]>
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 | { |
df049a5f | 59 | unsigned long bytes = (pages + 7) / 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 | { |
636cc40c | 80 | struct list_head *iter; |
f71bf0ca | 81 | |
636cc40c JW |
82 | list_for_each(iter, &bdata_list) { |
83 | bootmem_data_t *ent; | |
84 | ||
85 | ent = list_entry(iter, bootmem_data_t, list); | |
3560e249 | 86 | if (bdata->node_min_pfn < ent->node_min_pfn) |
636cc40c | 87 | break; |
679bc9fb | 88 | } |
636cc40c | 89 | list_add_tail(&bdata->list, iter); |
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 | |
150 | * @addr: starting address of the range | |
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 | */ | |
157 | void __init free_bootmem_late(unsigned long addr, unsigned long size) | |
158 | { | |
159 | unsigned long cursor, end; | |
160 | ||
161 | kmemleak_free_part(__va(addr), size); | |
162 | ||
163 | cursor = PFN_UP(addr); | |
164 | end = PFN_DOWN(addr + size); | |
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 | { | |
41546c17 | 174 | int aligned; |
223e8dc9 | 175 | struct page *page; |
41546c17 JW |
176 | unsigned long start, end, pages, count = 0; |
177 | ||
178 | if (!bdata->node_bootmem_map) | |
179 | return 0; | |
180 | ||
3560e249 | 181 | start = bdata->node_min_pfn; |
41546c17 JW |
182 | end = bdata->node_low_pfn; |
183 | ||
223e8dc9 | 184 | /* |
41546c17 JW |
185 | * If the start is aligned to the machines wordsize, we might |
186 | * be able to free pages in bulks of that order. | |
223e8dc9 | 187 | */ |
41546c17 | 188 | aligned = !(start & (BITS_PER_LONG - 1)); |
223e8dc9 | 189 | |
41546c17 JW |
190 | bdebug("nid=%td start=%lx end=%lx aligned=%d\n", |
191 | bdata - bootmem_node_data, start, end, aligned); | |
223e8dc9 | 192 | |
41546c17 JW |
193 | while (start < end) { |
194 | unsigned long *map, idx, vec; | |
223e8dc9 | 195 | |
41546c17 | 196 | map = bdata->node_bootmem_map; |
3560e249 | 197 | idx = start - bdata->node_min_pfn; |
41546c17 JW |
198 | vec = ~map[idx / BITS_PER_LONG]; |
199 | ||
200 | if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) { | |
201 | int order = ilog2(BITS_PER_LONG); | |
202 | ||
203 | __free_pages_bootmem(pfn_to_page(start), order); | |
223e8dc9 | 204 | count += BITS_PER_LONG; |
41546c17 JW |
205 | } else { |
206 | unsigned long off = 0; | |
207 | ||
208 | while (vec && off < BITS_PER_LONG) { | |
209 | if (vec & 1) { | |
210 | page = pfn_to_page(start + off); | |
223e8dc9 | 211 | __free_pages_bootmem(page, 0); |
41546c17 | 212 | count++; |
223e8dc9 | 213 | } |
41546c17 JW |
214 | vec >>= 1; |
215 | off++; | |
223e8dc9 | 216 | } |
223e8dc9 | 217 | } |
41546c17 | 218 | start += BITS_PER_LONG; |
223e8dc9 JW |
219 | } |
220 | ||
223e8dc9 | 221 | page = virt_to_page(bdata->node_bootmem_map); |
3560e249 | 222 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
41546c17 JW |
223 | pages = bootmem_bootmap_pages(pages); |
224 | count += pages; | |
225 | while (pages--) | |
226 | __free_pages_bootmem(page++, 0); | |
223e8dc9 | 227 | |
2e5237da JW |
228 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
229 | ||
223e8dc9 JW |
230 | return count; |
231 | } | |
232 | ||
a66fd7da JW |
233 | /** |
234 | * free_all_bootmem_node - release a node's free pages to the buddy allocator | |
235 | * @pgdat: node to be released | |
236 | * | |
237 | * Returns the number of pages actually released. | |
238 | */ | |
223e8dc9 JW |
239 | unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) |
240 | { | |
241 | register_page_bootmem_info_node(pgdat); | |
242 | return free_all_bootmem_core(pgdat->bdata); | |
243 | } | |
244 | ||
a66fd7da JW |
245 | /** |
246 | * free_all_bootmem - release free pages to the buddy allocator | |
247 | * | |
248 | * Returns the number of pages actually released. | |
249 | */ | |
223e8dc9 JW |
250 | unsigned long __init free_all_bootmem(void) |
251 | { | |
aa235fc7 YL |
252 | unsigned long total_pages = 0; |
253 | bootmem_data_t *bdata; | |
254 | ||
255 | list_for_each_entry(bdata, &bdata_list, list) | |
256 | total_pages += free_all_bootmem_core(bdata); | |
257 | ||
258 | return total_pages; | |
223e8dc9 JW |
259 | } |
260 | ||
d747fa4b JW |
261 | static void __init __free(bootmem_data_t *bdata, |
262 | unsigned long sidx, unsigned long eidx) | |
263 | { | |
264 | unsigned long idx; | |
265 | ||
266 | bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data, | |
3560e249 JW |
267 | sidx + bdata->node_min_pfn, |
268 | eidx + bdata->node_min_pfn); | |
d747fa4b | 269 | |
e2bf3cae JW |
270 | if (bdata->hint_idx > sidx) |
271 | bdata->hint_idx = sidx; | |
272 | ||
d747fa4b JW |
273 | for (idx = sidx; idx < eidx; idx++) |
274 | if (!test_and_clear_bit(idx, bdata->node_bootmem_map)) | |
275 | BUG(); | |
276 | } | |
277 | ||
278 | static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx, | |
279 | unsigned long eidx, int flags) | |
280 | { | |
281 | unsigned long idx; | |
282 | int exclusive = flags & BOOTMEM_EXCLUSIVE; | |
283 | ||
284 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", | |
285 | bdata - bootmem_node_data, | |
3560e249 JW |
286 | sidx + bdata->node_min_pfn, |
287 | eidx + bdata->node_min_pfn, | |
d747fa4b JW |
288 | flags); |
289 | ||
290 | for (idx = sidx; idx < eidx; idx++) | |
291 | if (test_and_set_bit(idx, bdata->node_bootmem_map)) { | |
292 | if (exclusive) { | |
293 | __free(bdata, sidx, idx); | |
294 | return -EBUSY; | |
295 | } | |
296 | bdebug("silent double reserve of PFN %lx\n", | |
3560e249 | 297 | idx + bdata->node_min_pfn); |
d747fa4b JW |
298 | } |
299 | return 0; | |
300 | } | |
301 | ||
e2bf3cae JW |
302 | static int __init mark_bootmem_node(bootmem_data_t *bdata, |
303 | unsigned long start, unsigned long end, | |
304 | int reserve, int flags) | |
223e8dc9 JW |
305 | { |
306 | unsigned long sidx, eidx; | |
223e8dc9 | 307 | |
e2bf3cae JW |
308 | bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n", |
309 | bdata - bootmem_node_data, start, end, reserve, flags); | |
223e8dc9 | 310 | |
3560e249 | 311 | BUG_ON(start < bdata->node_min_pfn); |
e2bf3cae | 312 | BUG_ON(end > bdata->node_low_pfn); |
223e8dc9 | 313 | |
3560e249 JW |
314 | sidx = start - bdata->node_min_pfn; |
315 | eidx = end - bdata->node_min_pfn; | |
223e8dc9 | 316 | |
e2bf3cae JW |
317 | if (reserve) |
318 | return __reserve(bdata, sidx, eidx, flags); | |
223e8dc9 | 319 | else |
e2bf3cae JW |
320 | __free(bdata, sidx, eidx); |
321 | return 0; | |
322 | } | |
323 | ||
324 | static int __init mark_bootmem(unsigned long start, unsigned long end, | |
325 | int reserve, int flags) | |
326 | { | |
327 | unsigned long pos; | |
328 | bootmem_data_t *bdata; | |
329 | ||
330 | pos = start; | |
331 | list_for_each_entry(bdata, &bdata_list, list) { | |
332 | int err; | |
333 | unsigned long max; | |
334 | ||
3560e249 JW |
335 | if (pos < bdata->node_min_pfn || |
336 | pos >= bdata->node_low_pfn) { | |
e2bf3cae JW |
337 | BUG_ON(pos != start); |
338 | continue; | |
339 | } | |
340 | ||
341 | max = min(bdata->node_low_pfn, end); | |
223e8dc9 | 342 | |
e2bf3cae JW |
343 | err = mark_bootmem_node(bdata, pos, max, reserve, flags); |
344 | if (reserve && err) { | |
345 | mark_bootmem(start, pos, 0, 0); | |
346 | return err; | |
347 | } | |
223e8dc9 | 348 | |
e2bf3cae JW |
349 | if (max == end) |
350 | return 0; | |
351 | pos = bdata->node_low_pfn; | |
352 | } | |
353 | BUG(); | |
223e8dc9 JW |
354 | } |
355 | ||
a66fd7da JW |
356 | /** |
357 | * free_bootmem_node - mark a page range as usable | |
358 | * @pgdat: node the range resides on | |
359 | * @physaddr: starting address of the range | |
360 | * @size: size of the range in bytes | |
361 | * | |
362 | * Partial pages will be considered reserved and left as they are. | |
363 | * | |
e2bf3cae | 364 | * The range must reside completely on the specified node. |
a66fd7da | 365 | */ |
223e8dc9 JW |
366 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
367 | unsigned long size) | |
368 | { | |
e2bf3cae JW |
369 | unsigned long start, end; |
370 | ||
ec3a354b CM |
371 | kmemleak_free_part(__va(physaddr), size); |
372 | ||
e2bf3cae JW |
373 | start = PFN_UP(physaddr); |
374 | end = PFN_DOWN(physaddr + size); | |
375 | ||
376 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); | |
223e8dc9 JW |
377 | } |
378 | ||
a66fd7da JW |
379 | /** |
380 | * free_bootmem - mark a page range as usable | |
381 | * @addr: starting address of the range | |
382 | * @size: size of the range in bytes | |
383 | * | |
384 | * Partial pages will be considered reserved and left as they are. | |
385 | * | |
e2bf3cae | 386 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 387 | */ |
223e8dc9 JW |
388 | void __init free_bootmem(unsigned long addr, unsigned long size) |
389 | { | |
e2bf3cae | 390 | unsigned long start, end; |
a5645a61 | 391 | |
ec3a354b CM |
392 | kmemleak_free_part(__va(addr), size); |
393 | ||
e2bf3cae JW |
394 | start = PFN_UP(addr); |
395 | end = PFN_DOWN(addr + size); | |
1da177e4 | 396 | |
e2bf3cae | 397 | mark_bootmem(start, end, 0, 0); |
1da177e4 LT |
398 | } |
399 | ||
a66fd7da JW |
400 | /** |
401 | * reserve_bootmem_node - mark a page range as reserved | |
402 | * @pgdat: node the range resides on | |
403 | * @physaddr: starting address of the range | |
404 | * @size: size of the range in bytes | |
405 | * @flags: reservation flags (see linux/bootmem.h) | |
406 | * | |
407 | * Partial pages will be reserved. | |
408 | * | |
e2bf3cae | 409 | * The range must reside completely on the specified node. |
a66fd7da | 410 | */ |
223e8dc9 JW |
411 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
412 | unsigned long size, int flags) | |
1da177e4 | 413 | { |
e2bf3cae | 414 | unsigned long start, end; |
1da177e4 | 415 | |
e2bf3cae JW |
416 | start = PFN_DOWN(physaddr); |
417 | end = PFN_UP(physaddr + size); | |
418 | ||
419 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); | |
223e8dc9 | 420 | } |
5a982cbc | 421 | |
a66fd7da JW |
422 | /** |
423 | * reserve_bootmem - mark a page range as usable | |
424 | * @addr: starting address of the range | |
425 | * @size: size of the range in bytes | |
426 | * @flags: reservation flags (see linux/bootmem.h) | |
427 | * | |
428 | * Partial pages will be reserved. | |
429 | * | |
e2bf3cae | 430 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 431 | */ |
223e8dc9 JW |
432 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
433 | int flags) | |
434 | { | |
e2bf3cae | 435 | unsigned long start, end; |
1da177e4 | 436 | |
e2bf3cae JW |
437 | start = PFN_DOWN(addr); |
438 | end = PFN_UP(addr + size); | |
223e8dc9 | 439 | |
e2bf3cae | 440 | return mark_bootmem(start, end, 1, flags); |
1da177e4 LT |
441 | } |
442 | ||
f88eff74 YL |
443 | int __weak __init reserve_bootmem_generic(unsigned long phys, unsigned long len, |
444 | int flags) | |
445 | { | |
446 | return reserve_bootmem(phys, len, flags); | |
447 | } | |
448 | ||
8aa043d7 JB |
449 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
450 | unsigned long idx, unsigned long step) | |
481ebd0d JW |
451 | { |
452 | unsigned long base = bdata->node_min_pfn; | |
453 | ||
454 | /* | |
455 | * Align the index with respect to the node start so that the | |
456 | * combination of both satisfies the requested alignment. | |
457 | */ | |
458 | ||
459 | return ALIGN(base + idx, step) - base; | |
460 | } | |
461 | ||
8aa043d7 JB |
462 | static unsigned long __init align_off(struct bootmem_data *bdata, |
463 | unsigned long off, unsigned long align) | |
481ebd0d JW |
464 | { |
465 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); | |
466 | ||
467 | /* Same as align_idx for byte offsets */ | |
468 | ||
469 | return ALIGN(base + off, align) - base; | |
470 | } | |
471 | ||
d0c4f570 TH |
472 | static void * __init alloc_bootmem_core(struct bootmem_data *bdata, |
473 | unsigned long size, unsigned long align, | |
474 | unsigned long goal, unsigned long limit) | |
1da177e4 | 475 | { |
0f3caba2 | 476 | unsigned long fallback = 0; |
5f2809e6 JW |
477 | unsigned long min, max, start, sidx, midx, step; |
478 | ||
594fe1a0 JW |
479 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
480 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, | |
481 | align, goal, limit); | |
482 | ||
5f2809e6 JW |
483 | BUG_ON(!size); |
484 | BUG_ON(align & (align - 1)); | |
485 | BUG_ON(limit && goal + size > limit); | |
1da177e4 | 486 | |
7c309a64 CK |
487 | if (!bdata->node_bootmem_map) |
488 | return NULL; | |
489 | ||
3560e249 | 490 | min = bdata->node_min_pfn; |
5f2809e6 | 491 | max = bdata->node_low_pfn; |
9a2dc04c | 492 | |
5f2809e6 JW |
493 | goal >>= PAGE_SHIFT; |
494 | limit >>= PAGE_SHIFT; | |
495 | ||
496 | if (limit && max > limit) | |
497 | max = limit; | |
498 | if (max <= min) | |
9a2dc04c YL |
499 | return NULL; |
500 | ||
5f2809e6 | 501 | step = max(align >> PAGE_SHIFT, 1UL); |
281dd25c | 502 | |
5f2809e6 JW |
503 | if (goal && min < goal && goal < max) |
504 | start = ALIGN(goal, step); | |
505 | else | |
506 | start = ALIGN(min, step); | |
1da177e4 | 507 | |
481ebd0d | 508 | sidx = start - bdata->node_min_pfn; |
3560e249 | 509 | midx = max - bdata->node_min_pfn; |
1da177e4 | 510 | |
5f2809e6 | 511 | if (bdata->hint_idx > sidx) { |
0f3caba2 JW |
512 | /* |
513 | * Handle the valid case of sidx being zero and still | |
514 | * catch the fallback below. | |
515 | */ | |
516 | fallback = sidx + 1; | |
481ebd0d | 517 | sidx = align_idx(bdata, bdata->hint_idx, step); |
5f2809e6 | 518 | } |
1da177e4 | 519 | |
5f2809e6 JW |
520 | while (1) { |
521 | int merge; | |
522 | void *region; | |
523 | unsigned long eidx, i, start_off, end_off; | |
524 | find_block: | |
525 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); | |
481ebd0d | 526 | sidx = align_idx(bdata, sidx, step); |
5f2809e6 | 527 | eidx = sidx + PFN_UP(size); |
ad09315c | 528 | |
5f2809e6 | 529 | if (sidx >= midx || eidx > midx) |
66d43e98 | 530 | break; |
1da177e4 | 531 | |
5f2809e6 JW |
532 | for (i = sidx; i < eidx; i++) |
533 | if (test_bit(i, bdata->node_bootmem_map)) { | |
481ebd0d | 534 | sidx = align_idx(bdata, i, step); |
5f2809e6 JW |
535 | if (sidx == i) |
536 | sidx += step; | |
537 | goto find_block; | |
538 | } | |
1da177e4 | 539 | |
627240aa | 540 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
5f2809e6 | 541 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
481ebd0d | 542 | start_off = align_off(bdata, bdata->last_end_off, align); |
5f2809e6 JW |
543 | else |
544 | start_off = PFN_PHYS(sidx); | |
545 | ||
546 | merge = PFN_DOWN(start_off) < sidx; | |
547 | end_off = start_off + size; | |
548 | ||
549 | bdata->last_end_off = end_off; | |
550 | bdata->hint_idx = PFN_UP(end_off); | |
551 | ||
552 | /* | |
553 | * Reserve the area now: | |
554 | */ | |
d747fa4b JW |
555 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
556 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) | |
557 | BUG(); | |
5f2809e6 | 558 | |
3560e249 JW |
559 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
560 | start_off); | |
5f2809e6 | 561 | memset(region, 0, size); |
008139d9 CM |
562 | /* |
563 | * The min_count is set to 0 so that bootmem allocated blocks | |
564 | * are never reported as leaks. | |
565 | */ | |
566 | kmemleak_alloc(region, size, 0, 0); | |
5f2809e6 | 567 | return region; |
1da177e4 LT |
568 | } |
569 | ||
0f3caba2 | 570 | if (fallback) { |
481ebd0d | 571 | sidx = align_idx(bdata, fallback - 1, step); |
0f3caba2 JW |
572 | fallback = 0; |
573 | goto find_block; | |
574 | } | |
575 | ||
576 | return NULL; | |
577 | } | |
578 | ||
d0c4f570 TH |
579 | static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata, |
580 | unsigned long size, unsigned long align, | |
581 | unsigned long goal, unsigned long limit) | |
582 | { | |
441c7e0a PE |
583 | if (WARN_ON_ONCE(slab_is_available())) |
584 | return kzalloc(size, GFP_NOWAIT); | |
585 | ||
d0c4f570 | 586 | #ifdef CONFIG_HAVE_ARCH_BOOTMEM |
433f13a7 JP |
587 | { |
588 | bootmem_data_t *p_bdata; | |
589 | ||
590 | p_bdata = bootmem_arch_preferred_node(bdata, size, align, | |
591 | goal, limit); | |
592 | if (p_bdata) | |
593 | return alloc_bootmem_core(p_bdata, size, align, | |
594 | goal, limit); | |
595 | } | |
d0c4f570 TH |
596 | #endif |
597 | return NULL; | |
598 | } | |
599 | ||
0f3caba2 JW |
600 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, |
601 | unsigned long align, | |
602 | unsigned long goal, | |
603 | unsigned long limit) | |
604 | { | |
605 | bootmem_data_t *bdata; | |
d0c4f570 | 606 | void *region; |
0f3caba2 JW |
607 | |
608 | restart: | |
d0c4f570 TH |
609 | region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit); |
610 | if (region) | |
611 | return region; | |
0f3caba2 | 612 | |
d0c4f570 | 613 | list_for_each_entry(bdata, &bdata_list, list) { |
0f3caba2 JW |
614 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
615 | continue; | |
3560e249 | 616 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
0f3caba2 JW |
617 | break; |
618 | ||
619 | region = alloc_bootmem_core(bdata, size, align, goal, limit); | |
620 | if (region) | |
621 | return region; | |
622 | } | |
623 | ||
5f2809e6 JW |
624 | if (goal) { |
625 | goal = 0; | |
0f3caba2 | 626 | goto restart; |
5f2809e6 | 627 | } |
2e5237da | 628 | |
5f2809e6 | 629 | return NULL; |
1da177e4 LT |
630 | } |
631 | ||
a66fd7da JW |
632 | /** |
633 | * __alloc_bootmem_nopanic - allocate boot memory without panicking | |
634 | * @size: size of the request in bytes | |
635 | * @align: alignment of the region | |
636 | * @goal: preferred starting address of the region | |
637 | * | |
638 | * The goal is dropped if it can not be satisfied and the allocation will | |
639 | * fall back to memory below @goal. | |
640 | * | |
641 | * Allocation may happen on any node in the system. | |
642 | * | |
643 | * Returns NULL on failure. | |
644 | */ | |
bb0923a6 | 645 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
0f3caba2 | 646 | unsigned long goal) |
1da177e4 | 647 | { |
08677214 YL |
648 | unsigned long limit = 0; |
649 | ||
08677214 | 650 | return ___alloc_bootmem_nopanic(size, align, goal, limit); |
0f3caba2 | 651 | } |
1da177e4 | 652 | |
0f3caba2 JW |
653 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
654 | unsigned long goal, unsigned long limit) | |
655 | { | |
656 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); | |
657 | ||
658 | if (mem) | |
659 | return mem; | |
660 | /* | |
661 | * Whoops, we cannot satisfy the allocation request. | |
662 | */ | |
663 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); | |
664 | panic("Out of memory"); | |
a8062231 AK |
665 | return NULL; |
666 | } | |
1da177e4 | 667 | |
a66fd7da JW |
668 | /** |
669 | * __alloc_bootmem - allocate boot memory | |
670 | * @size: size of the request in bytes | |
671 | * @align: alignment of the region | |
672 | * @goal: preferred starting address of the region | |
673 | * | |
674 | * The goal is dropped if it can not be satisfied and the allocation will | |
675 | * fall back to memory below @goal. | |
676 | * | |
677 | * Allocation may happen on any node in the system. | |
678 | * | |
679 | * The function panics if the request can not be satisfied. | |
680 | */ | |
bb0923a6 FBH |
681 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
682 | unsigned long goal) | |
a8062231 | 683 | { |
08677214 YL |
684 | unsigned long limit = 0; |
685 | ||
08677214 | 686 | return ___alloc_bootmem(size, align, goal, limit); |
1da177e4 LT |
687 | } |
688 | ||
4cc278b7 JW |
689 | static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata, |
690 | unsigned long size, unsigned long align, | |
691 | unsigned long goal, unsigned long limit) | |
692 | { | |
693 | void *ptr; | |
694 | ||
d0c4f570 TH |
695 | ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit); |
696 | if (ptr) | |
697 | return ptr; | |
698 | ||
4cc278b7 JW |
699 | ptr = alloc_bootmem_core(bdata, size, align, goal, limit); |
700 | if (ptr) | |
701 | return ptr; | |
702 | ||
703 | return ___alloc_bootmem(size, align, goal, limit); | |
704 | } | |
705 | ||
a66fd7da JW |
706 | /** |
707 | * __alloc_bootmem_node - allocate boot memory from a specific node | |
708 | * @pgdat: node to allocate from | |
709 | * @size: size of the request in bytes | |
710 | * @align: alignment of the region | |
711 | * @goal: preferred starting address of the region | |
712 | * | |
713 | * The goal is dropped if it can not be satisfied and the allocation will | |
714 | * fall back to memory below @goal. | |
715 | * | |
716 | * Allocation may fall back to any node in the system if the specified node | |
717 | * can not hold the requested memory. | |
718 | * | |
719 | * The function panics if the request can not be satisfied. | |
720 | */ | |
bb0923a6 FBH |
721 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
722 | unsigned long align, unsigned long goal) | |
1da177e4 | 723 | { |
c91c4773 PE |
724 | if (WARN_ON_ONCE(slab_is_available())) |
725 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
726 | ||
09325873 | 727 | return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0); |
08677214 YL |
728 | } |
729 | ||
730 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, | |
731 | unsigned long align, unsigned long goal) | |
732 | { | |
733 | #ifdef MAX_DMA32_PFN | |
734 | unsigned long end_pfn; | |
735 | ||
736 | if (WARN_ON_ONCE(slab_is_available())) | |
737 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
738 | ||
739 | /* update goal according ...MAX_DMA32_PFN */ | |
740 | end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
741 | ||
742 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && | |
743 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { | |
744 | void *ptr; | |
745 | unsigned long new_goal; | |
746 | ||
747 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; | |
08677214 YL |
748 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, |
749 | new_goal, 0); | |
08677214 YL |
750 | if (ptr) |
751 | return ptr; | |
752 | } | |
753 | #endif | |
754 | ||
755 | return __alloc_bootmem_node(pgdat, size, align, goal); | |
756 | ||
1da177e4 LT |
757 | } |
758 | ||
e70260aa | 759 | #ifdef CONFIG_SPARSEMEM |
a66fd7da JW |
760 | /** |
761 | * alloc_bootmem_section - allocate boot memory from a specific section | |
762 | * @size: size of the request in bytes | |
763 | * @section_nr: sparse map section to allocate from | |
764 | * | |
765 | * Return NULL on failure. | |
766 | */ | |
e70260aa YG |
767 | void * __init alloc_bootmem_section(unsigned long size, |
768 | unsigned long section_nr) | |
769 | { | |
75a56cfe JW |
770 | bootmem_data_t *bdata; |
771 | unsigned long pfn, goal, limit; | |
e70260aa YG |
772 | |
773 | pfn = section_nr_to_pfn(section_nr); | |
75a56cfe JW |
774 | goal = pfn << PAGE_SHIFT; |
775 | limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT; | |
776 | bdata = &bootmem_node_data[early_pfn_to_nid(pfn)]; | |
e70260aa | 777 | |
75a56cfe | 778 | return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit); |
e70260aa YG |
779 | } |
780 | #endif | |
781 | ||
b54bbf7b AK |
782 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, |
783 | unsigned long align, unsigned long goal) | |
784 | { | |
785 | void *ptr; | |
786 | ||
c91c4773 PE |
787 | if (WARN_ON_ONCE(slab_is_available())) |
788 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
789 | ||
d0c4f570 TH |
790 | ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0); |
791 | if (ptr) | |
792 | return ptr; | |
793 | ||
b54bbf7b AK |
794 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0); |
795 | if (ptr) | |
796 | return ptr; | |
797 | ||
798 | return __alloc_bootmem_nopanic(size, align, goal); | |
799 | } | |
800 | ||
dfd54cbc HC |
801 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
802 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL | |
803 | #endif | |
008857c1 | 804 | |
a66fd7da JW |
805 | /** |
806 | * __alloc_bootmem_low - allocate low boot memory | |
807 | * @size: size of the request in bytes | |
808 | * @align: alignment of the region | |
809 | * @goal: preferred starting address of the region | |
810 | * | |
811 | * The goal is dropped if it can not be satisfied and the allocation will | |
812 | * fall back to memory below @goal. | |
813 | * | |
814 | * Allocation may happen on any node in the system. | |
815 | * | |
816 | * The function panics if the request can not be satisfied. | |
817 | */ | |
bb0923a6 FBH |
818 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
819 | unsigned long goal) | |
008857c1 | 820 | { |
0f3caba2 | 821 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
008857c1 RT |
822 | } |
823 | ||
a66fd7da JW |
824 | /** |
825 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node | |
826 | * @pgdat: node to allocate from | |
827 | * @size: size of the request in bytes | |
828 | * @align: alignment of the region | |
829 | * @goal: preferred starting address of the region | |
830 | * | |
831 | * The goal is dropped if it can not be satisfied and the allocation will | |
832 | * fall back to memory below @goal. | |
833 | * | |
834 | * Allocation may fall back to any node in the system if the specified node | |
835 | * can not hold the requested memory. | |
836 | * | |
837 | * The function panics if the request can not be satisfied. | |
838 | */ | |
008857c1 RT |
839 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
840 | unsigned long align, unsigned long goal) | |
841 | { | |
c91c4773 PE |
842 | if (WARN_ON_ONCE(slab_is_available())) |
843 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
844 | ||
09325873 | 845 | return ___alloc_bootmem_node(pgdat->bdata, size, align, |
b8ab9f82 | 846 | goal, ARCH_LOW_ADDRESS_LIMIT); |
008857c1 | 847 | } |