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