485 | ms = __nr_to_section(pnum); | |
486 | nodeid_begin = sparse_early_nid(ms); | |
487 | pnum_begin = pnum; | |
488 | break; | |
489 | } | |
490 | map_count = 1; | |
491 | for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) { | |
492 | struct mem_section *ms; | |
493 | int nodeid; | |
494 | ||
495 | if (!present_section_nr(pnum)) | |
496 | continue; | |
497 | ms = __nr_to_section(pnum); | |
498 | nodeid = sparse_early_nid(ms); | |
499 | if (nodeid == nodeid_begin) { | |
500 | map_count++; | |
501 | continue; | |
502 | } | |
503 | /* ok, we need to take cake of from pnum_begin to pnum - 1*/ | |
504 | alloc_func(data, pnum_begin, pnum, | |
505 | map_count, nodeid_begin); | |
506 | /* new start, update count etc*/ | |
507 | nodeid_begin = nodeid; | |
508 | pnum_begin = pnum; | |
509 | map_count = 1; | |
510 | } | |
511 | /* ok, last chunk */ | |
512 | alloc_func(data, pnum_begin, NR_MEM_SECTIONS, | |
513 | map_count, nodeid_begin); | |
514 | } | |
515 | ||
193faea9 SR |
516 | /* |
517 | * Allocate the accumulated non-linear sections, allocate a mem_map | |
518 | * for each and record the physical to section mapping. | |
519 | */ | |
520 | void __init sparse_init(void) | |
521 | { | |
522 | unsigned long pnum; | |
523 | struct page *map; | |
5c0e3066 | 524 | unsigned long *usemap; |
e123dd3f | 525 | unsigned long **usemap_map; |
81d0d950 | 526 | int size; |
81d0d950 | 527 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
81d0d950 YL |
528 | int size2; |
529 | struct page **map_map; | |
530 | #endif | |
e123dd3f | 531 | |
55878e88 CS |
532 | /* see include/linux/mmzone.h 'struct mem_section' definition */ |
533 | BUILD_BUG_ON(!is_power_of_2(sizeof(struct mem_section))); | |
534 | ||
ca57df79 XQ |
535 | /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */ |
536 | set_pageblock_order(); | |
537 | ||
e123dd3f YL |
538 | /* |
539 | * map is using big page (aka 2M in x86 64 bit) | |
540 | * usemap is less one page (aka 24 bytes) | |
541 | * so alloc 2M (with 2M align) and 24 bytes in turn will | |
542 | * make next 2M slip to one more 2M later. | |
543 | * then in big system, the memory will have a lot of holes... | |
25985edc | 544 | * here try to allocate 2M pages continuously. |
e123dd3f YL |
545 | * |
546 | * powerpc need to call sparse_init_one_section right after each | |
547 | * sparse_early_mem_map_alloc, so allocate usemap_map at first. | |
548 | */ | |
549 | size = sizeof(unsigned long *) * NR_MEM_SECTIONS; | |
bb016b84 | 550 | usemap_map = memblock_virt_alloc(size, 0); |
e123dd3f YL |
551 | if (!usemap_map) |
552 | panic("can not allocate usemap_map\n"); | |
18732093 WL |
553 | alloc_usemap_and_memmap(sparse_early_usemaps_alloc_node, |
554 | (void *)usemap_map); | |
193faea9 | 555 | |
9bdac914 YL |
556 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
557 | size2 = sizeof(struct page *) * NR_MEM_SECTIONS; | |
bb016b84 | 558 | map_map = memblock_virt_alloc(size2, 0); |
9bdac914 YL |
559 | if (!map_map) |
560 | panic("can not allocate map_map\n"); | |
18732093 WL |
561 | alloc_usemap_and_memmap(sparse_early_mem_maps_alloc_node, |
562 | (void *)map_map); | |
9bdac914 YL |
563 | #endif |
564 | ||
e123dd3f YL |
565 | for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) { |
566 | if (!present_section_nr(pnum)) | |
193faea9 | 567 | continue; |
5c0e3066 | 568 | |
e123dd3f | 569 | usemap = usemap_map[pnum]; |
5c0e3066 MG |
570 | if (!usemap) |
571 | continue; | |
572 | ||
9bdac914 YL |
573 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
574 | map = map_map[pnum]; | |
575 | #else | |
e123dd3f | 576 | map = sparse_early_mem_map_alloc(pnum); |
9bdac914 | 577 | #endif |
e123dd3f YL |
578 | if (!map) |
579 | continue; | |
580 | ||
5c0e3066 MG |
581 | sparse_init_one_section(__nr_to_section(pnum), pnum, map, |
582 | usemap); | |
193faea9 | 583 | } |
e123dd3f | 584 | |
c2b91e2e YL |
585 | vmemmap_populate_print_last(); |
586 | ||
9bdac914 | 587 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
bb016b84 | 588 | memblock_free_early(__pa(map_map), size2); |
9bdac914 | 589 | #endif |
bb016b84 | 590 | memblock_free_early(__pa(usemap_map), size); |
193faea9 SR |
591 | } |
592 | ||
593 | #ifdef CONFIG_MEMORY_HOTPLUG | |
98f3cfc1 | 594 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
85b35fea | 595 | static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid) |
98f3cfc1 YG |
596 | { |
597 | /* This will make the necessary allocations eventually. */ | |
598 | return sparse_mem_map_populate(pnum, nid); | |
599 | } | |
85b35fea | 600 | static void __kfree_section_memmap(struct page *memmap) |
98f3cfc1 | 601 | { |
0aad818b | 602 | unsigned long start = (unsigned long)memmap; |
85b35fea | 603 | unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION); |
0aad818b JW |
604 | |
605 | vmemmap_free(start, end); | |
98f3cfc1 | 606 | } |
4edd7cef | 607 | #ifdef CONFIG_MEMORY_HOTREMOVE |
81556b02 | 608 | static void free_map_bootmem(struct page *memmap) |
0c0a4a51 | 609 | { |
0aad818b | 610 | unsigned long start = (unsigned long)memmap; |
81556b02 | 611 | unsigned long end = (unsigned long)(memmap + PAGES_PER_SECTION); |
0aad818b JW |
612 | |
613 | vmemmap_free(start, end); | |
0c0a4a51 | 614 | } |
4edd7cef | 615 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
98f3cfc1 | 616 | #else |
85b35fea | 617 | static struct page *__kmalloc_section_memmap(void) |
0b0acbec DH |
618 | { |
619 | struct page *page, *ret; | |
85b35fea | 620 | unsigned long memmap_size = sizeof(struct page) * PAGES_PER_SECTION; |
0b0acbec | 621 | |
f2d0aa5b | 622 | page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size)); |
0b0acbec DH |
623 | if (page) |
624 | goto got_map_page; | |
625 | ||
626 | ret = vmalloc(memmap_size); | |
627 | if (ret) | |
628 | goto got_map_ptr; | |
629 | ||
630 | return NULL; | |
631 | got_map_page: | |
632 | ret = (struct page *)pfn_to_kaddr(page_to_pfn(page)); | |
633 | got_map_ptr: | |
0b0acbec DH |
634 | |
635 | return ret; | |
636 | } | |
637 | ||
85b35fea | 638 | static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid) |
98f3cfc1 | 639 | { |
85b35fea | 640 | return __kmalloc_section_memmap(); |
98f3cfc1 YG |
641 | } |
642 | ||
85b35fea | 643 | static void __kfree_section_memmap(struct page *memmap) |
0b0acbec | 644 | { |
9e2779fa | 645 | if (is_vmalloc_addr(memmap)) |
0b0acbec DH |
646 | vfree(memmap); |
647 | else | |
648 | free_pages((unsigned long)memmap, | |
85b35fea | 649 | get_order(sizeof(struct page) * PAGES_PER_SECTION)); |
0b0acbec | 650 | } |
0c0a4a51 | 651 | |
4edd7cef | 652 | #ifdef CONFIG_MEMORY_HOTREMOVE |
81556b02 | 653 | static void free_map_bootmem(struct page *memmap) |
0c0a4a51 YG |
654 | { |
655 | unsigned long maps_section_nr, removing_section_nr, i; | |
81556b02 | 656 | unsigned long magic, nr_pages; |
ae64ffca | 657 | struct page *page = virt_to_page(memmap); |
0c0a4a51 | 658 | |
81556b02 ZY |
659 | nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page)) |
660 | >> PAGE_SHIFT; | |
661 | ||
0c0a4a51 | 662 | for (i = 0; i < nr_pages; i++, page++) { |
5f24ce5f | 663 | magic = (unsigned long) page->lru.next; |
0c0a4a51 YG |
664 | |
665 | BUG_ON(magic == NODE_INFO); | |
666 | ||
667 | maps_section_nr = pfn_to_section_nr(page_to_pfn(page)); | |
668 | removing_section_nr = page->private; | |
669 | ||
670 | /* | |
671 | * When this function is called, the removing section is | |
672 | * logical offlined state. This means all pages are isolated | |
673 | * from page allocator. If removing section's memmap is placed | |
674 | * on the same section, it must not be freed. | |
675 | * If it is freed, page allocator may allocate it which will | |
676 | * be removed physically soon. | |
677 | */ | |
678 | if (maps_section_nr != removing_section_nr) | |
679 | put_page_bootmem(page); | |
680 | } | |
681 | } | |
4edd7cef | 682 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
98f3cfc1 | 683 | #endif /* CONFIG_SPARSEMEM_VMEMMAP */ |
0b0acbec | 684 | |
29751f69 AW |
685 | /* |
686 | * returns the number of sections whose mem_maps were properly | |
687 | * set. If this is <=0, then that means that the passed-in | |
688 | * map was not consumed and must be freed. | |
689 | */ | |
85b35fea | 690 | int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn) |
29751f69 | 691 | { |
0b0acbec DH |
692 | unsigned long section_nr = pfn_to_section_nr(start_pfn); |
693 | struct pglist_data *pgdat = zone->zone_pgdat; | |
694 | struct mem_section *ms; | |
695 | struct page *memmap; | |
5c0e3066 | 696 | unsigned long *usemap; |
0b0acbec DH |
697 | unsigned long flags; |
698 | int ret; | |
29751f69 | 699 | |
0b0acbec DH |
700 | /* |
701 | * no locking for this, because it does its own | |
702 | * plus, it does a kmalloc | |
703 | */ | |
bbd06825 WC |
704 | ret = sparse_index_init(section_nr, pgdat->node_id); |
705 | if (ret < 0 && ret != -EEXIST) | |
706 | return ret; | |
85b35fea | 707 | memmap = kmalloc_section_memmap(section_nr, pgdat->node_id); |
bbd06825 WC |
708 | if (!memmap) |
709 | return -ENOMEM; | |
5c0e3066 | 710 | usemap = __kmalloc_section_usemap(); |
bbd06825 | 711 | if (!usemap) { |
85b35fea | 712 | __kfree_section_memmap(memmap); |
bbd06825 WC |
713 | return -ENOMEM; |
714 | } | |
0b0acbec DH |
715 | |
716 | pgdat_resize_lock(pgdat, &flags); | |
29751f69 | 717 | |
0b0acbec DH |
718 | ms = __pfn_to_section(start_pfn); |
719 | if (ms->section_mem_map & SECTION_MARKED_PRESENT) { | |
720 | ret = -EEXIST; | |
721 | goto out; | |
722 | } | |
5c0e3066 | 723 | |
85b35fea | 724 | memset(memmap, 0, sizeof(struct page) * PAGES_PER_SECTION); |
3ac19f8e | 725 | |
29751f69 AW |
726 | ms->section_mem_map |= SECTION_MARKED_PRESENT; |
727 | ||
5c0e3066 | 728 | ret = sparse_init_one_section(ms, section_nr, memmap, usemap); |
0b0acbec | 729 | |
0b0acbec DH |
730 | out: |
731 | pgdat_resize_unlock(pgdat, &flags); | |
bbd06825 WC |
732 | if (ret <= 0) { |
733 | kfree(usemap); | |
85b35fea | 734 | __kfree_section_memmap(memmap); |
bbd06825 | 735 | } |
0b0acbec | 736 | return ret; |
29751f69 | 737 | } |
ea01ea93 | 738 | |
f3deb687 | 739 | #ifdef CONFIG_MEMORY_HOTREMOVE |
95a4774d WC |
740 | #ifdef CONFIG_MEMORY_FAILURE |
741 | static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages) | |
742 | { | |
743 | int i; | |
744 | ||
745 | if (!memmap) | |
746 | return; | |
747 | ||
4b94ffdc | 748 | for (i = 0; i < nr_pages; i++) { |
95a4774d | 749 | if (PageHWPoison(&memmap[i])) { |
293c07e3 | 750 | atomic_long_sub(1, &num_poisoned_pages); |
95a4774d WC |
751 | ClearPageHWPoison(&memmap[i]); |
752 | } | |
753 | } | |
754 | } | |
755 | #else | |
756 | static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages) | |
757 | { | |
758 | } | |
759 | #endif | |
760 | ||
4edd7cef DR |
761 | static void free_section_usemap(struct page *memmap, unsigned long *usemap) |
762 | { | |
763 | struct page *usemap_page; | |
4edd7cef DR |
764 | |
765 | if (!usemap) | |
766 | return; | |
767 | ||
768 | usemap_page = virt_to_page(usemap); | |
769 | /* | |
770 | * Check to see if allocation came from hot-plug-add | |
771 | */ | |
772 | if (PageSlab(usemap_page) || PageCompound(usemap_page)) { | |
773 | kfree(usemap); | |
774 | if (memmap) | |
85b35fea | 775 | __kfree_section_memmap(memmap); |
4edd7cef DR |
776 | return; |
777 | } | |
778 | ||
779 | /* | |
780 | * The usemap came from bootmem. This is packed with other usemaps | |
781 | * on the section which has pgdat at boot time. Just keep it as is now. | |
782 | */ | |
783 | ||
81556b02 ZY |
784 | if (memmap) |
785 | free_map_bootmem(memmap); | |
4edd7cef DR |
786 | } |
787 | ||
4b94ffdc DW |
788 | void sparse_remove_one_section(struct zone *zone, struct mem_section *ms, |
789 | unsigned long map_offset) | |
ea01ea93 BP |
790 | { |
791 | struct page *memmap = NULL; | |
cd099682 TC |
792 | unsigned long *usemap = NULL, flags; |
793 | struct pglist_data *pgdat = zone->zone_pgdat; | |
ea01ea93 | 794 | |
cd099682 | 795 | pgdat_resize_lock(pgdat, &flags); |
ea01ea93 BP |
796 | if (ms->section_mem_map) { |
797 | usemap = ms->pageblock_flags; | |
798 | memmap = sparse_decode_mem_map(ms->section_mem_map, | |
799 | __section_nr(ms)); | |
800 | ms->section_mem_map = 0; | |
801 | ms->pageblock_flags = NULL; | |
802 | } | |
cd099682 | 803 | pgdat_resize_unlock(pgdat, &flags); |
ea01ea93 | 804 | |
4b94ffdc DW |
805 | clear_hwpoisoned_pages(memmap + map_offset, |
806 | PAGES_PER_SECTION - map_offset); | |
ea01ea93 BP |
807 | free_section_usemap(memmap, usemap); |
808 | } | |
4edd7cef DR |
809 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
810 | #endif /* CONFIG_MEMORY_HOTPLUG */ |