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