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