]>
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> |
e786e86a FBH |
15 | |
16 | #include <asm/bug.h> | |
1da177e4 | 17 | #include <asm/io.h> |
dfd54cbc | 18 | #include <asm/processor.h> |
e786e86a | 19 | |
1da177e4 LT |
20 | #include "internal.h" |
21 | ||
1da177e4 LT |
22 | unsigned long max_low_pfn; |
23 | unsigned long min_low_pfn; | |
24 | unsigned long max_pfn; | |
25 | ||
92aa63a5 VG |
26 | #ifdef CONFIG_CRASH_DUMP |
27 | /* | |
28 | * If we have booted due to a crash, max_pfn will be a very low value. We need | |
29 | * to know the amount of memory that the previous kernel used. | |
30 | */ | |
31 | unsigned long saved_max_pfn; | |
32 | #endif | |
33 | ||
b61bfa3c JW |
34 | bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata; |
35 | ||
636cc40c JW |
36 | static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list); |
37 | ||
2e5237da JW |
38 | static int bootmem_debug; |
39 | ||
40 | static int __init bootmem_debug_setup(char *buf) | |
41 | { | |
42 | bootmem_debug = 1; | |
43 | return 0; | |
44 | } | |
45 | early_param("bootmem_debug", bootmem_debug_setup); | |
46 | ||
47 | #define bdebug(fmt, args...) ({ \ | |
48 | if (unlikely(bootmem_debug)) \ | |
49 | printk(KERN_INFO \ | |
50 | "bootmem::%s " fmt, \ | |
51 | __FUNCTION__, ## args); \ | |
52 | }) | |
53 | ||
df049a5f | 54 | static unsigned long __init bootmap_bytes(unsigned long pages) |
223e8dc9 | 55 | { |
df049a5f | 56 | unsigned long bytes = (pages + 7) / 8; |
223e8dc9 | 57 | |
df049a5f | 58 | return ALIGN(bytes, sizeof(long)); |
223e8dc9 JW |
59 | } |
60 | ||
a66fd7da JW |
61 | /** |
62 | * bootmem_bootmap_pages - calculate bitmap size in pages | |
63 | * @pages: number of pages the bitmap has to represent | |
64 | */ | |
f71bf0ca | 65 | unsigned long __init bootmem_bootmap_pages(unsigned long pages) |
1da177e4 | 66 | { |
df049a5f | 67 | unsigned long bytes = bootmap_bytes(pages); |
1da177e4 | 68 | |
df049a5f | 69 | return PAGE_ALIGN(bytes) >> PAGE_SHIFT; |
1da177e4 | 70 | } |
f71bf0ca | 71 | |
679bc9fb KH |
72 | /* |
73 | * link bdata in order | |
74 | */ | |
69d49e68 | 75 | static void __init link_bootmem(bootmem_data_t *bdata) |
679bc9fb | 76 | { |
636cc40c | 77 | struct list_head *iter; |
f71bf0ca | 78 | |
636cc40c JW |
79 | list_for_each(iter, &bdata_list) { |
80 | bootmem_data_t *ent; | |
81 | ||
82 | ent = list_entry(iter, bootmem_data_t, list); | |
83 | if (bdata->node_boot_start < ent->node_boot_start) | |
84 | break; | |
679bc9fb | 85 | } |
636cc40c | 86 | list_add_tail(&bdata->list, iter); |
679bc9fb KH |
87 | } |
88 | ||
1da177e4 LT |
89 | /* |
90 | * Called once to set up the allocator itself. | |
91 | */ | |
8ae04463 | 92 | static unsigned long __init init_bootmem_core(bootmem_data_t *bdata, |
1da177e4 LT |
93 | unsigned long mapstart, unsigned long start, unsigned long end) |
94 | { | |
bbc7b92e | 95 | unsigned long mapsize; |
1da177e4 | 96 | |
2dbb51c4 | 97 | mminit_validate_memmodel_limits(&start, &end); |
bbc7b92e FBH |
98 | bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart)); |
99 | bdata->node_boot_start = PFN_PHYS(start); | |
1da177e4 | 100 | bdata->node_low_pfn = end; |
679bc9fb | 101 | link_bootmem(bdata); |
1da177e4 LT |
102 | |
103 | /* | |
104 | * Initially all pages are reserved - setup_arch() has to | |
105 | * register free RAM areas explicitly. | |
106 | */ | |
df049a5f | 107 | mapsize = bootmap_bytes(end - start); |
1da177e4 LT |
108 | memset(bdata->node_bootmem_map, 0xff, mapsize); |
109 | ||
2e5237da JW |
110 | bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n", |
111 | bdata - bootmem_node_data, start, mapstart, end, mapsize); | |
112 | ||
1da177e4 LT |
113 | return mapsize; |
114 | } | |
115 | ||
a66fd7da JW |
116 | /** |
117 | * init_bootmem_node - register a node as boot memory | |
118 | * @pgdat: node to register | |
119 | * @freepfn: pfn where the bitmap for this node is to be placed | |
120 | * @startpfn: first pfn on the node | |
121 | * @endpfn: first pfn after the node | |
122 | * | |
123 | * Returns the number of bytes needed to hold the bitmap for this node. | |
124 | */ | |
223e8dc9 JW |
125 | unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn, |
126 | unsigned long startpfn, unsigned long endpfn) | |
127 | { | |
128 | return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn); | |
129 | } | |
130 | ||
a66fd7da JW |
131 | /** |
132 | * init_bootmem - register boot memory | |
133 | * @start: pfn where the bitmap is to be placed | |
134 | * @pages: number of available physical pages | |
135 | * | |
136 | * Returns the number of bytes needed to hold the bitmap. | |
137 | */ | |
223e8dc9 JW |
138 | unsigned long __init init_bootmem(unsigned long start, unsigned long pages) |
139 | { | |
140 | max_low_pfn = pages; | |
141 | min_low_pfn = start; | |
142 | return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages); | |
143 | } | |
144 | ||
145 | static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) | |
146 | { | |
41546c17 | 147 | int aligned; |
223e8dc9 | 148 | struct page *page; |
41546c17 JW |
149 | unsigned long start, end, pages, count = 0; |
150 | ||
151 | if (!bdata->node_bootmem_map) | |
152 | return 0; | |
153 | ||
154 | start = PFN_DOWN(bdata->node_boot_start); | |
155 | end = bdata->node_low_pfn; | |
156 | ||
223e8dc9 | 157 | /* |
41546c17 JW |
158 | * If the start is aligned to the machines wordsize, we might |
159 | * be able to free pages in bulks of that order. | |
223e8dc9 | 160 | */ |
41546c17 | 161 | aligned = !(start & (BITS_PER_LONG - 1)); |
223e8dc9 | 162 | |
41546c17 JW |
163 | bdebug("nid=%td start=%lx end=%lx aligned=%d\n", |
164 | bdata - bootmem_node_data, start, end, aligned); | |
223e8dc9 | 165 | |
41546c17 JW |
166 | while (start < end) { |
167 | unsigned long *map, idx, vec; | |
223e8dc9 | 168 | |
41546c17 JW |
169 | map = bdata->node_bootmem_map; |
170 | idx = start - PFN_DOWN(bdata->node_boot_start); | |
171 | vec = ~map[idx / BITS_PER_LONG]; | |
172 | ||
173 | if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) { | |
174 | int order = ilog2(BITS_PER_LONG); | |
175 | ||
176 | __free_pages_bootmem(pfn_to_page(start), order); | |
223e8dc9 | 177 | count += BITS_PER_LONG; |
41546c17 JW |
178 | } else { |
179 | unsigned long off = 0; | |
180 | ||
181 | while (vec && off < BITS_PER_LONG) { | |
182 | if (vec & 1) { | |
183 | page = pfn_to_page(start + off); | |
223e8dc9 | 184 | __free_pages_bootmem(page, 0); |
41546c17 | 185 | count++; |
223e8dc9 | 186 | } |
41546c17 JW |
187 | vec >>= 1; |
188 | off++; | |
223e8dc9 | 189 | } |
223e8dc9 | 190 | } |
41546c17 | 191 | start += BITS_PER_LONG; |
223e8dc9 JW |
192 | } |
193 | ||
223e8dc9 | 194 | page = virt_to_page(bdata->node_bootmem_map); |
df049a5f | 195 | pages = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start); |
41546c17 JW |
196 | pages = bootmem_bootmap_pages(pages); |
197 | count += pages; | |
198 | while (pages--) | |
199 | __free_pages_bootmem(page++, 0); | |
223e8dc9 | 200 | |
2e5237da JW |
201 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
202 | ||
223e8dc9 JW |
203 | return count; |
204 | } | |
205 | ||
a66fd7da JW |
206 | /** |
207 | * free_all_bootmem_node - release a node's free pages to the buddy allocator | |
208 | * @pgdat: node to be released | |
209 | * | |
210 | * Returns the number of pages actually released. | |
211 | */ | |
223e8dc9 JW |
212 | unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) |
213 | { | |
214 | register_page_bootmem_info_node(pgdat); | |
215 | return free_all_bootmem_core(pgdat->bdata); | |
216 | } | |
217 | ||
a66fd7da JW |
218 | /** |
219 | * free_all_bootmem - release free pages to the buddy allocator | |
220 | * | |
221 | * Returns the number of pages actually released. | |
222 | */ | |
223e8dc9 JW |
223 | unsigned long __init free_all_bootmem(void) |
224 | { | |
225 | return free_all_bootmem_core(NODE_DATA(0)->bdata); | |
226 | } | |
227 | ||
228 | static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr, | |
229 | unsigned long size) | |
230 | { | |
231 | unsigned long sidx, eidx; | |
232 | unsigned long i; | |
233 | ||
234 | BUG_ON(!size); | |
235 | ||
236 | /* out range */ | |
237 | if (addr + size < bdata->node_boot_start || | |
238 | PFN_DOWN(addr) > bdata->node_low_pfn) | |
239 | return; | |
240 | /* | |
241 | * round down end of usable mem, partially free pages are | |
242 | * considered reserved. | |
243 | */ | |
244 | ||
245 | if (addr >= bdata->node_boot_start && addr < bdata->last_success) | |
246 | bdata->last_success = addr; | |
247 | ||
248 | /* | |
249 | * Round up to index to the range. | |
250 | */ | |
251 | if (PFN_UP(addr) > PFN_DOWN(bdata->node_boot_start)) | |
252 | sidx = PFN_UP(addr) - PFN_DOWN(bdata->node_boot_start); | |
253 | else | |
254 | sidx = 0; | |
255 | ||
256 | eidx = PFN_DOWN(addr + size - bdata->node_boot_start); | |
257 | if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start)) | |
258 | eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start); | |
259 | ||
2e5237da JW |
260 | bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data, |
261 | sidx + PFN_DOWN(bdata->node_boot_start), | |
262 | eidx + PFN_DOWN(bdata->node_boot_start)); | |
263 | ||
223e8dc9 JW |
264 | for (i = sidx; i < eidx; i++) { |
265 | if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map))) | |
266 | BUG(); | |
267 | } | |
268 | } | |
269 | ||
a66fd7da JW |
270 | /** |
271 | * free_bootmem_node - mark a page range as usable | |
272 | * @pgdat: node the range resides on | |
273 | * @physaddr: starting address of the range | |
274 | * @size: size of the range in bytes | |
275 | * | |
276 | * Partial pages will be considered reserved and left as they are. | |
277 | * | |
278 | * Only physical pages that actually reside on @pgdat are marked. | |
279 | */ | |
223e8dc9 JW |
280 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
281 | unsigned long size) | |
282 | { | |
283 | free_bootmem_core(pgdat->bdata, physaddr, size); | |
284 | } | |
285 | ||
a66fd7da JW |
286 | /** |
287 | * free_bootmem - mark a page range as usable | |
288 | * @addr: starting address of the range | |
289 | * @size: size of the range in bytes | |
290 | * | |
291 | * Partial pages will be considered reserved and left as they are. | |
292 | * | |
293 | * All physical pages within the range are marked, no matter what | |
294 | * node they reside on. | |
295 | */ | |
223e8dc9 JW |
296 | void __init free_bootmem(unsigned long addr, unsigned long size) |
297 | { | |
298 | bootmem_data_t *bdata; | |
299 | list_for_each_entry(bdata, &bdata_list, list) | |
300 | free_bootmem_core(bdata, addr, size); | |
301 | } | |
302 | ||
1da177e4 LT |
303 | /* |
304 | * Marks a particular physical memory range as unallocatable. Usable RAM | |
305 | * might be used for boot-time allocations - or it might get added | |
306 | * to the free page pool later on. | |
307 | */ | |
a5645a61 | 308 | static int __init can_reserve_bootmem_core(bootmem_data_t *bdata, |
72a7fe39 | 309 | unsigned long addr, unsigned long size, int flags) |
1da177e4 | 310 | { |
bbc7b92e | 311 | unsigned long sidx, eidx; |
1da177e4 | 312 | unsigned long i; |
a5645a61 YL |
313 | |
314 | BUG_ON(!size); | |
315 | ||
316 | /* out of range, don't hold other */ | |
317 | if (addr + size < bdata->node_boot_start || | |
318 | PFN_DOWN(addr) > bdata->node_low_pfn) | |
319 | return 0; | |
bbc7b92e | 320 | |
1da177e4 | 321 | /* |
a5645a61 | 322 | * Round up to index to the range. |
1da177e4 | 323 | */ |
a5645a61 YL |
324 | if (addr > bdata->node_boot_start) |
325 | sidx= PFN_DOWN(addr - bdata->node_boot_start); | |
326 | else | |
327 | sidx = 0; | |
328 | ||
329 | eidx = PFN_UP(addr + size - bdata->node_boot_start); | |
330 | if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start)) | |
331 | eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start); | |
332 | ||
333 | for (i = sidx; i < eidx; i++) { | |
334 | if (test_bit(i, bdata->node_bootmem_map)) { | |
335 | if (flags & BOOTMEM_EXCLUSIVE) | |
336 | return -EBUSY; | |
337 | } | |
338 | } | |
339 | ||
340 | return 0; | |
341 | ||
342 | } | |
343 | ||
344 | static void __init reserve_bootmem_core(bootmem_data_t *bdata, | |
345 | unsigned long addr, unsigned long size, int flags) | |
346 | { | |
347 | unsigned long sidx, eidx; | |
348 | unsigned long i; | |
349 | ||
1da177e4 | 350 | BUG_ON(!size); |
bbc7b92e | 351 | |
a5645a61 YL |
352 | /* out of range */ |
353 | if (addr + size < bdata->node_boot_start || | |
354 | PFN_DOWN(addr) > bdata->node_low_pfn) | |
355 | return; | |
356 | ||
357 | /* | |
358 | * Round up to index to the range. | |
359 | */ | |
360 | if (addr > bdata->node_boot_start) | |
361 | sidx= PFN_DOWN(addr - bdata->node_boot_start); | |
362 | else | |
363 | sidx = 0; | |
364 | ||
bbc7b92e | 365 | eidx = PFN_UP(addr + size - bdata->node_boot_start); |
a5645a61 YL |
366 | if (eidx > bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start)) |
367 | eidx = bdata->node_low_pfn - PFN_DOWN(bdata->node_boot_start); | |
1da177e4 | 368 | |
2e5237da JW |
369 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", |
370 | bdata - bootmem_node_data, | |
371 | sidx + PFN_DOWN(bdata->node_boot_start), | |
372 | eidx + PFN_DOWN(bdata->node_boot_start), | |
373 | flags); | |
374 | ||
375 | for (i = sidx; i < eidx; i++) | |
376 | if (test_and_set_bit(i, bdata->node_bootmem_map)) | |
377 | bdebug("hm, page %lx reserved twice.\n", | |
378 | PFN_DOWN(bdata->node_boot_start) + i); | |
1da177e4 LT |
379 | } |
380 | ||
a66fd7da JW |
381 | /** |
382 | * reserve_bootmem_node - mark a page range as reserved | |
383 | * @pgdat: node the range resides on | |
384 | * @physaddr: starting address of the range | |
385 | * @size: size of the range in bytes | |
386 | * @flags: reservation flags (see linux/bootmem.h) | |
387 | * | |
388 | * Partial pages will be reserved. | |
389 | * | |
390 | * Only physical pages that actually reside on @pgdat are marked. | |
391 | */ | |
223e8dc9 JW |
392 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
393 | unsigned long size, int flags) | |
1da177e4 | 394 | { |
223e8dc9 | 395 | int ret; |
1da177e4 | 396 | |
223e8dc9 JW |
397 | ret = can_reserve_bootmem_core(pgdat->bdata, physaddr, size, flags); |
398 | if (ret < 0) | |
399 | return -ENOMEM; | |
400 | reserve_bootmem_core(pgdat->bdata, physaddr, size, flags); | |
401 | return 0; | |
402 | } | |
5a982cbc | 403 | |
223e8dc9 | 404 | #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE |
a66fd7da JW |
405 | /** |
406 | * reserve_bootmem - mark a page range as usable | |
407 | * @addr: starting address of the range | |
408 | * @size: size of the range in bytes | |
409 | * @flags: reservation flags (see linux/bootmem.h) | |
410 | * | |
411 | * Partial pages will be reserved. | |
412 | * | |
413 | * All physical pages within the range are marked, no matter what | |
414 | * node they reside on. | |
415 | */ | |
223e8dc9 JW |
416 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
417 | int flags) | |
418 | { | |
419 | bootmem_data_t *bdata; | |
420 | int ret; | |
1da177e4 | 421 | |
223e8dc9 JW |
422 | list_for_each_entry(bdata, &bdata_list, list) { |
423 | ret = can_reserve_bootmem_core(bdata, addr, size, flags); | |
424 | if (ret < 0) | |
425 | return ret; | |
1da177e4 | 426 | } |
223e8dc9 JW |
427 | list_for_each_entry(bdata, &bdata_list, list) |
428 | reserve_bootmem_core(bdata, addr, size, flags); | |
429 | ||
430 | return 0; | |
1da177e4 | 431 | } |
223e8dc9 | 432 | #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */ |
1da177e4 LT |
433 | |
434 | /* | |
435 | * We 'merge' subsequent allocations to save space. We might 'lose' | |
436 | * some fraction of a page if allocations cannot be satisfied due to | |
437 | * size constraints on boxes where there is physical RAM space | |
438 | * fragmentation - in these cases (mostly large memory boxes) this | |
439 | * is not a problem. | |
440 | * | |
441 | * On low memory boxes we get it right in 100% of the cases. | |
442 | * | |
443 | * alignment has to be a power of 2 value. | |
444 | * | |
445 | * NOTE: This function is _not_ reentrant. | |
446 | */ | |
ffc6421f JW |
447 | static void * __init |
448 | alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size, | |
449 | unsigned long align, unsigned long goal, unsigned long limit) | |
1da177e4 | 450 | { |
9a2dc04c | 451 | unsigned long areasize, preferred; |
bbc7b92e | 452 | unsigned long i, start = 0, incr, eidx, end_pfn; |
1da177e4 | 453 | void *ret; |
9a2dc04c YL |
454 | unsigned long node_boot_start; |
455 | void *node_bootmem_map; | |
1da177e4 | 456 | |
f71bf0ca | 457 | if (!size) { |
ffc6421f | 458 | printk("alloc_bootmem_core(): zero-sized request\n"); |
1da177e4 LT |
459 | BUG(); |
460 | } | |
461 | BUG_ON(align & (align-1)); | |
462 | ||
7c309a64 CK |
463 | /* on nodes without memory - bootmem_map is NULL */ |
464 | if (!bdata->node_bootmem_map) | |
465 | return NULL; | |
466 | ||
2e5237da JW |
467 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
468 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, | |
469 | align, goal, limit); | |
470 | ||
9a2dc04c YL |
471 | /* bdata->node_boot_start is supposed to be (12+6)bits alignment on x86_64 ? */ |
472 | node_boot_start = bdata->node_boot_start; | |
473 | node_bootmem_map = bdata->node_bootmem_map; | |
474 | if (align) { | |
475 | node_boot_start = ALIGN(bdata->node_boot_start, align); | |
476 | if (node_boot_start > bdata->node_boot_start) | |
477 | node_bootmem_map = (unsigned long *)bdata->node_bootmem_map + | |
478 | PFN_DOWN(node_boot_start - bdata->node_boot_start)/BITS_PER_LONG; | |
479 | } | |
480 | ||
481 | if (limit && node_boot_start >= limit) | |
482 | return NULL; | |
483 | ||
bbc7b92e FBH |
484 | end_pfn = bdata->node_low_pfn; |
485 | limit = PFN_DOWN(limit); | |
281dd25c YG |
486 | if (limit && end_pfn > limit) |
487 | end_pfn = limit; | |
488 | ||
9a2dc04c | 489 | eidx = end_pfn - PFN_DOWN(node_boot_start); |
1da177e4 LT |
490 | |
491 | /* | |
492 | * We try to allocate bootmem pages above 'goal' | |
493 | * first, then we try to allocate lower pages. | |
494 | */ | |
ad09315c YL |
495 | preferred = 0; |
496 | if (goal && PFN_DOWN(goal) < end_pfn) { | |
9a2dc04c YL |
497 | if (goal > node_boot_start) |
498 | preferred = goal - node_boot_start; | |
1da177e4 | 499 | |
9a2dc04c YL |
500 | if (bdata->last_success > node_boot_start && |
501 | bdata->last_success - node_boot_start >= preferred) | |
281dd25c | 502 | if (!limit || (limit && limit > bdata->last_success)) |
9a2dc04c | 503 | preferred = bdata->last_success - node_boot_start; |
ad09315c | 504 | } |
1da177e4 | 505 | |
9a2dc04c | 506 | preferred = PFN_DOWN(ALIGN(preferred, align)); |
bbc7b92e | 507 | areasize = (size + PAGE_SIZE-1) / PAGE_SIZE; |
1da177e4 LT |
508 | incr = align >> PAGE_SHIFT ? : 1; |
509 | ||
510 | restart_scan: | |
ad09315c | 511 | for (i = preferred; i < eidx;) { |
1da177e4 | 512 | unsigned long j; |
ad09315c | 513 | |
9a2dc04c | 514 | i = find_next_zero_bit(node_bootmem_map, eidx, i); |
1da177e4 | 515 | i = ALIGN(i, incr); |
66d43e98 HM |
516 | if (i >= eidx) |
517 | break; | |
9a2dc04c | 518 | if (test_bit(i, node_bootmem_map)) { |
ad09315c | 519 | i += incr; |
1da177e4 | 520 | continue; |
ad09315c | 521 | } |
1da177e4 LT |
522 | for (j = i + 1; j < i + areasize; ++j) { |
523 | if (j >= eidx) | |
524 | goto fail_block; | |
9a2dc04c | 525 | if (test_bit(j, node_bootmem_map)) |
1da177e4 LT |
526 | goto fail_block; |
527 | } | |
528 | start = i; | |
529 | goto found; | |
530 | fail_block: | |
531 | i = ALIGN(j, incr); | |
ad09315c YL |
532 | if (i == j) |
533 | i += incr; | |
1da177e4 LT |
534 | } |
535 | ||
9a2dc04c YL |
536 | if (preferred > 0) { |
537 | preferred = 0; | |
1da177e4 LT |
538 | goto restart_scan; |
539 | } | |
540 | return NULL; | |
541 | ||
542 | found: | |
9a2dc04c | 543 | bdata->last_success = PFN_PHYS(start) + node_boot_start; |
1da177e4 LT |
544 | BUG_ON(start >= eidx); |
545 | ||
546 | /* | |
547 | * Is the next page of the previous allocation-end the start | |
548 | * of this allocation's buffer? If yes then we can 'merge' | |
549 | * the previous partial page with this allocation. | |
550 | */ | |
551 | if (align < PAGE_SIZE && | |
552 | bdata->last_offset && bdata->last_pos+1 == start) { | |
9a2dc04c | 553 | unsigned long offset, remaining_size; |
8c0e33c1 | 554 | offset = ALIGN(bdata->last_offset, align); |
1da177e4 | 555 | BUG_ON(offset > PAGE_SIZE); |
f71bf0ca | 556 | remaining_size = PAGE_SIZE - offset; |
1da177e4 LT |
557 | if (size < remaining_size) { |
558 | areasize = 0; | |
559 | /* last_pos unchanged */ | |
f71bf0ca FBH |
560 | bdata->last_offset = offset + size; |
561 | ret = phys_to_virt(bdata->last_pos * PAGE_SIZE + | |
9a2dc04c | 562 | offset + node_boot_start); |
1da177e4 LT |
563 | } else { |
564 | remaining_size = size - remaining_size; | |
f71bf0ca FBH |
565 | areasize = (remaining_size + PAGE_SIZE-1) / PAGE_SIZE; |
566 | ret = phys_to_virt(bdata->last_pos * PAGE_SIZE + | |
9a2dc04c | 567 | offset + node_boot_start); |
f71bf0ca | 568 | bdata->last_pos = start + areasize - 1; |
1da177e4 LT |
569 | bdata->last_offset = remaining_size; |
570 | } | |
571 | bdata->last_offset &= ~PAGE_MASK; | |
572 | } else { | |
573 | bdata->last_pos = start + areasize - 1; | |
574 | bdata->last_offset = size & ~PAGE_MASK; | |
9a2dc04c | 575 | ret = phys_to_virt(start * PAGE_SIZE + node_boot_start); |
1da177e4 LT |
576 | } |
577 | ||
2e5237da JW |
578 | bdebug("nid=%td start=%lx end=%lx\n", |
579 | bdata - bootmem_node_data, | |
580 | start + PFN_DOWN(bdata->node_boot_start), | |
581 | start + areasize + PFN_DOWN(bdata->node_boot_start)); | |
582 | ||
1da177e4 LT |
583 | /* |
584 | * Reserve the area now: | |
585 | */ | |
f71bf0ca | 586 | for (i = start; i < start + areasize; i++) |
9a2dc04c | 587 | if (unlikely(test_and_set_bit(i, node_bootmem_map))) |
1da177e4 LT |
588 | BUG(); |
589 | memset(ret, 0, size); | |
590 | return ret; | |
591 | } | |
592 | ||
a66fd7da JW |
593 | /** |
594 | * __alloc_bootmem_nopanic - allocate boot memory without panicking | |
595 | * @size: size of the request in bytes | |
596 | * @align: alignment of the region | |
597 | * @goal: preferred starting address of the region | |
598 | * | |
599 | * The goal is dropped if it can not be satisfied and the allocation will | |
600 | * fall back to memory below @goal. | |
601 | * | |
602 | * Allocation may happen on any node in the system. | |
603 | * | |
604 | * Returns NULL on failure. | |
605 | */ | |
bb0923a6 FBH |
606 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
607 | unsigned long goal) | |
1da177e4 | 608 | { |
679bc9fb | 609 | bootmem_data_t *bdata; |
1da177e4 LT |
610 | void *ptr; |
611 | ||
f71bf0ca | 612 | list_for_each_entry(bdata, &bdata_list, list) { |
ffc6421f | 613 | ptr = alloc_bootmem_core(bdata, size, align, goal, 0); |
f71bf0ca FBH |
614 | if (ptr) |
615 | return ptr; | |
616 | } | |
a8062231 AK |
617 | return NULL; |
618 | } | |
1da177e4 | 619 | |
a66fd7da JW |
620 | /** |
621 | * __alloc_bootmem - allocate boot memory | |
622 | * @size: size of the request in bytes | |
623 | * @align: alignment of the region | |
624 | * @goal: preferred starting address of the region | |
625 | * | |
626 | * The goal is dropped if it can not be satisfied and the allocation will | |
627 | * fall back to memory below @goal. | |
628 | * | |
629 | * Allocation may happen on any node in the system. | |
630 | * | |
631 | * The function panics if the request can not be satisfied. | |
632 | */ | |
bb0923a6 FBH |
633 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
634 | unsigned long goal) | |
a8062231 AK |
635 | { |
636 | void *mem = __alloc_bootmem_nopanic(size,align,goal); | |
f71bf0ca | 637 | |
a8062231 AK |
638 | if (mem) |
639 | return mem; | |
1da177e4 LT |
640 | /* |
641 | * Whoops, we cannot satisfy the allocation request. | |
642 | */ | |
643 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); | |
644 | panic("Out of memory"); | |
645 | return NULL; | |
646 | } | |
647 | ||
a66fd7da JW |
648 | /** |
649 | * __alloc_bootmem_node - allocate boot memory from a specific node | |
650 | * @pgdat: node to allocate from | |
651 | * @size: size of the request in bytes | |
652 | * @align: alignment of the region | |
653 | * @goal: preferred starting address of the region | |
654 | * | |
655 | * The goal is dropped if it can not be satisfied and the allocation will | |
656 | * fall back to memory below @goal. | |
657 | * | |
658 | * Allocation may fall back to any node in the system if the specified node | |
659 | * can not hold the requested memory. | |
660 | * | |
661 | * The function panics if the request can not be satisfied. | |
662 | */ | |
bb0923a6 FBH |
663 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
664 | unsigned long align, unsigned long goal) | |
1da177e4 LT |
665 | { |
666 | void *ptr; | |
667 | ||
ffc6421f | 668 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0); |
1da177e4 | 669 | if (ptr) |
f71bf0ca | 670 | return ptr; |
1da177e4 | 671 | |
008857c1 | 672 | return __alloc_bootmem(size, align, goal); |
1da177e4 LT |
673 | } |
674 | ||
e70260aa | 675 | #ifdef CONFIG_SPARSEMEM |
a66fd7da JW |
676 | /** |
677 | * alloc_bootmem_section - allocate boot memory from a specific section | |
678 | * @size: size of the request in bytes | |
679 | * @section_nr: sparse map section to allocate from | |
680 | * | |
681 | * Return NULL on failure. | |
682 | */ | |
e70260aa YG |
683 | void * __init alloc_bootmem_section(unsigned long size, |
684 | unsigned long section_nr) | |
685 | { | |
686 | void *ptr; | |
687 | unsigned long limit, goal, start_nr, end_nr, pfn; | |
688 | struct pglist_data *pgdat; | |
689 | ||
690 | pfn = section_nr_to_pfn(section_nr); | |
691 | goal = PFN_PHYS(pfn); | |
692 | limit = PFN_PHYS(section_nr_to_pfn(section_nr + 1)) - 1; | |
693 | pgdat = NODE_DATA(early_pfn_to_nid(pfn)); | |
ffc6421f JW |
694 | ptr = alloc_bootmem_core(pgdat->bdata, size, SMP_CACHE_BYTES, goal, |
695 | limit); | |
e70260aa YG |
696 | |
697 | if (!ptr) | |
698 | return NULL; | |
699 | ||
700 | start_nr = pfn_to_section_nr(PFN_DOWN(__pa(ptr))); | |
701 | end_nr = pfn_to_section_nr(PFN_DOWN(__pa(ptr) + size)); | |
702 | if (start_nr != section_nr || end_nr != section_nr) { | |
703 | printk(KERN_WARNING "alloc_bootmem failed on section %ld.\n", | |
704 | section_nr); | |
705 | free_bootmem_core(pgdat->bdata, __pa(ptr), size); | |
706 | ptr = NULL; | |
707 | } | |
708 | ||
709 | return ptr; | |
710 | } | |
711 | #endif | |
712 | ||
b54bbf7b AK |
713 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, |
714 | unsigned long align, unsigned long goal) | |
715 | { | |
716 | void *ptr; | |
717 | ||
718 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0); | |
719 | if (ptr) | |
720 | return ptr; | |
721 | ||
722 | return __alloc_bootmem_nopanic(size, align, goal); | |
723 | } | |
724 | ||
dfd54cbc HC |
725 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
726 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL | |
727 | #endif | |
008857c1 | 728 | |
a66fd7da JW |
729 | /** |
730 | * __alloc_bootmem_low - allocate low boot memory | |
731 | * @size: size of the request in bytes | |
732 | * @align: alignment of the region | |
733 | * @goal: preferred starting address of the region | |
734 | * | |
735 | * The goal is dropped if it can not be satisfied and the allocation will | |
736 | * fall back to memory below @goal. | |
737 | * | |
738 | * Allocation may happen on any node in the system. | |
739 | * | |
740 | * The function panics if the request can not be satisfied. | |
741 | */ | |
bb0923a6 FBH |
742 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
743 | unsigned long goal) | |
008857c1 | 744 | { |
679bc9fb | 745 | bootmem_data_t *bdata; |
008857c1 RT |
746 | void *ptr; |
747 | ||
f71bf0ca | 748 | list_for_each_entry(bdata, &bdata_list, list) { |
ffc6421f JW |
749 | ptr = alloc_bootmem_core(bdata, size, align, goal, |
750 | ARCH_LOW_ADDRESS_LIMIT); | |
f71bf0ca FBH |
751 | if (ptr) |
752 | return ptr; | |
753 | } | |
008857c1 RT |
754 | |
755 | /* | |
756 | * Whoops, we cannot satisfy the allocation request. | |
757 | */ | |
758 | printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size); | |
759 | panic("Out of low memory"); | |
760 | return NULL; | |
761 | } | |
762 | ||
a66fd7da JW |
763 | /** |
764 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node | |
765 | * @pgdat: node to allocate from | |
766 | * @size: size of the request in bytes | |
767 | * @align: alignment of the region | |
768 | * @goal: preferred starting address of the region | |
769 | * | |
770 | * The goal is dropped if it can not be satisfied and the allocation will | |
771 | * fall back to memory below @goal. | |
772 | * | |
773 | * Allocation may fall back to any node in the system if the specified node | |
774 | * can not hold the requested memory. | |
775 | * | |
776 | * The function panics if the request can not be satisfied. | |
777 | */ | |
008857c1 RT |
778 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
779 | unsigned long align, unsigned long goal) | |
780 | { | |
ffc6421f JW |
781 | return alloc_bootmem_core(pgdat->bdata, size, align, goal, |
782 | ARCH_LOW_ADDRESS_LIMIT); | |
008857c1 | 783 | } |