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> |
1da177e4 | 15 | #include <linux/module.h> |
ec3a354b | 16 | #include <linux/kmemleak.h> |
08677214 | 17 | #include <linux/range.h> |
e786e86a FBH |
18 | |
19 | #include <asm/bug.h> | |
1da177e4 | 20 | #include <asm/io.h> |
dfd54cbc | 21 | #include <asm/processor.h> |
e786e86a | 22 | |
1da177e4 LT |
23 | #include "internal.h" |
24 | ||
1da177e4 LT |
25 | unsigned long max_low_pfn; |
26 | unsigned long min_low_pfn; | |
27 | unsigned long max_pfn; | |
28 | ||
92aa63a5 VG |
29 | #ifdef CONFIG_CRASH_DUMP |
30 | /* | |
31 | * If we have booted due to a crash, max_pfn will be a very low value. We need | |
32 | * to know the amount of memory that the previous kernel used. | |
33 | */ | |
34 | unsigned long saved_max_pfn; | |
35 | #endif | |
36 | ||
08677214 | 37 | #ifndef CONFIG_NO_BOOTMEM |
b61bfa3c JW |
38 | bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata; |
39 | ||
636cc40c JW |
40 | static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list); |
41 | ||
2e5237da JW |
42 | static int bootmem_debug; |
43 | ||
44 | static int __init bootmem_debug_setup(char *buf) | |
45 | { | |
46 | bootmem_debug = 1; | |
47 | return 0; | |
48 | } | |
49 | early_param("bootmem_debug", bootmem_debug_setup); | |
50 | ||
51 | #define bdebug(fmt, args...) ({ \ | |
52 | if (unlikely(bootmem_debug)) \ | |
53 | printk(KERN_INFO \ | |
54 | "bootmem::%s " fmt, \ | |
80a914dc | 55 | __func__, ## args); \ |
2e5237da JW |
56 | }) |
57 | ||
df049a5f | 58 | static unsigned long __init bootmap_bytes(unsigned long pages) |
223e8dc9 | 59 | { |
df049a5f | 60 | unsigned long bytes = (pages + 7) / 8; |
223e8dc9 | 61 | |
df049a5f | 62 | return ALIGN(bytes, sizeof(long)); |
223e8dc9 JW |
63 | } |
64 | ||
a66fd7da JW |
65 | /** |
66 | * bootmem_bootmap_pages - calculate bitmap size in pages | |
67 | * @pages: number of pages the bitmap has to represent | |
68 | */ | |
f71bf0ca | 69 | unsigned long __init bootmem_bootmap_pages(unsigned long pages) |
1da177e4 | 70 | { |
df049a5f | 71 | unsigned long bytes = bootmap_bytes(pages); |
1da177e4 | 72 | |
df049a5f | 73 | return PAGE_ALIGN(bytes) >> PAGE_SHIFT; |
1da177e4 | 74 | } |
f71bf0ca | 75 | |
679bc9fb KH |
76 | /* |
77 | * link bdata in order | |
78 | */ | |
69d49e68 | 79 | static void __init link_bootmem(bootmem_data_t *bdata) |
679bc9fb | 80 | { |
636cc40c | 81 | struct list_head *iter; |
f71bf0ca | 82 | |
636cc40c JW |
83 | list_for_each(iter, &bdata_list) { |
84 | bootmem_data_t *ent; | |
85 | ||
86 | ent = list_entry(iter, bootmem_data_t, list); | |
3560e249 | 87 | if (bdata->node_min_pfn < ent->node_min_pfn) |
636cc40c | 88 | break; |
679bc9fb | 89 | } |
636cc40c | 90 | list_add_tail(&bdata->list, iter); |
679bc9fb KH |
91 | } |
92 | ||
1da177e4 LT |
93 | /* |
94 | * Called once to set up the allocator itself. | |
95 | */ | |
8ae04463 | 96 | static unsigned long __init init_bootmem_core(bootmem_data_t *bdata, |
1da177e4 LT |
97 | unsigned long mapstart, unsigned long start, unsigned long end) |
98 | { | |
bbc7b92e | 99 | unsigned long mapsize; |
1da177e4 | 100 | |
2dbb51c4 | 101 | mminit_validate_memmodel_limits(&start, &end); |
bbc7b92e | 102 | bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart)); |
3560e249 | 103 | bdata->node_min_pfn = start; |
1da177e4 | 104 | bdata->node_low_pfn = end; |
679bc9fb | 105 | link_bootmem(bdata); |
1da177e4 LT |
106 | |
107 | /* | |
108 | * Initially all pages are reserved - setup_arch() has to | |
109 | * register free RAM areas explicitly. | |
110 | */ | |
df049a5f | 111 | mapsize = bootmap_bytes(end - start); |
1da177e4 LT |
112 | memset(bdata->node_bootmem_map, 0xff, mapsize); |
113 | ||
2e5237da JW |
114 | bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n", |
115 | bdata - bootmem_node_data, start, mapstart, end, mapsize); | |
116 | ||
1da177e4 LT |
117 | return mapsize; |
118 | } | |
119 | ||
a66fd7da JW |
120 | /** |
121 | * init_bootmem_node - register a node as boot memory | |
122 | * @pgdat: node to register | |
123 | * @freepfn: pfn where the bitmap for this node is to be placed | |
124 | * @startpfn: first pfn on the node | |
125 | * @endpfn: first pfn after the node | |
126 | * | |
127 | * Returns the number of bytes needed to hold the bitmap for this node. | |
128 | */ | |
223e8dc9 JW |
129 | unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn, |
130 | unsigned long startpfn, unsigned long endpfn) | |
131 | { | |
132 | return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn); | |
133 | } | |
134 | ||
a66fd7da JW |
135 | /** |
136 | * init_bootmem - register boot memory | |
137 | * @start: pfn where the bitmap is to be placed | |
138 | * @pages: number of available physical pages | |
139 | * | |
140 | * Returns the number of bytes needed to hold the bitmap. | |
141 | */ | |
223e8dc9 JW |
142 | unsigned long __init init_bootmem(unsigned long start, unsigned long pages) |
143 | { | |
144 | max_low_pfn = pages; | |
145 | min_low_pfn = start; | |
146 | return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages); | |
147 | } | |
08677214 | 148 | #endif |
9f993ac3 FT |
149 | /* |
150 | * free_bootmem_late - free bootmem pages directly to page allocator | |
151 | * @addr: starting address of the range | |
152 | * @size: size of the range in bytes | |
153 | * | |
154 | * This is only useful when the bootmem allocator has already been torn | |
155 | * down, but we are still initializing the system. Pages are given directly | |
156 | * to the page allocator, no bootmem metadata is updated because it is gone. | |
157 | */ | |
158 | void __init free_bootmem_late(unsigned long addr, unsigned long size) | |
159 | { | |
160 | unsigned long cursor, end; | |
161 | ||
162 | kmemleak_free_part(__va(addr), size); | |
163 | ||
164 | cursor = PFN_UP(addr); | |
165 | end = PFN_DOWN(addr + size); | |
166 | ||
167 | for (; cursor < end; cursor++) { | |
168 | __free_pages_bootmem(pfn_to_page(cursor), 0); | |
169 | totalram_pages++; | |
170 | } | |
171 | } | |
172 | ||
08677214 YL |
173 | #ifdef CONFIG_NO_BOOTMEM |
174 | static void __init __free_pages_memory(unsigned long start, unsigned long end) | |
175 | { | |
176 | int i; | |
177 | unsigned long start_aligned, end_aligned; | |
178 | int order = ilog2(BITS_PER_LONG); | |
179 | ||
180 | start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1); | |
181 | end_aligned = end & ~(BITS_PER_LONG - 1); | |
182 | ||
183 | if (end_aligned <= start_aligned) { | |
08677214 YL |
184 | for (i = start; i < end; i++) |
185 | __free_pages_bootmem(pfn_to_page(i), 0); | |
186 | ||
187 | return; | |
188 | } | |
189 | ||
08677214 YL |
190 | for (i = start; i < start_aligned; i++) |
191 | __free_pages_bootmem(pfn_to_page(i), 0); | |
192 | ||
193 | for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG) | |
194 | __free_pages_bootmem(pfn_to_page(i), order); | |
195 | ||
196 | for (i = end_aligned; i < end; i++) | |
197 | __free_pages_bootmem(pfn_to_page(i), 0); | |
198 | } | |
199 | ||
200 | unsigned long __init free_all_memory_core_early(int nodeid) | |
201 | { | |
202 | int i; | |
203 | u64 start, end; | |
204 | unsigned long count = 0; | |
205 | struct range *range = NULL; | |
206 | int nr_range; | |
207 | ||
208 | nr_range = get_free_all_memory_range(&range, nodeid); | |
209 | ||
210 | for (i = 0; i < nr_range; i++) { | |
211 | start = range[i].start; | |
212 | end = range[i].end; | |
213 | count += end - start; | |
214 | __free_pages_memory(start, end); | |
215 | } | |
216 | ||
217 | return count; | |
218 | } | |
219 | #else | |
223e8dc9 JW |
220 | static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) |
221 | { | |
41546c17 | 222 | int aligned; |
223e8dc9 | 223 | struct page *page; |
41546c17 JW |
224 | unsigned long start, end, pages, count = 0; |
225 | ||
226 | if (!bdata->node_bootmem_map) | |
227 | return 0; | |
228 | ||
3560e249 | 229 | start = bdata->node_min_pfn; |
41546c17 JW |
230 | end = bdata->node_low_pfn; |
231 | ||
223e8dc9 | 232 | /* |
41546c17 JW |
233 | * If the start is aligned to the machines wordsize, we might |
234 | * be able to free pages in bulks of that order. | |
223e8dc9 | 235 | */ |
41546c17 | 236 | aligned = !(start & (BITS_PER_LONG - 1)); |
223e8dc9 | 237 | |
41546c17 JW |
238 | bdebug("nid=%td start=%lx end=%lx aligned=%d\n", |
239 | bdata - bootmem_node_data, start, end, aligned); | |
223e8dc9 | 240 | |
41546c17 JW |
241 | while (start < end) { |
242 | unsigned long *map, idx, vec; | |
223e8dc9 | 243 | |
41546c17 | 244 | map = bdata->node_bootmem_map; |
3560e249 | 245 | idx = start - bdata->node_min_pfn; |
41546c17 JW |
246 | vec = ~map[idx / BITS_PER_LONG]; |
247 | ||
248 | if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) { | |
249 | int order = ilog2(BITS_PER_LONG); | |
250 | ||
251 | __free_pages_bootmem(pfn_to_page(start), order); | |
223e8dc9 | 252 | count += BITS_PER_LONG; |
41546c17 JW |
253 | } else { |
254 | unsigned long off = 0; | |
255 | ||
256 | while (vec && off < BITS_PER_LONG) { | |
257 | if (vec & 1) { | |
258 | page = pfn_to_page(start + off); | |
223e8dc9 | 259 | __free_pages_bootmem(page, 0); |
41546c17 | 260 | count++; |
223e8dc9 | 261 | } |
41546c17 JW |
262 | vec >>= 1; |
263 | off++; | |
223e8dc9 | 264 | } |
223e8dc9 | 265 | } |
41546c17 | 266 | start += BITS_PER_LONG; |
223e8dc9 JW |
267 | } |
268 | ||
223e8dc9 | 269 | page = virt_to_page(bdata->node_bootmem_map); |
3560e249 | 270 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
41546c17 JW |
271 | pages = bootmem_bootmap_pages(pages); |
272 | count += pages; | |
273 | while (pages--) | |
274 | __free_pages_bootmem(page++, 0); | |
223e8dc9 | 275 | |
2e5237da JW |
276 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
277 | ||
223e8dc9 JW |
278 | return count; |
279 | } | |
08677214 | 280 | #endif |
223e8dc9 | 281 | |
a66fd7da JW |
282 | /** |
283 | * free_all_bootmem_node - release a node's free pages to the buddy allocator | |
284 | * @pgdat: node to be released | |
285 | * | |
286 | * Returns the number of pages actually released. | |
287 | */ | |
223e8dc9 JW |
288 | unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) |
289 | { | |
290 | register_page_bootmem_info_node(pgdat); | |
08677214 YL |
291 | #ifdef CONFIG_NO_BOOTMEM |
292 | /* free_all_memory_core_early(MAX_NUMNODES) will be called later */ | |
293 | return 0; | |
294 | #else | |
223e8dc9 | 295 | return free_all_bootmem_core(pgdat->bdata); |
08677214 | 296 | #endif |
223e8dc9 JW |
297 | } |
298 | ||
a66fd7da JW |
299 | /** |
300 | * free_all_bootmem - release free pages to the buddy allocator | |
301 | * | |
302 | * Returns the number of pages actually released. | |
303 | */ | |
223e8dc9 JW |
304 | unsigned long __init free_all_bootmem(void) |
305 | { | |
08677214 | 306 | #ifdef CONFIG_NO_BOOTMEM |
33799858 YL |
307 | /* |
308 | * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id | |
309 | * because in some case like Node0 doesnt have RAM installed | |
310 | * low ram will be on Node1 | |
311 | * Use MAX_NUMNODES will make sure all ranges in early_node_map[] | |
312 | * will be used instead of only Node0 related | |
313 | */ | |
314 | return free_all_memory_core_early(MAX_NUMNODES); | |
08677214 | 315 | #else |
aa235fc7 YL |
316 | unsigned long total_pages = 0; |
317 | bootmem_data_t *bdata; | |
318 | ||
319 | list_for_each_entry(bdata, &bdata_list, list) | |
320 | total_pages += free_all_bootmem_core(bdata); | |
321 | ||
322 | return total_pages; | |
08677214 | 323 | #endif |
223e8dc9 JW |
324 | } |
325 | ||
08677214 | 326 | #ifndef CONFIG_NO_BOOTMEM |
d747fa4b JW |
327 | static void __init __free(bootmem_data_t *bdata, |
328 | unsigned long sidx, unsigned long eidx) | |
329 | { | |
330 | unsigned long idx; | |
331 | ||
332 | bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data, | |
3560e249 JW |
333 | sidx + bdata->node_min_pfn, |
334 | eidx + bdata->node_min_pfn); | |
d747fa4b | 335 | |
e2bf3cae JW |
336 | if (bdata->hint_idx > sidx) |
337 | bdata->hint_idx = sidx; | |
338 | ||
d747fa4b JW |
339 | for (idx = sidx; idx < eidx; idx++) |
340 | if (!test_and_clear_bit(idx, bdata->node_bootmem_map)) | |
341 | BUG(); | |
342 | } | |
343 | ||
344 | static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx, | |
345 | unsigned long eidx, int flags) | |
346 | { | |
347 | unsigned long idx; | |
348 | int exclusive = flags & BOOTMEM_EXCLUSIVE; | |
349 | ||
350 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", | |
351 | bdata - bootmem_node_data, | |
3560e249 JW |
352 | sidx + bdata->node_min_pfn, |
353 | eidx + bdata->node_min_pfn, | |
d747fa4b JW |
354 | flags); |
355 | ||
356 | for (idx = sidx; idx < eidx; idx++) | |
357 | if (test_and_set_bit(idx, bdata->node_bootmem_map)) { | |
358 | if (exclusive) { | |
359 | __free(bdata, sidx, idx); | |
360 | return -EBUSY; | |
361 | } | |
362 | bdebug("silent double reserve of PFN %lx\n", | |
3560e249 | 363 | idx + bdata->node_min_pfn); |
d747fa4b JW |
364 | } |
365 | return 0; | |
366 | } | |
367 | ||
e2bf3cae JW |
368 | static int __init mark_bootmem_node(bootmem_data_t *bdata, |
369 | unsigned long start, unsigned long end, | |
370 | int reserve, int flags) | |
223e8dc9 JW |
371 | { |
372 | unsigned long sidx, eidx; | |
223e8dc9 | 373 | |
e2bf3cae JW |
374 | bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n", |
375 | bdata - bootmem_node_data, start, end, reserve, flags); | |
223e8dc9 | 376 | |
3560e249 | 377 | BUG_ON(start < bdata->node_min_pfn); |
e2bf3cae | 378 | BUG_ON(end > bdata->node_low_pfn); |
223e8dc9 | 379 | |
3560e249 JW |
380 | sidx = start - bdata->node_min_pfn; |
381 | eidx = end - bdata->node_min_pfn; | |
223e8dc9 | 382 | |
e2bf3cae JW |
383 | if (reserve) |
384 | return __reserve(bdata, sidx, eidx, flags); | |
223e8dc9 | 385 | else |
e2bf3cae JW |
386 | __free(bdata, sidx, eidx); |
387 | return 0; | |
388 | } | |
389 | ||
390 | static int __init mark_bootmem(unsigned long start, unsigned long end, | |
391 | int reserve, int flags) | |
392 | { | |
393 | unsigned long pos; | |
394 | bootmem_data_t *bdata; | |
395 | ||
396 | pos = start; | |
397 | list_for_each_entry(bdata, &bdata_list, list) { | |
398 | int err; | |
399 | unsigned long max; | |
400 | ||
3560e249 JW |
401 | if (pos < bdata->node_min_pfn || |
402 | pos >= bdata->node_low_pfn) { | |
e2bf3cae JW |
403 | BUG_ON(pos != start); |
404 | continue; | |
405 | } | |
406 | ||
407 | max = min(bdata->node_low_pfn, end); | |
223e8dc9 | 408 | |
e2bf3cae JW |
409 | err = mark_bootmem_node(bdata, pos, max, reserve, flags); |
410 | if (reserve && err) { | |
411 | mark_bootmem(start, pos, 0, 0); | |
412 | return err; | |
413 | } | |
223e8dc9 | 414 | |
e2bf3cae JW |
415 | if (max == end) |
416 | return 0; | |
417 | pos = bdata->node_low_pfn; | |
418 | } | |
419 | BUG(); | |
223e8dc9 | 420 | } |
08677214 | 421 | #endif |
223e8dc9 | 422 | |
a66fd7da JW |
423 | /** |
424 | * free_bootmem_node - mark a page range as usable | |
425 | * @pgdat: node the range resides on | |
426 | * @physaddr: starting address of the range | |
427 | * @size: size of the range in bytes | |
428 | * | |
429 | * Partial pages will be considered reserved and left as they are. | |
430 | * | |
e2bf3cae | 431 | * The range must reside completely on the specified node. |
a66fd7da | 432 | */ |
223e8dc9 JW |
433 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
434 | unsigned long size) | |
435 | { | |
08677214 YL |
436 | #ifdef CONFIG_NO_BOOTMEM |
437 | free_early(physaddr, physaddr + size); | |
08677214 | 438 | #else |
e2bf3cae JW |
439 | unsigned long start, end; |
440 | ||
ec3a354b CM |
441 | kmemleak_free_part(__va(physaddr), size); |
442 | ||
e2bf3cae JW |
443 | start = PFN_UP(physaddr); |
444 | end = PFN_DOWN(physaddr + size); | |
445 | ||
446 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); | |
08677214 | 447 | #endif |
223e8dc9 JW |
448 | } |
449 | ||
a66fd7da JW |
450 | /** |
451 | * free_bootmem - mark a page range as usable | |
452 | * @addr: starting address of the range | |
453 | * @size: size of the range in bytes | |
454 | * | |
455 | * Partial pages will be considered reserved and left as they are. | |
456 | * | |
e2bf3cae | 457 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 458 | */ |
223e8dc9 JW |
459 | void __init free_bootmem(unsigned long addr, unsigned long size) |
460 | { | |
08677214 YL |
461 | #ifdef CONFIG_NO_BOOTMEM |
462 | free_early(addr, addr + size); | |
08677214 | 463 | #else |
e2bf3cae | 464 | unsigned long start, end; |
a5645a61 | 465 | |
ec3a354b CM |
466 | kmemleak_free_part(__va(addr), size); |
467 | ||
e2bf3cae JW |
468 | start = PFN_UP(addr); |
469 | end = PFN_DOWN(addr + size); | |
1da177e4 | 470 | |
e2bf3cae | 471 | mark_bootmem(start, end, 0, 0); |
08677214 | 472 | #endif |
1da177e4 LT |
473 | } |
474 | ||
a66fd7da JW |
475 | /** |
476 | * reserve_bootmem_node - mark a page range as reserved | |
477 | * @pgdat: node the range resides on | |
478 | * @physaddr: starting address of the range | |
479 | * @size: size of the range in bytes | |
480 | * @flags: reservation flags (see linux/bootmem.h) | |
481 | * | |
482 | * Partial pages will be reserved. | |
483 | * | |
e2bf3cae | 484 | * The range must reside completely on the specified node. |
a66fd7da | 485 | */ |
223e8dc9 JW |
486 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
487 | unsigned long size, int flags) | |
1da177e4 | 488 | { |
08677214 YL |
489 | #ifdef CONFIG_NO_BOOTMEM |
490 | panic("no bootmem"); | |
491 | return 0; | |
492 | #else | |
e2bf3cae | 493 | unsigned long start, end; |
1da177e4 | 494 | |
e2bf3cae JW |
495 | start = PFN_DOWN(physaddr); |
496 | end = PFN_UP(physaddr + size); | |
497 | ||
498 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); | |
08677214 | 499 | #endif |
223e8dc9 | 500 | } |
5a982cbc | 501 | |
a66fd7da JW |
502 | /** |
503 | * reserve_bootmem - mark a page range as usable | |
504 | * @addr: starting address of the range | |
505 | * @size: size of the range in bytes | |
506 | * @flags: reservation flags (see linux/bootmem.h) | |
507 | * | |
508 | * Partial pages will be reserved. | |
509 | * | |
e2bf3cae | 510 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 511 | */ |
223e8dc9 JW |
512 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
513 | int flags) | |
514 | { | |
08677214 YL |
515 | #ifdef CONFIG_NO_BOOTMEM |
516 | panic("no bootmem"); | |
517 | return 0; | |
518 | #else | |
e2bf3cae | 519 | unsigned long start, end; |
1da177e4 | 520 | |
e2bf3cae JW |
521 | start = PFN_DOWN(addr); |
522 | end = PFN_UP(addr + size); | |
223e8dc9 | 523 | |
e2bf3cae | 524 | return mark_bootmem(start, end, 1, flags); |
08677214 | 525 | #endif |
1da177e4 LT |
526 | } |
527 | ||
08677214 | 528 | #ifndef CONFIG_NO_BOOTMEM |
8aa043d7 JB |
529 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
530 | unsigned long idx, unsigned long step) | |
481ebd0d JW |
531 | { |
532 | unsigned long base = bdata->node_min_pfn; | |
533 | ||
534 | /* | |
535 | * Align the index with respect to the node start so that the | |
536 | * combination of both satisfies the requested alignment. | |
537 | */ | |
538 | ||
539 | return ALIGN(base + idx, step) - base; | |
540 | } | |
541 | ||
8aa043d7 JB |
542 | static unsigned long __init align_off(struct bootmem_data *bdata, |
543 | unsigned long off, unsigned long align) | |
481ebd0d JW |
544 | { |
545 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); | |
546 | ||
547 | /* Same as align_idx for byte offsets */ | |
548 | ||
549 | return ALIGN(base + off, align) - base; | |
550 | } | |
551 | ||
d0c4f570 TH |
552 | static void * __init alloc_bootmem_core(struct bootmem_data *bdata, |
553 | unsigned long size, unsigned long align, | |
554 | unsigned long goal, unsigned long limit) | |
1da177e4 | 555 | { |
0f3caba2 | 556 | unsigned long fallback = 0; |
5f2809e6 JW |
557 | unsigned long min, max, start, sidx, midx, step; |
558 | ||
594fe1a0 JW |
559 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
560 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, | |
561 | align, goal, limit); | |
562 | ||
5f2809e6 JW |
563 | BUG_ON(!size); |
564 | BUG_ON(align & (align - 1)); | |
565 | BUG_ON(limit && goal + size > limit); | |
1da177e4 | 566 | |
7c309a64 CK |
567 | if (!bdata->node_bootmem_map) |
568 | return NULL; | |
569 | ||
3560e249 | 570 | min = bdata->node_min_pfn; |
5f2809e6 | 571 | max = bdata->node_low_pfn; |
9a2dc04c | 572 | |
5f2809e6 JW |
573 | goal >>= PAGE_SHIFT; |
574 | limit >>= PAGE_SHIFT; | |
575 | ||
576 | if (limit && max > limit) | |
577 | max = limit; | |
578 | if (max <= min) | |
9a2dc04c YL |
579 | return NULL; |
580 | ||
5f2809e6 | 581 | step = max(align >> PAGE_SHIFT, 1UL); |
281dd25c | 582 | |
5f2809e6 JW |
583 | if (goal && min < goal && goal < max) |
584 | start = ALIGN(goal, step); | |
585 | else | |
586 | start = ALIGN(min, step); | |
1da177e4 | 587 | |
481ebd0d | 588 | sidx = start - bdata->node_min_pfn; |
3560e249 | 589 | midx = max - bdata->node_min_pfn; |
1da177e4 | 590 | |
5f2809e6 | 591 | if (bdata->hint_idx > sidx) { |
0f3caba2 JW |
592 | /* |
593 | * Handle the valid case of sidx being zero and still | |
594 | * catch the fallback below. | |
595 | */ | |
596 | fallback = sidx + 1; | |
481ebd0d | 597 | sidx = align_idx(bdata, bdata->hint_idx, step); |
5f2809e6 | 598 | } |
1da177e4 | 599 | |
5f2809e6 JW |
600 | while (1) { |
601 | int merge; | |
602 | void *region; | |
603 | unsigned long eidx, i, start_off, end_off; | |
604 | find_block: | |
605 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); | |
481ebd0d | 606 | sidx = align_idx(bdata, sidx, step); |
5f2809e6 | 607 | eidx = sidx + PFN_UP(size); |
ad09315c | 608 | |
5f2809e6 | 609 | if (sidx >= midx || eidx > midx) |
66d43e98 | 610 | break; |
1da177e4 | 611 | |
5f2809e6 JW |
612 | for (i = sidx; i < eidx; i++) |
613 | if (test_bit(i, bdata->node_bootmem_map)) { | |
481ebd0d | 614 | sidx = align_idx(bdata, i, step); |
5f2809e6 JW |
615 | if (sidx == i) |
616 | sidx += step; | |
617 | goto find_block; | |
618 | } | |
1da177e4 | 619 | |
627240aa | 620 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
5f2809e6 | 621 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
481ebd0d | 622 | start_off = align_off(bdata, bdata->last_end_off, align); |
5f2809e6 JW |
623 | else |
624 | start_off = PFN_PHYS(sidx); | |
625 | ||
626 | merge = PFN_DOWN(start_off) < sidx; | |
627 | end_off = start_off + size; | |
628 | ||
629 | bdata->last_end_off = end_off; | |
630 | bdata->hint_idx = PFN_UP(end_off); | |
631 | ||
632 | /* | |
633 | * Reserve the area now: | |
634 | */ | |
d747fa4b JW |
635 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
636 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) | |
637 | BUG(); | |
5f2809e6 | 638 | |
3560e249 JW |
639 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
640 | start_off); | |
5f2809e6 | 641 | memset(region, 0, size); |
008139d9 CM |
642 | /* |
643 | * The min_count is set to 0 so that bootmem allocated blocks | |
644 | * are never reported as leaks. | |
645 | */ | |
646 | kmemleak_alloc(region, size, 0, 0); | |
5f2809e6 | 647 | return region; |
1da177e4 LT |
648 | } |
649 | ||
0f3caba2 | 650 | if (fallback) { |
481ebd0d | 651 | sidx = align_idx(bdata, fallback - 1, step); |
0f3caba2 JW |
652 | fallback = 0; |
653 | goto find_block; | |
654 | } | |
655 | ||
656 | return NULL; | |
657 | } | |
658 | ||
d0c4f570 TH |
659 | static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata, |
660 | unsigned long size, unsigned long align, | |
661 | unsigned long goal, unsigned long limit) | |
662 | { | |
441c7e0a PE |
663 | if (WARN_ON_ONCE(slab_is_available())) |
664 | return kzalloc(size, GFP_NOWAIT); | |
665 | ||
d0c4f570 | 666 | #ifdef CONFIG_HAVE_ARCH_BOOTMEM |
433f13a7 JP |
667 | { |
668 | bootmem_data_t *p_bdata; | |
669 | ||
670 | p_bdata = bootmem_arch_preferred_node(bdata, size, align, | |
671 | goal, limit); | |
672 | if (p_bdata) | |
673 | return alloc_bootmem_core(p_bdata, size, align, | |
674 | goal, limit); | |
675 | } | |
d0c4f570 TH |
676 | #endif |
677 | return NULL; | |
678 | } | |
08677214 | 679 | #endif |
d0c4f570 | 680 | |
0f3caba2 JW |
681 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, |
682 | unsigned long align, | |
683 | unsigned long goal, | |
684 | unsigned long limit) | |
685 | { | |
08677214 YL |
686 | #ifdef CONFIG_NO_BOOTMEM |
687 | void *ptr; | |
688 | ||
689 | if (WARN_ON_ONCE(slab_is_available())) | |
690 | return kzalloc(size, GFP_NOWAIT); | |
691 | ||
692 | restart: | |
693 | ||
694 | ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit); | |
695 | ||
696 | if (ptr) | |
697 | return ptr; | |
698 | ||
699 | if (goal != 0) { | |
700 | goal = 0; | |
701 | goto restart; | |
702 | } | |
703 | ||
704 | return NULL; | |
705 | #else | |
0f3caba2 | 706 | bootmem_data_t *bdata; |
d0c4f570 | 707 | void *region; |
0f3caba2 JW |
708 | |
709 | restart: | |
d0c4f570 TH |
710 | region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit); |
711 | if (region) | |
712 | return region; | |
0f3caba2 | 713 | |
d0c4f570 | 714 | list_for_each_entry(bdata, &bdata_list, list) { |
0f3caba2 JW |
715 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
716 | continue; | |
3560e249 | 717 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
0f3caba2 JW |
718 | break; |
719 | ||
720 | region = alloc_bootmem_core(bdata, size, align, goal, limit); | |
721 | if (region) | |
722 | return region; | |
723 | } | |
724 | ||
5f2809e6 JW |
725 | if (goal) { |
726 | goal = 0; | |
0f3caba2 | 727 | goto restart; |
5f2809e6 | 728 | } |
2e5237da | 729 | |
5f2809e6 | 730 | return NULL; |
08677214 | 731 | #endif |
1da177e4 LT |
732 | } |
733 | ||
a66fd7da JW |
734 | /** |
735 | * __alloc_bootmem_nopanic - allocate boot memory without panicking | |
736 | * @size: size of the request in bytes | |
737 | * @align: alignment of the region | |
738 | * @goal: preferred starting address of the region | |
739 | * | |
740 | * The goal is dropped if it can not be satisfied and the allocation will | |
741 | * fall back to memory below @goal. | |
742 | * | |
743 | * Allocation may happen on any node in the system. | |
744 | * | |
745 | * Returns NULL on failure. | |
746 | */ | |
bb0923a6 | 747 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
0f3caba2 | 748 | unsigned long goal) |
1da177e4 | 749 | { |
08677214 YL |
750 | unsigned long limit = 0; |
751 | ||
752 | #ifdef CONFIG_NO_BOOTMEM | |
753 | limit = -1UL; | |
754 | #endif | |
755 | ||
756 | return ___alloc_bootmem_nopanic(size, align, goal, limit); | |
0f3caba2 | 757 | } |
1da177e4 | 758 | |
0f3caba2 JW |
759 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
760 | unsigned long goal, unsigned long limit) | |
761 | { | |
762 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); | |
763 | ||
764 | if (mem) | |
765 | return mem; | |
766 | /* | |
767 | * Whoops, we cannot satisfy the allocation request. | |
768 | */ | |
769 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); | |
770 | panic("Out of memory"); | |
a8062231 AK |
771 | return NULL; |
772 | } | |
1da177e4 | 773 | |
a66fd7da JW |
774 | /** |
775 | * __alloc_bootmem - allocate boot memory | |
776 | * @size: size of the request in bytes | |
777 | * @align: alignment of the region | |
778 | * @goal: preferred starting address of the region | |
779 | * | |
780 | * The goal is dropped if it can not be satisfied and the allocation will | |
781 | * fall back to memory below @goal. | |
782 | * | |
783 | * Allocation may happen on any node in the system. | |
784 | * | |
785 | * The function panics if the request can not be satisfied. | |
786 | */ | |
bb0923a6 FBH |
787 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
788 | unsigned long goal) | |
a8062231 | 789 | { |
08677214 YL |
790 | unsigned long limit = 0; |
791 | ||
792 | #ifdef CONFIG_NO_BOOTMEM | |
793 | limit = -1UL; | |
794 | #endif | |
795 | ||
796 | return ___alloc_bootmem(size, align, goal, limit); | |
1da177e4 LT |
797 | } |
798 | ||
08677214 | 799 | #ifndef CONFIG_NO_BOOTMEM |
4cc278b7 JW |
800 | static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata, |
801 | unsigned long size, unsigned long align, | |
802 | unsigned long goal, unsigned long limit) | |
803 | { | |
804 | void *ptr; | |
805 | ||
d0c4f570 TH |
806 | ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit); |
807 | if (ptr) | |
808 | return ptr; | |
809 | ||
4cc278b7 JW |
810 | ptr = alloc_bootmem_core(bdata, size, align, goal, limit); |
811 | if (ptr) | |
812 | return ptr; | |
813 | ||
814 | return ___alloc_bootmem(size, align, goal, limit); | |
815 | } | |
08677214 | 816 | #endif |
4cc278b7 | 817 | |
a66fd7da JW |
818 | /** |
819 | * __alloc_bootmem_node - allocate boot memory from a specific node | |
820 | * @pgdat: node to allocate from | |
821 | * @size: size of the request in bytes | |
822 | * @align: alignment of the region | |
823 | * @goal: preferred starting address of the region | |
824 | * | |
825 | * The goal is dropped if it can not be satisfied and the allocation will | |
826 | * fall back to memory below @goal. | |
827 | * | |
828 | * Allocation may fall back to any node in the system if the specified node | |
829 | * can not hold the requested memory. | |
830 | * | |
831 | * The function panics if the request can not be satisfied. | |
832 | */ | |
bb0923a6 FBH |
833 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
834 | unsigned long align, unsigned long goal) | |
1da177e4 | 835 | { |
c91c4773 PE |
836 | if (WARN_ON_ONCE(slab_is_available())) |
837 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
838 | ||
08677214 YL |
839 | #ifdef CONFIG_NO_BOOTMEM |
840 | return __alloc_memory_core_early(pgdat->node_id, size, align, | |
841 | goal, -1ULL); | |
842 | #else | |
4cc278b7 | 843 | return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0); |
08677214 YL |
844 | #endif |
845 | } | |
846 | ||
847 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, | |
848 | unsigned long align, unsigned long goal) | |
849 | { | |
850 | #ifdef MAX_DMA32_PFN | |
851 | unsigned long end_pfn; | |
852 | ||
853 | if (WARN_ON_ONCE(slab_is_available())) | |
854 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
855 | ||
856 | /* update goal according ...MAX_DMA32_PFN */ | |
857 | end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
858 | ||
859 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && | |
860 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { | |
861 | void *ptr; | |
862 | unsigned long new_goal; | |
863 | ||
864 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; | |
865 | #ifdef CONFIG_NO_BOOTMEM | |
866 | ptr = __alloc_memory_core_early(pgdat->node_id, size, align, | |
867 | new_goal, -1ULL); | |
868 | #else | |
869 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, | |
870 | new_goal, 0); | |
871 | #endif | |
872 | if (ptr) | |
873 | return ptr; | |
874 | } | |
875 | #endif | |
876 | ||
877 | return __alloc_bootmem_node(pgdat, size, align, goal); | |
878 | ||
1da177e4 LT |
879 | } |
880 | ||
e70260aa | 881 | #ifdef CONFIG_SPARSEMEM |
a66fd7da JW |
882 | /** |
883 | * alloc_bootmem_section - allocate boot memory from a specific section | |
884 | * @size: size of the request in bytes | |
885 | * @section_nr: sparse map section to allocate from | |
886 | * | |
887 | * Return NULL on failure. | |
888 | */ | |
e70260aa YG |
889 | void * __init alloc_bootmem_section(unsigned long size, |
890 | unsigned long section_nr) | |
891 | { | |
08677214 YL |
892 | #ifdef CONFIG_NO_BOOTMEM |
893 | unsigned long pfn, goal, limit; | |
894 | ||
895 | pfn = section_nr_to_pfn(section_nr); | |
896 | goal = pfn << PAGE_SHIFT; | |
897 | limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT; | |
898 | ||
899 | return __alloc_memory_core_early(early_pfn_to_nid(pfn), size, | |
900 | SMP_CACHE_BYTES, goal, limit); | |
901 | #else | |
75a56cfe JW |
902 | bootmem_data_t *bdata; |
903 | unsigned long pfn, goal, limit; | |
e70260aa YG |
904 | |
905 | pfn = section_nr_to_pfn(section_nr); | |
75a56cfe JW |
906 | goal = pfn << PAGE_SHIFT; |
907 | limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT; | |
908 | bdata = &bootmem_node_data[early_pfn_to_nid(pfn)]; | |
e70260aa | 909 | |
75a56cfe | 910 | return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit); |
08677214 | 911 | #endif |
e70260aa YG |
912 | } |
913 | #endif | |
914 | ||
b54bbf7b AK |
915 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, |
916 | unsigned long align, unsigned long goal) | |
917 | { | |
918 | void *ptr; | |
919 | ||
c91c4773 PE |
920 | if (WARN_ON_ONCE(slab_is_available())) |
921 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
922 | ||
08677214 YL |
923 | #ifdef CONFIG_NO_BOOTMEM |
924 | ptr = __alloc_memory_core_early(pgdat->node_id, size, align, | |
925 | goal, -1ULL); | |
926 | #else | |
d0c4f570 TH |
927 | ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0); |
928 | if (ptr) | |
929 | return ptr; | |
930 | ||
b54bbf7b | 931 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0); |
08677214 | 932 | #endif |
b54bbf7b AK |
933 | if (ptr) |
934 | return ptr; | |
935 | ||
936 | return __alloc_bootmem_nopanic(size, align, goal); | |
937 | } | |
938 | ||
dfd54cbc HC |
939 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
940 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL | |
941 | #endif | |
008857c1 | 942 | |
a66fd7da JW |
943 | /** |
944 | * __alloc_bootmem_low - allocate low boot memory | |
945 | * @size: size of the request in bytes | |
946 | * @align: alignment of the region | |
947 | * @goal: preferred starting address of the region | |
948 | * | |
949 | * The goal is dropped if it can not be satisfied and the allocation will | |
950 | * fall back to memory below @goal. | |
951 | * | |
952 | * Allocation may happen on any node in the system. | |
953 | * | |
954 | * The function panics if the request can not be satisfied. | |
955 | */ | |
bb0923a6 FBH |
956 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
957 | unsigned long goal) | |
008857c1 | 958 | { |
0f3caba2 | 959 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
008857c1 RT |
960 | } |
961 | ||
a66fd7da JW |
962 | /** |
963 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node | |
964 | * @pgdat: node to allocate from | |
965 | * @size: size of the request in bytes | |
966 | * @align: alignment of the region | |
967 | * @goal: preferred starting address of the region | |
968 | * | |
969 | * The goal is dropped if it can not be satisfied and the allocation will | |
970 | * fall back to memory below @goal. | |
971 | * | |
972 | * Allocation may fall back to any node in the system if the specified node | |
973 | * can not hold the requested memory. | |
974 | * | |
975 | * The function panics if the request can not be satisfied. | |
976 | */ | |
008857c1 RT |
977 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
978 | unsigned long align, unsigned long goal) | |
979 | { | |
c91c4773 PE |
980 | if (WARN_ON_ONCE(slab_is_available())) |
981 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
982 | ||
08677214 YL |
983 | #ifdef CONFIG_NO_BOOTMEM |
984 | return __alloc_memory_core_early(pgdat->node_id, size, align, | |
985 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
986 | #else | |
4cc278b7 JW |
987 | return ___alloc_bootmem_node(pgdat->bdata, size, align, |
988 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
08677214 | 989 | #endif |
008857c1 | 990 | } |