]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/bootmem.c | |
3 | * | |
4 | * Copyright (C) 1999 Ingo Molnar | |
5 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
6 | * | |
7 | * simple boot-time physical memory area allocator and | |
8 | * free memory collector. It's used to deal with reserved | |
9 | * system memory and memory holes as well. | |
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> |
e786e86a | 18 | |
1da177e4 LT |
19 | #include "internal.h" |
20 | ||
21 | /* | |
22 | * Access to this subsystem has to be serialized externally. (this is | |
23 | * true for the boot process anyway) | |
24 | */ | |
25 | unsigned long max_low_pfn; | |
26 | unsigned long min_low_pfn; | |
27 | unsigned long max_pfn; | |
28 | ||
6d46cc6b | 29 | EXPORT_UNUSED_SYMBOL(max_pfn); /* June 2006 */ |
1da177e4 | 30 | |
679bc9fb | 31 | static LIST_HEAD(bdata_list); |
92aa63a5 VG |
32 | #ifdef CONFIG_CRASH_DUMP |
33 | /* | |
34 | * If we have booted due to a crash, max_pfn will be a very low value. We need | |
35 | * to know the amount of memory that the previous kernel used. | |
36 | */ | |
37 | unsigned long saved_max_pfn; | |
38 | #endif | |
39 | ||
1da177e4 | 40 | /* return the number of _pages_ that will be allocated for the boot bitmap */ |
f71bf0ca | 41 | unsigned long __init bootmem_bootmap_pages(unsigned long pages) |
1da177e4 LT |
42 | { |
43 | unsigned long mapsize; | |
44 | ||
45 | mapsize = (pages+7)/8; | |
46 | mapsize = (mapsize + ~PAGE_MASK) & PAGE_MASK; | |
47 | mapsize >>= PAGE_SHIFT; | |
48 | ||
49 | return mapsize; | |
50 | } | |
f71bf0ca | 51 | |
679bc9fb KH |
52 | /* |
53 | * link bdata in order | |
54 | */ | |
69d49e68 | 55 | static void __init link_bootmem(bootmem_data_t *bdata) |
679bc9fb KH |
56 | { |
57 | bootmem_data_t *ent; | |
f71bf0ca | 58 | |
679bc9fb KH |
59 | if (list_empty(&bdata_list)) { |
60 | list_add(&bdata->list, &bdata_list); | |
61 | return; | |
62 | } | |
63 | /* insert in order */ | |
64 | list_for_each_entry(ent, &bdata_list, list) { | |
65 | if (bdata->node_boot_start < ent->node_boot_start) { | |
66 | list_add_tail(&bdata->list, &ent->list); | |
67 | return; | |
68 | } | |
69 | } | |
70 | list_add_tail(&bdata->list, &bdata_list); | |
679bc9fb KH |
71 | } |
72 | ||
bbc7b92e FBH |
73 | /* |
74 | * Given an initialised bdata, it returns the size of the boot bitmap | |
75 | */ | |
76 | static unsigned long __init get_mapsize(bootmem_data_t *bdata) | |
77 | { | |
78 | unsigned long mapsize; | |
79 | unsigned long start = PFN_DOWN(bdata->node_boot_start); | |
80 | unsigned long end = bdata->node_low_pfn; | |
81 | ||
82 | mapsize = ((end - start) + 7) / 8; | |
83 | return ALIGN(mapsize, sizeof(long)); | |
84 | } | |
1da177e4 LT |
85 | |
86 | /* | |
87 | * Called once to set up the allocator itself. | |
88 | */ | |
f71bf0ca | 89 | static unsigned long __init init_bootmem_core(pg_data_t *pgdat, |
1da177e4 LT |
90 | unsigned long mapstart, unsigned long start, unsigned long end) |
91 | { | |
92 | bootmem_data_t *bdata = pgdat->bdata; | |
bbc7b92e | 93 | unsigned long mapsize; |
1da177e4 | 94 | |
bbc7b92e FBH |
95 | bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart)); |
96 | bdata->node_boot_start = PFN_PHYS(start); | |
1da177e4 | 97 | bdata->node_low_pfn = end; |
679bc9fb | 98 | link_bootmem(bdata); |
1da177e4 LT |
99 | |
100 | /* | |
101 | * Initially all pages are reserved - setup_arch() has to | |
102 | * register free RAM areas explicitly. | |
103 | */ | |
bbc7b92e | 104 | mapsize = get_mapsize(bdata); |
1da177e4 LT |
105 | memset(bdata->node_bootmem_map, 0xff, mapsize); |
106 | ||
107 | return mapsize; | |
108 | } | |
109 | ||
110 | /* | |
111 | * Marks a particular physical memory range as unallocatable. Usable RAM | |
112 | * might be used for boot-time allocations - or it might get added | |
113 | * to the free page pool later on. | |
114 | */ | |
bb0923a6 FBH |
115 | static void __init reserve_bootmem_core(bootmem_data_t *bdata, unsigned long addr, |
116 | unsigned long size) | |
1da177e4 | 117 | { |
bbc7b92e | 118 | unsigned long sidx, eidx; |
1da177e4 | 119 | unsigned long i; |
bbc7b92e | 120 | |
1da177e4 LT |
121 | /* |
122 | * round up, partially reserved pages are considered | |
123 | * fully reserved. | |
124 | */ | |
1da177e4 | 125 | BUG_ON(!size); |
bbc7b92e FBH |
126 | BUG_ON(PFN_DOWN(addr) >= bdata->node_low_pfn); |
127 | BUG_ON(PFN_UP(addr + size) > bdata->node_low_pfn); | |
128 | ||
129 | sidx = PFN_DOWN(addr - bdata->node_boot_start); | |
130 | eidx = PFN_UP(addr + size - bdata->node_boot_start); | |
1da177e4 LT |
131 | |
132 | for (i = sidx; i < eidx; i++) | |
133 | if (test_and_set_bit(i, bdata->node_bootmem_map)) { | |
134 | #ifdef CONFIG_DEBUG_BOOTMEM | |
135 | printk("hm, page %08lx reserved twice.\n", i*PAGE_SIZE); | |
136 | #endif | |
137 | } | |
138 | } | |
139 | ||
bb0923a6 FBH |
140 | static void __init free_bootmem_core(bootmem_data_t *bdata, unsigned long addr, |
141 | unsigned long size) | |
1da177e4 | 142 | { |
bbc7b92e | 143 | unsigned long sidx, eidx; |
1da177e4 | 144 | unsigned long i; |
bbc7b92e | 145 | |
1da177e4 LT |
146 | /* |
147 | * round down end of usable mem, partially free pages are | |
148 | * considered reserved. | |
149 | */ | |
1da177e4 | 150 | BUG_ON(!size); |
bbc7b92e | 151 | BUG_ON(PFN_DOWN(addr + size) > bdata->node_low_pfn); |
1da177e4 LT |
152 | |
153 | if (addr < bdata->last_success) | |
154 | bdata->last_success = addr; | |
155 | ||
156 | /* | |
157 | * Round up the beginning of the address. | |
158 | */ | |
bbc7b92e FBH |
159 | sidx = PFN_UP(addr) - PFN_DOWN(bdata->node_boot_start); |
160 | eidx = PFN_DOWN(addr + size - bdata->node_boot_start); | |
1da177e4 LT |
161 | |
162 | for (i = sidx; i < eidx; i++) { | |
163 | if (unlikely(!test_and_clear_bit(i, bdata->node_bootmem_map))) | |
164 | BUG(); | |
165 | } | |
166 | } | |
167 | ||
168 | /* | |
169 | * We 'merge' subsequent allocations to save space. We might 'lose' | |
170 | * some fraction of a page if allocations cannot be satisfied due to | |
171 | * size constraints on boxes where there is physical RAM space | |
172 | * fragmentation - in these cases (mostly large memory boxes) this | |
173 | * is not a problem. | |
174 | * | |
175 | * On low memory boxes we get it right in 100% of the cases. | |
176 | * | |
177 | * alignment has to be a power of 2 value. | |
178 | * | |
179 | * NOTE: This function is _not_ reentrant. | |
180 | */ | |
267b4801 | 181 | void * __init |
1da177e4 | 182 | __alloc_bootmem_core(struct bootmem_data *bdata, unsigned long size, |
281dd25c | 183 | unsigned long align, unsigned long goal, unsigned long limit) |
1da177e4 LT |
184 | { |
185 | unsigned long offset, remaining_size, areasize, preferred; | |
bbc7b92e | 186 | unsigned long i, start = 0, incr, eidx, end_pfn; |
1da177e4 LT |
187 | void *ret; |
188 | ||
f71bf0ca | 189 | if (!size) { |
1da177e4 LT |
190 | printk("__alloc_bootmem_core(): zero-sized request\n"); |
191 | BUG(); | |
192 | } | |
193 | BUG_ON(align & (align-1)); | |
194 | ||
281dd25c YG |
195 | if (limit && bdata->node_boot_start >= limit) |
196 | return NULL; | |
197 | ||
bbc7b92e FBH |
198 | end_pfn = bdata->node_low_pfn; |
199 | limit = PFN_DOWN(limit); | |
281dd25c YG |
200 | if (limit && end_pfn > limit) |
201 | end_pfn = limit; | |
202 | ||
bbc7b92e | 203 | eidx = end_pfn - PFN_DOWN(bdata->node_boot_start); |
1da177e4 | 204 | offset = 0; |
bbc7b92e FBH |
205 | if (align && (bdata->node_boot_start & (align - 1UL)) != 0) |
206 | offset = align - (bdata->node_boot_start & (align - 1UL)); | |
207 | offset = PFN_DOWN(offset); | |
1da177e4 LT |
208 | |
209 | /* | |
210 | * We try to allocate bootmem pages above 'goal' | |
211 | * first, then we try to allocate lower pages. | |
212 | */ | |
bbc7b92e | 213 | if (goal && goal >= bdata->node_boot_start && PFN_DOWN(goal) < end_pfn) { |
1da177e4 LT |
214 | preferred = goal - bdata->node_boot_start; |
215 | ||
216 | if (bdata->last_success >= preferred) | |
281dd25c YG |
217 | if (!limit || (limit && limit > bdata->last_success)) |
218 | preferred = bdata->last_success; | |
1da177e4 LT |
219 | } else |
220 | preferred = 0; | |
221 | ||
bbc7b92e FBH |
222 | preferred = PFN_DOWN(ALIGN(preferred, align)) + offset; |
223 | areasize = (size + PAGE_SIZE-1) / PAGE_SIZE; | |
1da177e4 LT |
224 | incr = align >> PAGE_SHIFT ? : 1; |
225 | ||
226 | restart_scan: | |
227 | for (i = preferred; i < eidx; i += incr) { | |
228 | unsigned long j; | |
229 | i = find_next_zero_bit(bdata->node_bootmem_map, eidx, i); | |
230 | i = ALIGN(i, incr); | |
66d43e98 HM |
231 | if (i >= eidx) |
232 | break; | |
1da177e4 LT |
233 | if (test_bit(i, bdata->node_bootmem_map)) |
234 | continue; | |
235 | for (j = i + 1; j < i + areasize; ++j) { | |
236 | if (j >= eidx) | |
237 | goto fail_block; | |
f71bf0ca | 238 | if (test_bit(j, bdata->node_bootmem_map)) |
1da177e4 LT |
239 | goto fail_block; |
240 | } | |
241 | start = i; | |
242 | goto found; | |
243 | fail_block: | |
244 | i = ALIGN(j, incr); | |
245 | } | |
246 | ||
247 | if (preferred > offset) { | |
248 | preferred = offset; | |
249 | goto restart_scan; | |
250 | } | |
251 | return NULL; | |
252 | ||
253 | found: | |
bbc7b92e | 254 | bdata->last_success = PFN_PHYS(start); |
1da177e4 LT |
255 | BUG_ON(start >= eidx); |
256 | ||
257 | /* | |
258 | * Is the next page of the previous allocation-end the start | |
259 | * of this allocation's buffer? If yes then we can 'merge' | |
260 | * the previous partial page with this allocation. | |
261 | */ | |
262 | if (align < PAGE_SIZE && | |
263 | bdata->last_offset && bdata->last_pos+1 == start) { | |
8c0e33c1 | 264 | offset = ALIGN(bdata->last_offset, align); |
1da177e4 | 265 | BUG_ON(offset > PAGE_SIZE); |
f71bf0ca | 266 | remaining_size = PAGE_SIZE - offset; |
1da177e4 LT |
267 | if (size < remaining_size) { |
268 | areasize = 0; | |
269 | /* last_pos unchanged */ | |
f71bf0ca FBH |
270 | bdata->last_offset = offset + size; |
271 | ret = phys_to_virt(bdata->last_pos * PAGE_SIZE + | |
272 | offset + | |
273 | bdata->node_boot_start); | |
1da177e4 LT |
274 | } else { |
275 | remaining_size = size - remaining_size; | |
f71bf0ca FBH |
276 | areasize = (remaining_size + PAGE_SIZE-1) / PAGE_SIZE; |
277 | ret = phys_to_virt(bdata->last_pos * PAGE_SIZE + | |
278 | offset + | |
279 | bdata->node_boot_start); | |
280 | bdata->last_pos = start + areasize - 1; | |
1da177e4 LT |
281 | bdata->last_offset = remaining_size; |
282 | } | |
283 | bdata->last_offset &= ~PAGE_MASK; | |
284 | } else { | |
285 | bdata->last_pos = start + areasize - 1; | |
286 | bdata->last_offset = size & ~PAGE_MASK; | |
287 | ret = phys_to_virt(start * PAGE_SIZE + bdata->node_boot_start); | |
288 | } | |
289 | ||
290 | /* | |
291 | * Reserve the area now: | |
292 | */ | |
f71bf0ca | 293 | for (i = start; i < start + areasize; i++) |
1da177e4 LT |
294 | if (unlikely(test_and_set_bit(i, bdata->node_bootmem_map))) |
295 | BUG(); | |
296 | memset(ret, 0, size); | |
297 | return ret; | |
298 | } | |
299 | ||
300 | static unsigned long __init free_all_bootmem_core(pg_data_t *pgdat) | |
301 | { | |
302 | struct page *page; | |
d41dee36 | 303 | unsigned long pfn; |
1da177e4 LT |
304 | bootmem_data_t *bdata = pgdat->bdata; |
305 | unsigned long i, count, total = 0; | |
306 | unsigned long idx; | |
307 | unsigned long *map; | |
308 | int gofast = 0; | |
309 | ||
310 | BUG_ON(!bdata->node_bootmem_map); | |
311 | ||
312 | count = 0; | |
313 | /* first extant page of the node */ | |
bbc7b92e FBH |
314 | pfn = PFN_DOWN(bdata->node_boot_start); |
315 | idx = bdata->node_low_pfn - pfn; | |
1da177e4 LT |
316 | map = bdata->node_bootmem_map; |
317 | /* Check physaddr is O(LOG2(BITS_PER_LONG)) page aligned */ | |
318 | if (bdata->node_boot_start == 0 || | |
319 | ffs(bdata->node_boot_start) - PAGE_SHIFT > ffs(BITS_PER_LONG)) | |
320 | gofast = 1; | |
321 | for (i = 0; i < idx; ) { | |
322 | unsigned long v = ~map[i / BITS_PER_LONG]; | |
d41dee36 | 323 | |
1da177e4 | 324 | if (gofast && v == ~0UL) { |
a226f6c8 | 325 | int order; |
1da177e4 | 326 | |
d41dee36 | 327 | page = pfn_to_page(pfn); |
1da177e4 | 328 | count += BITS_PER_LONG; |
1da177e4 | 329 | order = ffs(BITS_PER_LONG) - 1; |
a226f6c8 | 330 | __free_pages_bootmem(page, order); |
1da177e4 LT |
331 | i += BITS_PER_LONG; |
332 | page += BITS_PER_LONG; | |
333 | } else if (v) { | |
334 | unsigned long m; | |
d41dee36 AW |
335 | |
336 | page = pfn_to_page(pfn); | |
1da177e4 LT |
337 | for (m = 1; m && i < idx; m<<=1, page++, i++) { |
338 | if (v & m) { | |
339 | count++; | |
a226f6c8 | 340 | __free_pages_bootmem(page, 0); |
1da177e4 LT |
341 | } |
342 | } | |
343 | } else { | |
f71bf0ca | 344 | i += BITS_PER_LONG; |
1da177e4 | 345 | } |
d41dee36 | 346 | pfn += BITS_PER_LONG; |
1da177e4 LT |
347 | } |
348 | total += count; | |
349 | ||
350 | /* | |
351 | * Now free the allocator bitmap itself, it's not | |
352 | * needed anymore: | |
353 | */ | |
354 | page = virt_to_page(bdata->node_bootmem_map); | |
355 | count = 0; | |
bbc7b92e FBH |
356 | idx = (get_mapsize(bdata) + PAGE_SIZE-1) >> PAGE_SHIFT; |
357 | for (i = 0; i < idx; i++, page++) { | |
a226f6c8 | 358 | __free_pages_bootmem(page, 0); |
bbc7b92e | 359 | count++; |
1da177e4 LT |
360 | } |
361 | total += count; | |
362 | bdata->node_bootmem_map = NULL; | |
363 | ||
364 | return total; | |
365 | } | |
366 | ||
f71bf0ca | 367 | unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn, |
bb0923a6 | 368 | unsigned long startpfn, unsigned long endpfn) |
1da177e4 | 369 | { |
f71bf0ca | 370 | return init_bootmem_core(pgdat, freepfn, startpfn, endpfn); |
1da177e4 LT |
371 | } |
372 | ||
f71bf0ca FBH |
373 | void __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
374 | unsigned long size) | |
1da177e4 LT |
375 | { |
376 | reserve_bootmem_core(pgdat->bdata, physaddr, size); | |
377 | } | |
378 | ||
f71bf0ca FBH |
379 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
380 | unsigned long size) | |
1da177e4 LT |
381 | { |
382 | free_bootmem_core(pgdat->bdata, physaddr, size); | |
383 | } | |
384 | ||
f71bf0ca | 385 | unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) |
1da177e4 | 386 | { |
f71bf0ca | 387 | return free_all_bootmem_core(pgdat); |
1da177e4 LT |
388 | } |
389 | ||
f71bf0ca | 390 | unsigned long __init init_bootmem(unsigned long start, unsigned long pages) |
1da177e4 LT |
391 | { |
392 | max_low_pfn = pages; | |
393 | min_low_pfn = start; | |
f71bf0ca | 394 | return init_bootmem_core(NODE_DATA(0), start, 0, pages); |
1da177e4 LT |
395 | } |
396 | ||
397 | #ifndef CONFIG_HAVE_ARCH_BOOTMEM_NODE | |
f71bf0ca | 398 | void __init reserve_bootmem(unsigned long addr, unsigned long size) |
1da177e4 LT |
399 | { |
400 | reserve_bootmem_core(NODE_DATA(0)->bdata, addr, size); | |
401 | } | |
402 | #endif /* !CONFIG_HAVE_ARCH_BOOTMEM_NODE */ | |
403 | ||
f71bf0ca | 404 | void __init free_bootmem(unsigned long addr, unsigned long size) |
1da177e4 LT |
405 | { |
406 | free_bootmem_core(NODE_DATA(0)->bdata, addr, size); | |
407 | } | |
408 | ||
f71bf0ca | 409 | unsigned long __init free_all_bootmem(void) |
1da177e4 | 410 | { |
f71bf0ca | 411 | return free_all_bootmem_core(NODE_DATA(0)); |
1da177e4 LT |
412 | } |
413 | ||
bb0923a6 FBH |
414 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
415 | unsigned long goal) | |
1da177e4 | 416 | { |
679bc9fb | 417 | bootmem_data_t *bdata; |
1da177e4 LT |
418 | void *ptr; |
419 | ||
f71bf0ca FBH |
420 | list_for_each_entry(bdata, &bdata_list, list) { |
421 | ptr = __alloc_bootmem_core(bdata, size, align, goal, 0); | |
422 | if (ptr) | |
423 | return ptr; | |
424 | } | |
a8062231 AK |
425 | return NULL; |
426 | } | |
1da177e4 | 427 | |
bb0923a6 FBH |
428 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
429 | unsigned long goal) | |
a8062231 AK |
430 | { |
431 | void *mem = __alloc_bootmem_nopanic(size,align,goal); | |
f71bf0ca | 432 | |
a8062231 AK |
433 | if (mem) |
434 | return mem; | |
1da177e4 LT |
435 | /* |
436 | * Whoops, we cannot satisfy the allocation request. | |
437 | */ | |
438 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); | |
439 | panic("Out of memory"); | |
440 | return NULL; | |
441 | } | |
442 | ||
281dd25c | 443 | |
bb0923a6 FBH |
444 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
445 | unsigned long align, unsigned long goal) | |
1da177e4 LT |
446 | { |
447 | void *ptr; | |
448 | ||
008857c1 | 449 | ptr = __alloc_bootmem_core(pgdat->bdata, size, align, goal, 0); |
1da177e4 | 450 | if (ptr) |
f71bf0ca | 451 | return ptr; |
1da177e4 | 452 | |
008857c1 | 453 | return __alloc_bootmem(size, align, goal); |
1da177e4 LT |
454 | } |
455 | ||
008857c1 RT |
456 | #define LOW32LIMIT 0xffffffff |
457 | ||
bb0923a6 FBH |
458 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
459 | unsigned long goal) | |
008857c1 | 460 | { |
679bc9fb | 461 | bootmem_data_t *bdata; |
008857c1 RT |
462 | void *ptr; |
463 | ||
f71bf0ca FBH |
464 | list_for_each_entry(bdata, &bdata_list, list) { |
465 | ptr = __alloc_bootmem_core(bdata, size, align, goal, LOW32LIMIT); | |
466 | if (ptr) | |
467 | return ptr; | |
468 | } | |
008857c1 RT |
469 | |
470 | /* | |
471 | * Whoops, we cannot satisfy the allocation request. | |
472 | */ | |
473 | printk(KERN_ALERT "low bootmem alloc of %lu bytes failed!\n", size); | |
474 | panic("Out of low memory"); | |
475 | return NULL; | |
476 | } | |
477 | ||
478 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, | |
479 | unsigned long align, unsigned long goal) | |
480 | { | |
481 | return __alloc_bootmem_core(pgdat->bdata, size, align, goal, LOW32LIMIT); | |
482 | } |