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> |
1da177e4 | 13 | #include <linux/bootmem.h> |
1da177e4 | 14 | #include <linux/module.h> |
ec3a354b | 15 | #include <linux/kmemleak.h> |
08677214 | 16 | #include <linux/range.h> |
e786e86a FBH |
17 | |
18 | #include <asm/bug.h> | |
1da177e4 | 19 | #include <asm/io.h> |
dfd54cbc | 20 | #include <asm/processor.h> |
e786e86a | 21 | |
1da177e4 LT |
22 | #include "internal.h" |
23 | ||
1da177e4 LT |
24 | unsigned long max_low_pfn; |
25 | unsigned long min_low_pfn; | |
26 | unsigned long max_pfn; | |
27 | ||
92aa63a5 VG |
28 | #ifdef CONFIG_CRASH_DUMP |
29 | /* | |
30 | * If we have booted due to a crash, max_pfn will be a very low value. We need | |
31 | * to know the amount of memory that the previous kernel used. | |
32 | */ | |
33 | unsigned long saved_max_pfn; | |
34 | #endif | |
35 | ||
08677214 | 36 | #ifndef CONFIG_NO_BOOTMEM |
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 | { |
df049a5f | 59 | unsigned long bytes = (pages + 7) / 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 | { |
636cc40c | 80 | struct list_head *iter; |
f71bf0ca | 81 | |
636cc40c JW |
82 | list_for_each(iter, &bdata_list) { |
83 | bootmem_data_t *ent; | |
84 | ||
85 | ent = list_entry(iter, bootmem_data_t, list); | |
3560e249 | 86 | if (bdata->node_min_pfn < ent->node_min_pfn) |
636cc40c | 87 | break; |
679bc9fb | 88 | } |
636cc40c | 89 | list_add_tail(&bdata->list, iter); |
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 | } | |
08677214 | 147 | #endif |
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 | ||
08677214 YL |
172 | #ifdef CONFIG_NO_BOOTMEM |
173 | static void __init __free_pages_memory(unsigned long start, unsigned long end) | |
174 | { | |
175 | int i; | |
176 | unsigned long start_aligned, end_aligned; | |
177 | int order = ilog2(BITS_PER_LONG); | |
178 | ||
179 | start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1); | |
180 | end_aligned = end & ~(BITS_PER_LONG - 1); | |
181 | ||
182 | if (end_aligned <= start_aligned) { | |
08677214 YL |
183 | for (i = start; i < end; i++) |
184 | __free_pages_bootmem(pfn_to_page(i), 0); | |
185 | ||
186 | return; | |
187 | } | |
188 | ||
08677214 YL |
189 | for (i = start; i < start_aligned; i++) |
190 | __free_pages_bootmem(pfn_to_page(i), 0); | |
191 | ||
192 | for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG) | |
193 | __free_pages_bootmem(pfn_to_page(i), order); | |
194 | ||
195 | for (i = end_aligned; i < end; i++) | |
196 | __free_pages_bootmem(pfn_to_page(i), 0); | |
197 | } | |
198 | ||
199 | unsigned long __init free_all_memory_core_early(int nodeid) | |
200 | { | |
201 | int i; | |
202 | u64 start, end; | |
203 | unsigned long count = 0; | |
204 | struct range *range = NULL; | |
205 | int nr_range; | |
206 | ||
207 | nr_range = get_free_all_memory_range(&range, nodeid); | |
208 | ||
209 | for (i = 0; i < nr_range; i++) { | |
210 | start = range[i].start; | |
211 | end = range[i].end; | |
212 | count += end - start; | |
213 | __free_pages_memory(start, end); | |
214 | } | |
215 | ||
216 | return count; | |
217 | } | |
218 | #else | |
223e8dc9 JW |
219 | static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) |
220 | { | |
41546c17 | 221 | int aligned; |
223e8dc9 | 222 | struct page *page; |
41546c17 JW |
223 | unsigned long start, end, pages, count = 0; |
224 | ||
225 | if (!bdata->node_bootmem_map) | |
226 | return 0; | |
227 | ||
3560e249 | 228 | start = bdata->node_min_pfn; |
41546c17 JW |
229 | end = bdata->node_low_pfn; |
230 | ||
223e8dc9 | 231 | /* |
41546c17 JW |
232 | * If the start is aligned to the machines wordsize, we might |
233 | * be able to free pages in bulks of that order. | |
223e8dc9 | 234 | */ |
41546c17 | 235 | aligned = !(start & (BITS_PER_LONG - 1)); |
223e8dc9 | 236 | |
41546c17 JW |
237 | bdebug("nid=%td start=%lx end=%lx aligned=%d\n", |
238 | bdata - bootmem_node_data, start, end, aligned); | |
223e8dc9 | 239 | |
41546c17 JW |
240 | while (start < end) { |
241 | unsigned long *map, idx, vec; | |
223e8dc9 | 242 | |
41546c17 | 243 | map = bdata->node_bootmem_map; |
3560e249 | 244 | idx = start - bdata->node_min_pfn; |
41546c17 JW |
245 | vec = ~map[idx / BITS_PER_LONG]; |
246 | ||
247 | if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) { | |
248 | int order = ilog2(BITS_PER_LONG); | |
249 | ||
250 | __free_pages_bootmem(pfn_to_page(start), order); | |
223e8dc9 | 251 | count += BITS_PER_LONG; |
41546c17 JW |
252 | } else { |
253 | unsigned long off = 0; | |
254 | ||
255 | while (vec && off < BITS_PER_LONG) { | |
256 | if (vec & 1) { | |
257 | page = pfn_to_page(start + off); | |
223e8dc9 | 258 | __free_pages_bootmem(page, 0); |
41546c17 | 259 | count++; |
223e8dc9 | 260 | } |
41546c17 JW |
261 | vec >>= 1; |
262 | off++; | |
223e8dc9 | 263 | } |
223e8dc9 | 264 | } |
41546c17 | 265 | start += BITS_PER_LONG; |
223e8dc9 JW |
266 | } |
267 | ||
223e8dc9 | 268 | page = virt_to_page(bdata->node_bootmem_map); |
3560e249 | 269 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
41546c17 JW |
270 | pages = bootmem_bootmap_pages(pages); |
271 | count += pages; | |
272 | while (pages--) | |
273 | __free_pages_bootmem(page++, 0); | |
223e8dc9 | 274 | |
2e5237da JW |
275 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
276 | ||
223e8dc9 JW |
277 | return count; |
278 | } | |
08677214 | 279 | #endif |
223e8dc9 | 280 | |
a66fd7da JW |
281 | /** |
282 | * free_all_bootmem_node - release a node's free pages to the buddy allocator | |
283 | * @pgdat: node to be released | |
284 | * | |
285 | * Returns the number of pages actually released. | |
286 | */ | |
223e8dc9 JW |
287 | unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) |
288 | { | |
289 | register_page_bootmem_info_node(pgdat); | |
08677214 YL |
290 | #ifdef CONFIG_NO_BOOTMEM |
291 | /* free_all_memory_core_early(MAX_NUMNODES) will be called later */ | |
292 | return 0; | |
293 | #else | |
223e8dc9 | 294 | return free_all_bootmem_core(pgdat->bdata); |
08677214 | 295 | #endif |
223e8dc9 JW |
296 | } |
297 | ||
a66fd7da JW |
298 | /** |
299 | * free_all_bootmem - release free pages to the buddy allocator | |
300 | * | |
301 | * Returns the number of pages actually released. | |
302 | */ | |
223e8dc9 JW |
303 | unsigned long __init free_all_bootmem(void) |
304 | { | |
08677214 YL |
305 | #ifdef CONFIG_NO_BOOTMEM |
306 | return free_all_memory_core_early(NODE_DATA(0)->node_id); | |
307 | #else | |
223e8dc9 | 308 | return free_all_bootmem_core(NODE_DATA(0)->bdata); |
08677214 | 309 | #endif |
223e8dc9 JW |
310 | } |
311 | ||
08677214 | 312 | #ifndef CONFIG_NO_BOOTMEM |
d747fa4b JW |
313 | static void __init __free(bootmem_data_t *bdata, |
314 | unsigned long sidx, unsigned long eidx) | |
315 | { | |
316 | unsigned long idx; | |
317 | ||
318 | bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data, | |
3560e249 JW |
319 | sidx + bdata->node_min_pfn, |
320 | eidx + bdata->node_min_pfn); | |
d747fa4b | 321 | |
e2bf3cae JW |
322 | if (bdata->hint_idx > sidx) |
323 | bdata->hint_idx = sidx; | |
324 | ||
d747fa4b JW |
325 | for (idx = sidx; idx < eidx; idx++) |
326 | if (!test_and_clear_bit(idx, bdata->node_bootmem_map)) | |
327 | BUG(); | |
328 | } | |
329 | ||
330 | static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx, | |
331 | unsigned long eidx, int flags) | |
332 | { | |
333 | unsigned long idx; | |
334 | int exclusive = flags & BOOTMEM_EXCLUSIVE; | |
335 | ||
336 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", | |
337 | bdata - bootmem_node_data, | |
3560e249 JW |
338 | sidx + bdata->node_min_pfn, |
339 | eidx + bdata->node_min_pfn, | |
d747fa4b JW |
340 | flags); |
341 | ||
342 | for (idx = sidx; idx < eidx; idx++) | |
343 | if (test_and_set_bit(idx, bdata->node_bootmem_map)) { | |
344 | if (exclusive) { | |
345 | __free(bdata, sidx, idx); | |
346 | return -EBUSY; | |
347 | } | |
348 | bdebug("silent double reserve of PFN %lx\n", | |
3560e249 | 349 | idx + bdata->node_min_pfn); |
d747fa4b JW |
350 | } |
351 | return 0; | |
352 | } | |
353 | ||
e2bf3cae JW |
354 | static int __init mark_bootmem_node(bootmem_data_t *bdata, |
355 | unsigned long start, unsigned long end, | |
356 | int reserve, int flags) | |
223e8dc9 JW |
357 | { |
358 | unsigned long sidx, eidx; | |
223e8dc9 | 359 | |
e2bf3cae JW |
360 | bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n", |
361 | bdata - bootmem_node_data, start, end, reserve, flags); | |
223e8dc9 | 362 | |
3560e249 | 363 | BUG_ON(start < bdata->node_min_pfn); |
e2bf3cae | 364 | BUG_ON(end > bdata->node_low_pfn); |
223e8dc9 | 365 | |
3560e249 JW |
366 | sidx = start - bdata->node_min_pfn; |
367 | eidx = end - bdata->node_min_pfn; | |
223e8dc9 | 368 | |
e2bf3cae JW |
369 | if (reserve) |
370 | return __reserve(bdata, sidx, eidx, flags); | |
223e8dc9 | 371 | else |
e2bf3cae JW |
372 | __free(bdata, sidx, eidx); |
373 | return 0; | |
374 | } | |
375 | ||
376 | static int __init mark_bootmem(unsigned long start, unsigned long end, | |
377 | int reserve, int flags) | |
378 | { | |
379 | unsigned long pos; | |
380 | bootmem_data_t *bdata; | |
381 | ||
382 | pos = start; | |
383 | list_for_each_entry(bdata, &bdata_list, list) { | |
384 | int err; | |
385 | unsigned long max; | |
386 | ||
3560e249 JW |
387 | if (pos < bdata->node_min_pfn || |
388 | pos >= bdata->node_low_pfn) { | |
e2bf3cae JW |
389 | BUG_ON(pos != start); |
390 | continue; | |
391 | } | |
392 | ||
393 | max = min(bdata->node_low_pfn, end); | |
223e8dc9 | 394 | |
e2bf3cae JW |
395 | err = mark_bootmem_node(bdata, pos, max, reserve, flags); |
396 | if (reserve && err) { | |
397 | mark_bootmem(start, pos, 0, 0); | |
398 | return err; | |
399 | } | |
223e8dc9 | 400 | |
e2bf3cae JW |
401 | if (max == end) |
402 | return 0; | |
403 | pos = bdata->node_low_pfn; | |
404 | } | |
405 | BUG(); | |
223e8dc9 | 406 | } |
08677214 | 407 | #endif |
223e8dc9 | 408 | |
a66fd7da JW |
409 | /** |
410 | * free_bootmem_node - mark a page range as usable | |
411 | * @pgdat: node the range resides on | |
412 | * @physaddr: starting address of the range | |
413 | * @size: size of the range in bytes | |
414 | * | |
415 | * Partial pages will be considered reserved and left as they are. | |
416 | * | |
e2bf3cae | 417 | * The range must reside completely on the specified node. |
a66fd7da | 418 | */ |
223e8dc9 JW |
419 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
420 | unsigned long size) | |
421 | { | |
08677214 YL |
422 | #ifdef CONFIG_NO_BOOTMEM |
423 | free_early(physaddr, physaddr + size); | |
08677214 | 424 | #else |
e2bf3cae JW |
425 | unsigned long start, end; |
426 | ||
ec3a354b CM |
427 | kmemleak_free_part(__va(physaddr), size); |
428 | ||
e2bf3cae JW |
429 | start = PFN_UP(physaddr); |
430 | end = PFN_DOWN(physaddr + size); | |
431 | ||
432 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); | |
08677214 | 433 | #endif |
223e8dc9 JW |
434 | } |
435 | ||
a66fd7da JW |
436 | /** |
437 | * free_bootmem - mark a page range as usable | |
438 | * @addr: starting address of the range | |
439 | * @size: size of the range in bytes | |
440 | * | |
441 | * Partial pages will be considered reserved and left as they are. | |
442 | * | |
e2bf3cae | 443 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 444 | */ |
223e8dc9 JW |
445 | void __init free_bootmem(unsigned long addr, unsigned long size) |
446 | { | |
08677214 YL |
447 | #ifdef CONFIG_NO_BOOTMEM |
448 | free_early(addr, addr + size); | |
08677214 | 449 | #else |
e2bf3cae | 450 | unsigned long start, end; |
a5645a61 | 451 | |
ec3a354b CM |
452 | kmemleak_free_part(__va(addr), size); |
453 | ||
e2bf3cae JW |
454 | start = PFN_UP(addr); |
455 | end = PFN_DOWN(addr + size); | |
1da177e4 | 456 | |
e2bf3cae | 457 | mark_bootmem(start, end, 0, 0); |
08677214 | 458 | #endif |
1da177e4 LT |
459 | } |
460 | ||
a66fd7da JW |
461 | /** |
462 | * reserve_bootmem_node - mark a page range as reserved | |
463 | * @pgdat: node the range resides on | |
464 | * @physaddr: starting address of the range | |
465 | * @size: size of the range in bytes | |
466 | * @flags: reservation flags (see linux/bootmem.h) | |
467 | * | |
468 | * Partial pages will be reserved. | |
469 | * | |
e2bf3cae | 470 | * The range must reside completely on the specified node. |
a66fd7da | 471 | */ |
223e8dc9 JW |
472 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
473 | unsigned long size, int flags) | |
1da177e4 | 474 | { |
08677214 YL |
475 | #ifdef CONFIG_NO_BOOTMEM |
476 | panic("no bootmem"); | |
477 | return 0; | |
478 | #else | |
e2bf3cae | 479 | unsigned long start, end; |
1da177e4 | 480 | |
e2bf3cae JW |
481 | start = PFN_DOWN(physaddr); |
482 | end = PFN_UP(physaddr + size); | |
483 | ||
484 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); | |
08677214 | 485 | #endif |
223e8dc9 | 486 | } |
5a982cbc | 487 | |
a66fd7da JW |
488 | /** |
489 | * reserve_bootmem - mark a page range as usable | |
490 | * @addr: starting address of the range | |
491 | * @size: size of the range in bytes | |
492 | * @flags: reservation flags (see linux/bootmem.h) | |
493 | * | |
494 | * Partial pages will be reserved. | |
495 | * | |
e2bf3cae | 496 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 497 | */ |
223e8dc9 JW |
498 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
499 | int flags) | |
500 | { | |
08677214 YL |
501 | #ifdef CONFIG_NO_BOOTMEM |
502 | panic("no bootmem"); | |
503 | return 0; | |
504 | #else | |
e2bf3cae | 505 | unsigned long start, end; |
1da177e4 | 506 | |
e2bf3cae JW |
507 | start = PFN_DOWN(addr); |
508 | end = PFN_UP(addr + size); | |
223e8dc9 | 509 | |
e2bf3cae | 510 | return mark_bootmem(start, end, 1, flags); |
08677214 | 511 | #endif |
1da177e4 LT |
512 | } |
513 | ||
08677214 | 514 | #ifndef CONFIG_NO_BOOTMEM |
8aa043d7 JB |
515 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
516 | unsigned long idx, unsigned long step) | |
481ebd0d JW |
517 | { |
518 | unsigned long base = bdata->node_min_pfn; | |
519 | ||
520 | /* | |
521 | * Align the index with respect to the node start so that the | |
522 | * combination of both satisfies the requested alignment. | |
523 | */ | |
524 | ||
525 | return ALIGN(base + idx, step) - base; | |
526 | } | |
527 | ||
8aa043d7 JB |
528 | static unsigned long __init align_off(struct bootmem_data *bdata, |
529 | unsigned long off, unsigned long align) | |
481ebd0d JW |
530 | { |
531 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); | |
532 | ||
533 | /* Same as align_idx for byte offsets */ | |
534 | ||
535 | return ALIGN(base + off, align) - base; | |
536 | } | |
537 | ||
d0c4f570 TH |
538 | static void * __init alloc_bootmem_core(struct bootmem_data *bdata, |
539 | unsigned long size, unsigned long align, | |
540 | unsigned long goal, unsigned long limit) | |
1da177e4 | 541 | { |
0f3caba2 | 542 | unsigned long fallback = 0; |
5f2809e6 JW |
543 | unsigned long min, max, start, sidx, midx, step; |
544 | ||
594fe1a0 JW |
545 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
546 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, | |
547 | align, goal, limit); | |
548 | ||
5f2809e6 JW |
549 | BUG_ON(!size); |
550 | BUG_ON(align & (align - 1)); | |
551 | BUG_ON(limit && goal + size > limit); | |
1da177e4 | 552 | |
7c309a64 CK |
553 | if (!bdata->node_bootmem_map) |
554 | return NULL; | |
555 | ||
3560e249 | 556 | min = bdata->node_min_pfn; |
5f2809e6 | 557 | max = bdata->node_low_pfn; |
9a2dc04c | 558 | |
5f2809e6 JW |
559 | goal >>= PAGE_SHIFT; |
560 | limit >>= PAGE_SHIFT; | |
561 | ||
562 | if (limit && max > limit) | |
563 | max = limit; | |
564 | if (max <= min) | |
9a2dc04c YL |
565 | return NULL; |
566 | ||
5f2809e6 | 567 | step = max(align >> PAGE_SHIFT, 1UL); |
281dd25c | 568 | |
5f2809e6 JW |
569 | if (goal && min < goal && goal < max) |
570 | start = ALIGN(goal, step); | |
571 | else | |
572 | start = ALIGN(min, step); | |
1da177e4 | 573 | |
481ebd0d | 574 | sidx = start - bdata->node_min_pfn; |
3560e249 | 575 | midx = max - bdata->node_min_pfn; |
1da177e4 | 576 | |
5f2809e6 | 577 | if (bdata->hint_idx > sidx) { |
0f3caba2 JW |
578 | /* |
579 | * Handle the valid case of sidx being zero and still | |
580 | * catch the fallback below. | |
581 | */ | |
582 | fallback = sidx + 1; | |
481ebd0d | 583 | sidx = align_idx(bdata, bdata->hint_idx, step); |
5f2809e6 | 584 | } |
1da177e4 | 585 | |
5f2809e6 JW |
586 | while (1) { |
587 | int merge; | |
588 | void *region; | |
589 | unsigned long eidx, i, start_off, end_off; | |
590 | find_block: | |
591 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); | |
481ebd0d | 592 | sidx = align_idx(bdata, sidx, step); |
5f2809e6 | 593 | eidx = sidx + PFN_UP(size); |
ad09315c | 594 | |
5f2809e6 | 595 | if (sidx >= midx || eidx > midx) |
66d43e98 | 596 | break; |
1da177e4 | 597 | |
5f2809e6 JW |
598 | for (i = sidx; i < eidx; i++) |
599 | if (test_bit(i, bdata->node_bootmem_map)) { | |
481ebd0d | 600 | sidx = align_idx(bdata, i, step); |
5f2809e6 JW |
601 | if (sidx == i) |
602 | sidx += step; | |
603 | goto find_block; | |
604 | } | |
1da177e4 | 605 | |
627240aa | 606 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
5f2809e6 | 607 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
481ebd0d | 608 | start_off = align_off(bdata, bdata->last_end_off, align); |
5f2809e6 JW |
609 | else |
610 | start_off = PFN_PHYS(sidx); | |
611 | ||
612 | merge = PFN_DOWN(start_off) < sidx; | |
613 | end_off = start_off + size; | |
614 | ||
615 | bdata->last_end_off = end_off; | |
616 | bdata->hint_idx = PFN_UP(end_off); | |
617 | ||
618 | /* | |
619 | * Reserve the area now: | |
620 | */ | |
d747fa4b JW |
621 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
622 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) | |
623 | BUG(); | |
5f2809e6 | 624 | |
3560e249 JW |
625 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
626 | start_off); | |
5f2809e6 | 627 | memset(region, 0, size); |
008139d9 CM |
628 | /* |
629 | * The min_count is set to 0 so that bootmem allocated blocks | |
630 | * are never reported as leaks. | |
631 | */ | |
632 | kmemleak_alloc(region, size, 0, 0); | |
5f2809e6 | 633 | return region; |
1da177e4 LT |
634 | } |
635 | ||
0f3caba2 | 636 | if (fallback) { |
481ebd0d | 637 | sidx = align_idx(bdata, fallback - 1, step); |
0f3caba2 JW |
638 | fallback = 0; |
639 | goto find_block; | |
640 | } | |
641 | ||
642 | return NULL; | |
643 | } | |
644 | ||
d0c4f570 TH |
645 | static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata, |
646 | unsigned long size, unsigned long align, | |
647 | unsigned long goal, unsigned long limit) | |
648 | { | |
441c7e0a PE |
649 | if (WARN_ON_ONCE(slab_is_available())) |
650 | return kzalloc(size, GFP_NOWAIT); | |
651 | ||
d0c4f570 | 652 | #ifdef CONFIG_HAVE_ARCH_BOOTMEM |
433f13a7 JP |
653 | { |
654 | bootmem_data_t *p_bdata; | |
655 | ||
656 | p_bdata = bootmem_arch_preferred_node(bdata, size, align, | |
657 | goal, limit); | |
658 | if (p_bdata) | |
659 | return alloc_bootmem_core(p_bdata, size, align, | |
660 | goal, limit); | |
661 | } | |
d0c4f570 TH |
662 | #endif |
663 | return NULL; | |
664 | } | |
08677214 | 665 | #endif |
d0c4f570 | 666 | |
0f3caba2 JW |
667 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, |
668 | unsigned long align, | |
669 | unsigned long goal, | |
670 | unsigned long limit) | |
671 | { | |
08677214 YL |
672 | #ifdef CONFIG_NO_BOOTMEM |
673 | void *ptr; | |
674 | ||
675 | if (WARN_ON_ONCE(slab_is_available())) | |
676 | return kzalloc(size, GFP_NOWAIT); | |
677 | ||
678 | restart: | |
679 | ||
680 | ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit); | |
681 | ||
682 | if (ptr) | |
683 | return ptr; | |
684 | ||
685 | if (goal != 0) { | |
686 | goal = 0; | |
687 | goto restart; | |
688 | } | |
689 | ||
690 | return NULL; | |
691 | #else | |
0f3caba2 | 692 | bootmem_data_t *bdata; |
d0c4f570 | 693 | void *region; |
0f3caba2 JW |
694 | |
695 | restart: | |
d0c4f570 TH |
696 | region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit); |
697 | if (region) | |
698 | return region; | |
0f3caba2 | 699 | |
d0c4f570 | 700 | list_for_each_entry(bdata, &bdata_list, list) { |
0f3caba2 JW |
701 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
702 | continue; | |
3560e249 | 703 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
0f3caba2 JW |
704 | break; |
705 | ||
706 | region = alloc_bootmem_core(bdata, size, align, goal, limit); | |
707 | if (region) | |
708 | return region; | |
709 | } | |
710 | ||
5f2809e6 JW |
711 | if (goal) { |
712 | goal = 0; | |
0f3caba2 | 713 | goto restart; |
5f2809e6 | 714 | } |
2e5237da | 715 | |
5f2809e6 | 716 | return NULL; |
08677214 | 717 | #endif |
1da177e4 LT |
718 | } |
719 | ||
a66fd7da JW |
720 | /** |
721 | * __alloc_bootmem_nopanic - allocate boot memory without panicking | |
722 | * @size: size of the request in bytes | |
723 | * @align: alignment of the region | |
724 | * @goal: preferred starting address of the region | |
725 | * | |
726 | * The goal is dropped if it can not be satisfied and the allocation will | |
727 | * fall back to memory below @goal. | |
728 | * | |
729 | * Allocation may happen on any node in the system. | |
730 | * | |
731 | * Returns NULL on failure. | |
732 | */ | |
bb0923a6 | 733 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
0f3caba2 | 734 | unsigned long goal) |
1da177e4 | 735 | { |
08677214 YL |
736 | unsigned long limit = 0; |
737 | ||
738 | #ifdef CONFIG_NO_BOOTMEM | |
739 | limit = -1UL; | |
740 | #endif | |
741 | ||
742 | return ___alloc_bootmem_nopanic(size, align, goal, limit); | |
0f3caba2 | 743 | } |
1da177e4 | 744 | |
0f3caba2 JW |
745 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
746 | unsigned long goal, unsigned long limit) | |
747 | { | |
748 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); | |
749 | ||
750 | if (mem) | |
751 | return mem; | |
752 | /* | |
753 | * Whoops, we cannot satisfy the allocation request. | |
754 | */ | |
755 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); | |
756 | panic("Out of memory"); | |
a8062231 AK |
757 | return NULL; |
758 | } | |
1da177e4 | 759 | |
a66fd7da JW |
760 | /** |
761 | * __alloc_bootmem - allocate boot memory | |
762 | * @size: size of the request in bytes | |
763 | * @align: alignment of the region | |
764 | * @goal: preferred starting address of the region | |
765 | * | |
766 | * The goal is dropped if it can not be satisfied and the allocation will | |
767 | * fall back to memory below @goal. | |
768 | * | |
769 | * Allocation may happen on any node in the system. | |
770 | * | |
771 | * The function panics if the request can not be satisfied. | |
772 | */ | |
bb0923a6 FBH |
773 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
774 | unsigned long goal) | |
a8062231 | 775 | { |
08677214 YL |
776 | unsigned long limit = 0; |
777 | ||
778 | #ifdef CONFIG_NO_BOOTMEM | |
779 | limit = -1UL; | |
780 | #endif | |
781 | ||
782 | return ___alloc_bootmem(size, align, goal, limit); | |
1da177e4 LT |
783 | } |
784 | ||
08677214 | 785 | #ifndef CONFIG_NO_BOOTMEM |
4cc278b7 JW |
786 | static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata, |
787 | unsigned long size, unsigned long align, | |
788 | unsigned long goal, unsigned long limit) | |
789 | { | |
790 | void *ptr; | |
791 | ||
d0c4f570 TH |
792 | ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit); |
793 | if (ptr) | |
794 | return ptr; | |
795 | ||
4cc278b7 JW |
796 | ptr = alloc_bootmem_core(bdata, size, align, goal, limit); |
797 | if (ptr) | |
798 | return ptr; | |
799 | ||
800 | return ___alloc_bootmem(size, align, goal, limit); | |
801 | } | |
08677214 | 802 | #endif |
4cc278b7 | 803 | |
a66fd7da JW |
804 | /** |
805 | * __alloc_bootmem_node - allocate boot memory from a specific node | |
806 | * @pgdat: node to allocate from | |
807 | * @size: size of the request in bytes | |
808 | * @align: alignment of the region | |
809 | * @goal: preferred starting address of the region | |
810 | * | |
811 | * The goal is dropped if it can not be satisfied and the allocation will | |
812 | * fall back to memory below @goal. | |
813 | * | |
814 | * Allocation may fall back to any node in the system if the specified node | |
815 | * can not hold the requested memory. | |
816 | * | |
817 | * The function panics if the request can not be satisfied. | |
818 | */ | |
bb0923a6 FBH |
819 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
820 | unsigned long align, unsigned long goal) | |
1da177e4 | 821 | { |
c91c4773 PE |
822 | if (WARN_ON_ONCE(slab_is_available())) |
823 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
824 | ||
08677214 YL |
825 | #ifdef CONFIG_NO_BOOTMEM |
826 | return __alloc_memory_core_early(pgdat->node_id, size, align, | |
827 | goal, -1ULL); | |
828 | #else | |
4cc278b7 | 829 | return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0); |
08677214 YL |
830 | #endif |
831 | } | |
832 | ||
833 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, | |
834 | unsigned long align, unsigned long goal) | |
835 | { | |
836 | #ifdef MAX_DMA32_PFN | |
837 | unsigned long end_pfn; | |
838 | ||
839 | if (WARN_ON_ONCE(slab_is_available())) | |
840 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
841 | ||
842 | /* update goal according ...MAX_DMA32_PFN */ | |
843 | end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
844 | ||
845 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && | |
846 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { | |
847 | void *ptr; | |
848 | unsigned long new_goal; | |
849 | ||
850 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; | |
851 | #ifdef CONFIG_NO_BOOTMEM | |
852 | ptr = __alloc_memory_core_early(pgdat->node_id, size, align, | |
853 | new_goal, -1ULL); | |
854 | #else | |
855 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, | |
856 | new_goal, 0); | |
857 | #endif | |
858 | if (ptr) | |
859 | return ptr; | |
860 | } | |
861 | #endif | |
862 | ||
863 | return __alloc_bootmem_node(pgdat, size, align, goal); | |
864 | ||
1da177e4 LT |
865 | } |
866 | ||
e70260aa | 867 | #ifdef CONFIG_SPARSEMEM |
a66fd7da JW |
868 | /** |
869 | * alloc_bootmem_section - allocate boot memory from a specific section | |
870 | * @size: size of the request in bytes | |
871 | * @section_nr: sparse map section to allocate from | |
872 | * | |
873 | * Return NULL on failure. | |
874 | */ | |
e70260aa YG |
875 | void * __init alloc_bootmem_section(unsigned long size, |
876 | unsigned long section_nr) | |
877 | { | |
08677214 YL |
878 | #ifdef CONFIG_NO_BOOTMEM |
879 | unsigned long pfn, goal, limit; | |
880 | ||
881 | pfn = section_nr_to_pfn(section_nr); | |
882 | goal = pfn << PAGE_SHIFT; | |
883 | limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT; | |
884 | ||
885 | return __alloc_memory_core_early(early_pfn_to_nid(pfn), size, | |
886 | SMP_CACHE_BYTES, goal, limit); | |
887 | #else | |
75a56cfe JW |
888 | bootmem_data_t *bdata; |
889 | unsigned long pfn, goal, limit; | |
e70260aa YG |
890 | |
891 | pfn = section_nr_to_pfn(section_nr); | |
75a56cfe JW |
892 | goal = pfn << PAGE_SHIFT; |
893 | limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT; | |
894 | bdata = &bootmem_node_data[early_pfn_to_nid(pfn)]; | |
e70260aa | 895 | |
75a56cfe | 896 | return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit); |
08677214 | 897 | #endif |
e70260aa YG |
898 | } |
899 | #endif | |
900 | ||
b54bbf7b AK |
901 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, |
902 | unsigned long align, unsigned long goal) | |
903 | { | |
904 | void *ptr; | |
905 | ||
c91c4773 PE |
906 | if (WARN_ON_ONCE(slab_is_available())) |
907 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
908 | ||
08677214 YL |
909 | #ifdef CONFIG_NO_BOOTMEM |
910 | ptr = __alloc_memory_core_early(pgdat->node_id, size, align, | |
911 | goal, -1ULL); | |
912 | #else | |
d0c4f570 TH |
913 | ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0); |
914 | if (ptr) | |
915 | return ptr; | |
916 | ||
b54bbf7b | 917 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0); |
08677214 | 918 | #endif |
b54bbf7b AK |
919 | if (ptr) |
920 | return ptr; | |
921 | ||
922 | return __alloc_bootmem_nopanic(size, align, goal); | |
923 | } | |
924 | ||
dfd54cbc HC |
925 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
926 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL | |
927 | #endif | |
008857c1 | 928 | |
a66fd7da JW |
929 | /** |
930 | * __alloc_bootmem_low - allocate low boot memory | |
931 | * @size: size of the request in bytes | |
932 | * @align: alignment of the region | |
933 | * @goal: preferred starting address of the region | |
934 | * | |
935 | * The goal is dropped if it can not be satisfied and the allocation will | |
936 | * fall back to memory below @goal. | |
937 | * | |
938 | * Allocation may happen on any node in the system. | |
939 | * | |
940 | * The function panics if the request can not be satisfied. | |
941 | */ | |
bb0923a6 FBH |
942 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
943 | unsigned long goal) | |
008857c1 | 944 | { |
0f3caba2 | 945 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
008857c1 RT |
946 | } |
947 | ||
a66fd7da JW |
948 | /** |
949 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node | |
950 | * @pgdat: node to allocate from | |
951 | * @size: size of the request in bytes | |
952 | * @align: alignment of the region | |
953 | * @goal: preferred starting address of the region | |
954 | * | |
955 | * The goal is dropped if it can not be satisfied and the allocation will | |
956 | * fall back to memory below @goal. | |
957 | * | |
958 | * Allocation may fall back to any node in the system if the specified node | |
959 | * can not hold the requested memory. | |
960 | * | |
961 | * The function panics if the request can not be satisfied. | |
962 | */ | |
008857c1 RT |
963 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
964 | unsigned long align, unsigned long goal) | |
965 | { | |
c91c4773 PE |
966 | if (WARN_ON_ONCE(slab_is_available())) |
967 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
968 | ||
08677214 YL |
969 | #ifdef CONFIG_NO_BOOTMEM |
970 | return __alloc_memory_core_early(pgdat->node_id, size, align, | |
971 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
972 | #else | |
4cc278b7 JW |
973 | return ___alloc_bootmem_node(pgdat->bdata, size, align, |
974 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
08677214 | 975 | #endif |
008857c1 | 976 | } |