Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
57cfc29e | 2 | * bootmem - A boot-time physical memory allocator and configurator |
1da177e4 LT |
3 | * |
4 | * Copyright (C) 1999 Ingo Molnar | |
57cfc29e JW |
5 | * 1999 Kanoj Sarcar, SGI |
6 | * 2008 Johannes Weiner | |
1da177e4 | 7 | * |
57cfc29e JW |
8 | * Access to this subsystem has to be serialized externally (which is true |
9 | * for the boot process anyway). | |
1da177e4 | 10 | */ |
1da177e4 | 11 | #include <linux/init.h> |
bbc7b92e | 12 | #include <linux/pfn.h> |
1da177e4 | 13 | #include <linux/bootmem.h> |
1da177e4 | 14 | #include <linux/module.h> |
ec3a354b | 15 | #include <linux/kmemleak.h> |
08677214 | 16 | #include <linux/range.h> |
e786e86a FBH |
17 | |
18 | #include <asm/bug.h> | |
1da177e4 | 19 | #include <asm/io.h> |
dfd54cbc | 20 | #include <asm/processor.h> |
e786e86a | 21 | |
1da177e4 LT |
22 | #include "internal.h" |
23 | ||
1da177e4 LT |
24 | unsigned long max_low_pfn; |
25 | unsigned long min_low_pfn; | |
26 | unsigned long max_pfn; | |
27 | ||
92aa63a5 VG |
28 | #ifdef CONFIG_CRASH_DUMP |
29 | /* | |
30 | * If we have booted due to a crash, max_pfn will be a very low value. We need | |
31 | * to know the amount of memory that the previous kernel used. | |
32 | */ | |
33 | unsigned long saved_max_pfn; | |
34 | #endif | |
35 | ||
08677214 | 36 | #ifndef CONFIG_NO_BOOTMEM |
b61bfa3c JW |
37 | bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata; |
38 | ||
636cc40c JW |
39 | static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list); |
40 | ||
2e5237da JW |
41 | static int bootmem_debug; |
42 | ||
43 | static int __init bootmem_debug_setup(char *buf) | |
44 | { | |
45 | bootmem_debug = 1; | |
46 | return 0; | |
47 | } | |
48 | early_param("bootmem_debug", bootmem_debug_setup); | |
49 | ||
50 | #define bdebug(fmt, args...) ({ \ | |
51 | if (unlikely(bootmem_debug)) \ | |
52 | printk(KERN_INFO \ | |
53 | "bootmem::%s " fmt, \ | |
80a914dc | 54 | __func__, ## args); \ |
2e5237da JW |
55 | }) |
56 | ||
df049a5f | 57 | static unsigned long __init bootmap_bytes(unsigned long pages) |
223e8dc9 | 58 | { |
df049a5f | 59 | unsigned long bytes = (pages + 7) / 8; |
223e8dc9 | 60 | |
df049a5f | 61 | return ALIGN(bytes, sizeof(long)); |
223e8dc9 JW |
62 | } |
63 | ||
a66fd7da JW |
64 | /** |
65 | * bootmem_bootmap_pages - calculate bitmap size in pages | |
66 | * @pages: number of pages the bitmap has to represent | |
67 | */ | |
f71bf0ca | 68 | unsigned long __init bootmem_bootmap_pages(unsigned long pages) |
1da177e4 | 69 | { |
df049a5f | 70 | unsigned long bytes = bootmap_bytes(pages); |
1da177e4 | 71 | |
df049a5f | 72 | return PAGE_ALIGN(bytes) >> PAGE_SHIFT; |
1da177e4 | 73 | } |
f71bf0ca | 74 | |
679bc9fb KH |
75 | /* |
76 | * link bdata in order | |
77 | */ | |
69d49e68 | 78 | static void __init link_bootmem(bootmem_data_t *bdata) |
679bc9fb | 79 | { |
636cc40c | 80 | struct list_head *iter; |
f71bf0ca | 81 | |
636cc40c JW |
82 | list_for_each(iter, &bdata_list) { |
83 | bootmem_data_t *ent; | |
84 | ||
85 | ent = list_entry(iter, bootmem_data_t, list); | |
3560e249 | 86 | if (bdata->node_min_pfn < ent->node_min_pfn) |
636cc40c | 87 | break; |
679bc9fb | 88 | } |
636cc40c | 89 | list_add_tail(&bdata->list, iter); |
679bc9fb KH |
90 | } |
91 | ||
1da177e4 LT |
92 | /* |
93 | * Called once to set up the allocator itself. | |
94 | */ | |
8ae04463 | 95 | static unsigned long __init init_bootmem_core(bootmem_data_t *bdata, |
1da177e4 LT |
96 | unsigned long mapstart, unsigned long start, unsigned long end) |
97 | { | |
bbc7b92e | 98 | unsigned long mapsize; |
1da177e4 | 99 | |
2dbb51c4 | 100 | mminit_validate_memmodel_limits(&start, &end); |
bbc7b92e | 101 | bdata->node_bootmem_map = phys_to_virt(PFN_PHYS(mapstart)); |
3560e249 | 102 | bdata->node_min_pfn = start; |
1da177e4 | 103 | bdata->node_low_pfn = end; |
679bc9fb | 104 | link_bootmem(bdata); |
1da177e4 LT |
105 | |
106 | /* | |
107 | * Initially all pages are reserved - setup_arch() has to | |
108 | * register free RAM areas explicitly. | |
109 | */ | |
df049a5f | 110 | mapsize = bootmap_bytes(end - start); |
1da177e4 LT |
111 | memset(bdata->node_bootmem_map, 0xff, mapsize); |
112 | ||
2e5237da JW |
113 | bdebug("nid=%td start=%lx map=%lx end=%lx mapsize=%lx\n", |
114 | bdata - bootmem_node_data, start, mapstart, end, mapsize); | |
115 | ||
1da177e4 LT |
116 | return mapsize; |
117 | } | |
118 | ||
a66fd7da JW |
119 | /** |
120 | * init_bootmem_node - register a node as boot memory | |
121 | * @pgdat: node to register | |
122 | * @freepfn: pfn where the bitmap for this node is to be placed | |
123 | * @startpfn: first pfn on the node | |
124 | * @endpfn: first pfn after the node | |
125 | * | |
126 | * Returns the number of bytes needed to hold the bitmap for this node. | |
127 | */ | |
223e8dc9 JW |
128 | unsigned long __init init_bootmem_node(pg_data_t *pgdat, unsigned long freepfn, |
129 | unsigned long startpfn, unsigned long endpfn) | |
130 | { | |
131 | return init_bootmem_core(pgdat->bdata, freepfn, startpfn, endpfn); | |
132 | } | |
133 | ||
a66fd7da JW |
134 | /** |
135 | * init_bootmem - register boot memory | |
136 | * @start: pfn where the bitmap is to be placed | |
137 | * @pages: number of available physical pages | |
138 | * | |
139 | * Returns the number of bytes needed to hold the bitmap. | |
140 | */ | |
223e8dc9 JW |
141 | unsigned long __init init_bootmem(unsigned long start, unsigned long pages) |
142 | { | |
143 | max_low_pfn = pages; | |
144 | min_low_pfn = start; | |
145 | return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages); | |
146 | } | |
08677214 | 147 | #endif |
9f993ac3 FT |
148 | /* |
149 | * free_bootmem_late - free bootmem pages directly to page allocator | |
150 | * @addr: starting address of the range | |
151 | * @size: size of the range in bytes | |
152 | * | |
153 | * This is only useful when the bootmem allocator has already been torn | |
154 | * down, but we are still initializing the system. Pages are given directly | |
155 | * to the page allocator, no bootmem metadata is updated because it is gone. | |
156 | */ | |
157 | void __init free_bootmem_late(unsigned long addr, unsigned long size) | |
158 | { | |
159 | unsigned long cursor, end; | |
160 | ||
161 | kmemleak_free_part(__va(addr), size); | |
162 | ||
163 | cursor = PFN_UP(addr); | |
164 | end = PFN_DOWN(addr + size); | |
165 | ||
166 | for (; cursor < end; cursor++) { | |
167 | __free_pages_bootmem(pfn_to_page(cursor), 0); | |
168 | totalram_pages++; | |
169 | } | |
170 | } | |
171 | ||
08677214 YL |
172 | #ifdef CONFIG_NO_BOOTMEM |
173 | static void __init __free_pages_memory(unsigned long start, unsigned long end) | |
174 | { | |
175 | int i; | |
176 | unsigned long start_aligned, end_aligned; | |
177 | int order = ilog2(BITS_PER_LONG); | |
178 | ||
179 | start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1); | |
180 | end_aligned = end & ~(BITS_PER_LONG - 1); | |
181 | ||
182 | if (end_aligned <= start_aligned) { | |
08677214 YL |
183 | for (i = start; i < end; i++) |
184 | __free_pages_bootmem(pfn_to_page(i), 0); | |
185 | ||
186 | return; | |
187 | } | |
188 | ||
08677214 YL |
189 | for (i = start; i < start_aligned; i++) |
190 | __free_pages_bootmem(pfn_to_page(i), 0); | |
191 | ||
192 | for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG) | |
193 | __free_pages_bootmem(pfn_to_page(i), order); | |
194 | ||
195 | for (i = end_aligned; i < end; i++) | |
196 | __free_pages_bootmem(pfn_to_page(i), 0); | |
197 | } | |
198 | ||
199 | unsigned long __init free_all_memory_core_early(int nodeid) | |
200 | { | |
201 | int i; | |
202 | u64 start, end; | |
203 | unsigned long count = 0; | |
204 | struct range *range = NULL; | |
205 | int nr_range; | |
206 | ||
207 | nr_range = get_free_all_memory_range(&range, nodeid); | |
208 | ||
209 | for (i = 0; i < nr_range; i++) { | |
210 | start = range[i].start; | |
211 | end = range[i].end; | |
212 | count += end - start; | |
213 | __free_pages_memory(start, end); | |
214 | } | |
215 | ||
216 | return count; | |
217 | } | |
218 | #else | |
223e8dc9 JW |
219 | static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) |
220 | { | |
41546c17 | 221 | int aligned; |
223e8dc9 | 222 | struct page *page; |
41546c17 JW |
223 | unsigned long start, end, pages, count = 0; |
224 | ||
225 | if (!bdata->node_bootmem_map) | |
226 | return 0; | |
227 | ||
3560e249 | 228 | start = bdata->node_min_pfn; |
41546c17 JW |
229 | end = bdata->node_low_pfn; |
230 | ||
223e8dc9 | 231 | /* |
41546c17 JW |
232 | * If the start is aligned to the machines wordsize, we might |
233 | * be able to free pages in bulks of that order. | |
223e8dc9 | 234 | */ |
41546c17 | 235 | aligned = !(start & (BITS_PER_LONG - 1)); |
223e8dc9 | 236 | |
41546c17 JW |
237 | bdebug("nid=%td start=%lx end=%lx aligned=%d\n", |
238 | bdata - bootmem_node_data, start, end, aligned); | |
223e8dc9 | 239 | |
41546c17 JW |
240 | while (start < end) { |
241 | unsigned long *map, idx, vec; | |
223e8dc9 | 242 | |
41546c17 | 243 | map = bdata->node_bootmem_map; |
3560e249 | 244 | idx = start - bdata->node_min_pfn; |
41546c17 JW |
245 | vec = ~map[idx / BITS_PER_LONG]; |
246 | ||
247 | if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) { | |
248 | int order = ilog2(BITS_PER_LONG); | |
249 | ||
250 | __free_pages_bootmem(pfn_to_page(start), order); | |
223e8dc9 | 251 | count += BITS_PER_LONG; |
41546c17 JW |
252 | } else { |
253 | unsigned long off = 0; | |
254 | ||
255 | while (vec && off < BITS_PER_LONG) { | |
256 | if (vec & 1) { | |
257 | page = pfn_to_page(start + off); | |
223e8dc9 | 258 | __free_pages_bootmem(page, 0); |
41546c17 | 259 | count++; |
223e8dc9 | 260 | } |
41546c17 JW |
261 | vec >>= 1; |
262 | off++; | |
223e8dc9 | 263 | } |
223e8dc9 | 264 | } |
41546c17 | 265 | start += BITS_PER_LONG; |
223e8dc9 JW |
266 | } |
267 | ||
223e8dc9 | 268 | page = virt_to_page(bdata->node_bootmem_map); |
3560e249 | 269 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
41546c17 JW |
270 | pages = bootmem_bootmap_pages(pages); |
271 | count += pages; | |
272 | while (pages--) | |
273 | __free_pages_bootmem(page++, 0); | |
223e8dc9 | 274 | |
2e5237da JW |
275 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
276 | ||
223e8dc9 JW |
277 | return count; |
278 | } | |
08677214 | 279 | #endif |
223e8dc9 | 280 | |
a66fd7da JW |
281 | /** |
282 | * free_all_bootmem_node - release a node's free pages to the buddy allocator | |
283 | * @pgdat: node to be released | |
284 | * | |
285 | * Returns the number of pages actually released. | |
286 | */ | |
223e8dc9 JW |
287 | unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) |
288 | { | |
289 | register_page_bootmem_info_node(pgdat); | |
08677214 YL |
290 | #ifdef CONFIG_NO_BOOTMEM |
291 | /* free_all_memory_core_early(MAX_NUMNODES) will be called later */ | |
292 | return 0; | |
293 | #else | |
223e8dc9 | 294 | return free_all_bootmem_core(pgdat->bdata); |
08677214 | 295 | #endif |
223e8dc9 JW |
296 | } |
297 | ||
a66fd7da JW |
298 | /** |
299 | * free_all_bootmem - release free pages to the buddy allocator | |
300 | * | |
301 | * Returns the number of pages actually released. | |
302 | */ | |
223e8dc9 JW |
303 | unsigned long __init free_all_bootmem(void) |
304 | { | |
08677214 | 305 | #ifdef CONFIG_NO_BOOTMEM |
33799858 YL |
306 | /* |
307 | * We need to use MAX_NUMNODES instead of NODE_DATA(0)->node_id | |
308 | * because in some case like Node0 doesnt have RAM installed | |
309 | * low ram will be on Node1 | |
310 | * Use MAX_NUMNODES will make sure all ranges in early_node_map[] | |
311 | * will be used instead of only Node0 related | |
312 | */ | |
313 | return free_all_memory_core_early(MAX_NUMNODES); | |
08677214 | 314 | #else |
223e8dc9 | 315 | return free_all_bootmem_core(NODE_DATA(0)->bdata); |
08677214 | 316 | #endif |
223e8dc9 JW |
317 | } |
318 | ||
08677214 | 319 | #ifndef CONFIG_NO_BOOTMEM |
d747fa4b JW |
320 | static void __init __free(bootmem_data_t *bdata, |
321 | unsigned long sidx, unsigned long eidx) | |
322 | { | |
323 | unsigned long idx; | |
324 | ||
325 | bdebug("nid=%td start=%lx end=%lx\n", bdata - bootmem_node_data, | |
3560e249 JW |
326 | sidx + bdata->node_min_pfn, |
327 | eidx + bdata->node_min_pfn); | |
d747fa4b | 328 | |
e2bf3cae JW |
329 | if (bdata->hint_idx > sidx) |
330 | bdata->hint_idx = sidx; | |
331 | ||
d747fa4b JW |
332 | for (idx = sidx; idx < eidx; idx++) |
333 | if (!test_and_clear_bit(idx, bdata->node_bootmem_map)) | |
334 | BUG(); | |
335 | } | |
336 | ||
337 | static int __init __reserve(bootmem_data_t *bdata, unsigned long sidx, | |
338 | unsigned long eidx, int flags) | |
339 | { | |
340 | unsigned long idx; | |
341 | int exclusive = flags & BOOTMEM_EXCLUSIVE; | |
342 | ||
343 | bdebug("nid=%td start=%lx end=%lx flags=%x\n", | |
344 | bdata - bootmem_node_data, | |
3560e249 JW |
345 | sidx + bdata->node_min_pfn, |
346 | eidx + bdata->node_min_pfn, | |
d747fa4b JW |
347 | flags); |
348 | ||
349 | for (idx = sidx; idx < eidx; idx++) | |
350 | if (test_and_set_bit(idx, bdata->node_bootmem_map)) { | |
351 | if (exclusive) { | |
352 | __free(bdata, sidx, idx); | |
353 | return -EBUSY; | |
354 | } | |
355 | bdebug("silent double reserve of PFN %lx\n", | |
3560e249 | 356 | idx + bdata->node_min_pfn); |
d747fa4b JW |
357 | } |
358 | return 0; | |
359 | } | |
360 | ||
e2bf3cae JW |
361 | static int __init mark_bootmem_node(bootmem_data_t *bdata, |
362 | unsigned long start, unsigned long end, | |
363 | int reserve, int flags) | |
223e8dc9 JW |
364 | { |
365 | unsigned long sidx, eidx; | |
223e8dc9 | 366 | |
e2bf3cae JW |
367 | bdebug("nid=%td start=%lx end=%lx reserve=%d flags=%x\n", |
368 | bdata - bootmem_node_data, start, end, reserve, flags); | |
223e8dc9 | 369 | |
3560e249 | 370 | BUG_ON(start < bdata->node_min_pfn); |
e2bf3cae | 371 | BUG_ON(end > bdata->node_low_pfn); |
223e8dc9 | 372 | |
3560e249 JW |
373 | sidx = start - bdata->node_min_pfn; |
374 | eidx = end - bdata->node_min_pfn; | |
223e8dc9 | 375 | |
e2bf3cae JW |
376 | if (reserve) |
377 | return __reserve(bdata, sidx, eidx, flags); | |
223e8dc9 | 378 | else |
e2bf3cae JW |
379 | __free(bdata, sidx, eidx); |
380 | return 0; | |
381 | } | |
382 | ||
383 | static int __init mark_bootmem(unsigned long start, unsigned long end, | |
384 | int reserve, int flags) | |
385 | { | |
386 | unsigned long pos; | |
387 | bootmem_data_t *bdata; | |
388 | ||
389 | pos = start; | |
390 | list_for_each_entry(bdata, &bdata_list, list) { | |
391 | int err; | |
392 | unsigned long max; | |
393 | ||
3560e249 JW |
394 | if (pos < bdata->node_min_pfn || |
395 | pos >= bdata->node_low_pfn) { | |
e2bf3cae JW |
396 | BUG_ON(pos != start); |
397 | continue; | |
398 | } | |
399 | ||
400 | max = min(bdata->node_low_pfn, end); | |
223e8dc9 | 401 | |
e2bf3cae JW |
402 | err = mark_bootmem_node(bdata, pos, max, reserve, flags); |
403 | if (reserve && err) { | |
404 | mark_bootmem(start, pos, 0, 0); | |
405 | return err; | |
406 | } | |
223e8dc9 | 407 | |
e2bf3cae JW |
408 | if (max == end) |
409 | return 0; | |
410 | pos = bdata->node_low_pfn; | |
411 | } | |
412 | BUG(); | |
223e8dc9 | 413 | } |
08677214 | 414 | #endif |
223e8dc9 | 415 | |
a66fd7da JW |
416 | /** |
417 | * free_bootmem_node - mark a page range as usable | |
418 | * @pgdat: node the range resides on | |
419 | * @physaddr: starting address of the range | |
420 | * @size: size of the range in bytes | |
421 | * | |
422 | * Partial pages will be considered reserved and left as they are. | |
423 | * | |
e2bf3cae | 424 | * The range must reside completely on the specified node. |
a66fd7da | 425 | */ |
223e8dc9 JW |
426 | void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
427 | unsigned long size) | |
428 | { | |
08677214 YL |
429 | #ifdef CONFIG_NO_BOOTMEM |
430 | free_early(physaddr, physaddr + size); | |
08677214 | 431 | #else |
e2bf3cae JW |
432 | unsigned long start, end; |
433 | ||
ec3a354b CM |
434 | kmemleak_free_part(__va(physaddr), size); |
435 | ||
e2bf3cae JW |
436 | start = PFN_UP(physaddr); |
437 | end = PFN_DOWN(physaddr + size); | |
438 | ||
439 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); | |
08677214 | 440 | #endif |
223e8dc9 JW |
441 | } |
442 | ||
a66fd7da JW |
443 | /** |
444 | * free_bootmem - mark a page range as usable | |
445 | * @addr: starting address of the range | |
446 | * @size: size of the range in bytes | |
447 | * | |
448 | * Partial pages will be considered reserved and left as they are. | |
449 | * | |
e2bf3cae | 450 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 451 | */ |
223e8dc9 JW |
452 | void __init free_bootmem(unsigned long addr, unsigned long size) |
453 | { | |
08677214 YL |
454 | #ifdef CONFIG_NO_BOOTMEM |
455 | free_early(addr, addr + size); | |
08677214 | 456 | #else |
e2bf3cae | 457 | unsigned long start, end; |
a5645a61 | 458 | |
ec3a354b CM |
459 | kmemleak_free_part(__va(addr), size); |
460 | ||
e2bf3cae JW |
461 | start = PFN_UP(addr); |
462 | end = PFN_DOWN(addr + size); | |
1da177e4 | 463 | |
e2bf3cae | 464 | mark_bootmem(start, end, 0, 0); |
08677214 | 465 | #endif |
1da177e4 LT |
466 | } |
467 | ||
a66fd7da JW |
468 | /** |
469 | * reserve_bootmem_node - mark a page range as reserved | |
470 | * @pgdat: node the range resides on | |
471 | * @physaddr: starting address of the range | |
472 | * @size: size of the range in bytes | |
473 | * @flags: reservation flags (see linux/bootmem.h) | |
474 | * | |
475 | * Partial pages will be reserved. | |
476 | * | |
e2bf3cae | 477 | * The range must reside completely on the specified node. |
a66fd7da | 478 | */ |
223e8dc9 JW |
479 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
480 | unsigned long size, int flags) | |
1da177e4 | 481 | { |
08677214 YL |
482 | #ifdef CONFIG_NO_BOOTMEM |
483 | panic("no bootmem"); | |
484 | return 0; | |
485 | #else | |
e2bf3cae | 486 | unsigned long start, end; |
1da177e4 | 487 | |
e2bf3cae JW |
488 | start = PFN_DOWN(physaddr); |
489 | end = PFN_UP(physaddr + size); | |
490 | ||
491 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); | |
08677214 | 492 | #endif |
223e8dc9 | 493 | } |
5a982cbc | 494 | |
a66fd7da JW |
495 | /** |
496 | * reserve_bootmem - mark a page range as usable | |
497 | * @addr: starting address of the range | |
498 | * @size: size of the range in bytes | |
499 | * @flags: reservation flags (see linux/bootmem.h) | |
500 | * | |
501 | * Partial pages will be reserved. | |
502 | * | |
e2bf3cae | 503 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 504 | */ |
223e8dc9 JW |
505 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
506 | int flags) | |
507 | { | |
08677214 YL |
508 | #ifdef CONFIG_NO_BOOTMEM |
509 | panic("no bootmem"); | |
510 | return 0; | |
511 | #else | |
e2bf3cae | 512 | unsigned long start, end; |
1da177e4 | 513 | |
e2bf3cae JW |
514 | start = PFN_DOWN(addr); |
515 | end = PFN_UP(addr + size); | |
223e8dc9 | 516 | |
e2bf3cae | 517 | return mark_bootmem(start, end, 1, flags); |
08677214 | 518 | #endif |
1da177e4 LT |
519 | } |
520 | ||
08677214 | 521 | #ifndef CONFIG_NO_BOOTMEM |
8aa043d7 JB |
522 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
523 | unsigned long idx, unsigned long step) | |
481ebd0d JW |
524 | { |
525 | unsigned long base = bdata->node_min_pfn; | |
526 | ||
527 | /* | |
528 | * Align the index with respect to the node start so that the | |
529 | * combination of both satisfies the requested alignment. | |
530 | */ | |
531 | ||
532 | return ALIGN(base + idx, step) - base; | |
533 | } | |
534 | ||
8aa043d7 JB |
535 | static unsigned long __init align_off(struct bootmem_data *bdata, |
536 | unsigned long off, unsigned long align) | |
481ebd0d JW |
537 | { |
538 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); | |
539 | ||
540 | /* Same as align_idx for byte offsets */ | |
541 | ||
542 | return ALIGN(base + off, align) - base; | |
543 | } | |
544 | ||
d0c4f570 TH |
545 | static void * __init alloc_bootmem_core(struct bootmem_data *bdata, |
546 | unsigned long size, unsigned long align, | |
547 | unsigned long goal, unsigned long limit) | |
1da177e4 | 548 | { |
0f3caba2 | 549 | unsigned long fallback = 0; |
5f2809e6 JW |
550 | unsigned long min, max, start, sidx, midx, step; |
551 | ||
594fe1a0 JW |
552 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
553 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, | |
554 | align, goal, limit); | |
555 | ||
5f2809e6 JW |
556 | BUG_ON(!size); |
557 | BUG_ON(align & (align - 1)); | |
558 | BUG_ON(limit && goal + size > limit); | |
1da177e4 | 559 | |
7c309a64 CK |
560 | if (!bdata->node_bootmem_map) |
561 | return NULL; | |
562 | ||
3560e249 | 563 | min = bdata->node_min_pfn; |
5f2809e6 | 564 | max = bdata->node_low_pfn; |
9a2dc04c | 565 | |
5f2809e6 JW |
566 | goal >>= PAGE_SHIFT; |
567 | limit >>= PAGE_SHIFT; | |
568 | ||
569 | if (limit && max > limit) | |
570 | max = limit; | |
571 | if (max <= min) | |
9a2dc04c YL |
572 | return NULL; |
573 | ||
5f2809e6 | 574 | step = max(align >> PAGE_SHIFT, 1UL); |
281dd25c | 575 | |
5f2809e6 JW |
576 | if (goal && min < goal && goal < max) |
577 | start = ALIGN(goal, step); | |
578 | else | |
579 | start = ALIGN(min, step); | |
1da177e4 | 580 | |
481ebd0d | 581 | sidx = start - bdata->node_min_pfn; |
3560e249 | 582 | midx = max - bdata->node_min_pfn; |
1da177e4 | 583 | |
5f2809e6 | 584 | if (bdata->hint_idx > sidx) { |
0f3caba2 JW |
585 | /* |
586 | * Handle the valid case of sidx being zero and still | |
587 | * catch the fallback below. | |
588 | */ | |
589 | fallback = sidx + 1; | |
481ebd0d | 590 | sidx = align_idx(bdata, bdata->hint_idx, step); |
5f2809e6 | 591 | } |
1da177e4 | 592 | |
5f2809e6 JW |
593 | while (1) { |
594 | int merge; | |
595 | void *region; | |
596 | unsigned long eidx, i, start_off, end_off; | |
597 | find_block: | |
598 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); | |
481ebd0d | 599 | sidx = align_idx(bdata, sidx, step); |
5f2809e6 | 600 | eidx = sidx + PFN_UP(size); |
ad09315c | 601 | |
5f2809e6 | 602 | if (sidx >= midx || eidx > midx) |
66d43e98 | 603 | break; |
1da177e4 | 604 | |
5f2809e6 JW |
605 | for (i = sidx; i < eidx; i++) |
606 | if (test_bit(i, bdata->node_bootmem_map)) { | |
481ebd0d | 607 | sidx = align_idx(bdata, i, step); |
5f2809e6 JW |
608 | if (sidx == i) |
609 | sidx += step; | |
610 | goto find_block; | |
611 | } | |
1da177e4 | 612 | |
627240aa | 613 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
5f2809e6 | 614 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
481ebd0d | 615 | start_off = align_off(bdata, bdata->last_end_off, align); |
5f2809e6 JW |
616 | else |
617 | start_off = PFN_PHYS(sidx); | |
618 | ||
619 | merge = PFN_DOWN(start_off) < sidx; | |
620 | end_off = start_off + size; | |
621 | ||
622 | bdata->last_end_off = end_off; | |
623 | bdata->hint_idx = PFN_UP(end_off); | |
624 | ||
625 | /* | |
626 | * Reserve the area now: | |
627 | */ | |
d747fa4b JW |
628 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
629 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) | |
630 | BUG(); | |
5f2809e6 | 631 | |
3560e249 JW |
632 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
633 | start_off); | |
5f2809e6 | 634 | memset(region, 0, size); |
008139d9 CM |
635 | /* |
636 | * The min_count is set to 0 so that bootmem allocated blocks | |
637 | * are never reported as leaks. | |
638 | */ | |
639 | kmemleak_alloc(region, size, 0, 0); | |
5f2809e6 | 640 | return region; |
1da177e4 LT |
641 | } |
642 | ||
0f3caba2 | 643 | if (fallback) { |
481ebd0d | 644 | sidx = align_idx(bdata, fallback - 1, step); |
0f3caba2 JW |
645 | fallback = 0; |
646 | goto find_block; | |
647 | } | |
648 | ||
649 | return NULL; | |
650 | } | |
651 | ||
d0c4f570 TH |
652 | static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata, |
653 | unsigned long size, unsigned long align, | |
654 | unsigned long goal, unsigned long limit) | |
655 | { | |
441c7e0a PE |
656 | if (WARN_ON_ONCE(slab_is_available())) |
657 | return kzalloc(size, GFP_NOWAIT); | |
658 | ||
d0c4f570 | 659 | #ifdef CONFIG_HAVE_ARCH_BOOTMEM |
433f13a7 JP |
660 | { |
661 | bootmem_data_t *p_bdata; | |
662 | ||
663 | p_bdata = bootmem_arch_preferred_node(bdata, size, align, | |
664 | goal, limit); | |
665 | if (p_bdata) | |
666 | return alloc_bootmem_core(p_bdata, size, align, | |
667 | goal, limit); | |
668 | } | |
d0c4f570 TH |
669 | #endif |
670 | return NULL; | |
671 | } | |
08677214 | 672 | #endif |
d0c4f570 | 673 | |
0f3caba2 JW |
674 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, |
675 | unsigned long align, | |
676 | unsigned long goal, | |
677 | unsigned long limit) | |
678 | { | |
08677214 YL |
679 | #ifdef CONFIG_NO_BOOTMEM |
680 | void *ptr; | |
681 | ||
682 | if (WARN_ON_ONCE(slab_is_available())) | |
683 | return kzalloc(size, GFP_NOWAIT); | |
684 | ||
685 | restart: | |
686 | ||
687 | ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit); | |
688 | ||
689 | if (ptr) | |
690 | return ptr; | |
691 | ||
692 | if (goal != 0) { | |
693 | goal = 0; | |
694 | goto restart; | |
695 | } | |
696 | ||
697 | return NULL; | |
698 | #else | |
0f3caba2 | 699 | bootmem_data_t *bdata; |
d0c4f570 | 700 | void *region; |
0f3caba2 JW |
701 | |
702 | restart: | |
d0c4f570 TH |
703 | region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit); |
704 | if (region) | |
705 | return region; | |
0f3caba2 | 706 | |
d0c4f570 | 707 | list_for_each_entry(bdata, &bdata_list, list) { |
0f3caba2 JW |
708 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
709 | continue; | |
3560e249 | 710 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
0f3caba2 JW |
711 | break; |
712 | ||
713 | region = alloc_bootmem_core(bdata, size, align, goal, limit); | |
714 | if (region) | |
715 | return region; | |
716 | } | |
717 | ||
5f2809e6 JW |
718 | if (goal) { |
719 | goal = 0; | |
0f3caba2 | 720 | goto restart; |
5f2809e6 | 721 | } |
2e5237da | 722 | |
5f2809e6 | 723 | return NULL; |
08677214 | 724 | #endif |
1da177e4 LT |
725 | } |
726 | ||
a66fd7da JW |
727 | /** |
728 | * __alloc_bootmem_nopanic - allocate boot memory without panicking | |
729 | * @size: size of the request in bytes | |
730 | * @align: alignment of the region | |
731 | * @goal: preferred starting address of the region | |
732 | * | |
733 | * The goal is dropped if it can not be satisfied and the allocation will | |
734 | * fall back to memory below @goal. | |
735 | * | |
736 | * Allocation may happen on any node in the system. | |
737 | * | |
738 | * Returns NULL on failure. | |
739 | */ | |
bb0923a6 | 740 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
0f3caba2 | 741 | unsigned long goal) |
1da177e4 | 742 | { |
08677214 YL |
743 | unsigned long limit = 0; |
744 | ||
745 | #ifdef CONFIG_NO_BOOTMEM | |
746 | limit = -1UL; | |
747 | #endif | |
748 | ||
749 | return ___alloc_bootmem_nopanic(size, align, goal, limit); | |
0f3caba2 | 750 | } |
1da177e4 | 751 | |
0f3caba2 JW |
752 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
753 | unsigned long goal, unsigned long limit) | |
754 | { | |
755 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); | |
756 | ||
757 | if (mem) | |
758 | return mem; | |
759 | /* | |
760 | * Whoops, we cannot satisfy the allocation request. | |
761 | */ | |
762 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); | |
763 | panic("Out of memory"); | |
a8062231 AK |
764 | return NULL; |
765 | } | |
1da177e4 | 766 | |
a66fd7da JW |
767 | /** |
768 | * __alloc_bootmem - allocate boot memory | |
769 | * @size: size of the request in bytes | |
770 | * @align: alignment of the region | |
771 | * @goal: preferred starting address of the region | |
772 | * | |
773 | * The goal is dropped if it can not be satisfied and the allocation will | |
774 | * fall back to memory below @goal. | |
775 | * | |
776 | * Allocation may happen on any node in the system. | |
777 | * | |
778 | * The function panics if the request can not be satisfied. | |
779 | */ | |
bb0923a6 FBH |
780 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
781 | unsigned long goal) | |
a8062231 | 782 | { |
08677214 YL |
783 | unsigned long limit = 0; |
784 | ||
785 | #ifdef CONFIG_NO_BOOTMEM | |
786 | limit = -1UL; | |
787 | #endif | |
788 | ||
789 | return ___alloc_bootmem(size, align, goal, limit); | |
1da177e4 LT |
790 | } |
791 | ||
08677214 | 792 | #ifndef CONFIG_NO_BOOTMEM |
4cc278b7 JW |
793 | static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata, |
794 | unsigned long size, unsigned long align, | |
795 | unsigned long goal, unsigned long limit) | |
796 | { | |
797 | void *ptr; | |
798 | ||
d0c4f570 TH |
799 | ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit); |
800 | if (ptr) | |
801 | return ptr; | |
802 | ||
4cc278b7 JW |
803 | ptr = alloc_bootmem_core(bdata, size, align, goal, limit); |
804 | if (ptr) | |
805 | return ptr; | |
806 | ||
807 | return ___alloc_bootmem(size, align, goal, limit); | |
808 | } | |
08677214 | 809 | #endif |
4cc278b7 | 810 | |
a66fd7da JW |
811 | /** |
812 | * __alloc_bootmem_node - allocate boot memory from a specific node | |
813 | * @pgdat: node to allocate from | |
814 | * @size: size of the request in bytes | |
815 | * @align: alignment of the region | |
816 | * @goal: preferred starting address of the region | |
817 | * | |
818 | * The goal is dropped if it can not be satisfied and the allocation will | |
819 | * fall back to memory below @goal. | |
820 | * | |
821 | * Allocation may fall back to any node in the system if the specified node | |
822 | * can not hold the requested memory. | |
823 | * | |
824 | * The function panics if the request can not be satisfied. | |
825 | */ | |
bb0923a6 FBH |
826 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
827 | unsigned long align, unsigned long goal) | |
1da177e4 | 828 | { |
c91c4773 PE |
829 | if (WARN_ON_ONCE(slab_is_available())) |
830 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
831 | ||
08677214 YL |
832 | #ifdef CONFIG_NO_BOOTMEM |
833 | return __alloc_memory_core_early(pgdat->node_id, size, align, | |
834 | goal, -1ULL); | |
835 | #else | |
4cc278b7 | 836 | return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0); |
08677214 YL |
837 | #endif |
838 | } | |
839 | ||
840 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, | |
841 | unsigned long align, unsigned long goal) | |
842 | { | |
843 | #ifdef MAX_DMA32_PFN | |
844 | unsigned long end_pfn; | |
845 | ||
846 | if (WARN_ON_ONCE(slab_is_available())) | |
847 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
848 | ||
849 | /* update goal according ...MAX_DMA32_PFN */ | |
850 | end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
851 | ||
852 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && | |
853 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { | |
854 | void *ptr; | |
855 | unsigned long new_goal; | |
856 | ||
857 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; | |
858 | #ifdef CONFIG_NO_BOOTMEM | |
859 | ptr = __alloc_memory_core_early(pgdat->node_id, size, align, | |
860 | new_goal, -1ULL); | |
861 | #else | |
862 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, | |
863 | new_goal, 0); | |
864 | #endif | |
865 | if (ptr) | |
866 | return ptr; | |
867 | } | |
868 | #endif | |
869 | ||
870 | return __alloc_bootmem_node(pgdat, size, align, goal); | |
871 | ||
1da177e4 LT |
872 | } |
873 | ||
e70260aa | 874 | #ifdef CONFIG_SPARSEMEM |
a66fd7da JW |
875 | /** |
876 | * alloc_bootmem_section - allocate boot memory from a specific section | |
877 | * @size: size of the request in bytes | |
878 | * @section_nr: sparse map section to allocate from | |
879 | * | |
880 | * Return NULL on failure. | |
881 | */ | |
e70260aa YG |
882 | void * __init alloc_bootmem_section(unsigned long size, |
883 | unsigned long section_nr) | |
884 | { | |
08677214 YL |
885 | #ifdef CONFIG_NO_BOOTMEM |
886 | unsigned long pfn, goal, limit; | |
887 | ||
888 | pfn = section_nr_to_pfn(section_nr); | |
889 | goal = pfn << PAGE_SHIFT; | |
890 | limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT; | |
891 | ||
892 | return __alloc_memory_core_early(early_pfn_to_nid(pfn), size, | |
893 | SMP_CACHE_BYTES, goal, limit); | |
894 | #else | |
75a56cfe JW |
895 | bootmem_data_t *bdata; |
896 | unsigned long pfn, goal, limit; | |
e70260aa YG |
897 | |
898 | pfn = section_nr_to_pfn(section_nr); | |
75a56cfe JW |
899 | goal = pfn << PAGE_SHIFT; |
900 | limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT; | |
901 | bdata = &bootmem_node_data[early_pfn_to_nid(pfn)]; | |
e70260aa | 902 | |
75a56cfe | 903 | return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit); |
08677214 | 904 | #endif |
e70260aa YG |
905 | } |
906 | #endif | |
907 | ||
b54bbf7b AK |
908 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, |
909 | unsigned long align, unsigned long goal) | |
910 | { | |
911 | void *ptr; | |
912 | ||
c91c4773 PE |
913 | if (WARN_ON_ONCE(slab_is_available())) |
914 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
915 | ||
08677214 YL |
916 | #ifdef CONFIG_NO_BOOTMEM |
917 | ptr = __alloc_memory_core_early(pgdat->node_id, size, align, | |
918 | goal, -1ULL); | |
919 | #else | |
d0c4f570 TH |
920 | ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0); |
921 | if (ptr) | |
922 | return ptr; | |
923 | ||
b54bbf7b | 924 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0); |
08677214 | 925 | #endif |
b54bbf7b AK |
926 | if (ptr) |
927 | return ptr; | |
928 | ||
929 | return __alloc_bootmem_nopanic(size, align, goal); | |
930 | } | |
931 | ||
dfd54cbc HC |
932 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
933 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL | |
934 | #endif | |
008857c1 | 935 | |
a66fd7da JW |
936 | /** |
937 | * __alloc_bootmem_low - allocate low boot memory | |
938 | * @size: size of the request in bytes | |
939 | * @align: alignment of the region | |
940 | * @goal: preferred starting address of the region | |
941 | * | |
942 | * The goal is dropped if it can not be satisfied and the allocation will | |
943 | * fall back to memory below @goal. | |
944 | * | |
945 | * Allocation may happen on any node in the system. | |
946 | * | |
947 | * The function panics if the request can not be satisfied. | |
948 | */ | |
bb0923a6 FBH |
949 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
950 | unsigned long goal) | |
008857c1 | 951 | { |
0f3caba2 | 952 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
008857c1 RT |
953 | } |
954 | ||
a66fd7da JW |
955 | /** |
956 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node | |
957 | * @pgdat: node to allocate from | |
958 | * @size: size of the request in bytes | |
959 | * @align: alignment of the region | |
960 | * @goal: preferred starting address of the region | |
961 | * | |
962 | * The goal is dropped if it can not be satisfied and the allocation will | |
963 | * fall back to memory below @goal. | |
964 | * | |
965 | * Allocation may fall back to any node in the system if the specified node | |
966 | * can not hold the requested memory. | |
967 | * | |
968 | * The function panics if the request can not be satisfied. | |
969 | */ | |
008857c1 RT |
970 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
971 | unsigned long align, unsigned long goal) | |
972 | { | |
c91c4773 PE |
973 | if (WARN_ON_ONCE(slab_is_available())) |
974 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
975 | ||
08677214 YL |
976 | #ifdef CONFIG_NO_BOOTMEM |
977 | return __alloc_memory_core_early(pgdat->node_id, size, align, | |
978 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
979 | #else | |
4cc278b7 JW |
980 | return ___alloc_bootmem_node(pgdat->bdata, size, align, |
981 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
08677214 | 982 | #endif |
008857c1 | 983 | } |