]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
57cfc29e | 2 | * bootmem - A boot-time physical memory allocator and configurator |
1da177e4 LT |
3 | * |
4 | * Copyright (C) 1999 Ingo Molnar | |
57cfc29e JW |
5 | * 1999 Kanoj Sarcar, SGI |
6 | * 2008 Johannes Weiner | |
1da177e4 | 7 | * |
57cfc29e JW |
8 | * Access to this subsystem has to be serialized externally (which is true |
9 | * for the boot process anyway). | |
1da177e4 | 10 | */ |
1da177e4 | 11 | #include <linux/init.h> |
bbc7b92e | 12 | #include <linux/pfn.h> |
5a0e3ad6 | 13 | #include <linux/slab.h> |
1da177e4 | 14 | #include <linux/bootmem.h> |
1da177e4 | 15 | #include <linux/module.h> |
ec3a354b | 16 | #include <linux/kmemleak.h> |
08677214 | 17 | #include <linux/range.h> |
e786e86a FBH |
18 | |
19 | #include <asm/bug.h> | |
1da177e4 | 20 | #include <asm/io.h> |
dfd54cbc | 21 | #include <asm/processor.h> |
e786e86a | 22 | |
1da177e4 LT |
23 | #include "internal.h" |
24 | ||
1da177e4 LT |
25 | unsigned long max_low_pfn; |
26 | unsigned long min_low_pfn; | |
27 | unsigned long max_pfn; | |
28 | ||
92aa63a5 VG |
29 | #ifdef CONFIG_CRASH_DUMP |
30 | /* | |
31 | * If we have booted due to a crash, max_pfn will be a very low value. We need | |
32 | * to know the amount of memory that the previous kernel used. | |
33 | */ | |
34 | unsigned long saved_max_pfn; | |
35 | #endif | |
36 | ||
08677214 | 37 | #ifndef CONFIG_NO_BOOTMEM |
b61bfa3c JW |
38 | bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata; |
39 | ||
636cc40c JW |
40 | static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list); |
41 | ||
2e5237da JW |
42 | static int bootmem_debug; |
43 | ||
44 | static int __init bootmem_debug_setup(char *buf) | |
45 | { | |
46 | bootmem_debug = 1; | |
47 | return 0; | |
48 | } | |
49 | early_param("bootmem_debug", bootmem_debug_setup); | |
50 | ||
51 | #define bdebug(fmt, args...) ({ \ | |
52 | if (unlikely(bootmem_debug)) \ | |
53 | printk(KERN_INFO \ | |
54 | "bootmem::%s " fmt, \ | |
80a914dc | 55 | __func__, ## args); \ |
2e5237da JW |
56 | }) |
57 | ||
df049a5f | 58 | static unsigned long __init bootmap_bytes(unsigned long pages) |
223e8dc9 | 59 | { |
df049a5f | 60 | unsigned long bytes = (pages + 7) / 8; |
223e8dc9 | 61 | |
df049a5f | 62 | return ALIGN(bytes, sizeof(long)); |
223e8dc9 JW |
63 | } |
64 | ||
a66fd7da JW |
65 | /** |
66 | * bootmem_bootmap_pages - calculate bitmap size in pages | |
67 | * @pages: number of pages the bitmap has to represent | |
68 | */ | |
f71bf0ca | 69 | unsigned long __init bootmem_bootmap_pages(unsigned long pages) |
1da177e4 | 70 | { |
df049a5f | 71 | unsigned long bytes = bootmap_bytes(pages); |
1da177e4 | 72 | |
df049a5f | 73 | return PAGE_ALIGN(bytes) >> PAGE_SHIFT; |
1da177e4 | 74 | } |
f71bf0ca | 75 | |
679bc9fb KH |
76 | /* |
77 | * link bdata in order | |
78 | */ | |
69d49e68 | 79 | static void __init link_bootmem(bootmem_data_t *bdata) |
679bc9fb | 80 | { |
636cc40c | 81 | struct list_head *iter; |
f71bf0ca | 82 | |
636cc40c JW |
83 | list_for_each(iter, &bdata_list) { |
84 | bootmem_data_t *ent; | |
85 | ||
86 | ent = list_entry(iter, bootmem_data_t, list); | |
3560e249 | 87 | if (bdata->node_min_pfn < ent->node_min_pfn) |
636cc40c | 88 | break; |
679bc9fb | 89 | } |
636cc40c | 90 | list_add_tail(&bdata->list, iter); |
679bc9fb KH |
91 | } |
92 | ||
1da177e4 LT |
93 | /* |
94 | * Called once to set up the allocator itself. | |
95 | */ | |
8ae04463 | 96 | static unsigned long __init init_bootmem_core(bootmem_data_t *bdata, |
1da177e4 LT |
97 | unsigned long mapstart, unsigned long start, unsigned long end) |
98 | { | |
bbc7b92e | 99 | unsigned long mapsize; |
1da177e4 | 100 | |
2dbb51c4 | 101 | mminit_validate_memmodel_limits(&start, &end); |
bbc7b92e | 102 | bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart)); |
3560e249 | 103 | bdata->node_min_pfn = start; |
1da177e4 | 104 | bdata->node_low_pfn = end; |
679bc9fb | 105 | link_bootmem(bdata); |
1da177e4 LT |
106 | |
107 | /* | |
108 | * Initially all pages are reserved - setup_arch() has to | |
109 | * register free RAM areas explicitly. | |
110 | */ | |
df049a5f | 111 | mapsize = bootmap_bytes(end - start); |
1da177e4 LT |
112 | memset(bdata->node_bootmem_map, 0xff, mapsize); |
113 | ||
2e5237da JW |
114 | bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n", |
115 | bdata - bootmem_node_data, start, mapstart, end, mapsize); | |
116 | ||
1da177e4 LT |
117 | return mapsize; |
118 | } | |
119 | ||
a66fd7da JW |
120 | /** |
121 | * init_bootmem_node - register a node as boot memory | |
122 | * @pgdat: node to register | |
123 | * @freepfn: pfn where the bitmap for this node is to be placed | |
124 | * @startpfn: first pfn on the node | |
125 | * @endpfn: first pfn after the node | |
126 | * | |
127 | * Returns the number of bytes needed to hold the bitmap for this node. | |
128 | */ | |
223e8dc9 JW |
129 | unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn, |
130 | unsigned long startpfn, unsigned long endpfn) | |
131 | { | |
132 | return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn); | |
133 | } | |
134 | ||
a66fd7da JW |
135 | /** |
136 | * init_bootmem - register boot memory | |
137 | * @start: pfn where the bitmap is to be placed | |
138 | * @pages: number of available physical pages | |
139 | * | |
140 | * Returns the number of bytes needed to hold the bitmap. | |
141 | */ | |
223e8dc9 JW |
142 | unsigned long __init init_bootmem(unsigned long start, unsigned long pages) |
143 | { | |
144 | max_low_pfn = pages; | |
145 | min_low_pfn = start; | |
146 | return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages); | |
147 | } | |
08677214 | 148 | #endif |
9f993ac3 FT |
149 | /* |
150 | * free_bootmem_late - free bootmem pages directly to page allocator | |
151 | * @addr: starting address of the range | |
152 | * @size: size of the range in bytes | |
153 | * | |
154 | * This is only useful when the bootmem allocator has already been torn | |
155 | * down, but we are still initializing the system. Pages are given directly | |
156 | * to the page allocator, no bootmem metadata is updated because it is gone. | |
157 | */ | |
158 | void __init free_bootmem_late(unsigned long addr, unsigned long size) | |
159 | { | |
160 | unsigned long cursor, end; | |
161 | ||
162 | kmemleak_free_part(__va(addr), size); | |
163 | ||
164 | cursor = PFN_UP(addr); | |
165 | end = PFN_DOWN(addr + size); | |
166 | ||
167 | for (; cursor < end; cursor++) { | |
168 | __free_pages_bootmem(pfn_to_page(cursor), 0); | |
169 | totalram_pages++; | |
170 | } | |
171 | } | |
172 | ||
08677214 YL |
173 | #ifdef CONFIG_NO_BOOTMEM |
174 | static void __init __free_pages_memory(unsigned long start, unsigned long end) | |
175 | { | |
176 | int i; | |
177 | unsigned long start_aligned, end_aligned; | |
178 | int order = ilog2(BITS_PER_LONG); | |
179 | ||
180 | start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1); | |
181 | end_aligned = end & ~(BITS_PER_LONG - 1); | |
182 | ||
183 | if (end_aligned <= start_aligned) { | |
08677214 YL |
184 | for (i = start; i < end; i++) |
185 | __free_pages_bootmem(pfn_to_page(i), 0); | |
186 | ||
187 | return; | |
188 | } | |
189 | ||
08677214 YL |
190 | for (i = start; i < start_aligned; i++) |
191 | __free_pages_bootmem(pfn_to_page(i), 0); | |
192 | ||
193 | for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG) | |
194 | __free_pages_bootmem(pfn_to_page(i), order); | |
195 | ||
196 | for (i = end_aligned; i < end; i++) | |
197 | __free_pages_bootmem(pfn_to_page(i), 0); | |
198 | } | |
199 | ||
200 | unsigned long __init free_all_memory_core_early(int nodeid) | |
201 | { | |
202 | int i; | |
203 | u64 start, end; | |
204 | unsigned long count = 0; | |
205 | struct range *range = NULL; | |
206 | int nr_range; | |
207 | ||
208 | nr_range = get_free_all_memory_range(&range, nodeid); | |
209 | ||
210 | for (i = 0; i < nr_range; i++) { | |
211 | start = range[i].start; | |
212 | end = range[i].end; | |
213 | count += end - start; | |
214 | __free_pages_memory(start, end); | |
215 | } | |
216 | ||
217 | return count; | |
218 | } | |
219 | #else | |
223e8dc9 JW |
220 | static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) |
221 | { | |
41546c17 | 222 | int aligned; |
223e8dc9 | 223 | struct page *page; |
41546c17 JW |
224 | unsigned long start, end, pages, count = 0; |
225 | ||
226 | if (!bdata->node_bootmem_map) | |
227 | return 0; | |
228 | ||
3560e249 | 229 | start = bdata->node_min_pfn; |
41546c17 JW |
230 | end = bdata->node_low_pfn; |
231 | ||
223e8dc9 | 232 | /* |
41546c17 JW |
233 | * If the start is aligned to the machines wordsize, we might |
234 | * be able to free pages in bulks of that order. | |
223e8dc9 | 235 | */ |
41546c17 | 236 | aligned = !(start & (BITS_PER_LONG - 1)); |
223e8dc9 | 237 | |
41546c17 JW |
238 | bdebug("nid=%td start=%lx end=%lx aligned=%d\n", |
239 | bdata - bootmem_node_data, start, end, aligned); | |
223e8dc9 | 240 | |
41546c17 JW |
241 | while (start < end) { |
242 | unsigned long *map, idx, vec; | |
223e8dc9 | 243 | |
41546c17 | 244 | map = bdata->node_bootmem_map; |
3560e249 | 245 | idx = start - bdata->node_min_pfn; |
41546c17 JW |
246 | vec = ~map[idx / BITS_PER_LONG]; |
247 | ||
248 | if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) { | |
249 | int order = ilog2(BITS_PER_LONG); | |
250 | ||
251 | __free_pages_bootmem(pfn_to_page(start), order); | |
223e8dc9 | 252 | count += BITS_PER_LONG; |
41546c17 JW |
253 | } else { |
254 | unsigned long off = 0; | |
255 | ||
256 | while (vec && off < BITS_PER_LONG) { | |
257 | if (vec & 1) { | |
258 | page = pfn_to_page(start + off); | |
223e8dc9 | 259 | __free_pages_bootmem(page, 0); |
41546c17 | 260 | count++; |
223e8dc9 | 261 | } |
41546c17 JW |
262 | vec >>= 1; |
263 | off++; | |
223e8dc9 | 264 | } |
223e8dc9 | 265 | } |
41546c17 | 266 | start += BITS_PER_LONG; |
223e8dc9 JW |
267 | } |
268 | ||
223e8dc9 | 269 | page = virt_to_page(bdata->node_bootmem_map); |
3560e249 | 270 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
41546c17 JW |
271 | pages = bootmem_bootmap_pages(pages); |
272 | count += pages; | |
273 | while (pages--) | |
274 | __free_pages_bootmem(page++, 0); | |
223e8dc9 | 275 | |
2e5237da JW |
276 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
277 | ||
223e8dc9 JW |
278 | return count; |
279 | } | |
08677214 | 280 | #endif |
223e8dc9 | 281 | |
a66fd7da JW |
282 | /** |
283 | * free_all_bootmem_node - release a node's free pages to the buddy allocator | |
284 | * @pgdat: node to be released | |
285 | * | |
286 | * Returns the number of pages actually released. | |
287 | */ | |
223e8dc9 JW |
288 | unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) |
289 | { | |
290 | register_page_bootmem_info_node(pgdat); | |
08677214 YL |
291 | #ifdef CONFIG_NO_BOOTMEM |
292 | /* free_all_memory_core_early(MAX_NUMNODES) will be called later */ | |
293 | return 0; | |
294 | #else | |
223e8dc9 | 295 | return free_all_bootmem_core(pgdat->bdata); |
08677214 | 296 | #endif |
223e8dc9 JW |
297 | } |
298 | ||
a66fd7da JW |
299 | /** |
300 | * free_all_bootmem - release free pages to the buddy allocator | |
301 | * | |
302 | * Returns the number of pages actually released. | |
303 | */ | |
223e8dc9 JW |
304 | unsigned long __init free_all_bootmem(void) |
305 | { | |
08677214 YL |
306 | #ifdef CONFIG_NO_BOOTMEM |
307 | return free_all_memory_core_early(NODE_DATA(0)->node_id); | |
308 | #else | |
223e8dc9 | 309 | return free_all_bootmem_core(NODE_DATA(0)->bdata); |
08677214 | 310 | #endif |
223e8dc9 JW |
311 | } |
312 | ||
08677214 | 313 | #ifndef CONFIG_NO_BOOTMEM |
d747fa4b JW |
314 | static void __init __free(bootmem_data_t *bdata, |
315 | unsigned long sidx, unsigned long eidx) | |
316 | { | |
317 | unsigned long idx; | |
318 | ||
319 | bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data, | |
3560e249 JW |
320 | sidx + bdata->node_min_pfn, |
321 | eidx + bdata->node_min_pfn); | |
d747fa4b | 322 | |
e2bf3cae JW |
323 | if (bdata->hint_idx > sidx) |
324 | bdata->hint_idx = sidx; | |
325 | ||
d747fa4b JW |
326 | for (idx = sidx; idx < eidx; idx++) |
327 | if (!test_and_clear_bit(idx, bdata->node_bootmem_map)) | |
328 | BUG(); | |
329 | } | |
330 | ||
331 | static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx, | |
332 | unsigned long eidx, int flags) | |
333 | { | |
334 | unsigned long idx; | |
335 | int exclusive = flags & BOOTMEM_EXCLUSIVE; | |
336 | ||
337 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", | |
338 | bdata - bootmem_node_data, | |
3560e249 JW |
339 | sidx + bdata->node_min_pfn, |
340 | eidx + bdata->node_min_pfn, | |
d747fa4b JW |
341 | flags); |
342 | ||
343 | for (idx = sidx; idx < eidx; idx++) | |
344 | if (test_and_set_bit(idx, bdata->node_bootmem_map)) { | |
345 | if (exclusive) { | |
346 | __free(bdata, sidx, idx); | |
347 | return -EBUSY; | |
348 | } | |
349 | bdebug("silent double reserve of PFN %lx\n", | |
3560e249 | 350 | idx + bdata->node_min_pfn); |
d747fa4b JW |
351 | } |
352 | return 0; | |
353 | } | |
354 | ||
e2bf3cae JW |
355 | static int __init mark_bootmem_node(bootmem_data_t *bdata, |
356 | unsigned long start, unsigned long end, | |
357 | int reserve, int flags) | |
223e8dc9 JW |
358 | { |
359 | unsigned long sidx, eidx; | |
223e8dc9 | 360 | |
e2bf3cae JW |
361 | bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n", |
362 | bdata - bootmem_node_data, start, end, reserve, flags); | |
223e8dc9 | 363 | |
3560e249 | 364 | BUG_ON(start < bdata->node_min_pfn); |
e2bf3cae | 365 | BUG_ON(end > bdata->node_low_pfn); |
223e8dc9 | 366 | |
3560e249 JW |
367 | sidx = start - bdata->node_min_pfn; |
368 | eidx = end - bdata->node_min_pfn; | |
223e8dc9 | 369 | |
e2bf3cae JW |
370 | if (reserve) |
371 | return __reserve(bdata, sidx, eidx, flags); | |
223e8dc9 | 372 | else |
e2bf3cae JW |
373 | __free(bdata, sidx, eidx); |
374 | return 0; | |
375 | } | |
376 | ||
377 | static int __init mark_bootmem(unsigned long start, unsigned long end, | |
378 | int reserve, int flags) | |
379 | { | |
380 | unsigned long pos; | |
381 | bootmem_data_t *bdata; | |
382 | ||
383 | pos = start; | |
384 | list_for_each_entry(bdata, &bdata_list, list) { | |
385 | int err; | |
386 | unsigned long max; | |
387 | ||
3560e249 JW |
388 | if (pos < bdata->node_min_pfn || |
389 | pos >= bdata->node_low_pfn) { | |
e2bf3cae JW |
390 | BUG_ON(pos != start); |
391 | continue; | |
392 | } | |
393 | ||
394 | max = min(bdata->node_low_pfn, end); | |
223e8dc9 | 395 | |
e2bf3cae JW |
396 | err = mark_bootmem_node(bdata, pos, max, reserve, flags); |
397 | if (reserve && err) { | |
398 | mark_bootmem(start, pos, 0, 0); | |
399 | return err; | |
400 | } | |
223e8dc9 | 401 | |
e2bf3cae JW |
402 | if (max == end) |
403 | return 0; | |
404 | pos = bdata->node_low_pfn; | |
405 | } | |
406 | BUG(); | |
223e8dc9 | 407 | } |
08677214 | 408 | #endif |
223e8dc9 | 409 | |
a66fd7da JW |
410 | /** |
411 | * free_bootmem_node - mark a page range as usable | |
412 | * @pgdat: node the range resides on | |
413 | * @physaddr: starting address of the range | |
414 | * @size: size of the range in bytes | |
415 | * | |
416 | * Partial pages will be considered reserved and left as they are. | |
417 | * | |
e2bf3cae | 418 | * The range must reside completely on the specified node. |
a66fd7da | 419 | */ |
223e8dc9 JW |
420 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
421 | unsigned long size) | |
422 | { | |
08677214 YL |
423 | #ifdef CONFIG_NO_BOOTMEM |
424 | free_early(physaddr, physaddr + size); | |
08677214 | 425 | #else |
e2bf3cae JW |
426 | unsigned long start, end; |
427 | ||
ec3a354b CM |
428 | kmemleak_free_part(__va(physaddr), size); |
429 | ||
e2bf3cae JW |
430 | start = PFN_UP(physaddr); |
431 | end = PFN_DOWN(physaddr + size); | |
432 | ||
433 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); | |
08677214 | 434 | #endif |
223e8dc9 JW |
435 | } |
436 | ||
a66fd7da JW |
437 | /** |
438 | * free_bootmem - mark a page range as usable | |
439 | * @addr: starting address of the range | |
440 | * @size: size of the range in bytes | |
441 | * | |
442 | * Partial pages will be considered reserved and left as they are. | |
443 | * | |
e2bf3cae | 444 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 445 | */ |
223e8dc9 JW |
446 | void __init free_bootmem(unsigned long addr, unsigned long size) |
447 | { | |
08677214 YL |
448 | #ifdef CONFIG_NO_BOOTMEM |
449 | free_early(addr, addr + size); | |
08677214 | 450 | #else |
e2bf3cae | 451 | unsigned long start, end; |
a5645a61 | 452 | |
ec3a354b CM |
453 | kmemleak_free_part(__va(addr), size); |
454 | ||
e2bf3cae JW |
455 | start = PFN_UP(addr); |
456 | end = PFN_DOWN(addr + size); | |
1da177e4 | 457 | |
e2bf3cae | 458 | mark_bootmem(start, end, 0, 0); |
08677214 | 459 | #endif |
1da177e4 LT |
460 | } |
461 | ||
a66fd7da JW |
462 | /** |
463 | * reserve_bootmem_node - mark a page range as reserved | |
464 | * @pgdat: node the range resides on | |
465 | * @physaddr: starting address of the range | |
466 | * @size: size of the range in bytes | |
467 | * @flags: reservation flags (see linux/bootmem.h) | |
468 | * | |
469 | * Partial pages will be reserved. | |
470 | * | |
e2bf3cae | 471 | * The range must reside completely on the specified node. |
a66fd7da | 472 | */ |
223e8dc9 JW |
473 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
474 | unsigned long size, int flags) | |
1da177e4 | 475 | { |
08677214 YL |
476 | #ifdef CONFIG_NO_BOOTMEM |
477 | panic("no bootmem"); | |
478 | return 0; | |
479 | #else | |
e2bf3cae | 480 | unsigned long start, end; |
1da177e4 | 481 | |
e2bf3cae JW |
482 | start = PFN_DOWN(physaddr); |
483 | end = PFN_UP(physaddr + size); | |
484 | ||
485 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); | |
08677214 | 486 | #endif |
223e8dc9 | 487 | } |
5a982cbc | 488 | |
a66fd7da JW |
489 | /** |
490 | * reserve_bootmem - mark a page range as usable | |
491 | * @addr: starting address of the range | |
492 | * @size: size of the range in bytes | |
493 | * @flags: reservation flags (see linux/bootmem.h) | |
494 | * | |
495 | * Partial pages will be reserved. | |
496 | * | |
e2bf3cae | 497 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 498 | */ |
223e8dc9 JW |
499 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
500 | int flags) | |
501 | { | |
08677214 YL |
502 | #ifdef CONFIG_NO_BOOTMEM |
503 | panic("no bootmem"); | |
504 | return 0; | |
505 | #else | |
e2bf3cae | 506 | unsigned long start, end; |
1da177e4 | 507 | |
e2bf3cae JW |
508 | start = PFN_DOWN(addr); |
509 | end = PFN_UP(addr + size); | |
223e8dc9 | 510 | |
e2bf3cae | 511 | return mark_bootmem(start, end, 1, flags); |
08677214 | 512 | #endif |
1da177e4 LT |
513 | } |
514 | ||
08677214 | 515 | #ifndef CONFIG_NO_BOOTMEM |
8aa043d7 JB |
516 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
517 | unsigned long idx, unsigned long step) | |
481ebd0d JW |
518 | { |
519 | unsigned long base = bdata->node_min_pfn; | |
520 | ||
521 | /* | |
522 | * Align the index with respect to the node start so that the | |
523 | * combination of both satisfies the requested alignment. | |
524 | */ | |
525 | ||
526 | return ALIGN(base + idx, step) - base; | |
527 | } | |
528 | ||
8aa043d7 JB |
529 | static unsigned long __init align_off(struct bootmem_data *bdata, |
530 | unsigned long off, unsigned long align) | |
481ebd0d JW |
531 | { |
532 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); | |
533 | ||
534 | /* Same as align_idx for byte offsets */ | |
535 | ||
536 | return ALIGN(base + off, align) - base; | |
537 | } | |
538 | ||
d0c4f570 TH |
539 | static void * __init alloc_bootmem_core(struct bootmem_data *bdata, |
540 | unsigned long size, unsigned long align, | |
541 | unsigned long goal, unsigned long limit) | |
1da177e4 | 542 | { |
0f3caba2 | 543 | unsigned long fallback = 0; |
5f2809e6 JW |
544 | unsigned long min, max, start, sidx, midx, step; |
545 | ||
594fe1a0 JW |
546 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
547 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, | |
548 | align, goal, limit); | |
549 | ||
5f2809e6 JW |
550 | BUG_ON(!size); |
551 | BUG_ON(align & (align - 1)); | |
552 | BUG_ON(limit && goal + size > limit); | |
1da177e4 | 553 | |
7c309a64 CK |
554 | if (!bdata->node_bootmem_map) |
555 | return NULL; | |
556 | ||
3560e249 | 557 | min = bdata->node_min_pfn; |
5f2809e6 | 558 | max = bdata->node_low_pfn; |
9a2dc04c | 559 | |
5f2809e6 JW |
560 | goal >>= PAGE_SHIFT; |
561 | limit >>= PAGE_SHIFT; | |
562 | ||
563 | if (limit && max > limit) | |
564 | max = limit; | |
565 | if (max <= min) | |
9a2dc04c YL |
566 | return NULL; |
567 | ||
5f2809e6 | 568 | step = max(align >> PAGE_SHIFT, 1UL); |
281dd25c | 569 | |
5f2809e6 JW |
570 | if (goal && min < goal && goal < max) |
571 | start = ALIGN(goal, step); | |
572 | else | |
573 | start = ALIGN(min, step); | |
1da177e4 | 574 | |
481ebd0d | 575 | sidx = start - bdata->node_min_pfn; |
3560e249 | 576 | midx = max - bdata->node_min_pfn; |
1da177e4 | 577 | |
5f2809e6 | 578 | if (bdata->hint_idx > sidx) { |
0f3caba2 JW |
579 | /* |
580 | * Handle the valid case of sidx being zero and still | |
581 | * catch the fallback below. | |
582 | */ | |
583 | fallback = sidx + 1; | |
481ebd0d | 584 | sidx = align_idx(bdata, bdata->hint_idx, step); |
5f2809e6 | 585 | } |
1da177e4 | 586 | |
5f2809e6 JW |
587 | while (1) { |
588 | int merge; | |
589 | void *region; | |
590 | unsigned long eidx, i, start_off, end_off; | |
591 | find_block: | |
592 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); | |
481ebd0d | 593 | sidx = align_idx(bdata, sidx, step); |
5f2809e6 | 594 | eidx = sidx + PFN_UP(size); |
ad09315c | 595 | |
5f2809e6 | 596 | if (sidx >= midx || eidx > midx) |
66d43e98 | 597 | break; |
1da177e4 | 598 | |
5f2809e6 JW |
599 | for (i = sidx; i < eidx; i++) |
600 | if (test_bit(i, bdata->node_bootmem_map)) { | |
481ebd0d | 601 | sidx = align_idx(bdata, i, step); |
5f2809e6 JW |
602 | if (sidx == i) |
603 | sidx += step; | |
604 | goto find_block; | |
605 | } | |
1da177e4 | 606 | |
627240aa | 607 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
5f2809e6 | 608 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
481ebd0d | 609 | start_off = align_off(bdata, bdata->last_end_off, align); |
5f2809e6 JW |
610 | else |
611 | start_off = PFN_PHYS(sidx); | |
612 | ||
613 | merge = PFN_DOWN(start_off) < sidx; | |
614 | end_off = start_off + size; | |
615 | ||
616 | bdata->last_end_off = end_off; | |
617 | bdata->hint_idx = PFN_UP(end_off); | |
618 | ||
619 | /* | |
620 | * Reserve the area now: | |
621 | */ | |
d747fa4b JW |
622 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
623 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) | |
624 | BUG(); | |
5f2809e6 | 625 | |
3560e249 JW |
626 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
627 | start_off); | |
5f2809e6 | 628 | memset(region, 0, size); |
008139d9 CM |
629 | /* |
630 | * The min_count is set to 0 so that bootmem allocated blocks | |
631 | * are never reported as leaks. | |
632 | */ | |
633 | kmemleak_alloc(region, size, 0, 0); | |
5f2809e6 | 634 | return region; |
1da177e4 LT |
635 | } |
636 | ||
0f3caba2 | 637 | if (fallback) { |
481ebd0d | 638 | sidx = align_idx(bdata, fallback - 1, step); |
0f3caba2 JW |
639 | fallback = 0; |
640 | goto find_block; | |
641 | } | |
642 | ||
643 | return NULL; | |
644 | } | |
645 | ||
d0c4f570 TH |
646 | static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata, |
647 | unsigned long size, unsigned long align, | |
648 | unsigned long goal, unsigned long limit) | |
649 | { | |
441c7e0a PE |
650 | if (WARN_ON_ONCE(slab_is_available())) |
651 | return kzalloc(size, GFP_NOWAIT); | |
652 | ||
d0c4f570 | 653 | #ifdef CONFIG_HAVE_ARCH_BOOTMEM |
433f13a7 JP |
654 | { |
655 | bootmem_data_t *p_bdata; | |
656 | ||
657 | p_bdata = bootmem_arch_preferred_node(bdata, size, align, | |
658 | goal, limit); | |
659 | if (p_bdata) | |
660 | return alloc_bootmem_core(p_bdata, size, align, | |
661 | goal, limit); | |
662 | } | |
d0c4f570 TH |
663 | #endif |
664 | return NULL; | |
665 | } | |
08677214 | 666 | #endif |
d0c4f570 | 667 | |
0f3caba2 JW |
668 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, |
669 | unsigned long align, | |
670 | unsigned long goal, | |
671 | unsigned long limit) | |
672 | { | |
08677214 YL |
673 | #ifdef CONFIG_NO_BOOTMEM |
674 | void *ptr; | |
675 | ||
676 | if (WARN_ON_ONCE(slab_is_available())) | |
677 | return kzalloc(size, GFP_NOWAIT); | |
678 | ||
679 | restart: | |
680 | ||
681 | ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit); | |
682 | ||
683 | if (ptr) | |
684 | return ptr; | |
685 | ||
686 | if (goal != 0) { | |
687 | goal = 0; | |
688 | goto restart; | |
689 | } | |
690 | ||
691 | return NULL; | |
692 | #else | |
0f3caba2 | 693 | bootmem_data_t *bdata; |
d0c4f570 | 694 | void *region; |
0f3caba2 JW |
695 | |
696 | restart: | |
d0c4f570 TH |
697 | region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit); |
698 | if (region) | |
699 | return region; | |
0f3caba2 | 700 | |
d0c4f570 | 701 | list_for_each_entry(bdata, &bdata_list, list) { |
0f3caba2 JW |
702 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
703 | continue; | |
3560e249 | 704 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
0f3caba2 JW |
705 | break; |
706 | ||
707 | region = alloc_bootmem_core(bdata, size, align, goal, limit); | |
708 | if (region) | |
709 | return region; | |
710 | } | |
711 | ||
5f2809e6 JW |
712 | if (goal) { |
713 | goal = 0; | |
0f3caba2 | 714 | goto restart; |
5f2809e6 | 715 | } |
2e5237da | 716 | |
5f2809e6 | 717 | return NULL; |
08677214 | 718 | #endif |
1da177e4 LT |
719 | } |
720 | ||
a66fd7da JW |
721 | /** |
722 | * __alloc_bootmem_nopanic - allocate boot memory without panicking | |
723 | * @size: size of the request in bytes | |
724 | * @align: alignment of the region | |
725 | * @goal: preferred starting address of the region | |
726 | * | |
727 | * The goal is dropped if it can not be satisfied and the allocation will | |
728 | * fall back to memory below @goal. | |
729 | * | |
730 | * Allocation may happen on any node in the system. | |
731 | * | |
732 | * Returns NULL on failure. | |
733 | */ | |
bb0923a6 | 734 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
0f3caba2 | 735 | unsigned long goal) |
1da177e4 | 736 | { |
08677214 YL |
737 | unsigned long limit = 0; |
738 | ||
739 | #ifdef CONFIG_NO_BOOTMEM | |
740 | limit = -1UL; | |
741 | #endif | |
742 | ||
743 | return ___alloc_bootmem_nopanic(size, align, goal, limit); | |
0f3caba2 | 744 | } |
1da177e4 | 745 | |
0f3caba2 JW |
746 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
747 | unsigned long goal, unsigned long limit) | |
748 | { | |
749 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); | |
750 | ||
751 | if (mem) | |
752 | return mem; | |
753 | /* | |
754 | * Whoops, we cannot satisfy the allocation request. | |
755 | */ | |
756 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); | |
757 | panic("Out of memory"); | |
a8062231 AK |
758 | return NULL; |
759 | } | |
1da177e4 | 760 | |
a66fd7da JW |
761 | /** |
762 | * __alloc_bootmem - allocate boot memory | |
763 | * @size: size of the request in bytes | |
764 | * @align: alignment of the region | |
765 | * @goal: preferred starting address of the region | |
766 | * | |
767 | * The goal is dropped if it can not be satisfied and the allocation will | |
768 | * fall back to memory below @goal. | |
769 | * | |
770 | * Allocation may happen on any node in the system. | |
771 | * | |
772 | * The function panics if the request can not be satisfied. | |
773 | */ | |
bb0923a6 FBH |
774 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
775 | unsigned long goal) | |
a8062231 | 776 | { |
08677214 YL |
777 | unsigned long limit = 0; |
778 | ||
779 | #ifdef CONFIG_NO_BOOTMEM | |
780 | limit = -1UL; | |
781 | #endif | |
782 | ||
783 | return ___alloc_bootmem(size, align, goal, limit); | |
1da177e4 LT |
784 | } |
785 | ||
08677214 | 786 | #ifndef CONFIG_NO_BOOTMEM |
4cc278b7 JW |
787 | static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata, |
788 | unsigned long size, unsigned long align, | |
789 | unsigned long goal, unsigned long limit) | |
790 | { | |
791 | void *ptr; | |
792 | ||
d0c4f570 TH |
793 | ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit); |
794 | if (ptr) | |
795 | return ptr; | |
796 | ||
4cc278b7 JW |
797 | ptr = alloc_bootmem_core(bdata, size, align, goal, limit); |
798 | if (ptr) | |
799 | return ptr; | |
800 | ||
801 | return ___alloc_bootmem(size, align, goal, limit); | |
802 | } | |
08677214 | 803 | #endif |
4cc278b7 | 804 | |
a66fd7da JW |
805 | /** |
806 | * __alloc_bootmem_node - allocate boot memory from a specific node | |
807 | * @pgdat: node to allocate from | |
808 | * @size: size of the request in bytes | |
809 | * @align: alignment of the region | |
810 | * @goal: preferred starting address of the region | |
811 | * | |
812 | * The goal is dropped if it can not be satisfied and the allocation will | |
813 | * fall back to memory below @goal. | |
814 | * | |
815 | * Allocation may fall back to any node in the system if the specified node | |
816 | * can not hold the requested memory. | |
817 | * | |
818 | * The function panics if the request can not be satisfied. | |
819 | */ | |
bb0923a6 FBH |
820 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
821 | unsigned long align, unsigned long goal) | |
1da177e4 | 822 | { |
c91c4773 PE |
823 | if (WARN_ON_ONCE(slab_is_available())) |
824 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
825 | ||
08677214 YL |
826 | #ifdef CONFIG_NO_BOOTMEM |
827 | return __alloc_memory_core_early(pgdat->node_id, size, align, | |
828 | goal, -1ULL); | |
829 | #else | |
4cc278b7 | 830 | return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0); |
08677214 YL |
831 | #endif |
832 | } | |
833 | ||
834 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, | |
835 | unsigned long align, unsigned long goal) | |
836 | { | |
837 | #ifdef MAX_DMA32_PFN | |
838 | unsigned long end_pfn; | |
839 | ||
840 | if (WARN_ON_ONCE(slab_is_available())) | |
841 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
842 | ||
843 | /* update goal according ...MAX_DMA32_PFN */ | |
844 | end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
845 | ||
846 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && | |
847 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { | |
848 | void *ptr; | |
849 | unsigned long new_goal; | |
850 | ||
851 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; | |
852 | #ifdef CONFIG_NO_BOOTMEM | |
853 | ptr = __alloc_memory_core_early(pgdat->node_id, size, align, | |
854 | new_goal, -1ULL); | |
855 | #else | |
856 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, | |
857 | new_goal, 0); | |
858 | #endif | |
859 | if (ptr) | |
860 | return ptr; | |
861 | } | |
862 | #endif | |
863 | ||
864 | return __alloc_bootmem_node(pgdat, size, align, goal); | |
865 | ||
1da177e4 LT |
866 | } |
867 | ||
e70260aa | 868 | #ifdef CONFIG_SPARSEMEM |
a66fd7da JW |
869 | /** |
870 | * alloc_bootmem_section - allocate boot memory from a specific section | |
871 | * @size: size of the request in bytes | |
872 | * @section_nr: sparse map section to allocate from | |
873 | * | |
874 | * Return NULL on failure. | |
875 | */ | |
e70260aa YG |
876 | void * __init alloc_bootmem_section(unsigned long size, |
877 | unsigned long section_nr) | |
878 | { | |
08677214 YL |
879 | #ifdef CONFIG_NO_BOOTMEM |
880 | unsigned long pfn, goal, limit; | |
881 | ||
882 | pfn = section_nr_to_pfn(section_nr); | |
883 | goal = pfn << PAGE_SHIFT; | |
884 | limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT; | |
885 | ||
886 | return __alloc_memory_core_early(early_pfn_to_nid(pfn), size, | |
887 | SMP_CACHE_BYTES, goal, limit); | |
888 | #else | |
75a56cfe JW |
889 | bootmem_data_t *bdata; |
890 | unsigned long pfn, goal, limit; | |
e70260aa YG |
891 | |
892 | pfn = section_nr_to_pfn(section_nr); | |
75a56cfe JW |
893 | goal = pfn << PAGE_SHIFT; |
894 | limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT; | |
895 | bdata = &bootmem_node_data[early_pfn_to_nid(pfn)]; | |
e70260aa | 896 | |
75a56cfe | 897 | return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit); |
08677214 | 898 | #endif |
e70260aa YG |
899 | } |
900 | #endif | |
901 | ||
b54bbf7b AK |
902 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, |
903 | unsigned long align, unsigned long goal) | |
904 | { | |
905 | void *ptr; | |
906 | ||
c91c4773 PE |
907 | if (WARN_ON_ONCE(slab_is_available())) |
908 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
909 | ||
08677214 YL |
910 | #ifdef CONFIG_NO_BOOTMEM |
911 | ptr = __alloc_memory_core_early(pgdat->node_id, size, align, | |
912 | goal, -1ULL); | |
913 | #else | |
d0c4f570 TH |
914 | ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0); |
915 | if (ptr) | |
916 | return ptr; | |
917 | ||
b54bbf7b | 918 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0); |
08677214 | 919 | #endif |
b54bbf7b AK |
920 | if (ptr) |
921 | return ptr; | |
922 | ||
923 | return __alloc_bootmem_nopanic(size, align, goal); | |
924 | } | |
925 | ||
dfd54cbc HC |
926 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
927 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL | |
928 | #endif | |
008857c1 | 929 | |
a66fd7da JW |
930 | /** |
931 | * __alloc_bootmem_low - allocate low boot memory | |
932 | * @size: size of the request in bytes | |
933 | * @align: alignment of the region | |
934 | * @goal: preferred starting address of the region | |
935 | * | |
936 | * The goal is dropped if it can not be satisfied and the allocation will | |
937 | * fall back to memory below @goal. | |
938 | * | |
939 | * Allocation may happen on any node in the system. | |
940 | * | |
941 | * The function panics if the request can not be satisfied. | |
942 | */ | |
bb0923a6 FBH |
943 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
944 | unsigned long goal) | |
008857c1 | 945 | { |
0f3caba2 | 946 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
008857c1 RT |
947 | } |
948 | ||
a66fd7da JW |
949 | /** |
950 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node | |
951 | * @pgdat: node to allocate from | |
952 | * @size: size of the request in bytes | |
953 | * @align: alignment of the region | |
954 | * @goal: preferred starting address of the region | |
955 | * | |
956 | * The goal is dropped if it can not be satisfied and the allocation will | |
957 | * fall back to memory below @goal. | |
958 | * | |
959 | * Allocation may fall back to any node in the system if the specified node | |
960 | * can not hold the requested memory. | |
961 | * | |
962 | * The function panics if the request can not be satisfied. | |
963 | */ | |
008857c1 RT |
964 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
965 | unsigned long align, unsigned long goal) | |
966 | { | |
c91c4773 PE |
967 | if (WARN_ON_ONCE(slab_is_available())) |
968 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
969 | ||
08677214 YL |
970 | #ifdef CONFIG_NO_BOOTMEM |
971 | return __alloc_memory_core_early(pgdat->node_id, size, align, | |
972 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
973 | #else | |
4cc278b7 JW |
974 | return ___alloc_bootmem_node(pgdat->bdata, size, align, |
975 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
08677214 | 976 | #endif |
008857c1 | 977 | } |