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 | |
150 | * @addr: starting address of the range | |
151 | * @size: size of the range in bytes | |
152 | * | |
153 | * This is only useful when the bootmem allocator has already been torn | |
154 | * down, but we are still initializing the system. Pages are given directly | |
155 | * to the page allocator, no bootmem metadata is updated because it is gone. | |
156 | */ | |
157 | void __init free_bootmem_late(unsigned long addr, unsigned long size) | |
158 | { | |
159 | unsigned long cursor, end; | |
160 | ||
161 | kmemleak_free_part(__va(addr), size); | |
162 | ||
163 | cursor = PFN_UP(addr); | |
164 | end = PFN_DOWN(addr + size); | |
165 | ||
166 | for (; cursor < end; cursor++) { | |
167 | __free_pages_bootmem(pfn_to_page(cursor), 0); | |
168 | totalram_pages++; | |
169 | } | |
170 | } | |
171 | ||
223e8dc9 JW |
172 | static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) |
173 | { | |
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); | |
7f1290f2 JW |
201 | fixup_zone_present_pages(page_to_nid(pfn_to_page(start)), |
202 | start, start + BITS_PER_LONG); | |
223e8dc9 | 203 | count += BITS_PER_LONG; |
799f933a | 204 | start += BITS_PER_LONG; |
41546c17 JW |
205 | } else { |
206 | unsigned long off = 0; | |
207 | ||
6dccdcbe | 208 | vec >>= start & (BITS_PER_LONG - 1); |
549381e1 | 209 | while (vec) { |
41546c17 JW |
210 | if (vec & 1) { |
211 | page = pfn_to_page(start + off); | |
223e8dc9 | 212 | __free_pages_bootmem(page, 0); |
7f1290f2 JW |
213 | fixup_zone_present_pages( |
214 | page_to_nid(page), | |
215 | start + off, start + off + 1); | |
41546c17 | 216 | count++; |
223e8dc9 | 217 | } |
41546c17 JW |
218 | vec >>= 1; |
219 | off++; | |
223e8dc9 | 220 | } |
799f933a | 221 | start = ALIGN(start + 1, BITS_PER_LONG); |
223e8dc9 | 222 | } |
223e8dc9 JW |
223 | } |
224 | ||
223e8dc9 | 225 | page = virt_to_page(bdata->node_bootmem_map); |
3560e249 | 226 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
41546c17 JW |
227 | pages = bootmem_bootmap_pages(pages); |
228 | count += pages; | |
7f1290f2 JW |
229 | while (pages--) { |
230 | fixup_zone_present_pages(page_to_nid(page), | |
231 | page_to_pfn(page), page_to_pfn(page) + 1); | |
41546c17 | 232 | __free_pages_bootmem(page++, 0); |
7f1290f2 | 233 | } |
223e8dc9 | 234 | |
2e5237da JW |
235 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
236 | ||
223e8dc9 JW |
237 | return count; |
238 | } | |
239 | ||
a66fd7da JW |
240 | /** |
241 | * free_all_bootmem_node - release a node's free pages to the buddy allocator | |
242 | * @pgdat: node to be released | |
243 | * | |
244 | * Returns the number of pages actually released. | |
245 | */ | |
223e8dc9 JW |
246 | unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) |
247 | { | |
248 | register_page_bootmem_info_node(pgdat); | |
249 | return free_all_bootmem_core(pgdat->bdata); | |
250 | } | |
251 | ||
a66fd7da JW |
252 | /** |
253 | * free_all_bootmem - release free pages to the buddy allocator | |
254 | * | |
255 | * Returns the number of pages actually released. | |
256 | */ | |
223e8dc9 JW |
257 | unsigned long __init free_all_bootmem(void) |
258 | { | |
aa235fc7 YL |
259 | unsigned long total_pages = 0; |
260 | bootmem_data_t *bdata; | |
261 | ||
262 | list_for_each_entry(bdata, &bdata_list, list) | |
263 | total_pages += free_all_bootmem_core(bdata); | |
264 | ||
265 | return total_pages; | |
223e8dc9 JW |
266 | } |
267 | ||
d747fa4b JW |
268 | static void __init __free(bootmem_data_t *bdata, |
269 | unsigned long sidx, unsigned long eidx) | |
270 | { | |
271 | unsigned long idx; | |
272 | ||
273 | bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data, | |
3560e249 JW |
274 | sidx + bdata->node_min_pfn, |
275 | eidx + bdata->node_min_pfn); | |
d747fa4b | 276 | |
e2bf3cae JW |
277 | if (bdata->hint_idx > sidx) |
278 | bdata->hint_idx = sidx; | |
279 | ||
d747fa4b JW |
280 | for (idx = sidx; idx < eidx; idx++) |
281 | if (!test_and_clear_bit(idx, bdata->node_bootmem_map)) | |
282 | BUG(); | |
283 | } | |
284 | ||
285 | static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx, | |
286 | unsigned long eidx, int flags) | |
287 | { | |
288 | unsigned long idx; | |
289 | int exclusive = flags & BOOTMEM_EXCLUSIVE; | |
290 | ||
291 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", | |
292 | bdata - bootmem_node_data, | |
3560e249 JW |
293 | sidx + bdata->node_min_pfn, |
294 | eidx + bdata->node_min_pfn, | |
d747fa4b JW |
295 | flags); |
296 | ||
297 | for (idx = sidx; idx < eidx; idx++) | |
298 | if (test_and_set_bit(idx, bdata->node_bootmem_map)) { | |
299 | if (exclusive) { | |
300 | __free(bdata, sidx, idx); | |
301 | return -EBUSY; | |
302 | } | |
303 | bdebug("silent double reserve of PFN %lx\n", | |
3560e249 | 304 | idx + bdata->node_min_pfn); |
d747fa4b JW |
305 | } |
306 | return 0; | |
307 | } | |
308 | ||
e2bf3cae JW |
309 | static int __init mark_bootmem_node(bootmem_data_t *bdata, |
310 | unsigned long start, unsigned long end, | |
311 | int reserve, int flags) | |
223e8dc9 JW |
312 | { |
313 | unsigned long sidx, eidx; | |
223e8dc9 | 314 | |
e2bf3cae JW |
315 | bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n", |
316 | bdata - bootmem_node_data, start, end, reserve, flags); | |
223e8dc9 | 317 | |
3560e249 | 318 | BUG_ON(start < bdata->node_min_pfn); |
e2bf3cae | 319 | BUG_ON(end > bdata->node_low_pfn); |
223e8dc9 | 320 | |
3560e249 JW |
321 | sidx = start - bdata->node_min_pfn; |
322 | eidx = end - bdata->node_min_pfn; | |
223e8dc9 | 323 | |
e2bf3cae JW |
324 | if (reserve) |
325 | return __reserve(bdata, sidx, eidx, flags); | |
223e8dc9 | 326 | else |
e2bf3cae JW |
327 | __free(bdata, sidx, eidx); |
328 | return 0; | |
329 | } | |
330 | ||
331 | static int __init mark_bootmem(unsigned long start, unsigned long end, | |
332 | int reserve, int flags) | |
333 | { | |
334 | unsigned long pos; | |
335 | bootmem_data_t *bdata; | |
336 | ||
337 | pos = start; | |
338 | list_for_each_entry(bdata, &bdata_list, list) { | |
339 | int err; | |
340 | unsigned long max; | |
341 | ||
3560e249 JW |
342 | if (pos < bdata->node_min_pfn || |
343 | pos >= bdata->node_low_pfn) { | |
e2bf3cae JW |
344 | BUG_ON(pos != start); |
345 | continue; | |
346 | } | |
347 | ||
348 | max = min(bdata->node_low_pfn, end); | |
223e8dc9 | 349 | |
e2bf3cae JW |
350 | err = mark_bootmem_node(bdata, pos, max, reserve, flags); |
351 | if (reserve && err) { | |
352 | mark_bootmem(start, pos, 0, 0); | |
353 | return err; | |
354 | } | |
223e8dc9 | 355 | |
e2bf3cae JW |
356 | if (max == end) |
357 | return 0; | |
358 | pos = bdata->node_low_pfn; | |
359 | } | |
360 | BUG(); | |
223e8dc9 JW |
361 | } |
362 | ||
a66fd7da JW |
363 | /** |
364 | * free_bootmem_node - mark a page range as usable | |
365 | * @pgdat: node the range resides on | |
366 | * @physaddr: starting address of the range | |
367 | * @size: size of the range in bytes | |
368 | * | |
369 | * Partial pages will be considered reserved and left as they are. | |
370 | * | |
e2bf3cae | 371 | * The range must reside completely on the specified node. |
a66fd7da | 372 | */ |
223e8dc9 JW |
373 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
374 | unsigned long size) | |
375 | { | |
e2bf3cae JW |
376 | unsigned long start, end; |
377 | ||
ec3a354b CM |
378 | kmemleak_free_part(__va(physaddr), size); |
379 | ||
e2bf3cae JW |
380 | start = PFN_UP(physaddr); |
381 | end = PFN_DOWN(physaddr + size); | |
382 | ||
383 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); | |
223e8dc9 JW |
384 | } |
385 | ||
a66fd7da JW |
386 | /** |
387 | * free_bootmem - mark a page range as usable | |
388 | * @addr: starting address of the range | |
389 | * @size: size of the range in bytes | |
390 | * | |
391 | * Partial pages will be considered reserved and left as they are. | |
392 | * | |
e2bf3cae | 393 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 394 | */ |
223e8dc9 JW |
395 | void __init free_bootmem(unsigned long addr, unsigned long size) |
396 | { | |
e2bf3cae | 397 | unsigned long start, end; |
a5645a61 | 398 | |
ec3a354b CM |
399 | kmemleak_free_part(__va(addr), size); |
400 | ||
e2bf3cae JW |
401 | start = PFN_UP(addr); |
402 | end = PFN_DOWN(addr + size); | |
1da177e4 | 403 | |
e2bf3cae | 404 | mark_bootmem(start, end, 0, 0); |
1da177e4 LT |
405 | } |
406 | ||
a66fd7da JW |
407 | /** |
408 | * reserve_bootmem_node - mark a page range as reserved | |
409 | * @pgdat: node the range resides on | |
410 | * @physaddr: starting address of the range | |
411 | * @size: size of the range in bytes | |
412 | * @flags: reservation flags (see linux/bootmem.h) | |
413 | * | |
414 | * Partial pages will be reserved. | |
415 | * | |
e2bf3cae | 416 | * The range must reside completely on the specified node. |
a66fd7da | 417 | */ |
223e8dc9 JW |
418 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
419 | unsigned long size, int flags) | |
1da177e4 | 420 | { |
e2bf3cae | 421 | unsigned long start, end; |
1da177e4 | 422 | |
e2bf3cae JW |
423 | start = PFN_DOWN(physaddr); |
424 | end = PFN_UP(physaddr + size); | |
425 | ||
426 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); | |
223e8dc9 | 427 | } |
5a982cbc | 428 | |
a66fd7da | 429 | /** |
0d4ba4d7 | 430 | * reserve_bootmem - mark a page range as reserved |
a66fd7da JW |
431 | * @addr: starting address of the range |
432 | * @size: size of the range in bytes | |
433 | * @flags: reservation flags (see linux/bootmem.h) | |
434 | * | |
435 | * Partial pages will be reserved. | |
436 | * | |
e2bf3cae | 437 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 438 | */ |
223e8dc9 JW |
439 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
440 | int flags) | |
441 | { | |
e2bf3cae | 442 | unsigned long start, end; |
1da177e4 | 443 | |
e2bf3cae JW |
444 | start = PFN_DOWN(addr); |
445 | end = PFN_UP(addr + size); | |
223e8dc9 | 446 | |
e2bf3cae | 447 | return mark_bootmem(start, end, 1, flags); |
1da177e4 LT |
448 | } |
449 | ||
f88eff74 YL |
450 | int __weak __init reserve_bootmem_generic(unsigned long phys, unsigned long len, |
451 | int flags) | |
452 | { | |
453 | return reserve_bootmem(phys, len, flags); | |
454 | } | |
455 | ||
8aa043d7 JB |
456 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
457 | unsigned long idx, unsigned long step) | |
481ebd0d JW |
458 | { |
459 | unsigned long base = bdata->node_min_pfn; | |
460 | ||
461 | /* | |
462 | * Align the index with respect to the node start so that the | |
463 | * combination of both satisfies the requested alignment. | |
464 | */ | |
465 | ||
466 | return ALIGN(base + idx, step) - base; | |
467 | } | |
468 | ||
8aa043d7 JB |
469 | static unsigned long __init align_off(struct bootmem_data *bdata, |
470 | unsigned long off, unsigned long align) | |
481ebd0d JW |
471 | { |
472 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); | |
473 | ||
474 | /* Same as align_idx for byte offsets */ | |
475 | ||
476 | return ALIGN(base + off, align) - base; | |
477 | } | |
478 | ||
c6785b6b | 479 | static void * __init alloc_bootmem_bdata(struct bootmem_data *bdata, |
d0c4f570 TH |
480 | unsigned long size, unsigned long align, |
481 | unsigned long goal, unsigned long limit) | |
1da177e4 | 482 | { |
0f3caba2 | 483 | unsigned long fallback = 0; |
5f2809e6 JW |
484 | unsigned long min, max, start, sidx, midx, step; |
485 | ||
594fe1a0 JW |
486 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
487 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, | |
488 | align, goal, limit); | |
489 | ||
5f2809e6 JW |
490 | BUG_ON(!size); |
491 | BUG_ON(align & (align - 1)); | |
492 | BUG_ON(limit && goal + size > limit); | |
1da177e4 | 493 | |
7c309a64 CK |
494 | if (!bdata->node_bootmem_map) |
495 | return NULL; | |
496 | ||
3560e249 | 497 | min = bdata->node_min_pfn; |
5f2809e6 | 498 | max = bdata->node_low_pfn; |
9a2dc04c | 499 | |
5f2809e6 JW |
500 | goal >>= PAGE_SHIFT; |
501 | limit >>= PAGE_SHIFT; | |
502 | ||
503 | if (limit && max > limit) | |
504 | max = limit; | |
505 | if (max <= min) | |
9a2dc04c YL |
506 | return NULL; |
507 | ||
5f2809e6 | 508 | step = max(align >> PAGE_SHIFT, 1UL); |
281dd25c | 509 | |
5f2809e6 JW |
510 | if (goal && min < goal && goal < max) |
511 | start = ALIGN(goal, step); | |
512 | else | |
513 | start = ALIGN(min, step); | |
1da177e4 | 514 | |
481ebd0d | 515 | sidx = start - bdata->node_min_pfn; |
3560e249 | 516 | midx = max - bdata->node_min_pfn; |
1da177e4 | 517 | |
5f2809e6 | 518 | if (bdata->hint_idx > sidx) { |
0f3caba2 JW |
519 | /* |
520 | * Handle the valid case of sidx being zero and still | |
521 | * catch the fallback below. | |
522 | */ | |
523 | fallback = sidx + 1; | |
481ebd0d | 524 | sidx = align_idx(bdata, bdata->hint_idx, step); |
5f2809e6 | 525 | } |
1da177e4 | 526 | |
5f2809e6 JW |
527 | while (1) { |
528 | int merge; | |
529 | void *region; | |
530 | unsigned long eidx, i, start_off, end_off; | |
531 | find_block: | |
532 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); | |
481ebd0d | 533 | sidx = align_idx(bdata, sidx, step); |
5f2809e6 | 534 | eidx = sidx + PFN_UP(size); |
ad09315c | 535 | |
5f2809e6 | 536 | if (sidx >= midx || eidx > midx) |
66d43e98 | 537 | break; |
1da177e4 | 538 | |
5f2809e6 JW |
539 | for (i = sidx; i < eidx; i++) |
540 | if (test_bit(i, bdata->node_bootmem_map)) { | |
481ebd0d | 541 | sidx = align_idx(bdata, i, step); |
5f2809e6 JW |
542 | if (sidx == i) |
543 | sidx += step; | |
544 | goto find_block; | |
545 | } | |
1da177e4 | 546 | |
627240aa | 547 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
5f2809e6 | 548 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
481ebd0d | 549 | start_off = align_off(bdata, bdata->last_end_off, align); |
5f2809e6 JW |
550 | else |
551 | start_off = PFN_PHYS(sidx); | |
552 | ||
553 | merge = PFN_DOWN(start_off) < sidx; | |
554 | end_off = start_off + size; | |
555 | ||
556 | bdata->last_end_off = end_off; | |
557 | bdata->hint_idx = PFN_UP(end_off); | |
558 | ||
559 | /* | |
560 | * Reserve the area now: | |
561 | */ | |
d747fa4b JW |
562 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
563 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) | |
564 | BUG(); | |
5f2809e6 | 565 | |
3560e249 JW |
566 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
567 | start_off); | |
5f2809e6 | 568 | memset(region, 0, size); |
008139d9 CM |
569 | /* |
570 | * The min_count is set to 0 so that bootmem allocated blocks | |
571 | * are never reported as leaks. | |
572 | */ | |
573 | kmemleak_alloc(region, size, 0, 0); | |
5f2809e6 | 574 | return region; |
1da177e4 LT |
575 | } |
576 | ||
0f3caba2 | 577 | if (fallback) { |
481ebd0d | 578 | sidx = align_idx(bdata, fallback - 1, step); |
0f3caba2 JW |
579 | fallback = 0; |
580 | goto find_block; | |
581 | } | |
582 | ||
583 | return NULL; | |
584 | } | |
585 | ||
d0c4f570 TH |
586 | static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata, |
587 | unsigned long size, unsigned long align, | |
588 | unsigned long goal, unsigned long limit) | |
589 | { | |
441c7e0a PE |
590 | if (WARN_ON_ONCE(slab_is_available())) |
591 | return kzalloc(size, GFP_NOWAIT); | |
592 | ||
d0c4f570 | 593 | #ifdef CONFIG_HAVE_ARCH_BOOTMEM |
433f13a7 JP |
594 | { |
595 | bootmem_data_t *p_bdata; | |
596 | ||
597 | p_bdata = bootmem_arch_preferred_node(bdata, size, align, | |
598 | goal, limit); | |
599 | if (p_bdata) | |
c6785b6b | 600 | return alloc_bootmem_bdata(p_bdata, size, align, |
433f13a7 JP |
601 | goal, limit); |
602 | } | |
d0c4f570 TH |
603 | #endif |
604 | return NULL; | |
605 | } | |
606 | ||
c12ab504 | 607 | static void * __init alloc_bootmem_core(unsigned long size, |
0f3caba2 JW |
608 | unsigned long align, |
609 | unsigned long goal, | |
610 | unsigned long limit) | |
611 | { | |
612 | bootmem_data_t *bdata; | |
d0c4f570 | 613 | void *region; |
0f3caba2 | 614 | |
d0c4f570 TH |
615 | region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit); |
616 | if (region) | |
617 | return region; | |
0f3caba2 | 618 | |
d0c4f570 | 619 | list_for_each_entry(bdata, &bdata_list, list) { |
0f3caba2 JW |
620 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
621 | continue; | |
3560e249 | 622 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
0f3caba2 JW |
623 | break; |
624 | ||
c6785b6b | 625 | region = alloc_bootmem_bdata(bdata, size, align, goal, limit); |
0f3caba2 JW |
626 | if (region) |
627 | return region; | |
628 | } | |
629 | ||
c12ab504 JW |
630 | return NULL; |
631 | } | |
632 | ||
633 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, | |
634 | unsigned long align, | |
635 | unsigned long goal, | |
636 | unsigned long limit) | |
637 | { | |
638 | void *ptr; | |
639 | ||
640 | restart: | |
641 | ptr = alloc_bootmem_core(size, align, goal, limit); | |
642 | if (ptr) | |
643 | return ptr; | |
5f2809e6 JW |
644 | if (goal) { |
645 | goal = 0; | |
0f3caba2 | 646 | goto restart; |
5f2809e6 | 647 | } |
2e5237da | 648 | |
5f2809e6 | 649 | return NULL; |
1da177e4 LT |
650 | } |
651 | ||
a66fd7da JW |
652 | /** |
653 | * __alloc_bootmem_nopanic - allocate boot memory without panicking | |
654 | * @size: size of the request in bytes | |
655 | * @align: alignment of the region | |
656 | * @goal: preferred starting address of the region | |
657 | * | |
658 | * The goal is dropped if it can not be satisfied and the allocation will | |
659 | * fall back to memory below @goal. | |
660 | * | |
661 | * Allocation may happen on any node in the system. | |
662 | * | |
663 | * Returns NULL on failure. | |
664 | */ | |
bb0923a6 | 665 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
0f3caba2 | 666 | unsigned long goal) |
1da177e4 | 667 | { |
08677214 YL |
668 | unsigned long limit = 0; |
669 | ||
08677214 | 670 | return ___alloc_bootmem_nopanic(size, align, goal, limit); |
0f3caba2 | 671 | } |
1da177e4 | 672 | |
0f3caba2 JW |
673 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
674 | unsigned long goal, unsigned long limit) | |
675 | { | |
676 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); | |
677 | ||
678 | if (mem) | |
679 | return mem; | |
680 | /* | |
681 | * Whoops, we cannot satisfy the allocation request. | |
682 | */ | |
683 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); | |
684 | panic("Out of memory"); | |
a8062231 AK |
685 | return NULL; |
686 | } | |
1da177e4 | 687 | |
a66fd7da JW |
688 | /** |
689 | * __alloc_bootmem - allocate boot memory | |
690 | * @size: size of the request in bytes | |
691 | * @align: alignment of the region | |
692 | * @goal: preferred starting address of the region | |
693 | * | |
694 | * The goal is dropped if it can not be satisfied and the allocation will | |
695 | * fall back to memory below @goal. | |
696 | * | |
697 | * Allocation may happen on any node in the system. | |
698 | * | |
699 | * The function panics if the request can not be satisfied. | |
700 | */ | |
bb0923a6 FBH |
701 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
702 | unsigned long goal) | |
a8062231 | 703 | { |
08677214 YL |
704 | unsigned long limit = 0; |
705 | ||
08677214 | 706 | return ___alloc_bootmem(size, align, goal, limit); |
1da177e4 LT |
707 | } |
708 | ||
99ab7b19 | 709 | void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat, |
4cc278b7 JW |
710 | unsigned long size, unsigned long align, |
711 | unsigned long goal, unsigned long limit) | |
712 | { | |
713 | void *ptr; | |
714 | ||
ab381843 | 715 | again: |
e9079911 JW |
716 | ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, |
717 | align, goal, limit); | |
d0c4f570 TH |
718 | if (ptr) |
719 | return ptr; | |
720 | ||
c8f4a2d0 YL |
721 | /* do not panic in alloc_bootmem_bdata() */ |
722 | if (limit && goal + size > limit) | |
723 | limit = 0; | |
724 | ||
e9079911 | 725 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, goal, limit); |
4cc278b7 JW |
726 | if (ptr) |
727 | return ptr; | |
728 | ||
ab381843 JW |
729 | ptr = alloc_bootmem_core(size, align, goal, limit); |
730 | if (ptr) | |
731 | return ptr; | |
732 | ||
733 | if (goal) { | |
734 | goal = 0; | |
735 | goto again; | |
736 | } | |
737 | ||
421456ed JW |
738 | return NULL; |
739 | } | |
740 | ||
741 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, | |
742 | unsigned long align, unsigned long goal) | |
743 | { | |
744 | if (WARN_ON_ONCE(slab_is_available())) | |
745 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
746 | ||
e9079911 | 747 | return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
421456ed JW |
748 | } |
749 | ||
e9079911 | 750 | void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
421456ed JW |
751 | unsigned long align, unsigned long goal, |
752 | unsigned long limit) | |
753 | { | |
754 | void *ptr; | |
755 | ||
e9079911 | 756 | ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0); |
421456ed JW |
757 | if (ptr) |
758 | return ptr; | |
759 | ||
ab381843 JW |
760 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); |
761 | panic("Out of memory"); | |
762 | return NULL; | |
4cc278b7 JW |
763 | } |
764 | ||
a66fd7da JW |
765 | /** |
766 | * __alloc_bootmem_node - allocate boot memory from a specific node | |
767 | * @pgdat: node to allocate from | |
768 | * @size: size of the request in bytes | |
769 | * @align: alignment of the region | |
770 | * @goal: preferred starting address of the region | |
771 | * | |
772 | * The goal is dropped if it can not be satisfied and the allocation will | |
773 | * fall back to memory below @goal. | |
774 | * | |
775 | * Allocation may fall back to any node in the system if the specified node | |
776 | * can not hold the requested memory. | |
777 | * | |
778 | * The function panics if the request can not be satisfied. | |
779 | */ | |
bb0923a6 FBH |
780 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
781 | unsigned long align, unsigned long goal) | |
1da177e4 | 782 | { |
c91c4773 PE |
783 | if (WARN_ON_ONCE(slab_is_available())) |
784 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
785 | ||
e9079911 | 786 | return ___alloc_bootmem_node(pgdat, size, align, goal, 0); |
08677214 YL |
787 | } |
788 | ||
789 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, | |
790 | unsigned long align, unsigned long goal) | |
791 | { | |
792 | #ifdef MAX_DMA32_PFN | |
793 | unsigned long end_pfn; | |
794 | ||
795 | if (WARN_ON_ONCE(slab_is_available())) | |
796 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
797 | ||
798 | /* update goal according ...MAX_DMA32_PFN */ | |
799 | end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
800 | ||
801 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && | |
802 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { | |
803 | void *ptr; | |
804 | unsigned long new_goal; | |
805 | ||
806 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; | |
c6785b6b | 807 | ptr = alloc_bootmem_bdata(pgdat->bdata, size, align, |
08677214 | 808 | new_goal, 0); |
08677214 YL |
809 | if (ptr) |
810 | return ptr; | |
811 | } | |
812 | #endif | |
813 | ||
814 | return __alloc_bootmem_node(pgdat, size, align, goal); | |
815 | ||
1da177e4 LT |
816 | } |
817 | ||
dfd54cbc HC |
818 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
819 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL | |
820 | #endif | |
008857c1 | 821 | |
a66fd7da JW |
822 | /** |
823 | * __alloc_bootmem_low - allocate low boot memory | |
824 | * @size: size of the request in bytes | |
825 | * @align: alignment of the region | |
826 | * @goal: preferred starting address of the region | |
827 | * | |
828 | * The goal is dropped if it can not be satisfied and the allocation will | |
829 | * fall back to memory below @goal. | |
830 | * | |
831 | * Allocation may happen on any node in the system. | |
832 | * | |
833 | * The function panics if the request can not be satisfied. | |
834 | */ | |
bb0923a6 FBH |
835 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
836 | unsigned long goal) | |
008857c1 | 837 | { |
0f3caba2 | 838 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
008857c1 RT |
839 | } |
840 | ||
a66fd7da JW |
841 | /** |
842 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node | |
843 | * @pgdat: node to allocate from | |
844 | * @size: size of the request in bytes | |
845 | * @align: alignment of the region | |
846 | * @goal: preferred starting address of the region | |
847 | * | |
848 | * The goal is dropped if it can not be satisfied and the allocation will | |
849 | * fall back to memory below @goal. | |
850 | * | |
851 | * Allocation may fall back to any node in the system if the specified node | |
852 | * can not hold the requested memory. | |
853 | * | |
854 | * The function panics if the request can not be satisfied. | |
855 | */ | |
008857c1 RT |
856 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
857 | unsigned long align, unsigned long goal) | |
858 | { | |
c91c4773 PE |
859 | if (WARN_ON_ONCE(slab_is_available())) |
860 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
861 | ||
e9079911 JW |
862 | return ___alloc_bootmem_node(pgdat, size, align, |
863 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
008857c1 | 864 | } |