]>
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> |
e786e86a FBH |
19 | |
20 | #include <asm/bug.h> | |
1da177e4 | 21 | #include <asm/io.h> |
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; | |
36 | ||
b61bfa3c JW |
37 | bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata; |
38 | ||
636cc40c JW |
39 | static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list); |
40 | ||
2e5237da JW |
41 | static int bootmem_debug; |
42 | ||
43 | static int __init bootmem_debug_setup(char *buf) | |
44 | { | |
45 | bootmem_debug = 1; | |
46 | return 0; | |
47 | } | |
48 | early_param("bootmem_debug", bootmem_debug_setup); | |
49 | ||
50 | #define bdebug(fmt, args...) ({ \ | |
51 | if (unlikely(bootmem_debug)) \ | |
52 | printk(KERN_INFO \ | |
53 | "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++) { | |
167 | __free_pages_bootmem(pfn_to_page(cursor), 0); | |
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; | |
41546c17 JW |
175 | unsigned long start, end, pages, count = 0; |
176 | ||
177 | if (!bdata->node_bootmem_map) | |
178 | return 0; | |
179 | ||
3560e249 | 180 | start = bdata->node_min_pfn; |
41546c17 JW |
181 | end = bdata->node_low_pfn; |
182 | ||
799f933a JW |
183 | bdebug("nid=%td start=%lx end=%lx\n", |
184 | bdata - bootmem_node_data, start, end); | |
223e8dc9 | 185 | |
41546c17 JW |
186 | while (start < end) { |
187 | unsigned long *map, idx, vec; | |
10d73e65 | 188 | unsigned shift; |
223e8dc9 | 189 | |
41546c17 | 190 | map = bdata->node_bootmem_map; |
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 | ||
213 | __free_pages_bootmem(pfn_to_page(start), order); | |
223e8dc9 | 214 | count += BITS_PER_LONG; |
799f933a | 215 | start += BITS_PER_LONG; |
41546c17 | 216 | } else { |
10d73e65 | 217 | unsigned long 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); |
223e8dc9 | 223 | __free_pages_bootmem(page, 0); |
41546c17 | 224 | count++; |
223e8dc9 | 225 | } |
41546c17 | 226 | vec >>= 1; |
10d73e65 | 227 | ++cur; |
223e8dc9 | 228 | } |
223e8dc9 | 229 | } |
223e8dc9 JW |
230 | } |
231 | ||
223e8dc9 | 232 | page = virt_to_page(bdata->node_bootmem_map); |
3560e249 | 233 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
41546c17 JW |
234 | pages = bootmem_bootmap_pages(pages); |
235 | count += pages; | |
5576646f | 236 | while (pages--) |
41546c17 | 237 | __free_pages_bootmem(page++, 0); |
223e8dc9 | 238 | |
2e5237da JW |
239 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
240 | ||
223e8dc9 JW |
241 | return count; |
242 | } | |
243 | ||
7b4b2a0d JL |
244 | static int reset_managed_pages_done __initdata; |
245 | ||
246 | static inline void __init reset_node_managed_pages(pg_data_t *pgdat) | |
9feedc9d JL |
247 | { |
248 | struct zone *z; | |
249 | ||
7b4b2a0d JL |
250 | if (reset_managed_pages_done) |
251 | return; | |
252 | ||
9feedc9d | 253 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) |
7b4b2a0d JL |
254 | z->managed_pages = 0; |
255 | } | |
256 | ||
257 | void __init reset_all_zones_managed_pages(void) | |
258 | { | |
259 | struct pglist_data *pgdat; | |
260 | ||
261 | for_each_online_pgdat(pgdat) | |
262 | reset_node_managed_pages(pgdat); | |
263 | reset_managed_pages_done = 1; | |
9feedc9d JL |
264 | } |
265 | ||
a66fd7da JW |
266 | /** |
267 | * free_all_bootmem_node - release a node's free pages to the buddy allocator | |
268 | * @pgdat: node to be released | |
269 | * | |
270 | * Returns the number of pages actually released. | |
271 | */ | |
223e8dc9 JW |
272 | unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) |
273 | { | |
274 | register_page_bootmem_info_node(pgdat); | |
7b4b2a0d | 275 | reset_node_managed_pages(pgdat); |
223e8dc9 JW |
276 | return free_all_bootmem_core(pgdat->bdata); |
277 | } | |
278 | ||
a66fd7da JW |
279 | /** |
280 | * free_all_bootmem - release free pages to the buddy allocator | |
281 | * | |
282 | * Returns the number of pages actually released. | |
283 | */ | |
223e8dc9 JW |
284 | unsigned long __init free_all_bootmem(void) |
285 | { | |
aa235fc7 YL |
286 | unsigned long total_pages = 0; |
287 | bootmem_data_t *bdata; | |
9feedc9d | 288 | |
7b4b2a0d | 289 | reset_all_zones_managed_pages(); |
aa235fc7 YL |
290 | |
291 | list_for_each_entry(bdata, &bdata_list, list) | |
292 | total_pages += free_all_bootmem_core(bdata); | |
293 | ||
294 | return total_pages; | |
223e8dc9 JW |
295 | } |
296 | ||
d747fa4b JW |
297 | static void __init __free(bootmem_data_t *bdata, |
298 | unsigned long sidx, unsigned long eidx) | |
299 | { | |
300 | unsigned long idx; | |
301 | ||
302 | bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data, | |
3560e249 JW |
303 | sidx + bdata->node_min_pfn, |
304 | eidx + bdata->node_min_pfn); | |
d747fa4b | 305 | |
e2bf3cae JW |
306 | if (bdata->hint_idx > sidx) |
307 | bdata->hint_idx = sidx; | |
308 | ||
d747fa4b JW |
309 | for (idx = sidx; idx < eidx; idx++) |
310 | if (!test_and_clear_bit(idx, bdata->node_bootmem_map)) | |
311 | BUG(); | |
312 | } | |
313 | ||
314 | static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx, | |
315 | unsigned long eidx, int flags) | |
316 | { | |
317 | unsigned long idx; | |
318 | int exclusive = flags & BOOTMEM_EXCLUSIVE; | |
319 | ||
320 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", | |
321 | bdata - bootmem_node_data, | |
3560e249 JW |
322 | sidx + bdata->node_min_pfn, |
323 | eidx + bdata->node_min_pfn, | |
d747fa4b JW |
324 | flags); |
325 | ||
326 | for (idx = sidx; idx < eidx; idx++) | |
327 | if (test_and_set_bit(idx, bdata->node_bootmem_map)) { | |
328 | if (exclusive) { | |
329 | __free(bdata, sidx, idx); | |
330 | return -EBUSY; | |
331 | } | |
332 | bdebug("silent double reserve of PFN %lx\n", | |
3560e249 | 333 | idx + bdata->node_min_pfn); |
d747fa4b JW |
334 | } |
335 | return 0; | |
336 | } | |
337 | ||
e2bf3cae JW |
338 | static int __init mark_bootmem_node(bootmem_data_t *bdata, |
339 | unsigned long start, unsigned long end, | |
340 | int reserve, int flags) | |
223e8dc9 JW |
341 | { |
342 | unsigned long sidx, eidx; | |
223e8dc9 | 343 | |
e2bf3cae JW |
344 | bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n", |
345 | bdata - bootmem_node_data, start, end, reserve, flags); | |
223e8dc9 | 346 | |
3560e249 | 347 | BUG_ON(start < bdata->node_min_pfn); |
e2bf3cae | 348 | BUG_ON(end > bdata->node_low_pfn); |
223e8dc9 | 349 | |
3560e249 JW |
350 | sidx = start - bdata->node_min_pfn; |
351 | eidx = end - bdata->node_min_pfn; | |
223e8dc9 | 352 | |
e2bf3cae JW |
353 | if (reserve) |
354 | return __reserve(bdata, sidx, eidx, flags); | |
223e8dc9 | 355 | else |
e2bf3cae JW |
356 | __free(bdata, sidx, eidx); |
357 | return 0; | |
358 | } | |
359 | ||
360 | static int __init mark_bootmem(unsigned long start, unsigned long end, | |
361 | int reserve, int flags) | |
362 | { | |
363 | unsigned long pos; | |
364 | bootmem_data_t *bdata; | |
365 | ||
366 | pos = start; | |
367 | list_for_each_entry(bdata, &bdata_list, list) { | |
368 | int err; | |
369 | unsigned long max; | |
370 | ||
3560e249 JW |
371 | if (pos < bdata->node_min_pfn || |
372 | pos >= bdata->node_low_pfn) { | |
e2bf3cae JW |
373 | BUG_ON(pos != start); |
374 | continue; | |
375 | } | |
376 | ||
377 | max = min(bdata->node_low_pfn, end); | |
223e8dc9 | 378 | |
e2bf3cae JW |
379 | err = mark_bootmem_node(bdata, pos, max, reserve, flags); |
380 | if (reserve && err) { | |
381 | mark_bootmem(start, pos, 0, 0); | |
382 | return err; | |
383 | } | |
223e8dc9 | 384 | |
e2bf3cae JW |
385 | if (max == end) |
386 | return 0; | |
387 | pos = bdata->node_low_pfn; | |
388 | } | |
389 | BUG(); | |
223e8dc9 JW |
390 | } |
391 | ||
a66fd7da JW |
392 | /** |
393 | * free_bootmem_node - mark a page range as usable | |
394 | * @pgdat: node the range resides on | |
395 | * @physaddr: starting address of the range | |
396 | * @size: size of the range in bytes | |
397 | * | |
398 | * Partial pages will be considered reserved and left as they are. | |
399 | * | |
e2bf3cae | 400 | * The range must reside completely on the specified node. |
a66fd7da | 401 | */ |
223e8dc9 JW |
402 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
403 | unsigned long size) | |
404 | { | |
e2bf3cae JW |
405 | unsigned long start, end; |
406 | ||
ec3a354b CM |
407 | kmemleak_free_part(__va(physaddr), size); |
408 | ||
e2bf3cae JW |
409 | start = PFN_UP(physaddr); |
410 | end = PFN_DOWN(physaddr + size); | |
411 | ||
412 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); | |
223e8dc9 JW |
413 | } |
414 | ||
a66fd7da JW |
415 | /** |
416 | * free_bootmem - mark a page range as usable | |
81df9bff | 417 | * @addr: starting physical address of the range |
a66fd7da JW |
418 | * @size: size of the range in bytes |
419 | * | |
420 | * Partial pages will be considered reserved and left as they are. | |
421 | * | |
e2bf3cae | 422 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 423 | */ |
81df9bff | 424 | void __init free_bootmem(unsigned long physaddr, unsigned long size) |
223e8dc9 | 425 | { |
e2bf3cae | 426 | unsigned long start, end; |
a5645a61 | 427 | |
81df9bff | 428 | kmemleak_free_part(__va(physaddr), size); |
ec3a354b | 429 | |
81df9bff JK |
430 | start = PFN_UP(physaddr); |
431 | end = PFN_DOWN(physaddr + size); | |
1da177e4 | 432 | |
e2bf3cae | 433 | mark_bootmem(start, end, 0, 0); |
1da177e4 LT |
434 | } |
435 | ||
a66fd7da JW |
436 | /** |
437 | * reserve_bootmem_node - mark a page range as reserved | |
438 | * @pgdat: node the range resides on | |
439 | * @physaddr: starting address of the range | |
440 | * @size: size of the range in bytes | |
441 | * @flags: reservation flags (see linux/bootmem.h) | |
442 | * | |
443 | * Partial pages will be reserved. | |
444 | * | |
e2bf3cae | 445 | * The range must reside completely on the specified node. |
a66fd7da | 446 | */ |
223e8dc9 JW |
447 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
448 | unsigned long size, int flags) | |
1da177e4 | 449 | { |
e2bf3cae | 450 | unsigned long start, end; |
1da177e4 | 451 | |
e2bf3cae JW |
452 | start = PFN_DOWN(physaddr); |
453 | end = PFN_UP(physaddr + size); | |
454 | ||
455 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); | |
223e8dc9 | 456 | } |
5a982cbc | 457 | |
a66fd7da | 458 | /** |
0d4ba4d7 | 459 | * reserve_bootmem - mark a page range as reserved |
a66fd7da JW |
460 | * @addr: starting address of the range |
461 | * @size: size of the range in bytes | |
462 | * @flags: reservation flags (see linux/bootmem.h) | |
463 | * | |
464 | * Partial pages will be reserved. | |
465 | * | |
e2bf3cae | 466 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 467 | */ |
223e8dc9 JW |
468 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
469 | int flags) | |
470 | { | |
e2bf3cae | 471 | unsigned long start, end; |
1da177e4 | 472 | |
e2bf3cae JW |
473 | start = PFN_DOWN(addr); |
474 | end = PFN_UP(addr + size); | |
223e8dc9 | 475 | |
e2bf3cae | 476 | return mark_bootmem(start, end, 1, flags); |
1da177e4 LT |
477 | } |
478 | ||
8aa043d7 JB |
479 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
480 | unsigned long idx, unsigned long step) | |
481ebd0d JW |
481 | { |
482 | unsigned long base = bdata->node_min_pfn; | |
483 | ||
484 | /* | |
485 | * Align the index with respect to the node start so that the | |
486 | * combination of both satisfies the requested alignment. | |
487 | */ | |
488 | ||
489 | return ALIGN(base + idx, step) - base; | |
490 | } | |
491 | ||
8aa043d7 JB |
492 | static unsigned long __init align_off(struct bootmem_data *bdata, |
493 | unsigned long off, unsigned long align) | |
481ebd0d JW |
494 | { |
495 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); | |
496 | ||
497 | /* Same as align_idx for byte offsets */ | |
498 | ||
499 | return ALIGN(base + off, align) - base; | |
500 | } | |
501 | ||
c6785b6b | 502 | static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata, |
d0c4f570 TH |
503 | unsigned long size, unsigned long align, |
504 | unsigned long goal, unsigned long limit) | |
1da177e4 | 505 | { |
0f3caba2 | 506 | unsigned long fallback = 0; |
5f2809e6 JW |
507 | unsigned long min, max, start, sidx, midx, step; |
508 | ||
594fe1a0 JW |
509 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
510 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, | |
511 | align, goal, limit); | |
512 | ||
5f2809e6 JW |
513 | BUG_ON(!size); |
514 | BUG_ON(align & (align - 1)); | |
515 | BUG_ON(limit && goal + size > limit); | |
1da177e4 | 516 | |
7c309a64 CK |
517 | if (!bdata->node_bootmem_map) |
518 | return NULL; | |
519 | ||
3560e249 | 520 | min = bdata->node_min_pfn; |
5f2809e6 | 521 | max = bdata->node_low_pfn; |
9a2dc04c | 522 | |
5f2809e6 JW |
523 | goal >>= PAGE_SHIFT; |
524 | limit >>= PAGE_SHIFT; | |
525 | ||
526 | if (limit && max > limit) | |
527 | max = limit; | |
528 | if (max <= min) | |
9a2dc04c YL |
529 | return NULL; |
530 | ||
5f2809e6 | 531 | step = max(align >> PAGE_SHIFT, 1UL); |
281dd25c | 532 | |
5f2809e6 JW |
533 | if (goal && min < goal && goal < max) |
534 | start = ALIGN(goal, step); | |
535 | else | |
536 | start = ALIGN(min, step); | |
1da177e4 | 537 | |
481ebd0d | 538 | sidx = start - bdata->node_min_pfn; |
3560e249 | 539 | midx = max - bdata->node_min_pfn; |
1da177e4 | 540 | |
5f2809e6 | 541 | if (bdata->hint_idx > sidx) { |
0f3caba2 JW |
542 | /* |
543 | * Handle the valid case of sidx being zero and still | |
544 | * catch the fallback below. | |
545 | */ | |
546 | fallback = sidx + 1; | |
481ebd0d | 547 | sidx = align_idx(bdata, bdata->hint_idx, step); |
5f2809e6 | 548 | } |
1da177e4 | 549 | |
5f2809e6 JW |
550 | while (1) { |
551 | int merge; | |
552 | void *region; | |
553 | unsigned long eidx, i, start_off, end_off; | |
554 | find_block: | |
555 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); | |
481ebd0d | 556 | sidx = align_idx(bdata, sidx, step); |
5f2809e6 | 557 | eidx = sidx + PFN_UP(size); |
ad09315c | 558 | |
5f2809e6 | 559 | if (sidx >= midx || eidx > midx) |
66d43e98 | 560 | break; |
1da177e4 | 561 | |
5f2809e6 JW |
562 | for (i = sidx; i < eidx; i++) |
563 | if (test_bit(i, bdata->node_bootmem_map)) { | |
481ebd0d | 564 | sidx = align_idx(bdata, i, step); |
5f2809e6 JW |
565 | if (sidx == i) |
566 | sidx += step; | |
567 | goto find_block; | |
568 | } | |
1da177e4 | 569 | |
627240aa | 570 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
5f2809e6 | 571 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
481ebd0d | 572 | start_off = align_off(bdata, bdata->last_end_off, align); |
5f2809e6 JW |
573 | else |
574 | start_off = PFN_PHYS(sidx); | |
575 | ||
576 | merge = PFN_DOWN(start_off) < sidx; | |
577 | end_off = start_off + size; | |
578 | ||
579 | bdata->last_end_off = end_off; | |
580 | bdata->hint_idx = PFN_UP(end_off); | |
581 | ||
582 | /* | |
583 | * Reserve the area now: | |
584 | */ | |
d747fa4b JW |
585 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
586 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) | |
587 | BUG(); | |
5f2809e6 | 588 | |
3560e249 JW |
589 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
590 | start_off); | |
5f2809e6 | 591 | memset(region, 0, size); |
008139d9 CM |
592 | /* |
593 | * The min_count is set to 0 so that bootmem allocated blocks | |
594 | * are never reported as leaks. | |
595 | */ | |
596 | kmemleak_alloc(region, size, 0, 0); | |
5f2809e6 | 597 | return region; |
1da177e4 LT |
598 | } |
599 | ||
0f3caba2 | 600 | if (fallback) { |
481ebd0d | 601 | sidx = align_idx(bdata, fallback - 1, step); |
0f3caba2 JW |
602 | fallback = 0; |
603 | goto find_block; | |
604 | } | |
605 | ||
606 | return NULL; | |
607 | } | |
608 | ||
c12ab504 | 609 | static void * __init alloc_bootmem_core(unsigned long size, |
0f3caba2 JW |
610 | unsigned long align, |
611 | unsigned long goal, | |
612 | unsigned long limit) | |
613 | { | |
614 | bootmem_data_t *bdata; | |
d0c4f570 | 615 | void *region; |
0f3caba2 | 616 | |
3f7dfe24 JK |
617 | if (WARN_ON_ONCE(slab_is_available())) |
618 | return kzalloc(size, GFP_NOWAIT); | |
0f3caba2 | 619 | |
d0c4f570 | 620 | list_for_each_entry(bdata, &bdata_list, list) { |
0f3caba2 JW |
621 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
622 | continue; | |
3560e249 | 623 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
0f3caba2 JW |
624 | break; |
625 | ||
c6785b6b | 626 | region = alloc_bootmem_bdata(bdata, size, align, goal, limit); |
0f3caba2 JW |
627 | if (region) |
628 | return region; | |
629 | } | |
630 | ||
c12ab504 JW |
631 | return NULL; |
632 | } | |
633 | ||
634 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, | |
635 | unsigned long align, | |
636 | unsigned long goal, | |
637 | unsigned long limit) | |
638 | { | |
639 | void *ptr; | |
640 | ||
641 | restart: | |
642 | ptr = alloc_bootmem_core(size, align, goal, limit); | |
643 | if (ptr) | |
644 | return ptr; | |
5f2809e6 JW |
645 | if (goal) { |
646 | goal = 0; | |
0f3caba2 | 647 | goto restart; |
5f2809e6 | 648 | } |
2e5237da | 649 | |
5f2809e6 | 650 | return NULL; |
1da177e4 LT |
651 | } |
652 | ||
a66fd7da JW |
653 | /** |
654 | * __alloc_bootmem_nopanic - allocate boot memory without panicking | |
655 | * @size: size of the request in bytes | |
656 | * @align: alignment of the region | |
657 | * @goal: preferred starting address of the region | |
658 | * | |
659 | * The goal is dropped if it can not be satisfied and the allocation will | |
660 | * fall back to memory below @goal. | |
661 | * | |
662 | * Allocation may happen on any node in the system. | |
663 | * | |
664 | * Returns NULL on failure. | |
665 | */ | |
bb0923a6 | 666 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
0f3caba2 | 667 | unsigned long goal) |
1da177e4 | 668 | { |
08677214 YL |
669 | unsigned long limit = 0; |
670 | ||
08677214 | 671 | return ___alloc_bootmem_nopanic(size, align, goal, limit); |
0f3caba2 | 672 | } |
1da177e4 | 673 | |
0f3caba2 JW |
674 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
675 | unsigned long goal, unsigned long limit) | |
676 | { | |
677 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); | |
678 | ||
679 | if (mem) | |
680 | return mem; | |
681 | /* | |
682 | * Whoops, we cannot satisfy the allocation request. | |
683 | */ | |
684 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); | |
685 | panic("Out of memory"); | |
a8062231 AK |
686 | return NULL; |
687 | } | |
1da177e4 | 688 | |
a66fd7da JW |
689 | /** |
690 | * __alloc_bootmem - allocate boot memory | |
691 | * @size: size of the request in bytes | |
692 | * @align: alignment of the region | |
693 | * @goal: preferred starting address of the region | |
694 | * | |
695 | * The goal is dropped if it can not be satisfied and the allocation will | |
696 | * fall back to memory below @goal. | |
697 | * | |
698 | * Allocation may happen on any node in the system. | |
699 | * | |
700 | * The function panics if the request can not be satisfied. | |
701 | */ | |
bb0923a6 FBH |
702 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
703 | unsigned long goal) | |
a8062231 | 704 | { |
08677214 YL |
705 | unsigned long limit = 0; |
706 | ||
08677214 | 707 | return ___alloc_bootmem(size, align, goal, limit); |
1da177e4 LT |
708 | } |
709 | ||
99ab7b19 | 710 | void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat, |
4cc278b7 JW |
711 | unsigned long size, unsigned long align, |
712 | unsigned long goal, unsigned long limit) | |
713 | { | |
714 | void *ptr; | |
715 | ||
3f7dfe24 JK |
716 | if (WARN_ON_ONCE(slab_is_available())) |
717 | return kzalloc(size, GFP_NOWAIT); | |
ab381843 | 718 | again: |
d0c4f570 | 719 | |
c8f4a2d0 YL |
720 | /* do not panic in alloc_bootmem_bdata() */ |
721 | if (limit && goal + size > limit) | |
722 | limit = 0; | |
723 | ||
e9079911 | 724 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit); |
4cc278b7 JW |
725 | if (ptr) |
726 | return ptr; | |
727 | ||
ab381843 JW |
728 | ptr = alloc_bootmem_core(size, align, goal, limit); |
729 | if (ptr) | |
730 | return ptr; | |
731 | ||
732 | if (goal) { | |
733 | goal = 0; | |
734 | goto again; | |
735 | } | |
736 | ||
421456ed JW |
737 | return NULL; |
738 | } | |
739 | ||
740 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, | |
741 | unsigned long align, unsigned long goal) | |
742 | { | |
743 | if (WARN_ON_ONCE(slab_is_available())) | |
744 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
745 | ||
e9079911 | 746 | return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
421456ed JW |
747 | } |
748 | ||
e9079911 | 749 | void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
421456ed JW |
750 | unsigned long align, unsigned long goal, |
751 | unsigned long limit) | |
752 | { | |
753 | void *ptr; | |
754 | ||
e9079911 | 755 | ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
421456ed JW |
756 | if (ptr) |
757 | return ptr; | |
758 | ||
ab381843 JW |
759 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); |
760 | panic("Out of memory"); | |
761 | return NULL; | |
4cc278b7 JW |
762 | } |
763 | ||
a66fd7da JW |
764 | /** |
765 | * __alloc_bootmem_node - allocate boot memory from a specific node | |
766 | * @pgdat: node to allocate from | |
767 | * @size: size of the request in bytes | |
768 | * @align: alignment of the region | |
769 | * @goal: preferred starting address of the region | |
770 | * | |
771 | * The goal is dropped if it can not be satisfied and the allocation will | |
772 | * fall back to memory below @goal. | |
773 | * | |
774 | * Allocation may fall back to any node in the system if the specified node | |
775 | * can not hold the requested memory. | |
776 | * | |
777 | * The function panics if the request can not be satisfied. | |
778 | */ | |
bb0923a6 FBH |
779 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
780 | unsigned long align, unsigned long goal) | |
1da177e4 | 781 | { |
c91c4773 PE |
782 | if (WARN_ON_ONCE(slab_is_available())) |
783 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
784 | ||
e9079911 | 785 | return ___alloc_bootmem_node(pgdat, size, align, goal, 0); |
08677214 YL |
786 | } |
787 | ||
788 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, | |
789 | unsigned long align, unsigned long goal) | |
790 | { | |
791 | #ifdef MAX_DMA32_PFN | |
792 | unsigned long end_pfn; | |
793 | ||
794 | if (WARN_ON_ONCE(slab_is_available())) | |
795 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
796 | ||
797 | /* update goal according ...MAX_DMA32_PFN */ | |
798 | end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
799 | ||
800 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && | |
801 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { | |
802 | void *ptr; | |
803 | unsigned long new_goal; | |
804 | ||
805 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; | |
c6785b6b | 806 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, |
08677214 | 807 | new_goal, 0); |
08677214 YL |
808 | if (ptr) |
809 | return ptr; | |
810 | } | |
811 | #endif | |
812 | ||
813 | return __alloc_bootmem_node(pgdat, size, align, goal); | |
814 | ||
1da177e4 LT |
815 | } |
816 | ||
dfd54cbc HC |
817 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
818 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL | |
819 | #endif | |
008857c1 | 820 | |
a66fd7da JW |
821 | /** |
822 | * __alloc_bootmem_low - allocate low boot memory | |
823 | * @size: size of the request in bytes | |
824 | * @align: alignment of the region | |
825 | * @goal: preferred starting address of the region | |
826 | * | |
827 | * The goal is dropped if it can not be satisfied and the allocation will | |
828 | * fall back to memory below @goal. | |
829 | * | |
830 | * Allocation may happen on any node in the system. | |
831 | * | |
832 | * The function panics if the request can not be satisfied. | |
833 | */ | |
bb0923a6 FBH |
834 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
835 | unsigned long goal) | |
008857c1 | 836 | { |
0f3caba2 | 837 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
008857c1 RT |
838 | } |
839 | ||
38fa4175 YL |
840 | void * __init __alloc_bootmem_low_nopanic(unsigned long size, |
841 | unsigned long align, | |
842 | unsigned long goal) | |
843 | { | |
844 | return ___alloc_bootmem_nopanic(size, align, goal, | |
845 | ARCH_LOW_ADDRESS_LIMIT); | |
846 | } | |
847 | ||
a66fd7da JW |
848 | /** |
849 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node | |
850 | * @pgdat: node to allocate from | |
851 | * @size: size of the request in bytes | |
852 | * @align: alignment of the region | |
853 | * @goal: preferred starting address of the region | |
854 | * | |
855 | * The goal is dropped if it can not be satisfied and the allocation will | |
856 | * fall back to memory below @goal. | |
857 | * | |
858 | * Allocation may fall back to any node in the system if the specified node | |
859 | * can not hold the requested memory. | |
860 | * | |
861 | * The function panics if the request can not be satisfied. | |
862 | */ | |
008857c1 RT |
863 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
864 | unsigned long align, unsigned long goal) | |
865 | { | |
c91c4773 PE |
866 | if (WARN_ON_ONCE(slab_is_available())) |
867 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
868 | ||
e9079911 JW |
869 | return ___alloc_bootmem_node(pgdat, size, align, |
870 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
008857c1 | 871 | } |