Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
57cfc29e | 2 | * bootmem - A boot-time physical memory allocator and configurator |
1da177e4 LT |
3 | * |
4 | * Copyright (C) 1999 Ingo Molnar | |
57cfc29e JW |
5 | * 1999 Kanoj Sarcar, SGI |
6 | * 2008 Johannes Weiner | |
1da177e4 | 7 | * |
57cfc29e JW |
8 | * Access to this subsystem has to be serialized externally (which is true |
9 | * for the boot process anyway). | |
1da177e4 | 10 | */ |
1da177e4 | 11 | #include <linux/init.h> |
bbc7b92e | 12 | #include <linux/pfn.h> |
5a0e3ad6 | 13 | #include <linux/slab.h> |
1da177e4 | 14 | #include <linux/bootmem.h> |
b95f1b31 | 15 | #include <linux/export.h> |
ec3a354b | 16 | #include <linux/kmemleak.h> |
08677214 | 17 | #include <linux/range.h> |
72d7c3b3 | 18 | #include <linux/memblock.h> |
d85fbee8 PM |
19 | #include <linux/bug.h> |
20 | #include <linux/io.h> | |
e786e86a | 21 | |
dfd54cbc | 22 | #include <asm/processor.h> |
e786e86a | 23 | |
1da177e4 LT |
24 | #include "internal.h" |
25 | ||
e782ab42 YL |
26 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
27 | struct pglist_data __refdata contig_page_data = { | |
28 | .bdata = &bootmem_node_data[0] | |
29 | }; | |
30 | EXPORT_SYMBOL(contig_page_data); | |
31 | #endif | |
32 | ||
1da177e4 LT |
33 | unsigned long max_low_pfn; |
34 | unsigned long min_low_pfn; | |
35 | unsigned long max_pfn; | |
8dd33030 | 36 | unsigned long long max_possible_pfn; |
1da177e4 | 37 | |
b61bfa3c JW |
38 | bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata; |
39 | ||
636cc40c JW |
40 | static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list); |
41 | ||
2e5237da JW |
42 | static int bootmem_debug; |
43 | ||
44 | static int __init bootmem_debug_setup(char *buf) | |
45 | { | |
46 | bootmem_debug = 1; | |
47 | return 0; | |
48 | } | |
49 | early_param("bootmem_debug", bootmem_debug_setup); | |
50 | ||
51 | #define bdebug(fmt, args...) ({ \ | |
52 | if (unlikely(bootmem_debug)) \ | |
1170532b | 53 | pr_info("bootmem::%s " fmt, \ |
80a914dc | 54 | __func__, ## args); \ |
2e5237da JW |
55 | }) |
56 | ||
df049a5f | 57 | static unsigned long __init bootmap_bytes(unsigned long pages) |
223e8dc9 | 58 | { |
9571a982 | 59 | unsigned long bytes = DIV_ROUND_UP(pages, 8); |
223e8dc9 | 60 | |
df049a5f | 61 | return ALIGN(bytes, sizeof(long)); |
223e8dc9 JW |
62 | } |
63 | ||
a66fd7da JW |
64 | /** |
65 | * bootmem_bootmap_pages - calculate bitmap size in pages | |
66 | * @pages: number of pages the bitmap has to represent | |
67 | */ | |
f71bf0ca | 68 | unsigned long __init bootmem_bootmap_pages(unsigned long pages) |
1da177e4 | 69 | { |
df049a5f | 70 | unsigned long bytes = bootmap_bytes(pages); |
1da177e4 | 71 | |
df049a5f | 72 | return PAGE_ALIGN(bytes) >> PAGE_SHIFT; |
1da177e4 | 73 | } |
f71bf0ca | 74 | |
679bc9fb KH |
75 | /* |
76 | * link bdata in order | |
77 | */ | |
69d49e68 | 78 | static void __init link_bootmem(bootmem_data_t *bdata) |
679bc9fb | 79 | { |
5c2b8a16 | 80 | bootmem_data_t *ent; |
f71bf0ca | 81 | |
5c2b8a16 GS |
82 | list_for_each_entry(ent, &bdata_list, list) { |
83 | if (bdata->node_min_pfn < ent->node_min_pfn) { | |
84 | list_add_tail(&bdata->list, &ent->list); | |
85 | return; | |
86 | } | |
679bc9fb | 87 | } |
5c2b8a16 GS |
88 | |
89 | list_add_tail(&bdata->list, &bdata_list); | |
679bc9fb KH |
90 | } |
91 | ||
1da177e4 LT |
92 | /* |
93 | * Called once to set up the allocator itself. | |
94 | */ | |
8ae04463 | 95 | static unsigned long __init init_bootmem_core(bootmem_data_t *bdata, |
1da177e4 LT |
96 | unsigned long mapstart, unsigned long start, unsigned long end) |
97 | { | |
bbc7b92e | 98 | unsigned long mapsize; |
1da177e4 | 99 | |
2dbb51c4 | 100 | mminit_validate_memmodel_limits(&start, &end); |
bbc7b92e | 101 | bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart)); |
3560e249 | 102 | bdata->node_min_pfn = start; |
1da177e4 | 103 | bdata->node_low_pfn = end; |
679bc9fb | 104 | link_bootmem(bdata); |
1da177e4 LT |
105 | |
106 | /* | |
107 | * Initially all pages are reserved - setup_arch() has to | |
108 | * register free RAM areas explicitly. | |
109 | */ | |
df049a5f | 110 | mapsize = bootmap_bytes(end - start); |
1da177e4 LT |
111 | memset(bdata->node_bootmem_map, 0xff, mapsize); |
112 | ||
2e5237da JW |
113 | bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n", |
114 | bdata - bootmem_node_data, start, mapstart, end, mapsize); | |
115 | ||
1da177e4 LT |
116 | return mapsize; |
117 | } | |
118 | ||
a66fd7da JW |
119 | /** |
120 | * init_bootmem_node - register a node as boot memory | |
121 | * @pgdat: node to register | |
122 | * @freepfn: pfn where the bitmap for this node is to be placed | |
123 | * @startpfn: first pfn on the node | |
124 | * @endpfn: first pfn after the node | |
125 | * | |
126 | * Returns the number of bytes needed to hold the bitmap for this node. | |
127 | */ | |
223e8dc9 JW |
128 | unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn, |
129 | unsigned long startpfn, unsigned long endpfn) | |
130 | { | |
131 | return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn); | |
132 | } | |
133 | ||
a66fd7da JW |
134 | /** |
135 | * init_bootmem - register boot memory | |
136 | * @start: pfn where the bitmap is to be placed | |
137 | * @pages: number of available physical pages | |
138 | * | |
139 | * Returns the number of bytes needed to hold the bitmap. | |
140 | */ | |
223e8dc9 JW |
141 | unsigned long __init init_bootmem(unsigned long start, unsigned long pages) |
142 | { | |
143 | max_low_pfn = pages; | |
144 | min_low_pfn = start; | |
145 | return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages); | |
146 | } | |
09325873 | 147 | |
9f993ac3 FT |
148 | /* |
149 | * free_bootmem_late - free bootmem pages directly to page allocator | |
81df9bff | 150 | * @addr: starting physical address of the range |
9f993ac3 FT |
151 | * @size: size of the range in bytes |
152 | * | |
153 | * This is only useful when the bootmem allocator has already been torn | |
154 | * down, but we are still initializing the system. Pages are given directly | |
155 | * to the page allocator, no bootmem metadata is updated because it is gone. | |
156 | */ | |
81df9bff | 157 | void __init free_bootmem_late(unsigned long physaddr, unsigned long size) |
9f993ac3 FT |
158 | { |
159 | unsigned long cursor, end; | |
160 | ||
81df9bff | 161 | kmemleak_free_part(__va(physaddr), size); |
9f993ac3 | 162 | |
81df9bff JK |
163 | cursor = PFN_UP(physaddr); |
164 | end = PFN_DOWN(physaddr + size); | |
9f993ac3 FT |
165 | |
166 | for (; cursor < end; cursor++) { | |
d70ddd7a | 167 | __free_pages_bootmem(pfn_to_page(cursor), cursor, 0); |
9f993ac3 FT |
168 | totalram_pages++; |
169 | } | |
170 | } | |
171 | ||
223e8dc9 JW |
172 | static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) |
173 | { | |
174 | struct page *page; | |
d70ddd7a | 175 | unsigned long *map, start, end, pages, cur, count = 0; |
41546c17 JW |
176 | |
177 | if (!bdata->node_bootmem_map) | |
178 | return 0; | |
179 | ||
4a099fb4 | 180 | map = bdata->node_bootmem_map; |
3560e249 | 181 | start = bdata->node_min_pfn; |
41546c17 JW |
182 | end = bdata->node_low_pfn; |
183 | ||
799f933a JW |
184 | bdebug("nid=%td start=%lx end=%lx\n", |
185 | bdata - bootmem_node_data, start, end); | |
223e8dc9 | 186 | |
41546c17 | 187 | while (start < end) { |
4a099fb4 | 188 | unsigned long idx, vec; |
10d73e65 | 189 | unsigned shift; |
223e8dc9 | 190 | |
3560e249 | 191 | idx = start - bdata->node_min_pfn; |
10d73e65 MF |
192 | shift = idx & (BITS_PER_LONG - 1); |
193 | /* | |
194 | * vec holds at most BITS_PER_LONG map bits, | |
195 | * bit 0 corresponds to start. | |
196 | */ | |
41546c17 | 197 | vec = ~map[idx / BITS_PER_LONG]; |
10d73e65 MF |
198 | |
199 | if (shift) { | |
200 | vec >>= shift; | |
201 | if (end - start >= BITS_PER_LONG) | |
202 | vec |= ~map[idx / BITS_PER_LONG + 1] << | |
203 | (BITS_PER_LONG - shift); | |
204 | } | |
799f933a JW |
205 | /* |
206 | * If we have a properly aligned and fully unreserved | |
207 | * BITS_PER_LONG block of pages in front of us, free | |
208 | * it in one go. | |
209 | */ | |
210 | if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) { | |
41546c17 JW |
211 | int order = ilog2(BITS_PER_LONG); |
212 | ||
d70ddd7a | 213 | __free_pages_bootmem(pfn_to_page(start), start, order); |
223e8dc9 | 214 | count += BITS_PER_LONG; |
799f933a | 215 | start += BITS_PER_LONG; |
41546c17 | 216 | } else { |
d70ddd7a | 217 | cur = start; |
41546c17 | 218 | |
10d73e65 MF |
219 | start = ALIGN(start + 1, BITS_PER_LONG); |
220 | while (vec && cur != start) { | |
41546c17 | 221 | if (vec & 1) { |
10d73e65 | 222 | page = pfn_to_page(cur); |
d70ddd7a | 223 | __free_pages_bootmem(page, cur, 0); |
41546c17 | 224 | count++; |
223e8dc9 | 225 | } |
41546c17 | 226 | vec >>= 1; |
10d73e65 | 227 | ++cur; |
223e8dc9 | 228 | } |
223e8dc9 | 229 | } |
223e8dc9 JW |
230 | } |
231 | ||
d70ddd7a | 232 | cur = bdata->node_min_pfn; |
223e8dc9 | 233 | page = virt_to_page(bdata->node_bootmem_map); |
3560e249 | 234 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
41546c17 JW |
235 | pages = bootmem_bootmap_pages(pages); |
236 | count += pages; | |
5576646f | 237 | while (pages--) |
d70ddd7a | 238 | __free_pages_bootmem(page++, cur++, 0); |
1b4ace41 | 239 | bdata->node_bootmem_map = NULL; |
223e8dc9 | 240 | |
2e5237da JW |
241 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
242 | ||
223e8dc9 JW |
243 | return count; |
244 | } | |
245 | ||
7b4b2a0d JL |
246 | static int reset_managed_pages_done __initdata; |
247 | ||
f784a3f1 | 248 | void reset_node_managed_pages(pg_data_t *pgdat) |
9feedc9d JL |
249 | { |
250 | struct zone *z; | |
251 | ||
9feedc9d | 252 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) |
7b4b2a0d JL |
253 | z->managed_pages = 0; |
254 | } | |
255 | ||
256 | void __init reset_all_zones_managed_pages(void) | |
257 | { | |
258 | struct pglist_data *pgdat; | |
259 | ||
f784a3f1 TC |
260 | if (reset_managed_pages_done) |
261 | return; | |
262 | ||
7b4b2a0d JL |
263 | for_each_online_pgdat(pgdat) |
264 | reset_node_managed_pages(pgdat); | |
f784a3f1 | 265 | |
7b4b2a0d | 266 | reset_managed_pages_done = 1; |
9feedc9d JL |
267 | } |
268 | ||
a66fd7da JW |
269 | /** |
270 | * free_all_bootmem - release free pages to the buddy allocator | |
271 | * | |
272 | * Returns the number of pages actually released. | |
273 | */ | |
223e8dc9 JW |
274 | unsigned long __init free_all_bootmem(void) |
275 | { | |
aa235fc7 YL |
276 | unsigned long total_pages = 0; |
277 | bootmem_data_t *bdata; | |
9feedc9d | 278 | |
7b4b2a0d | 279 | reset_all_zones_managed_pages(); |
aa235fc7 YL |
280 | |
281 | list_for_each_entry(bdata, &bdata_list, list) | |
282 | total_pages += free_all_bootmem_core(bdata); | |
283 | ||
0c988534 JL |
284 | totalram_pages += total_pages; |
285 | ||
aa235fc7 | 286 | return total_pages; |
223e8dc9 JW |
287 | } |
288 | ||
d747fa4b JW |
289 | static void __init __free(bootmem_data_t *bdata, |
290 | unsigned long sidx, unsigned long eidx) | |
291 | { | |
292 | unsigned long idx; | |
293 | ||
294 | bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data, | |
3560e249 JW |
295 | sidx + bdata->node_min_pfn, |
296 | eidx + bdata->node_min_pfn); | |
d747fa4b | 297 | |
1b4ace41 CM |
298 | if (WARN_ON(bdata->node_bootmem_map == NULL)) |
299 | return; | |
300 | ||
e2bf3cae JW |
301 | if (bdata->hint_idx > sidx) |
302 | bdata->hint_idx = sidx; | |
303 | ||
d747fa4b JW |
304 | for (idx = sidx; idx < eidx; idx++) |
305 | if (!test_and_clear_bit(idx, bdata->node_bootmem_map)) | |
306 | BUG(); | |
307 | } | |
308 | ||
309 | static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx, | |
310 | unsigned long eidx, int flags) | |
311 | { | |
312 | unsigned long idx; | |
313 | int exclusive = flags & BOOTMEM_EXCLUSIVE; | |
314 | ||
315 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", | |
316 | bdata - bootmem_node_data, | |
3560e249 JW |
317 | sidx + bdata->node_min_pfn, |
318 | eidx + bdata->node_min_pfn, | |
d747fa4b JW |
319 | flags); |
320 | ||
1b4ace41 CM |
321 | if (WARN_ON(bdata->node_bootmem_map == NULL)) |
322 | return 0; | |
323 | ||
d747fa4b JW |
324 | for (idx = sidx; idx < eidx; idx++) |
325 | if (test_and_set_bit(idx, bdata->node_bootmem_map)) { | |
326 | if (exclusive) { | |
327 | __free(bdata, sidx, idx); | |
328 | return -EBUSY; | |
329 | } | |
330 | bdebug("silent double reserve of PFN %lx\n", | |
3560e249 | 331 | idx + bdata->node_min_pfn); |
d747fa4b JW |
332 | } |
333 | return 0; | |
334 | } | |
335 | ||
e2bf3cae JW |
336 | static int __init mark_bootmem_node(bootmem_data_t *bdata, |
337 | unsigned long start, unsigned long end, | |
338 | int reserve, int flags) | |
223e8dc9 JW |
339 | { |
340 | unsigned long sidx, eidx; | |
223e8dc9 | 341 | |
e2bf3cae JW |
342 | bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n", |
343 | bdata - bootmem_node_data, start, end, reserve, flags); | |
223e8dc9 | 344 | |
3560e249 | 345 | BUG_ON(start < bdata->node_min_pfn); |
e2bf3cae | 346 | BUG_ON(end > bdata->node_low_pfn); |
223e8dc9 | 347 | |
3560e249 JW |
348 | sidx = start - bdata->node_min_pfn; |
349 | eidx = end - bdata->node_min_pfn; | |
223e8dc9 | 350 | |
e2bf3cae JW |
351 | if (reserve) |
352 | return __reserve(bdata, sidx, eidx, flags); | |
223e8dc9 | 353 | else |
e2bf3cae JW |
354 | __free(bdata, sidx, eidx); |
355 | return 0; | |
356 | } | |
357 | ||
358 | static int __init mark_bootmem(unsigned long start, unsigned long end, | |
359 | int reserve, int flags) | |
360 | { | |
361 | unsigned long pos; | |
362 | bootmem_data_t *bdata; | |
363 | ||
364 | pos = start; | |
365 | list_for_each_entry(bdata, &bdata_list, list) { | |
366 | int err; | |
367 | unsigned long max; | |
368 | ||
3560e249 JW |
369 | if (pos < bdata->node_min_pfn || |
370 | pos >= bdata->node_low_pfn) { | |
e2bf3cae JW |
371 | BUG_ON(pos != start); |
372 | continue; | |
373 | } | |
374 | ||
375 | max = min(bdata->node_low_pfn, end); | |
223e8dc9 | 376 | |
e2bf3cae JW |
377 | err = mark_bootmem_node(bdata, pos, max, reserve, flags); |
378 | if (reserve && err) { | |
379 | mark_bootmem(start, pos, 0, 0); | |
380 | return err; | |
381 | } | |
223e8dc9 | 382 | |
e2bf3cae JW |
383 | if (max == end) |
384 | return 0; | |
385 | pos = bdata->node_low_pfn; | |
386 | } | |
387 | BUG(); | |
223e8dc9 JW |
388 | } |
389 | ||
a66fd7da JW |
390 | /** |
391 | * free_bootmem_node - mark a page range as usable | |
392 | * @pgdat: node the range resides on | |
393 | * @physaddr: starting address of the range | |
394 | * @size: size of the range in bytes | |
395 | * | |
396 | * Partial pages will be considered reserved and left as they are. | |
397 | * | |
e2bf3cae | 398 | * The range must reside completely on the specified node. |
a66fd7da | 399 | */ |
223e8dc9 JW |
400 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
401 | unsigned long size) | |
402 | { | |
e2bf3cae JW |
403 | unsigned long start, end; |
404 | ||
ec3a354b CM |
405 | kmemleak_free_part(__va(physaddr), size); |
406 | ||
e2bf3cae JW |
407 | start = PFN_UP(physaddr); |
408 | end = PFN_DOWN(physaddr + size); | |
409 | ||
410 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); | |
223e8dc9 JW |
411 | } |
412 | ||
a66fd7da JW |
413 | /** |
414 | * free_bootmem - mark a page range as usable | |
81df9bff | 415 | * @addr: starting physical address of the range |
a66fd7da JW |
416 | * @size: size of the range in bytes |
417 | * | |
418 | * Partial pages will be considered reserved and left as they are. | |
419 | * | |
e2bf3cae | 420 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 421 | */ |
81df9bff | 422 | void __init free_bootmem(unsigned long physaddr, unsigned long size) |
223e8dc9 | 423 | { |
e2bf3cae | 424 | unsigned long start, end; |
a5645a61 | 425 | |
81df9bff | 426 | kmemleak_free_part(__va(physaddr), size); |
ec3a354b | 427 | |
81df9bff JK |
428 | start = PFN_UP(physaddr); |
429 | end = PFN_DOWN(physaddr + size); | |
1da177e4 | 430 | |
e2bf3cae | 431 | mark_bootmem(start, end, 0, 0); |
1da177e4 LT |
432 | } |
433 | ||
a66fd7da JW |
434 | /** |
435 | * reserve_bootmem_node - mark a page range as reserved | |
436 | * @pgdat: node the range resides on | |
437 | * @physaddr: starting address of the range | |
438 | * @size: size of the range in bytes | |
439 | * @flags: reservation flags (see linux/bootmem.h) | |
440 | * | |
441 | * Partial pages will be reserved. | |
442 | * | |
e2bf3cae | 443 | * The range must reside completely on the specified node. |
a66fd7da | 444 | */ |
223e8dc9 JW |
445 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
446 | unsigned long size, int flags) | |
1da177e4 | 447 | { |
e2bf3cae | 448 | unsigned long start, end; |
1da177e4 | 449 | |
e2bf3cae JW |
450 | start = PFN_DOWN(physaddr); |
451 | end = PFN_UP(physaddr + size); | |
452 | ||
453 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); | |
223e8dc9 | 454 | } |
5a982cbc | 455 | |
a66fd7da | 456 | /** |
0d4ba4d7 | 457 | * reserve_bootmem - mark a page range as reserved |
a66fd7da JW |
458 | * @addr: starting address of the range |
459 | * @size: size of the range in bytes | |
460 | * @flags: reservation flags (see linux/bootmem.h) | |
461 | * | |
462 | * Partial pages will be reserved. | |
463 | * | |
e2bf3cae | 464 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 465 | */ |
223e8dc9 JW |
466 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
467 | int flags) | |
468 | { | |
e2bf3cae | 469 | unsigned long start, end; |
1da177e4 | 470 | |
e2bf3cae JW |
471 | start = PFN_DOWN(addr); |
472 | end = PFN_UP(addr + size); | |
223e8dc9 | 473 | |
e2bf3cae | 474 | return mark_bootmem(start, end, 1, flags); |
1da177e4 LT |
475 | } |
476 | ||
8aa043d7 JB |
477 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
478 | unsigned long idx, unsigned long step) | |
481ebd0d JW |
479 | { |
480 | unsigned long base = bdata->node_min_pfn; | |
481 | ||
482 | /* | |
483 | * Align the index with respect to the node start so that the | |
484 | * combination of both satisfies the requested alignment. | |
485 | */ | |
486 | ||
487 | return ALIGN(base + idx, step) - base; | |
488 | } | |
489 | ||
8aa043d7 JB |
490 | static unsigned long __init align_off(struct bootmem_data *bdata, |
491 | unsigned long off, unsigned long align) | |
481ebd0d JW |
492 | { |
493 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); | |
494 | ||
495 | /* Same as align_idx for byte offsets */ | |
496 | ||
497 | return ALIGN(base + off, align) - base; | |
498 | } | |
499 | ||
c6785b6b | 500 | static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata, |
d0c4f570 TH |
501 | unsigned long size, unsigned long align, |
502 | unsigned long goal, unsigned long limit) | |
1da177e4 | 503 | { |
0f3caba2 | 504 | unsigned long fallback = 0; |
5f2809e6 JW |
505 | unsigned long min, max, start, sidx, midx, step; |
506 | ||
594fe1a0 JW |
507 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
508 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, | |
509 | align, goal, limit); | |
510 | ||
5f2809e6 JW |
511 | BUG_ON(!size); |
512 | BUG_ON(align & (align - 1)); | |
513 | BUG_ON(limit && goal + size > limit); | |
1da177e4 | 514 | |
7c309a64 CK |
515 | if (!bdata->node_bootmem_map) |
516 | return NULL; | |
517 | ||
3560e249 | 518 | min = bdata->node_min_pfn; |
5f2809e6 | 519 | max = bdata->node_low_pfn; |
9a2dc04c | 520 | |
5f2809e6 JW |
521 | goal >>= PAGE_SHIFT; |
522 | limit >>= PAGE_SHIFT; | |
523 | ||
524 | if (limit && max > limit) | |
525 | max = limit; | |
526 | if (max <= min) | |
9a2dc04c YL |
527 | return NULL; |
528 | ||
5f2809e6 | 529 | step = max(align >> PAGE_SHIFT, 1UL); |
281dd25c | 530 | |
5f2809e6 JW |
531 | if (goal && min < goal && goal < max) |
532 | start = ALIGN(goal, step); | |
533 | else | |
534 | start = ALIGN(min, step); | |
1da177e4 | 535 | |
481ebd0d | 536 | sidx = start - bdata->node_min_pfn; |
3560e249 | 537 | midx = max - bdata->node_min_pfn; |
1da177e4 | 538 | |
5f2809e6 | 539 | if (bdata->hint_idx > sidx) { |
0f3caba2 JW |
540 | /* |
541 | * Handle the valid case of sidx being zero and still | |
542 | * catch the fallback below. | |
543 | */ | |
544 | fallback = sidx + 1; | |
481ebd0d | 545 | sidx = align_idx(bdata, bdata->hint_idx, step); |
5f2809e6 | 546 | } |
1da177e4 | 547 | |
5f2809e6 JW |
548 | while (1) { |
549 | int merge; | |
550 | void *region; | |
551 | unsigned long eidx, i, start_off, end_off; | |
552 | find_block: | |
553 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); | |
481ebd0d | 554 | sidx = align_idx(bdata, sidx, step); |
5f2809e6 | 555 | eidx = sidx + PFN_UP(size); |
ad09315c | 556 | |
5f2809e6 | 557 | if (sidx >= midx || eidx > midx) |
66d43e98 | 558 | break; |
1da177e4 | 559 | |
5f2809e6 JW |
560 | for (i = sidx; i < eidx; i++) |
561 | if (test_bit(i, bdata->node_bootmem_map)) { | |
481ebd0d | 562 | sidx = align_idx(bdata, i, step); |
5f2809e6 JW |
563 | if (sidx == i) |
564 | sidx += step; | |
565 | goto find_block; | |
566 | } | |
1da177e4 | 567 | |
627240aa | 568 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
5f2809e6 | 569 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
481ebd0d | 570 | start_off = align_off(bdata, bdata->last_end_off, align); |
5f2809e6 JW |
571 | else |
572 | start_off = PFN_PHYS(sidx); | |
573 | ||
574 | merge = PFN_DOWN(start_off) < sidx; | |
575 | end_off = start_off + size; | |
576 | ||
577 | bdata->last_end_off = end_off; | |
578 | bdata->hint_idx = PFN_UP(end_off); | |
579 | ||
580 | /* | |
581 | * Reserve the area now: | |
582 | */ | |
d747fa4b JW |
583 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
584 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) | |
585 | BUG(); | |
5f2809e6 | 586 | |
3560e249 JW |
587 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
588 | start_off); | |
5f2809e6 | 589 | memset(region, 0, size); |
008139d9 CM |
590 | /* |
591 | * The min_count is set to 0 so that bootmem allocated blocks | |
592 | * are never reported as leaks. | |
593 | */ | |
594 | kmemleak_alloc(region, size, 0, 0); | |
5f2809e6 | 595 | return region; |
1da177e4 LT |
596 | } |
597 | ||
0f3caba2 | 598 | if (fallback) { |
481ebd0d | 599 | sidx = align_idx(bdata, fallback - 1, step); |
0f3caba2 JW |
600 | fallback = 0; |
601 | goto find_block; | |
602 | } | |
603 | ||
604 | return NULL; | |
605 | } | |
606 | ||
c12ab504 | 607 | static void * __init alloc_bootmem_core(unsigned long size, |
0f3caba2 JW |
608 | unsigned long align, |
609 | unsigned long goal, | |
610 | unsigned long limit) | |
611 | { | |
612 | bootmem_data_t *bdata; | |
d0c4f570 | 613 | void *region; |
0f3caba2 | 614 | |
3f7dfe24 JK |
615 | if (WARN_ON_ONCE(slab_is_available())) |
616 | return kzalloc(size, GFP_NOWAIT); | |
0f3caba2 | 617 | |
d0c4f570 | 618 | list_for_each_entry(bdata, &bdata_list, list) { |
0f3caba2 JW |
619 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
620 | continue; | |
3560e249 | 621 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
0f3caba2 JW |
622 | break; |
623 | ||
c6785b6b | 624 | region = alloc_bootmem_bdata(bdata, size, align, goal, limit); |
0f3caba2 JW |
625 | if (region) |
626 | return region; | |
627 | } | |
628 | ||
c12ab504 JW |
629 | return NULL; |
630 | } | |
631 | ||
632 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, | |
633 | unsigned long align, | |
634 | unsigned long goal, | |
635 | unsigned long limit) | |
636 | { | |
637 | void *ptr; | |
638 | ||
639 | restart: | |
640 | ptr = alloc_bootmem_core(size, align, goal, limit); | |
641 | if (ptr) | |
642 | return ptr; | |
5f2809e6 JW |
643 | if (goal) { |
644 | goal = 0; | |
0f3caba2 | 645 | goto restart; |
5f2809e6 | 646 | } |
2e5237da | 647 | |
5f2809e6 | 648 | return NULL; |
1da177e4 LT |
649 | } |
650 | ||
a66fd7da JW |
651 | /** |
652 | * __alloc_bootmem_nopanic - allocate boot memory without panicking | |
653 | * @size: size of the request in bytes | |
654 | * @align: alignment of the region | |
655 | * @goal: preferred starting address of the region | |
656 | * | |
657 | * The goal is dropped if it can not be satisfied and the allocation will | |
658 | * fall back to memory below @goal. | |
659 | * | |
660 | * Allocation may happen on any node in the system. | |
661 | * | |
662 | * Returns NULL on failure. | |
663 | */ | |
bb0923a6 | 664 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
0f3caba2 | 665 | unsigned long goal) |
1da177e4 | 666 | { |
08677214 YL |
667 | unsigned long limit = 0; |
668 | ||
08677214 | 669 | return ___alloc_bootmem_nopanic(size, align, goal, limit); |
0f3caba2 | 670 | } |
1da177e4 | 671 | |
0f3caba2 JW |
672 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
673 | unsigned long goal, unsigned long limit) | |
674 | { | |
675 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); | |
676 | ||
677 | if (mem) | |
678 | return mem; | |
679 | /* | |
680 | * Whoops, we cannot satisfy the allocation request. | |
681 | */ | |
1170532b | 682 | pr_alert("bootmem alloc of %lu bytes failed!\n", size); |
0f3caba2 | 683 | panic("Out of memory"); |
a8062231 AK |
684 | return NULL; |
685 | } | |
1da177e4 | 686 | |
a66fd7da JW |
687 | /** |
688 | * __alloc_bootmem - allocate boot memory | |
689 | * @size: size of the request in bytes | |
690 | * @align: alignment of the region | |
691 | * @goal: preferred starting address of the region | |
692 | * | |
693 | * The goal is dropped if it can not be satisfied and the allocation will | |
694 | * fall back to memory below @goal. | |
695 | * | |
696 | * Allocation may happen on any node in the system. | |
697 | * | |
698 | * The function panics if the request can not be satisfied. | |
699 | */ | |
bb0923a6 FBH |
700 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
701 | unsigned long goal) | |
a8062231 | 702 | { |
08677214 YL |
703 | unsigned long limit = 0; |
704 | ||
08677214 | 705 | return ___alloc_bootmem(size, align, goal, limit); |
1da177e4 LT |
706 | } |
707 | ||
99ab7b19 | 708 | void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat, |
4cc278b7 JW |
709 | unsigned long size, unsigned long align, |
710 | unsigned long goal, unsigned long limit) | |
711 | { | |
712 | void *ptr; | |
713 | ||
3f7dfe24 JK |
714 | if (WARN_ON_ONCE(slab_is_available())) |
715 | return kzalloc(size, GFP_NOWAIT); | |
ab381843 | 716 | again: |
d0c4f570 | 717 | |
c8f4a2d0 YL |
718 | /* do not panic in alloc_bootmem_bdata() */ |
719 | if (limit && goal + size > limit) | |
720 | limit = 0; | |
721 | ||
e9079911 | 722 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit); |
4cc278b7 JW |
723 | if (ptr) |
724 | return ptr; | |
725 | ||
ab381843 JW |
726 | ptr = alloc_bootmem_core(size, align, goal, limit); |
727 | if (ptr) | |
728 | return ptr; | |
729 | ||
730 | if (goal) { | |
731 | goal = 0; | |
732 | goto again; | |
733 | } | |
734 | ||
421456ed JW |
735 | return NULL; |
736 | } | |
737 | ||
738 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, | |
739 | unsigned long align, unsigned long goal) | |
740 | { | |
741 | if (WARN_ON_ONCE(slab_is_available())) | |
742 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
743 | ||
e9079911 | 744 | return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
421456ed JW |
745 | } |
746 | ||
e9079911 | 747 | void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
421456ed JW |
748 | unsigned long align, unsigned long goal, |
749 | unsigned long limit) | |
750 | { | |
751 | void *ptr; | |
752 | ||
e9079911 | 753 | ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
421456ed JW |
754 | if (ptr) |
755 | return ptr; | |
756 | ||
1170532b | 757 | pr_alert("bootmem alloc of %lu bytes failed!\n", size); |
ab381843 JW |
758 | panic("Out of memory"); |
759 | return NULL; | |
4cc278b7 JW |
760 | } |
761 | ||
a66fd7da JW |
762 | /** |
763 | * __alloc_bootmem_node - allocate boot memory from a specific node | |
764 | * @pgdat: node to allocate from | |
765 | * @size: size of the request in bytes | |
766 | * @align: alignment of the region | |
767 | * @goal: preferred starting address of the region | |
768 | * | |
769 | * The goal is dropped if it can not be satisfied and the allocation will | |
770 | * fall back to memory below @goal. | |
771 | * | |
772 | * Allocation may fall back to any node in the system if the specified node | |
773 | * can not hold the requested memory. | |
774 | * | |
775 | * The function panics if the request can not be satisfied. | |
776 | */ | |
bb0923a6 FBH |
777 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
778 | unsigned long align, unsigned long goal) | |
1da177e4 | 779 | { |
c91c4773 PE |
780 | if (WARN_ON_ONCE(slab_is_available())) |
781 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
782 | ||
e9079911 | 783 | return ___alloc_bootmem_node(pgdat, size, align, goal, 0); |
08677214 YL |
784 | } |
785 | ||
786 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, | |
787 | unsigned long align, unsigned long goal) | |
788 | { | |
789 | #ifdef MAX_DMA32_PFN | |
790 | unsigned long end_pfn; | |
791 | ||
792 | if (WARN_ON_ONCE(slab_is_available())) | |
793 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
794 | ||
795 | /* update goal according ...MAX_DMA32_PFN */ | |
83285c72 | 796 | end_pfn = pgdat_end_pfn(pgdat); |
08677214 YL |
797 | |
798 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && | |
799 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { | |
800 | void *ptr; | |
801 | unsigned long new_goal; | |
802 | ||
803 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; | |
c6785b6b | 804 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, |
08677214 | 805 | new_goal, 0); |
08677214 YL |
806 | if (ptr) |
807 | return ptr; | |
808 | } | |
809 | #endif | |
810 | ||
811 | return __alloc_bootmem_node(pgdat, size, align, goal); | |
812 | ||
1da177e4 LT |
813 | } |
814 | ||
dfd54cbc HC |
815 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
816 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL | |
817 | #endif | |
008857c1 | 818 | |
a66fd7da JW |
819 | /** |
820 | * __alloc_bootmem_low - allocate low boot memory | |
821 | * @size: size of the request in bytes | |
822 | * @align: alignment of the region | |
823 | * @goal: preferred starting address of the region | |
824 | * | |
825 | * The goal is dropped if it can not be satisfied and the allocation will | |
826 | * fall back to memory below @goal. | |
827 | * | |
828 | * Allocation may happen on any node in the system. | |
829 | * | |
830 | * The function panics if the request can not be satisfied. | |
831 | */ | |
bb0923a6 FBH |
832 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
833 | unsigned long goal) | |
008857c1 | 834 | { |
0f3caba2 | 835 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
008857c1 RT |
836 | } |
837 | ||
38fa4175 YL |
838 | void * __init __alloc_bootmem_low_nopanic(unsigned long size, |
839 | unsigned long align, | |
840 | unsigned long goal) | |
841 | { | |
842 | return ___alloc_bootmem_nopanic(size, align, goal, | |
843 | ARCH_LOW_ADDRESS_LIMIT); | |
844 | } | |
845 | ||
a66fd7da JW |
846 | /** |
847 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node | |
848 | * @pgdat: node to allocate from | |
849 | * @size: size of the request in bytes | |
850 | * @align: alignment of the region | |
851 | * @goal: preferred starting address of the region | |
852 | * | |
853 | * The goal is dropped if it can not be satisfied and the allocation will | |
854 | * fall back to memory below @goal. | |
855 | * | |
856 | * Allocation may fall back to any node in the system if the specified node | |
857 | * can not hold the requested memory. | |
858 | * | |
859 | * The function panics if the request can not be satisfied. | |
860 | */ | |
008857c1 RT |
861 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
862 | unsigned long align, unsigned long goal) | |
863 | { | |
c91c4773 PE |
864 | if (WARN_ON_ONCE(slab_is_available())) |
865 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
866 | ||
e9079911 JW |
867 | return ___alloc_bootmem_node(pgdat, size, align, |
868 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
008857c1 | 869 | } |