]>
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; | |
223e8dc9 | 188 | |
41546c17 | 189 | map = bdata->node_bootmem_map; |
3560e249 | 190 | idx = start - bdata->node_min_pfn; |
41546c17 | 191 | vec = ~map[idx / BITS_PER_LONG]; |
799f933a JW |
192 | /* |
193 | * If we have a properly aligned and fully unreserved | |
194 | * BITS_PER_LONG block of pages in front of us, free | |
195 | * it in one go. | |
196 | */ | |
197 | if (IS_ALIGNED(start, BITS_PER_LONG) && vec == ~0UL) { | |
41546c17 JW |
198 | int order = ilog2(BITS_PER_LONG); |
199 | ||
200 | __free_pages_bootmem(pfn_to_page(start), order); | |
223e8dc9 | 201 | count += BITS_PER_LONG; |
799f933a | 202 | start += BITS_PER_LONG; |
41546c17 JW |
203 | } else { |
204 | unsigned long off = 0; | |
205 | ||
6dccdcbe | 206 | vec >>= start & (BITS_PER_LONG - 1); |
549381e1 | 207 | while (vec) { |
41546c17 JW |
208 | if (vec & 1) { |
209 | page = pfn_to_page(start + off); | |
223e8dc9 | 210 | __free_pages_bootmem(page, 0); |
41546c17 | 211 | count++; |
223e8dc9 | 212 | } |
41546c17 JW |
213 | vec >>= 1; |
214 | off++; | |
223e8dc9 | 215 | } |
799f933a | 216 | start = ALIGN(start + 1, BITS_PER_LONG); |
223e8dc9 | 217 | } |
223e8dc9 JW |
218 | } |
219 | ||
223e8dc9 | 220 | page = virt_to_page(bdata->node_bootmem_map); |
3560e249 | 221 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
41546c17 JW |
222 | pages = bootmem_bootmap_pages(pages); |
223 | count += pages; | |
5576646f | 224 | while (pages--) |
41546c17 | 225 | __free_pages_bootmem(page++, 0); |
223e8dc9 | 226 | |
2e5237da JW |
227 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
228 | ||
223e8dc9 JW |
229 | return count; |
230 | } | |
231 | ||
9feedc9d JL |
232 | static void reset_node_lowmem_managed_pages(pg_data_t *pgdat) |
233 | { | |
234 | struct zone *z; | |
235 | ||
236 | /* | |
237 | * In free_area_init_core(), highmem zone's managed_pages is set to | |
238 | * present_pages, and bootmem allocator doesn't allocate from highmem | |
239 | * zones. So there's no need to recalculate managed_pages because all | |
240 | * highmem pages will be managed by the buddy system. Here highmem | |
241 | * zone also includes highmem movable zone. | |
242 | */ | |
243 | for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++) | |
244 | if (!is_highmem(z)) | |
245 | z->managed_pages = 0; | |
246 | } | |
247 | ||
a66fd7da JW |
248 | /** |
249 | * free_all_bootmem_node - release a node's free pages to the buddy allocator | |
250 | * @pgdat: node to be released | |
251 | * | |
252 | * Returns the number of pages actually released. | |
253 | */ | |
223e8dc9 JW |
254 | unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) |
255 | { | |
256 | register_page_bootmem_info_node(pgdat); | |
9feedc9d | 257 | reset_node_lowmem_managed_pages(pgdat); |
223e8dc9 JW |
258 | return free_all_bootmem_core(pgdat->bdata); |
259 | } | |
260 | ||
a66fd7da JW |
261 | /** |
262 | * free_all_bootmem - release free pages to the buddy allocator | |
263 | * | |
264 | * Returns the number of pages actually released. | |
265 | */ | |
223e8dc9 JW |
266 | unsigned long __init free_all_bootmem(void) |
267 | { | |
aa235fc7 YL |
268 | unsigned long total_pages = 0; |
269 | bootmem_data_t *bdata; | |
9feedc9d JL |
270 | struct pglist_data *pgdat; |
271 | ||
272 | for_each_online_pgdat(pgdat) | |
273 | reset_node_lowmem_managed_pages(pgdat); | |
aa235fc7 YL |
274 | |
275 | list_for_each_entry(bdata, &bdata_list, list) | |
276 | total_pages += free_all_bootmem_core(bdata); | |
277 | ||
278 | return total_pages; | |
223e8dc9 JW |
279 | } |
280 | ||
d747fa4b JW |
281 | static void __init __free(bootmem_data_t *bdata, |
282 | unsigned long sidx, unsigned long eidx) | |
283 | { | |
284 | unsigned long idx; | |
285 | ||
286 | bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data, | |
3560e249 JW |
287 | sidx + bdata->node_min_pfn, |
288 | eidx + bdata->node_min_pfn); | |
d747fa4b | 289 | |
e2bf3cae JW |
290 | if (bdata->hint_idx > sidx) |
291 | bdata->hint_idx = sidx; | |
292 | ||
d747fa4b JW |
293 | for (idx = sidx; idx < eidx; idx++) |
294 | if (!test_and_clear_bit(idx, bdata->node_bootmem_map)) | |
295 | BUG(); | |
296 | } | |
297 | ||
298 | static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx, | |
299 | unsigned long eidx, int flags) | |
300 | { | |
301 | unsigned long idx; | |
302 | int exclusive = flags & BOOTMEM_EXCLUSIVE; | |
303 | ||
304 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", | |
305 | bdata - bootmem_node_data, | |
3560e249 JW |
306 | sidx + bdata->node_min_pfn, |
307 | eidx + bdata->node_min_pfn, | |
d747fa4b JW |
308 | flags); |
309 | ||
310 | for (idx = sidx; idx < eidx; idx++) | |
311 | if (test_and_set_bit(idx, bdata->node_bootmem_map)) { | |
312 | if (exclusive) { | |
313 | __free(bdata, sidx, idx); | |
314 | return -EBUSY; | |
315 | } | |
316 | bdebug("silent double reserve of PFN %lx\n", | |
3560e249 | 317 | idx + bdata->node_min_pfn); |
d747fa4b JW |
318 | } |
319 | return 0; | |
320 | } | |
321 | ||
e2bf3cae JW |
322 | static int __init mark_bootmem_node(bootmem_data_t *bdata, |
323 | unsigned long start, unsigned long end, | |
324 | int reserve, int flags) | |
223e8dc9 JW |
325 | { |
326 | unsigned long sidx, eidx; | |
223e8dc9 | 327 | |
e2bf3cae JW |
328 | bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n", |
329 | bdata - bootmem_node_data, start, end, reserve, flags); | |
223e8dc9 | 330 | |
3560e249 | 331 | BUG_ON(start < bdata->node_min_pfn); |
e2bf3cae | 332 | BUG_ON(end > bdata->node_low_pfn); |
223e8dc9 | 333 | |
3560e249 JW |
334 | sidx = start - bdata->node_min_pfn; |
335 | eidx = end - bdata->node_min_pfn; | |
223e8dc9 | 336 | |
e2bf3cae JW |
337 | if (reserve) |
338 | return __reserve(bdata, sidx, eidx, flags); | |
223e8dc9 | 339 | else |
e2bf3cae JW |
340 | __free(bdata, sidx, eidx); |
341 | return 0; | |
342 | } | |
343 | ||
344 | static int __init mark_bootmem(unsigned long start, unsigned long end, | |
345 | int reserve, int flags) | |
346 | { | |
347 | unsigned long pos; | |
348 | bootmem_data_t *bdata; | |
349 | ||
350 | pos = start; | |
351 | list_for_each_entry(bdata, &bdata_list, list) { | |
352 | int err; | |
353 | unsigned long max; | |
354 | ||
3560e249 JW |
355 | if (pos < bdata->node_min_pfn || |
356 | pos >= bdata->node_low_pfn) { | |
e2bf3cae JW |
357 | BUG_ON(pos != start); |
358 | continue; | |
359 | } | |
360 | ||
361 | max = min(bdata->node_low_pfn, end); | |
223e8dc9 | 362 | |
e2bf3cae JW |
363 | err = mark_bootmem_node(bdata, pos, max, reserve, flags); |
364 | if (reserve && err) { | |
365 | mark_bootmem(start, pos, 0, 0); | |
366 | return err; | |
367 | } | |
223e8dc9 | 368 | |
e2bf3cae JW |
369 | if (max == end) |
370 | return 0; | |
371 | pos = bdata->node_low_pfn; | |
372 | } | |
373 | BUG(); | |
223e8dc9 JW |
374 | } |
375 | ||
a66fd7da JW |
376 | /** |
377 | * free_bootmem_node - mark a page range as usable | |
378 | * @pgdat: node the range resides on | |
379 | * @physaddr: starting address of the range | |
380 | * @size: size of the range in bytes | |
381 | * | |
382 | * Partial pages will be considered reserved and left as they are. | |
383 | * | |
e2bf3cae | 384 | * The range must reside completely on the specified node. |
a66fd7da | 385 | */ |
223e8dc9 JW |
386 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
387 | unsigned long size) | |
388 | { | |
e2bf3cae JW |
389 | unsigned long start, end; |
390 | ||
ec3a354b CM |
391 | kmemleak_free_part(__va(physaddr), size); |
392 | ||
e2bf3cae JW |
393 | start = PFN_UP(physaddr); |
394 | end = PFN_DOWN(physaddr + size); | |
395 | ||
396 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); | |
223e8dc9 JW |
397 | } |
398 | ||
a66fd7da JW |
399 | /** |
400 | * free_bootmem - mark a page range as usable | |
81df9bff | 401 | * @addr: starting physical address of the range |
a66fd7da JW |
402 | * @size: size of the range in bytes |
403 | * | |
404 | * Partial pages will be considered reserved and left as they are. | |
405 | * | |
e2bf3cae | 406 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 407 | */ |
81df9bff | 408 | void __init free_bootmem(unsigned long physaddr, unsigned long size) |
223e8dc9 | 409 | { |
e2bf3cae | 410 | unsigned long start, end; |
a5645a61 | 411 | |
81df9bff | 412 | kmemleak_free_part(__va(physaddr), size); |
ec3a354b | 413 | |
81df9bff JK |
414 | start = PFN_UP(physaddr); |
415 | end = PFN_DOWN(physaddr + size); | |
1da177e4 | 416 | |
e2bf3cae | 417 | mark_bootmem(start, end, 0, 0); |
1da177e4 LT |
418 | } |
419 | ||
a66fd7da JW |
420 | /** |
421 | * reserve_bootmem_node - mark a page range as reserved | |
422 | * @pgdat: node the range resides on | |
423 | * @physaddr: starting address of the range | |
424 | * @size: size of the range in bytes | |
425 | * @flags: reservation flags (see linux/bootmem.h) | |
426 | * | |
427 | * Partial pages will be reserved. | |
428 | * | |
e2bf3cae | 429 | * The range must reside completely on the specified node. |
a66fd7da | 430 | */ |
223e8dc9 JW |
431 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
432 | unsigned long size, int flags) | |
1da177e4 | 433 | { |
e2bf3cae | 434 | unsigned long start, end; |
1da177e4 | 435 | |
e2bf3cae JW |
436 | start = PFN_DOWN(physaddr); |
437 | end = PFN_UP(physaddr + size); | |
438 | ||
439 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); | |
223e8dc9 | 440 | } |
5a982cbc | 441 | |
a66fd7da | 442 | /** |
0d4ba4d7 | 443 | * reserve_bootmem - mark a page range as reserved |
a66fd7da JW |
444 | * @addr: starting address of the range |
445 | * @size: size of the range in bytes | |
446 | * @flags: reservation flags (see linux/bootmem.h) | |
447 | * | |
448 | * Partial pages will be reserved. | |
449 | * | |
e2bf3cae | 450 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 451 | */ |
223e8dc9 JW |
452 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
453 | int flags) | |
454 | { | |
e2bf3cae | 455 | unsigned long start, end; |
1da177e4 | 456 | |
e2bf3cae JW |
457 | start = PFN_DOWN(addr); |
458 | end = PFN_UP(addr + size); | |
223e8dc9 | 459 | |
e2bf3cae | 460 | return mark_bootmem(start, end, 1, flags); |
1da177e4 LT |
461 | } |
462 | ||
8aa043d7 JB |
463 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
464 | unsigned long idx, unsigned long step) | |
481ebd0d JW |
465 | { |
466 | unsigned long base = bdata->node_min_pfn; | |
467 | ||
468 | /* | |
469 | * Align the index with respect to the node start so that the | |
470 | * combination of both satisfies the requested alignment. | |
471 | */ | |
472 | ||
473 | return ALIGN(base + idx, step) - base; | |
474 | } | |
475 | ||
8aa043d7 JB |
476 | static unsigned long __init align_off(struct bootmem_data *bdata, |
477 | unsigned long off, unsigned long align) | |
481ebd0d JW |
478 | { |
479 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); | |
480 | ||
481 | /* Same as align_idx for byte offsets */ | |
482 | ||
483 | return ALIGN(base + off, align) - base; | |
484 | } | |
485 | ||
c6785b6b | 486 | static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata, |
d0c4f570 TH |
487 | unsigned long size, unsigned long align, |
488 | unsigned long goal, unsigned long limit) | |
1da177e4 | 489 | { |
0f3caba2 | 490 | unsigned long fallback = 0; |
5f2809e6 JW |
491 | unsigned long min, max, start, sidx, midx, step; |
492 | ||
594fe1a0 JW |
493 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
494 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, | |
495 | align, goal, limit); | |
496 | ||
5f2809e6 JW |
497 | BUG_ON(!size); |
498 | BUG_ON(align & (align - 1)); | |
499 | BUG_ON(limit && goal + size > limit); | |
1da177e4 | 500 | |
7c309a64 CK |
501 | if (!bdata->node_bootmem_map) |
502 | return NULL; | |
503 | ||
3560e249 | 504 | min = bdata->node_min_pfn; |
5f2809e6 | 505 | max = bdata->node_low_pfn; |
9a2dc04c | 506 | |
5f2809e6 JW |
507 | goal >>= PAGE_SHIFT; |
508 | limit >>= PAGE_SHIFT; | |
509 | ||
510 | if (limit && max > limit) | |
511 | max = limit; | |
512 | if (max <= min) | |
9a2dc04c YL |
513 | return NULL; |
514 | ||
5f2809e6 | 515 | step = max(align >> PAGE_SHIFT, 1UL); |
281dd25c | 516 | |
5f2809e6 JW |
517 | if (goal && min < goal && goal < max) |
518 | start = ALIGN(goal, step); | |
519 | else | |
520 | start = ALIGN(min, step); | |
1da177e4 | 521 | |
481ebd0d | 522 | sidx = start - bdata->node_min_pfn; |
3560e249 | 523 | midx = max - bdata->node_min_pfn; |
1da177e4 | 524 | |
5f2809e6 | 525 | if (bdata->hint_idx > sidx) { |
0f3caba2 JW |
526 | /* |
527 | * Handle the valid case of sidx being zero and still | |
528 | * catch the fallback below. | |
529 | */ | |
530 | fallback = sidx + 1; | |
481ebd0d | 531 | sidx = align_idx(bdata, bdata->hint_idx, step); |
5f2809e6 | 532 | } |
1da177e4 | 533 | |
5f2809e6 JW |
534 | while (1) { |
535 | int merge; | |
536 | void *region; | |
537 | unsigned long eidx, i, start_off, end_off; | |
538 | find_block: | |
539 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); | |
481ebd0d | 540 | sidx = align_idx(bdata, sidx, step); |
5f2809e6 | 541 | eidx = sidx + PFN_UP(size); |
ad09315c | 542 | |
5f2809e6 | 543 | if (sidx >= midx || eidx > midx) |
66d43e98 | 544 | break; |
1da177e4 | 545 | |
5f2809e6 JW |
546 | for (i = sidx; i < eidx; i++) |
547 | if (test_bit(i, bdata->node_bootmem_map)) { | |
481ebd0d | 548 | sidx = align_idx(bdata, i, step); |
5f2809e6 JW |
549 | if (sidx == i) |
550 | sidx += step; | |
551 | goto find_block; | |
552 | } | |
1da177e4 | 553 | |
627240aa | 554 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
5f2809e6 | 555 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
481ebd0d | 556 | start_off = align_off(bdata, bdata->last_end_off, align); |
5f2809e6 JW |
557 | else |
558 | start_off = PFN_PHYS(sidx); | |
559 | ||
560 | merge = PFN_DOWN(start_off) < sidx; | |
561 | end_off = start_off + size; | |
562 | ||
563 | bdata->last_end_off = end_off; | |
564 | bdata->hint_idx = PFN_UP(end_off); | |
565 | ||
566 | /* | |
567 | * Reserve the area now: | |
568 | */ | |
d747fa4b JW |
569 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
570 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) | |
571 | BUG(); | |
5f2809e6 | 572 | |
3560e249 JW |
573 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
574 | start_off); | |
5f2809e6 | 575 | memset(region, 0, size); |
008139d9 CM |
576 | /* |
577 | * The min_count is set to 0 so that bootmem allocated blocks | |
578 | * are never reported as leaks. | |
579 | */ | |
580 | kmemleak_alloc(region, size, 0, 0); | |
5f2809e6 | 581 | return region; |
1da177e4 LT |
582 | } |
583 | ||
0f3caba2 | 584 | if (fallback) { |
481ebd0d | 585 | sidx = align_idx(bdata, fallback - 1, step); |
0f3caba2 JW |
586 | fallback = 0; |
587 | goto find_block; | |
588 | } | |
589 | ||
590 | return NULL; | |
591 | } | |
592 | ||
c12ab504 | 593 | static void * __init alloc_bootmem_core(unsigned long size, |
0f3caba2 JW |
594 | unsigned long align, |
595 | unsigned long goal, | |
596 | unsigned long limit) | |
597 | { | |
598 | bootmem_data_t *bdata; | |
d0c4f570 | 599 | void *region; |
0f3caba2 | 600 | |
3f7dfe24 JK |
601 | if (WARN_ON_ONCE(slab_is_available())) |
602 | return kzalloc(size, GFP_NOWAIT); | |
0f3caba2 | 603 | |
d0c4f570 | 604 | list_for_each_entry(bdata, &bdata_list, list) { |
0f3caba2 JW |
605 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
606 | continue; | |
3560e249 | 607 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
0f3caba2 JW |
608 | break; |
609 | ||
c6785b6b | 610 | region = alloc_bootmem_bdata(bdata, size, align, goal, limit); |
0f3caba2 JW |
611 | if (region) |
612 | return region; | |
613 | } | |
614 | ||
c12ab504 JW |
615 | return NULL; |
616 | } | |
617 | ||
618 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, | |
619 | unsigned long align, | |
620 | unsigned long goal, | |
621 | unsigned long limit) | |
622 | { | |
623 | void *ptr; | |
624 | ||
625 | restart: | |
626 | ptr = alloc_bootmem_core(size, align, goal, limit); | |
627 | if (ptr) | |
628 | return ptr; | |
5f2809e6 JW |
629 | if (goal) { |
630 | goal = 0; | |
0f3caba2 | 631 | goto restart; |
5f2809e6 | 632 | } |
2e5237da | 633 | |
5f2809e6 | 634 | return NULL; |
1da177e4 LT |
635 | } |
636 | ||
a66fd7da JW |
637 | /** |
638 | * __alloc_bootmem_nopanic - allocate boot memory without panicking | |
639 | * @size: size of the request in bytes | |
640 | * @align: alignment of the region | |
641 | * @goal: preferred starting address of the region | |
642 | * | |
643 | * The goal is dropped if it can not be satisfied and the allocation will | |
644 | * fall back to memory below @goal. | |
645 | * | |
646 | * Allocation may happen on any node in the system. | |
647 | * | |
648 | * Returns NULL on failure. | |
649 | */ | |
bb0923a6 | 650 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
0f3caba2 | 651 | unsigned long goal) |
1da177e4 | 652 | { |
08677214 YL |
653 | unsigned long limit = 0; |
654 | ||
08677214 | 655 | return ___alloc_bootmem_nopanic(size, align, goal, limit); |
0f3caba2 | 656 | } |
1da177e4 | 657 | |
0f3caba2 JW |
658 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
659 | unsigned long goal, unsigned long limit) | |
660 | { | |
661 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); | |
662 | ||
663 | if (mem) | |
664 | return mem; | |
665 | /* | |
666 | * Whoops, we cannot satisfy the allocation request. | |
667 | */ | |
668 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); | |
669 | panic("Out of memory"); | |
a8062231 AK |
670 | return NULL; |
671 | } | |
1da177e4 | 672 | |
a66fd7da JW |
673 | /** |
674 | * __alloc_bootmem - allocate boot memory | |
675 | * @size: size of the request in bytes | |
676 | * @align: alignment of the region | |
677 | * @goal: preferred starting address of the region | |
678 | * | |
679 | * The goal is dropped if it can not be satisfied and the allocation will | |
680 | * fall back to memory below @goal. | |
681 | * | |
682 | * Allocation may happen on any node in the system. | |
683 | * | |
684 | * The function panics if the request can not be satisfied. | |
685 | */ | |
bb0923a6 FBH |
686 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
687 | unsigned long goal) | |
a8062231 | 688 | { |
08677214 YL |
689 | unsigned long limit = 0; |
690 | ||
08677214 | 691 | return ___alloc_bootmem(size, align, goal, limit); |
1da177e4 LT |
692 | } |
693 | ||
99ab7b19 | 694 | void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat, |
4cc278b7 JW |
695 | unsigned long size, unsigned long align, |
696 | unsigned long goal, unsigned long limit) | |
697 | { | |
698 | void *ptr; | |
699 | ||
3f7dfe24 JK |
700 | if (WARN_ON_ONCE(slab_is_available())) |
701 | return kzalloc(size, GFP_NOWAIT); | |
ab381843 | 702 | again: |
d0c4f570 | 703 | |
c8f4a2d0 YL |
704 | /* do not panic in alloc_bootmem_bdata() */ |
705 | if (limit && goal + size > limit) | |
706 | limit = 0; | |
707 | ||
e9079911 | 708 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit); |
4cc278b7 JW |
709 | if (ptr) |
710 | return ptr; | |
711 | ||
ab381843 JW |
712 | ptr = alloc_bootmem_core(size, align, goal, limit); |
713 | if (ptr) | |
714 | return ptr; | |
715 | ||
716 | if (goal) { | |
717 | goal = 0; | |
718 | goto again; | |
719 | } | |
720 | ||
421456ed JW |
721 | return NULL; |
722 | } | |
723 | ||
724 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, | |
725 | unsigned long align, unsigned long goal) | |
726 | { | |
727 | if (WARN_ON_ONCE(slab_is_available())) | |
728 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
729 | ||
e9079911 | 730 | return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
421456ed JW |
731 | } |
732 | ||
e9079911 | 733 | void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
421456ed JW |
734 | unsigned long align, unsigned long goal, |
735 | unsigned long limit) | |
736 | { | |
737 | void *ptr; | |
738 | ||
e9079911 | 739 | ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
421456ed JW |
740 | if (ptr) |
741 | return ptr; | |
742 | ||
ab381843 JW |
743 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); |
744 | panic("Out of memory"); | |
745 | return NULL; | |
4cc278b7 JW |
746 | } |
747 | ||
a66fd7da JW |
748 | /** |
749 | * __alloc_bootmem_node - allocate boot memory from a specific node | |
750 | * @pgdat: node to allocate from | |
751 | * @size: size of the request in bytes | |
752 | * @align: alignment of the region | |
753 | * @goal: preferred starting address of the region | |
754 | * | |
755 | * The goal is dropped if it can not be satisfied and the allocation will | |
756 | * fall back to memory below @goal. | |
757 | * | |
758 | * Allocation may fall back to any node in the system if the specified node | |
759 | * can not hold the requested memory. | |
760 | * | |
761 | * The function panics if the request can not be satisfied. | |
762 | */ | |
bb0923a6 FBH |
763 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
764 | unsigned long align, unsigned long goal) | |
1da177e4 | 765 | { |
c91c4773 PE |
766 | if (WARN_ON_ONCE(slab_is_available())) |
767 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
768 | ||
e9079911 | 769 | return ___alloc_bootmem_node(pgdat, size, align, goal, 0); |
08677214 YL |
770 | } |
771 | ||
772 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, | |
773 | unsigned long align, unsigned long goal) | |
774 | { | |
775 | #ifdef MAX_DMA32_PFN | |
776 | unsigned long end_pfn; | |
777 | ||
778 | if (WARN_ON_ONCE(slab_is_available())) | |
779 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
780 | ||
781 | /* update goal according ...MAX_DMA32_PFN */ | |
782 | end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
783 | ||
784 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && | |
785 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { | |
786 | void *ptr; | |
787 | unsigned long new_goal; | |
788 | ||
789 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; | |
c6785b6b | 790 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, |
08677214 | 791 | new_goal, 0); |
08677214 YL |
792 | if (ptr) |
793 | return ptr; | |
794 | } | |
795 | #endif | |
796 | ||
797 | return __alloc_bootmem_node(pgdat, size, align, goal); | |
798 | ||
1da177e4 LT |
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 | ||
e9079911 JW |
845 | return ___alloc_bootmem_node(pgdat, size, align, |
846 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
008857c1 | 847 | } |