]>
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> | |
0b0acbec | 8 | #include <linux/highmem.h> |
b95f1b31 | 9 | #include <linux/export.h> |
28ae55c9 | 10 | #include <linux/spinlock.h> |
0b0acbec | 11 | #include <linux/vmalloc.h> |
0c0a4a51 | 12 | #include "internal.h" |
d41dee36 | 13 | #include <asm/dma.h> |
8f6aac41 CL |
14 | #include <asm/pgalloc.h> |
15 | #include <asm/pgtable.h> | |
d41dee36 AW |
16 | |
17 | /* | |
18 | * Permanent SPARSEMEM data: | |
19 | * | |
20 | * 1) mem_section - memory sections, mem_map's for valid memory | |
21 | */ | |
3e347261 | 22 | #ifdef CONFIG_SPARSEMEM_EXTREME |
802f192e | 23 | struct mem_section *mem_section[NR_SECTION_ROOTS] |
22fc6ecc | 24 | ____cacheline_internodealigned_in_smp; |
3e347261 BP |
25 | #else |
26 | struct mem_section mem_section[NR_SECTION_ROOTS][SECTIONS_PER_ROOT] | |
22fc6ecc | 27 | ____cacheline_internodealigned_in_smp; |
3e347261 BP |
28 | #endif |
29 | EXPORT_SYMBOL(mem_section); | |
30 | ||
89689ae7 CL |
31 | #ifdef NODE_NOT_IN_PAGE_FLAGS |
32 | /* | |
33 | * If we did not store the node number in the page then we have to | |
34 | * do a lookup in the section_to_node_table in order to find which | |
35 | * node the page belongs to. | |
36 | */ | |
37 | #if MAX_NUMNODES <= 256 | |
38 | static u8 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned; | |
39 | #else | |
40 | static u16 section_to_node_table[NR_MEM_SECTIONS] __cacheline_aligned; | |
41 | #endif | |
42 | ||
33dd4e0e | 43 | int page_to_nid(const struct page *page) |
89689ae7 CL |
44 | { |
45 | return section_to_node_table[page_to_section(page)]; | |
46 | } | |
47 | EXPORT_SYMBOL(page_to_nid); | |
85770ffe AW |
48 | |
49 | static void set_section_nid(unsigned long section_nr, int nid) | |
50 | { | |
51 | section_to_node_table[section_nr] = nid; | |
52 | } | |
53 | #else /* !NODE_NOT_IN_PAGE_FLAGS */ | |
54 | static inline void set_section_nid(unsigned long section_nr, int nid) | |
55 | { | |
56 | } | |
89689ae7 CL |
57 | #endif |
58 | ||
3e347261 | 59 | #ifdef CONFIG_SPARSEMEM_EXTREME |
577a32f6 | 60 | static struct mem_section noinline __init_refok *sparse_index_alloc(int nid) |
28ae55c9 DH |
61 | { |
62 | struct mem_section *section = NULL; | |
63 | unsigned long array_size = SECTIONS_PER_ROOT * | |
64 | sizeof(struct mem_section); | |
65 | ||
f52407ce SL |
66 | if (slab_is_available()) { |
67 | if (node_state(nid, N_HIGH_MEMORY)) | |
5b760e64 | 68 | section = kzalloc_node(array_size, GFP_KERNEL, nid); |
f52407ce | 69 | else |
5b760e64 GS |
70 | section = kzalloc(array_size, GFP_KERNEL); |
71 | } else { | |
46a66eec | 72 | section = alloc_bootmem_node(NODE_DATA(nid), array_size); |
5b760e64 | 73 | } |
28ae55c9 DH |
74 | |
75 | return section; | |
3e347261 | 76 | } |
802f192e | 77 | |
a3142c8e | 78 | static int __meminit sparse_index_init(unsigned long section_nr, int nid) |
802f192e | 79 | { |
28ae55c9 DH |
80 | unsigned long root = SECTION_NR_TO_ROOT(section_nr); |
81 | struct mem_section *section; | |
802f192e BP |
82 | |
83 | if (mem_section[root]) | |
28ae55c9 | 84 | return -EEXIST; |
3e347261 | 85 | |
28ae55c9 | 86 | section = sparse_index_alloc(nid); |
af0cd5a7 WC |
87 | if (!section) |
88 | return -ENOMEM; | |
28ae55c9 DH |
89 | |
90 | mem_section[root] = section; | |
c1c95183 | 91 | |
9d1936cf | 92 | return 0; |
28ae55c9 DH |
93 | } |
94 | #else /* !SPARSEMEM_EXTREME */ | |
95 | static inline int sparse_index_init(unsigned long section_nr, int nid) | |
96 | { | |
97 | return 0; | |
802f192e | 98 | } |
28ae55c9 DH |
99 | #endif |
100 | ||
4ca644d9 DH |
101 | /* |
102 | * Although written for the SPARSEMEM_EXTREME case, this happens | |
cd881a6b | 103 | * to also work for the flat array case because |
4ca644d9 DH |
104 | * NR_SECTION_ROOTS==NR_MEM_SECTIONS. |
105 | */ | |
106 | int __section_nr(struct mem_section* ms) | |
107 | { | |
108 | unsigned long root_nr; | |
109 | struct mem_section* root; | |
110 | ||
12783b00 MK |
111 | for (root_nr = 0; root_nr < NR_SECTION_ROOTS; root_nr++) { |
112 | root = __nr_to_section(root_nr * SECTIONS_PER_ROOT); | |
4ca644d9 DH |
113 | if (!root) |
114 | continue; | |
115 | ||
116 | if ((ms >= root) && (ms < (root + SECTIONS_PER_ROOT))) | |
117 | break; | |
118 | } | |
119 | ||
db36a461 GS |
120 | VM_BUG_ON(root_nr == NR_SECTION_ROOTS); |
121 | ||
4ca644d9 DH |
122 | return (root_nr * SECTIONS_PER_ROOT) + (ms - root); |
123 | } | |
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 | ||
167 | /* Record a memory area against a node. */ | |
168 | void __init memory_present(int nid, unsigned long start, unsigned long end) | |
169 | { | |
170 | unsigned long pfn; | |
bead9a3a | 171 | |
d41dee36 | 172 | start &= PAGE_SECTION_MASK; |
2dbb51c4 | 173 | mminit_validate_memmodel_limits(&start, &end); |
d41dee36 AW |
174 | for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) { |
175 | unsigned long section = pfn_to_section_nr(pfn); | |
802f192e BP |
176 | struct mem_section *ms; |
177 | ||
178 | sparse_index_init(section, nid); | |
85770ffe | 179 | set_section_nid(section, nid); |
802f192e BP |
180 | |
181 | ms = __nr_to_section(section); | |
182 | if (!ms->section_mem_map) | |
30c253e6 AW |
183 | ms->section_mem_map = sparse_encode_early_nid(nid) | |
184 | SECTION_MARKED_PRESENT; | |
d41dee36 AW |
185 | } |
186 | } | |
187 | ||
188 | /* | |
189 | * Only used by the i386 NUMA architecures, but relatively | |
190 | * generic code. | |
191 | */ | |
192 | unsigned long __init node_memmap_size_bytes(int nid, unsigned long start_pfn, | |
193 | unsigned long end_pfn) | |
194 | { | |
195 | unsigned long pfn; | |
196 | unsigned long nr_pages = 0; | |
197 | ||
2dbb51c4 | 198 | mminit_validate_memmodel_limits(&start_pfn, &end_pfn); |
d41dee36 AW |
199 | for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) { |
200 | if (nid != early_pfn_to_nid(pfn)) | |
201 | continue; | |
202 | ||
540557b9 | 203 | if (pfn_present(pfn)) |
d41dee36 AW |
204 | nr_pages += PAGES_PER_SECTION; |
205 | } | |
206 | ||
207 | return nr_pages * sizeof(struct page); | |
208 | } | |
209 | ||
29751f69 AW |
210 | /* |
211 | * Subtle, we encode the real pfn into the mem_map such that | |
212 | * the identity pfn - section_mem_map will return the actual | |
213 | * physical page frame number. | |
214 | */ | |
215 | static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum) | |
216 | { | |
217 | return (unsigned long)(mem_map - (section_nr_to_pfn(pnum))); | |
218 | } | |
219 | ||
220 | /* | |
ea01ea93 | 221 | * Decode mem_map from the coded memmap |
29751f69 | 222 | */ |
29751f69 AW |
223 | struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum) |
224 | { | |
ea01ea93 BP |
225 | /* mask off the extra low bits of information */ |
226 | coded_mem_map &= SECTION_MAP_MASK; | |
29751f69 AW |
227 | return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum); |
228 | } | |
229 | ||
a3142c8e | 230 | static int __meminit sparse_init_one_section(struct mem_section *ms, |
5c0e3066 MG |
231 | unsigned long pnum, struct page *mem_map, |
232 | unsigned long *pageblock_bitmap) | |
29751f69 | 233 | { |
540557b9 | 234 | if (!present_section(ms)) |
29751f69 AW |
235 | return -EINVAL; |
236 | ||
30c253e6 | 237 | ms->section_mem_map &= ~SECTION_MAP_MASK; |
540557b9 AW |
238 | ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum) | |
239 | SECTION_HAS_MEM_MAP; | |
5c0e3066 | 240 | ms->pageblock_flags = pageblock_bitmap; |
29751f69 AW |
241 | |
242 | return 1; | |
243 | } | |
244 | ||
04753278 | 245 | unsigned long usemap_size(void) |
5c0e3066 MG |
246 | { |
247 | unsigned long size_bytes; | |
248 | size_bytes = roundup(SECTION_BLOCKFLAGS_BITS, 8) / 8; | |
249 | size_bytes = roundup(size_bytes, sizeof(unsigned long)); | |
250 | return size_bytes; | |
251 | } | |
252 | ||
253 | #ifdef CONFIG_MEMORY_HOTPLUG | |
254 | static unsigned long *__kmalloc_section_usemap(void) | |
255 | { | |
256 | return kmalloc(usemap_size(), GFP_KERNEL); | |
257 | } | |
258 | #endif /* CONFIG_MEMORY_HOTPLUG */ | |
259 | ||
48c90682 YG |
260 | #ifdef CONFIG_MEMORY_HOTREMOVE |
261 | static unsigned long * __init | |
a4322e1b | 262 | sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat, |
238305bb | 263 | unsigned long size) |
48c90682 | 264 | { |
99ab7b19 YL |
265 | unsigned long goal, limit; |
266 | unsigned long *p; | |
267 | int nid; | |
48c90682 YG |
268 | /* |
269 | * A page may contain usemaps for other sections preventing the | |
270 | * page being freed and making a section unremovable while | |
271 | * other sections referencing the usemap retmain active. Similarly, | |
272 | * a pgdat can prevent a section being removed. If section A | |
273 | * contains a pgdat and section B contains the usemap, both | |
274 | * sections become inter-dependent. This allocates usemaps | |
275 | * from the same section as the pgdat where possible to avoid | |
276 | * this problem. | |
277 | */ | |
07b4e2bc | 278 | goal = __pa(pgdat) & (PAGE_SECTION_MASK << PAGE_SHIFT); |
99ab7b19 YL |
279 | limit = goal + (1UL << PA_SECTION_SHIFT); |
280 | nid = early_pfn_to_nid(goal >> PAGE_SHIFT); | |
281 | again: | |
282 | p = ___alloc_bootmem_node_nopanic(NODE_DATA(nid), size, | |
283 | SMP_CACHE_BYTES, goal, limit); | |
284 | if (!p && limit) { | |
285 | limit = 0; | |
286 | goto again; | |
287 | } | |
288 | return p; | |
48c90682 YG |
289 | } |
290 | ||
291 | static void __init check_usemap_section_nr(int nid, unsigned long *usemap) | |
292 | { | |
293 | unsigned long usemap_snr, pgdat_snr; | |
294 | static unsigned long old_usemap_snr = NR_MEM_SECTIONS; | |
295 | static unsigned long old_pgdat_snr = NR_MEM_SECTIONS; | |
296 | struct pglist_data *pgdat = NODE_DATA(nid); | |
297 | int usemap_nid; | |
298 | ||
299 | usemap_snr = pfn_to_section_nr(__pa(usemap) >> PAGE_SHIFT); | |
300 | pgdat_snr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT); | |
301 | if (usemap_snr == pgdat_snr) | |
302 | return; | |
303 | ||
304 | if (old_usemap_snr == usemap_snr && old_pgdat_snr == pgdat_snr) | |
305 | /* skip redundant message */ | |
306 | return; | |
307 | ||
308 | old_usemap_snr = usemap_snr; | |
309 | old_pgdat_snr = pgdat_snr; | |
310 | ||
311 | usemap_nid = sparse_early_nid(__nr_to_section(usemap_snr)); | |
312 | if (usemap_nid != nid) { | |
313 | printk(KERN_INFO | |
314 | "node %d must be removed before remove section %ld\n", | |
315 | nid, usemap_snr); | |
316 | return; | |
317 | } | |
318 | /* | |
319 | * There is a circular dependency. | |
320 | * Some platforms allow un-removable section because they will just | |
321 | * gather other removable sections for dynamic partitioning. | |
322 | * Just notify un-removable section's number here. | |
323 | */ | |
324 | printk(KERN_INFO "Section %ld and %ld (node %d)", usemap_snr, | |
325 | pgdat_snr, nid); | |
326 | printk(KERN_CONT | |
327 | " have a circular dependency on usemap and pgdat allocations\n"); | |
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 | { |
238305bb | 334 | return alloc_bootmem_node_nopanic(pgdat, size); |
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 | ||
a4322e1b YL |
342 | static void __init sparse_early_usemaps_alloc_node(unsigned long**usemap_map, |
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; | |
349 | int size = usemap_size(); | |
5c0e3066 | 350 | |
a4322e1b | 351 | usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid), |
238305bb | 352 | size * usemap_count); |
f5bf18fa | 353 | if (!usemap) { |
238305bb JW |
354 | printk(KERN_WARNING "%s: allocation failed\n", __func__); |
355 | return; | |
48c90682 YG |
356 | } |
357 | ||
f5bf18fa NA |
358 | for (pnum = pnum_begin; pnum < pnum_end; pnum++) { |
359 | if (!present_section_nr(pnum)) | |
360 | continue; | |
361 | usemap_map[pnum] = usemap; | |
362 | usemap += size; | |
363 | check_usemap_section_nr(nodeid, usemap_map[pnum]); | |
a4322e1b | 364 | } |
5c0e3066 MG |
365 | } |
366 | ||
8f6aac41 | 367 | #ifndef CONFIG_SPARSEMEM_VMEMMAP |
98f3cfc1 | 368 | struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid) |
29751f69 AW |
369 | { |
370 | struct page *map; | |
e48e67e0 | 371 | unsigned long size; |
29751f69 AW |
372 | |
373 | map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION); | |
374 | if (map) | |
375 | return map; | |
376 | ||
e48e67e0 YL |
377 | size = PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION); |
378 | map = __alloc_bootmem_node_high(NODE_DATA(nid), size, | |
379 | PAGE_SIZE, __pa(MAX_DMA_ADDRESS)); | |
8f6aac41 CL |
380 | return map; |
381 | } | |
9bdac914 YL |
382 | void __init sparse_mem_maps_populate_node(struct page **map_map, |
383 | unsigned long pnum_begin, | |
384 | unsigned long pnum_end, | |
385 | unsigned long map_count, int nodeid) | |
386 | { | |
387 | void *map; | |
388 | unsigned long pnum; | |
389 | unsigned long size = sizeof(struct page) * PAGES_PER_SECTION; | |
390 | ||
391 | map = alloc_remap(nodeid, size * map_count); | |
392 | if (map) { | |
393 | for (pnum = pnum_begin; pnum < pnum_end; pnum++) { | |
394 | if (!present_section_nr(pnum)) | |
395 | continue; | |
396 | map_map[pnum] = map; | |
397 | map += size; | |
398 | } | |
399 | return; | |
400 | } | |
401 | ||
402 | size = PAGE_ALIGN(size); | |
e48e67e0 YL |
403 | map = __alloc_bootmem_node_high(NODE_DATA(nodeid), size * map_count, |
404 | PAGE_SIZE, __pa(MAX_DMA_ADDRESS)); | |
9bdac914 YL |
405 | if (map) { |
406 | for (pnum = pnum_begin; pnum < pnum_end; pnum++) { | |
407 | if (!present_section_nr(pnum)) | |
408 | continue; | |
409 | map_map[pnum] = map; | |
410 | map += size; | |
411 | } | |
412 | return; | |
413 | } | |
414 | ||
415 | /* fallback */ | |
416 | for (pnum = pnum_begin; pnum < pnum_end; pnum++) { | |
417 | struct mem_section *ms; | |
418 | ||
419 | if (!present_section_nr(pnum)) | |
420 | continue; | |
421 | map_map[pnum] = sparse_mem_map_populate(pnum, nodeid); | |
422 | if (map_map[pnum]) | |
423 | continue; | |
424 | ms = __nr_to_section(pnum); | |
425 | printk(KERN_ERR "%s: sparsemem memory map backing failed " | |
426 | "some memory will not be available.\n", __func__); | |
427 | ms->section_mem_map = 0; | |
428 | } | |
429 | } | |
8f6aac41 CL |
430 | #endif /* !CONFIG_SPARSEMEM_VMEMMAP */ |
431 | ||
81d0d950 | 432 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
9bdac914 YL |
433 | static void __init sparse_early_mem_maps_alloc_node(struct page **map_map, |
434 | unsigned long pnum_begin, | |
435 | unsigned long pnum_end, | |
436 | unsigned long map_count, int nodeid) | |
437 | { | |
438 | sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end, | |
439 | map_count, nodeid); | |
440 | } | |
81d0d950 | 441 | #else |
9e5c6da7 | 442 | static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum) |
8f6aac41 CL |
443 | { |
444 | struct page *map; | |
445 | struct mem_section *ms = __nr_to_section(pnum); | |
446 | int nid = sparse_early_nid(ms); | |
447 | ||
98f3cfc1 | 448 | map = sparse_mem_map_populate(pnum, nid); |
29751f69 AW |
449 | if (map) |
450 | return map; | |
451 | ||
8f6aac41 | 452 | printk(KERN_ERR "%s: sparsemem memory map backing failed " |
d40cee24 | 453 | "some memory will not be available.\n", __func__); |
802f192e | 454 | ms->section_mem_map = 0; |
29751f69 AW |
455 | return NULL; |
456 | } | |
9bdac914 | 457 | #endif |
29751f69 | 458 | |
c2b91e2e YL |
459 | void __attribute__((weak)) __meminit vmemmap_populate_print_last(void) |
460 | { | |
461 | } | |
a4322e1b | 462 | |
193faea9 SR |
463 | /* |
464 | * Allocate the accumulated non-linear sections, allocate a mem_map | |
465 | * for each and record the physical to section mapping. | |
466 | */ | |
467 | void __init sparse_init(void) | |
468 | { | |
469 | unsigned long pnum; | |
470 | struct page *map; | |
5c0e3066 | 471 | unsigned long *usemap; |
e123dd3f | 472 | unsigned long **usemap_map; |
81d0d950 | 473 | int size; |
a4322e1b YL |
474 | int nodeid_begin = 0; |
475 | unsigned long pnum_begin = 0; | |
476 | unsigned long usemap_count; | |
81d0d950 | 477 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
9bdac914 | 478 | unsigned long map_count; |
81d0d950 YL |
479 | int size2; |
480 | struct page **map_map; | |
481 | #endif | |
e123dd3f | 482 | |
55878e88 CS |
483 | /* see include/linux/mmzone.h 'struct mem_section' definition */ |
484 | BUILD_BUG_ON(!is_power_of_2(sizeof(struct mem_section))); | |
485 | ||
ca57df79 XQ |
486 | /* Setup pageblock_order for HUGETLB_PAGE_SIZE_VARIABLE */ |
487 | set_pageblock_order(); | |
488 | ||
e123dd3f YL |
489 | /* |
490 | * map is using big page (aka 2M in x86 64 bit) | |
491 | * usemap is less one page (aka 24 bytes) | |
492 | * so alloc 2M (with 2M align) and 24 bytes in turn will | |
493 | * make next 2M slip to one more 2M later. | |
494 | * then in big system, the memory will have a lot of holes... | |
25985edc | 495 | * here try to allocate 2M pages continuously. |
e123dd3f YL |
496 | * |
497 | * powerpc need to call sparse_init_one_section right after each | |
498 | * sparse_early_mem_map_alloc, so allocate usemap_map at first. | |
499 | */ | |
500 | size = sizeof(unsigned long *) * NR_MEM_SECTIONS; | |
501 | usemap_map = alloc_bootmem(size); | |
502 | if (!usemap_map) | |
503 | panic("can not allocate usemap_map\n"); | |
193faea9 SR |
504 | |
505 | for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) { | |
a4322e1b YL |
506 | struct mem_section *ms; |
507 | ||
540557b9 | 508 | if (!present_section_nr(pnum)) |
193faea9 | 509 | continue; |
a4322e1b YL |
510 | ms = __nr_to_section(pnum); |
511 | nodeid_begin = sparse_early_nid(ms); | |
512 | pnum_begin = pnum; | |
513 | break; | |
514 | } | |
515 | usemap_count = 1; | |
516 | for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) { | |
517 | struct mem_section *ms; | |
518 | int nodeid; | |
519 | ||
520 | if (!present_section_nr(pnum)) | |
521 | continue; | |
522 | ms = __nr_to_section(pnum); | |
523 | nodeid = sparse_early_nid(ms); | |
524 | if (nodeid == nodeid_begin) { | |
525 | usemap_count++; | |
526 | continue; | |
527 | } | |
528 | /* ok, we need to take cake of from pnum_begin to pnum - 1*/ | |
529 | sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, pnum, | |
530 | usemap_count, nodeid_begin); | |
531 | /* new start, update count etc*/ | |
532 | nodeid_begin = nodeid; | |
533 | pnum_begin = pnum; | |
534 | usemap_count = 1; | |
e123dd3f | 535 | } |
a4322e1b YL |
536 | /* ok, last chunk */ |
537 | sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, NR_MEM_SECTIONS, | |
538 | usemap_count, nodeid_begin); | |
193faea9 | 539 | |
9bdac914 YL |
540 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
541 | size2 = sizeof(struct page *) * NR_MEM_SECTIONS; | |
542 | map_map = alloc_bootmem(size2); | |
543 | if (!map_map) | |
544 | panic("can not allocate map_map\n"); | |
545 | ||
546 | for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) { | |
547 | struct mem_section *ms; | |
548 | ||
549 | if (!present_section_nr(pnum)) | |
550 | continue; | |
551 | ms = __nr_to_section(pnum); | |
552 | nodeid_begin = sparse_early_nid(ms); | |
553 | pnum_begin = pnum; | |
554 | break; | |
555 | } | |
556 | map_count = 1; | |
557 | for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) { | |
558 | struct mem_section *ms; | |
559 | int nodeid; | |
560 | ||
561 | if (!present_section_nr(pnum)) | |
562 | continue; | |
563 | ms = __nr_to_section(pnum); | |
564 | nodeid = sparse_early_nid(ms); | |
565 | if (nodeid == nodeid_begin) { | |
566 | map_count++; | |
567 | continue; | |
568 | } | |
569 | /* ok, we need to take cake of from pnum_begin to pnum - 1*/ | |
570 | sparse_early_mem_maps_alloc_node(map_map, pnum_begin, pnum, | |
571 | map_count, nodeid_begin); | |
572 | /* new start, update count etc*/ | |
573 | nodeid_begin = nodeid; | |
574 | pnum_begin = pnum; | |
575 | map_count = 1; | |
576 | } | |
577 | /* ok, last chunk */ | |
578 | sparse_early_mem_maps_alloc_node(map_map, pnum_begin, NR_MEM_SECTIONS, | |
579 | map_count, nodeid_begin); | |
580 | #endif | |
581 | ||
e123dd3f YL |
582 | for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) { |
583 | if (!present_section_nr(pnum)) | |
193faea9 | 584 | continue; |
5c0e3066 | 585 | |
e123dd3f | 586 | usemap = usemap_map[pnum]; |
5c0e3066 MG |
587 | if (!usemap) |
588 | continue; | |
589 | ||
9bdac914 YL |
590 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
591 | map = map_map[pnum]; | |
592 | #else | |
e123dd3f | 593 | map = sparse_early_mem_map_alloc(pnum); |
9bdac914 | 594 | #endif |
e123dd3f YL |
595 | if (!map) |
596 | continue; | |
597 | ||
5c0e3066 MG |
598 | sparse_init_one_section(__nr_to_section(pnum), pnum, map, |
599 | usemap); | |
193faea9 | 600 | } |
e123dd3f | 601 | |
c2b91e2e YL |
602 | vmemmap_populate_print_last(); |
603 | ||
9bdac914 YL |
604 | #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER |
605 | free_bootmem(__pa(map_map), size2); | |
606 | #endif | |
e123dd3f | 607 | free_bootmem(__pa(usemap_map), size); |
193faea9 SR |
608 | } |
609 | ||
610 | #ifdef CONFIG_MEMORY_HOTPLUG | |
98f3cfc1 YG |
611 | #ifdef CONFIG_SPARSEMEM_VMEMMAP |
612 | static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid, | |
613 | unsigned long nr_pages) | |
614 | { | |
615 | /* This will make the necessary allocations eventually. */ | |
616 | return sparse_mem_map_populate(pnum, nid); | |
617 | } | |
618 | static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages) | |
619 | { | |
0aad818b JW |
620 | unsigned long start = (unsigned long)memmap; |
621 | unsigned long end = (unsigned long)(memmap + nr_pages); | |
622 | ||
623 | vmemmap_free(start, end); | |
98f3cfc1 | 624 | } |
4edd7cef | 625 | #ifdef CONFIG_MEMORY_HOTREMOVE |
ae64ffca | 626 | static void free_map_bootmem(struct page *memmap, unsigned long nr_pages) |
0c0a4a51 | 627 | { |
0aad818b JW |
628 | unsigned long start = (unsigned long)memmap; |
629 | unsigned long end = (unsigned long)(memmap + nr_pages); | |
630 | ||
631 | vmemmap_free(start, end); | |
0c0a4a51 | 632 | } |
4edd7cef | 633 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
98f3cfc1 | 634 | #else |
0b0acbec DH |
635 | static struct page *__kmalloc_section_memmap(unsigned long nr_pages) |
636 | { | |
637 | struct page *page, *ret; | |
638 | unsigned long memmap_size = sizeof(struct page) * nr_pages; | |
639 | ||
f2d0aa5b | 640 | page = alloc_pages(GFP_KERNEL|__GFP_NOWARN, get_order(memmap_size)); |
0b0acbec DH |
641 | if (page) |
642 | goto got_map_page; | |
643 | ||
644 | ret = vmalloc(memmap_size); | |
645 | if (ret) | |
646 | goto got_map_ptr; | |
647 | ||
648 | return NULL; | |
649 | got_map_page: | |
650 | ret = (struct page *)pfn_to_kaddr(page_to_pfn(page)); | |
651 | got_map_ptr: | |
0b0acbec DH |
652 | |
653 | return ret; | |
654 | } | |
655 | ||
98f3cfc1 YG |
656 | static inline struct page *kmalloc_section_memmap(unsigned long pnum, int nid, |
657 | unsigned long nr_pages) | |
658 | { | |
659 | return __kmalloc_section_memmap(nr_pages); | |
660 | } | |
661 | ||
0b0acbec DH |
662 | static void __kfree_section_memmap(struct page *memmap, unsigned long nr_pages) |
663 | { | |
9e2779fa | 664 | if (is_vmalloc_addr(memmap)) |
0b0acbec DH |
665 | vfree(memmap); |
666 | else | |
667 | free_pages((unsigned long)memmap, | |
668 | get_order(sizeof(struct page) * nr_pages)); | |
669 | } | |
0c0a4a51 | 670 | |
4edd7cef | 671 | #ifdef CONFIG_MEMORY_HOTREMOVE |
ae64ffca | 672 | static void free_map_bootmem(struct page *memmap, unsigned long nr_pages) |
0c0a4a51 YG |
673 | { |
674 | unsigned long maps_section_nr, removing_section_nr, i; | |
5f24ce5f | 675 | unsigned long magic; |
ae64ffca | 676 | struct page *page = virt_to_page(memmap); |
0c0a4a51 YG |
677 | |
678 | for (i = 0; i < nr_pages; i++, page++) { | |
5f24ce5f | 679 | magic = (unsigned long) page->lru.next; |
0c0a4a51 YG |
680 | |
681 | BUG_ON(magic == NODE_INFO); | |
682 | ||
683 | maps_section_nr = pfn_to_section_nr(page_to_pfn(page)); | |
684 | removing_section_nr = page->private; | |
685 | ||
686 | /* | |
687 | * When this function is called, the removing section is | |
688 | * logical offlined state. This means all pages are isolated | |
689 | * from page allocator. If removing section's memmap is placed | |
690 | * on the same section, it must not be freed. | |
691 | * If it is freed, page allocator may allocate it which will | |
692 | * be removed physically soon. | |
693 | */ | |
694 | if (maps_section_nr != removing_section_nr) | |
695 | put_page_bootmem(page); | |
696 | } | |
697 | } | |
4edd7cef | 698 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
98f3cfc1 | 699 | #endif /* CONFIG_SPARSEMEM_VMEMMAP */ |
0b0acbec | 700 | |
29751f69 AW |
701 | /* |
702 | * returns the number of sections whose mem_maps were properly | |
703 | * set. If this is <=0, then that means that the passed-in | |
704 | * map was not consumed and must be freed. | |
705 | */ | |
31168481 | 706 | int __meminit sparse_add_one_section(struct zone *zone, unsigned long start_pfn, |
0b0acbec | 707 | int nr_pages) |
29751f69 | 708 | { |
0b0acbec DH |
709 | unsigned long section_nr = pfn_to_section_nr(start_pfn); |
710 | struct pglist_data *pgdat = zone->zone_pgdat; | |
711 | struct mem_section *ms; | |
712 | struct page *memmap; | |
5c0e3066 | 713 | unsigned long *usemap; |
0b0acbec DH |
714 | unsigned long flags; |
715 | int ret; | |
29751f69 | 716 | |
0b0acbec DH |
717 | /* |
718 | * no locking for this, because it does its own | |
719 | * plus, it does a kmalloc | |
720 | */ | |
bbd06825 WC |
721 | ret = sparse_index_init(section_nr, pgdat->node_id); |
722 | if (ret < 0 && ret != -EEXIST) | |
723 | return ret; | |
98f3cfc1 | 724 | memmap = kmalloc_section_memmap(section_nr, pgdat->node_id, nr_pages); |
bbd06825 WC |
725 | if (!memmap) |
726 | return -ENOMEM; | |
5c0e3066 | 727 | usemap = __kmalloc_section_usemap(); |
bbd06825 WC |
728 | if (!usemap) { |
729 | __kfree_section_memmap(memmap, nr_pages); | |
730 | return -ENOMEM; | |
731 | } | |
0b0acbec DH |
732 | |
733 | pgdat_resize_lock(pgdat, &flags); | |
29751f69 | 734 | |
0b0acbec DH |
735 | ms = __pfn_to_section(start_pfn); |
736 | if (ms->section_mem_map & SECTION_MARKED_PRESENT) { | |
737 | ret = -EEXIST; | |
738 | goto out; | |
739 | } | |
5c0e3066 | 740 | |
3ac19f8e WC |
741 | memset(memmap, 0, sizeof(struct page) * nr_pages); |
742 | ||
29751f69 AW |
743 | ms->section_mem_map |= SECTION_MARKED_PRESENT; |
744 | ||
5c0e3066 | 745 | ret = sparse_init_one_section(ms, section_nr, memmap, usemap); |
0b0acbec | 746 | |
0b0acbec DH |
747 | out: |
748 | pgdat_resize_unlock(pgdat, &flags); | |
bbd06825 WC |
749 | if (ret <= 0) { |
750 | kfree(usemap); | |
46a66eec | 751 | __kfree_section_memmap(memmap, nr_pages); |
bbd06825 | 752 | } |
0b0acbec | 753 | return ret; |
29751f69 | 754 | } |
ea01ea93 | 755 | |
f3deb687 | 756 | #ifdef CONFIG_MEMORY_HOTREMOVE |
95a4774d WC |
757 | #ifdef CONFIG_MEMORY_FAILURE |
758 | static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages) | |
759 | { | |
760 | int i; | |
761 | ||
762 | if (!memmap) | |
763 | return; | |
764 | ||
765 | for (i = 0; i < PAGES_PER_SECTION; i++) { | |
766 | if (PageHWPoison(&memmap[i])) { | |
293c07e3 | 767 | atomic_long_sub(1, &num_poisoned_pages); |
95a4774d WC |
768 | ClearPageHWPoison(&memmap[i]); |
769 | } | |
770 | } | |
771 | } | |
772 | #else | |
773 | static inline void clear_hwpoisoned_pages(struct page *memmap, int nr_pages) | |
774 | { | |
775 | } | |
776 | #endif | |
777 | ||
4edd7cef DR |
778 | static void free_section_usemap(struct page *memmap, unsigned long *usemap) |
779 | { | |
780 | struct page *usemap_page; | |
781 | unsigned long nr_pages; | |
782 | ||
783 | if (!usemap) | |
784 | return; | |
785 | ||
786 | usemap_page = virt_to_page(usemap); | |
787 | /* | |
788 | * Check to see if allocation came from hot-plug-add | |
789 | */ | |
790 | if (PageSlab(usemap_page) || PageCompound(usemap_page)) { | |
791 | kfree(usemap); | |
792 | if (memmap) | |
793 | __kfree_section_memmap(memmap, PAGES_PER_SECTION); | |
794 | return; | |
795 | } | |
796 | ||
797 | /* | |
798 | * The usemap came from bootmem. This is packed with other usemaps | |
799 | * on the section which has pgdat at boot time. Just keep it as is now. | |
800 | */ | |
801 | ||
802 | if (memmap) { | |
803 | nr_pages = PAGE_ALIGN(PAGES_PER_SECTION * sizeof(struct page)) | |
804 | >> PAGE_SHIFT; | |
805 | ||
806 | free_map_bootmem(memmap, nr_pages); | |
807 | } | |
808 | } | |
809 | ||
ea01ea93 BP |
810 | void sparse_remove_one_section(struct zone *zone, struct mem_section *ms) |
811 | { | |
812 | struct page *memmap = NULL; | |
cd099682 TC |
813 | unsigned long *usemap = NULL, flags; |
814 | struct pglist_data *pgdat = zone->zone_pgdat; | |
ea01ea93 | 815 | |
cd099682 | 816 | pgdat_resize_lock(pgdat, &flags); |
ea01ea93 BP |
817 | if (ms->section_mem_map) { |
818 | usemap = ms->pageblock_flags; | |
819 | memmap = sparse_decode_mem_map(ms->section_mem_map, | |
820 | __section_nr(ms)); | |
821 | ms->section_mem_map = 0; | |
822 | ms->pageblock_flags = NULL; | |
823 | } | |
cd099682 | 824 | pgdat_resize_unlock(pgdat, &flags); |
ea01ea93 | 825 | |
95a4774d | 826 | clear_hwpoisoned_pages(memmap, PAGES_PER_SECTION); |
ea01ea93 BP |
827 | free_section_usemap(memmap, usemap); |
828 | } | |
4edd7cef DR |
829 | #endif /* CONFIG_MEMORY_HOTREMOVE */ |
830 | #endif /* CONFIG_MEMORY_HOTPLUG */ |