]>
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) { | |
183 | #if 1 | |
184 | printk(KERN_DEBUG " %lx - %lx\n", start, end); | |
185 | #endif | |
186 | for (i = start; i < end; i++) | |
187 | __free_pages_bootmem(pfn_to_page(i), 0); | |
188 | ||
189 | return; | |
190 | } | |
191 | ||
192 | #if 1 | |
193 | printk(KERN_DEBUG " %lx %lx - %lx %lx\n", | |
194 | start, start_aligned, end_aligned, end); | |
195 | #endif | |
196 | for (i = start; i < start_aligned; i++) | |
197 | __free_pages_bootmem(pfn_to_page(i), 0); | |
198 | ||
199 | for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG) | |
200 | __free_pages_bootmem(pfn_to_page(i), order); | |
201 | ||
202 | for (i = end_aligned; i < end; i++) | |
203 | __free_pages_bootmem(pfn_to_page(i), 0); | |
204 | } | |
205 | ||
206 | unsigned long __init free_all_memory_core_early(int nodeid) | |
207 | { | |
208 | int i; | |
209 | u64 start, end; | |
210 | unsigned long count = 0; | |
211 | struct range *range = NULL; | |
212 | int nr_range; | |
213 | ||
214 | nr_range = get_free_all_memory_range(&range, nodeid); | |
215 | ||
216 | for (i = 0; i < nr_range; i++) { | |
217 | start = range[i].start; | |
218 | end = range[i].end; | |
219 | count += end - start; | |
220 | __free_pages_memory(start, end); | |
221 | } | |
222 | ||
223 | return count; | |
224 | } | |
225 | #else | |
223e8dc9 JW |
226 | static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) |
227 | { | |
41546c17 | 228 | int aligned; |
223e8dc9 | 229 | struct page *page; |
41546c17 JW |
230 | unsigned long start, end, pages, count = 0; |
231 | ||
232 | if (!bdata->node_bootmem_map) | |
233 | return 0; | |
234 | ||
3560e249 | 235 | start = bdata->node_min_pfn; |
41546c17 JW |
236 | end = bdata->node_low_pfn; |
237 | ||
223e8dc9 | 238 | /* |
41546c17 JW |
239 | * If the start is aligned to the machines wordsize, we might |
240 | * be able to free pages in bulks of that order. | |
223e8dc9 | 241 | */ |
41546c17 | 242 | aligned = !(start & (BITS_PER_LONG - 1)); |
223e8dc9 | 243 | |
41546c17 JW |
244 | bdebug("nid=%td start=%lx end=%lx aligned=%d\n", |
245 | bdata - bootmem_node_data, start, end, aligned); | |
223e8dc9 | 246 | |
41546c17 JW |
247 | while (start < end) { |
248 | unsigned long *map, idx, vec; | |
223e8dc9 | 249 | |
41546c17 | 250 | map = bdata->node_bootmem_map; |
3560e249 | 251 | idx = start - bdata->node_min_pfn; |
41546c17 JW |
252 | vec = ~map[idx / BITS_PER_LONG]; |
253 | ||
254 | if (aligned && vec == ~0UL && start + BITS_PER_LONG < end) { | |
255 | int order = ilog2(BITS_PER_LONG); | |
256 | ||
257 | __free_pages_bootmem(pfn_to_page(start), order); | |
223e8dc9 | 258 | count += BITS_PER_LONG; |
41546c17 JW |
259 | } else { |
260 | unsigned long off = 0; | |
261 | ||
262 | while (vec && off < BITS_PER_LONG) { | |
263 | if (vec & 1) { | |
264 | page = pfn_to_page(start + off); | |
223e8dc9 | 265 | __free_pages_bootmem(page, 0); |
41546c17 | 266 | count++; |
223e8dc9 | 267 | } |
41546c17 JW |
268 | vec >>= 1; |
269 | off++; | |
223e8dc9 | 270 | } |
223e8dc9 | 271 | } |
41546c17 | 272 | start += BITS_PER_LONG; |
223e8dc9 JW |
273 | } |
274 | ||
223e8dc9 | 275 | page = virt_to_page(bdata->node_bootmem_map); |
3560e249 | 276 | pages = bdata->node_low_pfn - bdata->node_min_pfn; |
41546c17 JW |
277 | pages = bootmem_bootmap_pages(pages); |
278 | count += pages; | |
279 | while (pages--) | |
280 | __free_pages_bootmem(page++, 0); | |
223e8dc9 | 281 | |
2e5237da JW |
282 | bdebug("nid=%td released=%lx\n", bdata - bootmem_node_data, count); |
283 | ||
223e8dc9 JW |
284 | return count; |
285 | } | |
08677214 | 286 | #endif |
223e8dc9 | 287 | |
a66fd7da JW |
288 | /** |
289 | * free_all_bootmem_node - release a node's free pages to the buddy allocator | |
290 | * @pgdat: node to be released | |
291 | * | |
292 | * Returns the number of pages actually released. | |
293 | */ | |
223e8dc9 JW |
294 | unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) |
295 | { | |
296 | register_page_bootmem_info_node(pgdat); | |
08677214 YL |
297 | #ifdef CONFIG_NO_BOOTMEM |
298 | /* free_all_memory_core_early(MAX_NUMNODES) will be called later */ | |
299 | return 0; | |
300 | #else | |
223e8dc9 | 301 | return free_all_bootmem_core(pgdat->bdata); |
08677214 | 302 | #endif |
223e8dc9 JW |
303 | } |
304 | ||
a66fd7da JW |
305 | /** |
306 | * free_all_bootmem - release free pages to the buddy allocator | |
307 | * | |
308 | * Returns the number of pages actually released. | |
309 | */ | |
223e8dc9 JW |
310 | unsigned long __init free_all_bootmem(void) |
311 | { | |
08677214 YL |
312 | #ifdef CONFIG_NO_BOOTMEM |
313 | return free_all_memory_core_early(NODE_DATA(0)->node_id); | |
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); | |
431 | #if 0 | |
432 | printk(KERN_DEBUG "free %lx %lx\n", physaddr, size); | |
433 | #endif | |
434 | #else | |
e2bf3cae JW |
435 | unsigned long start, end; |
436 | ||
ec3a354b CM |
437 | kmemleak_free_part(__va(physaddr), size); |
438 | ||
e2bf3cae JW |
439 | start = PFN_UP(physaddr); |
440 | end = PFN_DOWN(physaddr + size); | |
441 | ||
442 | mark_bootmem_node(pgdat->bdata, start, end, 0, 0); | |
08677214 | 443 | #endif |
223e8dc9 JW |
444 | } |
445 | ||
a66fd7da JW |
446 | /** |
447 | * free_bootmem - mark a page range as usable | |
448 | * @addr: starting address of the range | |
449 | * @size: size of the range in bytes | |
450 | * | |
451 | * Partial pages will be considered reserved and left as they are. | |
452 | * | |
e2bf3cae | 453 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 454 | */ |
223e8dc9 JW |
455 | void __init free_bootmem(unsigned long addr, unsigned long size) |
456 | { | |
08677214 YL |
457 | #ifdef CONFIG_NO_BOOTMEM |
458 | free_early(addr, addr + size); | |
459 | #if 0 | |
460 | printk(KERN_DEBUG "free %lx %lx\n", addr, size); | |
461 | #endif | |
462 | #else | |
e2bf3cae | 463 | unsigned long start, end; |
a5645a61 | 464 | |
ec3a354b CM |
465 | kmemleak_free_part(__va(addr), size); |
466 | ||
e2bf3cae JW |
467 | start = PFN_UP(addr); |
468 | end = PFN_DOWN(addr + size); | |
1da177e4 | 469 | |
e2bf3cae | 470 | mark_bootmem(start, end, 0, 0); |
08677214 | 471 | #endif |
1da177e4 LT |
472 | } |
473 | ||
a66fd7da JW |
474 | /** |
475 | * reserve_bootmem_node - mark a page range as reserved | |
476 | * @pgdat: node the range resides on | |
477 | * @physaddr: starting address of the range | |
478 | * @size: size of the range in bytes | |
479 | * @flags: reservation flags (see linux/bootmem.h) | |
480 | * | |
481 | * Partial pages will be reserved. | |
482 | * | |
e2bf3cae | 483 | * The range must reside completely on the specified node. |
a66fd7da | 484 | */ |
223e8dc9 JW |
485 | int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, |
486 | unsigned long size, int flags) | |
1da177e4 | 487 | { |
08677214 YL |
488 | #ifdef CONFIG_NO_BOOTMEM |
489 | panic("no bootmem"); | |
490 | return 0; | |
491 | #else | |
e2bf3cae | 492 | unsigned long start, end; |
1da177e4 | 493 | |
e2bf3cae JW |
494 | start = PFN_DOWN(physaddr); |
495 | end = PFN_UP(physaddr + size); | |
496 | ||
497 | return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); | |
08677214 | 498 | #endif |
223e8dc9 | 499 | } |
5a982cbc | 500 | |
a66fd7da JW |
501 | /** |
502 | * reserve_bootmem - mark a page range as usable | |
503 | * @addr: starting address of the range | |
504 | * @size: size of the range in bytes | |
505 | * @flags: reservation flags (see linux/bootmem.h) | |
506 | * | |
507 | * Partial pages will be reserved. | |
508 | * | |
e2bf3cae | 509 | * The range must be contiguous but may span node boundaries. |
a66fd7da | 510 | */ |
223e8dc9 JW |
511 | int __init reserve_bootmem(unsigned long addr, unsigned long size, |
512 | int flags) | |
513 | { | |
08677214 YL |
514 | #ifdef CONFIG_NO_BOOTMEM |
515 | panic("no bootmem"); | |
516 | return 0; | |
517 | #else | |
e2bf3cae | 518 | unsigned long start, end; |
1da177e4 | 519 | |
e2bf3cae JW |
520 | start = PFN_DOWN(addr); |
521 | end = PFN_UP(addr + size); | |
223e8dc9 | 522 | |
e2bf3cae | 523 | return mark_bootmem(start, end, 1, flags); |
08677214 | 524 | #endif |
1da177e4 LT |
525 | } |
526 | ||
08677214 | 527 | #ifndef CONFIG_NO_BOOTMEM |
8aa043d7 JB |
528 | static unsigned long __init align_idx(struct bootmem_data *bdata, |
529 | unsigned long idx, unsigned long step) | |
481ebd0d JW |
530 | { |
531 | unsigned long base = bdata->node_min_pfn; | |
532 | ||
533 | /* | |
534 | * Align the index with respect to the node start so that the | |
535 | * combination of both satisfies the requested alignment. | |
536 | */ | |
537 | ||
538 | return ALIGN(base + idx, step) - base; | |
539 | } | |
540 | ||
8aa043d7 JB |
541 | static unsigned long __init align_off(struct bootmem_data *bdata, |
542 | unsigned long off, unsigned long align) | |
481ebd0d JW |
543 | { |
544 | unsigned long base = PFN_PHYS(bdata->node_min_pfn); | |
545 | ||
546 | /* Same as align_idx for byte offsets */ | |
547 | ||
548 | return ALIGN(base + off, align) - base; | |
549 | } | |
550 | ||
d0c4f570 TH |
551 | static void * __init alloc_bootmem_core(struct bootmem_data *bdata, |
552 | unsigned long size, unsigned long align, | |
553 | unsigned long goal, unsigned long limit) | |
1da177e4 | 554 | { |
0f3caba2 | 555 | unsigned long fallback = 0; |
5f2809e6 JW |
556 | unsigned long min, max, start, sidx, midx, step; |
557 | ||
594fe1a0 JW |
558 | bdebug("nid=%td size=%lx [%lu pages] align=%lx goal=%lx limit=%lx\n", |
559 | bdata - bootmem_node_data, size, PAGE_ALIGN(size) >> PAGE_SHIFT, | |
560 | align, goal, limit); | |
561 | ||
5f2809e6 JW |
562 | BUG_ON(!size); |
563 | BUG_ON(align & (align - 1)); | |
564 | BUG_ON(limit && goal + size > limit); | |
1da177e4 | 565 | |
7c309a64 CK |
566 | if (!bdata->node_bootmem_map) |
567 | return NULL; | |
568 | ||
3560e249 | 569 | min = bdata->node_min_pfn; |
5f2809e6 | 570 | max = bdata->node_low_pfn; |
9a2dc04c | 571 | |
5f2809e6 JW |
572 | goal >>= PAGE_SHIFT; |
573 | limit >>= PAGE_SHIFT; | |
574 | ||
575 | if (limit && max > limit) | |
576 | max = limit; | |
577 | if (max <= min) | |
9a2dc04c YL |
578 | return NULL; |
579 | ||
5f2809e6 | 580 | step = max(align >> PAGE_SHIFT, 1UL); |
281dd25c | 581 | |
5f2809e6 JW |
582 | if (goal && min < goal && goal < max) |
583 | start = ALIGN(goal, step); | |
584 | else | |
585 | start = ALIGN(min, step); | |
1da177e4 | 586 | |
481ebd0d | 587 | sidx = start - bdata->node_min_pfn; |
3560e249 | 588 | midx = max - bdata->node_min_pfn; |
1da177e4 | 589 | |
5f2809e6 | 590 | if (bdata->hint_idx > sidx) { |
0f3caba2 JW |
591 | /* |
592 | * Handle the valid case of sidx being zero and still | |
593 | * catch the fallback below. | |
594 | */ | |
595 | fallback = sidx + 1; | |
481ebd0d | 596 | sidx = align_idx(bdata, bdata->hint_idx, step); |
5f2809e6 | 597 | } |
1da177e4 | 598 | |
5f2809e6 JW |
599 | while (1) { |
600 | int merge; | |
601 | void *region; | |
602 | unsigned long eidx, i, start_off, end_off; | |
603 | find_block: | |
604 | sidx = find_next_zero_bit(bdata->node_bootmem_map, midx, sidx); | |
481ebd0d | 605 | sidx = align_idx(bdata, sidx, step); |
5f2809e6 | 606 | eidx = sidx + PFN_UP(size); |
ad09315c | 607 | |
5f2809e6 | 608 | if (sidx >= midx || eidx > midx) |
66d43e98 | 609 | break; |
1da177e4 | 610 | |
5f2809e6 JW |
611 | for (i = sidx; i < eidx; i++) |
612 | if (test_bit(i, bdata->node_bootmem_map)) { | |
481ebd0d | 613 | sidx = align_idx(bdata, i, step); |
5f2809e6 JW |
614 | if (sidx == i) |
615 | sidx += step; | |
616 | goto find_block; | |
617 | } | |
1da177e4 | 618 | |
627240aa | 619 | if (bdata->last_end_off & (PAGE_SIZE - 1) && |
5f2809e6 | 620 | PFN_DOWN(bdata->last_end_off) + 1 == sidx) |
481ebd0d | 621 | start_off = align_off(bdata, bdata->last_end_off, align); |
5f2809e6 JW |
622 | else |
623 | start_off = PFN_PHYS(sidx); | |
624 | ||
625 | merge = PFN_DOWN(start_off) < sidx; | |
626 | end_off = start_off + size; | |
627 | ||
628 | bdata->last_end_off = end_off; | |
629 | bdata->hint_idx = PFN_UP(end_off); | |
630 | ||
631 | /* | |
632 | * Reserve the area now: | |
633 | */ | |
d747fa4b JW |
634 | if (__reserve(bdata, PFN_DOWN(start_off) + merge, |
635 | PFN_UP(end_off), BOOTMEM_EXCLUSIVE)) | |
636 | BUG(); | |
5f2809e6 | 637 | |
3560e249 JW |
638 | region = phys_to_virt(PFN_PHYS(bdata->node_min_pfn) + |
639 | start_off); | |
5f2809e6 | 640 | memset(region, 0, size); |
008139d9 CM |
641 | /* |
642 | * The min_count is set to 0 so that bootmem allocated blocks | |
643 | * are never reported as leaks. | |
644 | */ | |
645 | kmemleak_alloc(region, size, 0, 0); | |
5f2809e6 | 646 | return region; |
1da177e4 LT |
647 | } |
648 | ||
0f3caba2 | 649 | if (fallback) { |
481ebd0d | 650 | sidx = align_idx(bdata, fallback - 1, step); |
0f3caba2 JW |
651 | fallback = 0; |
652 | goto find_block; | |
653 | } | |
654 | ||
655 | return NULL; | |
656 | } | |
657 | ||
d0c4f570 TH |
658 | static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata, |
659 | unsigned long size, unsigned long align, | |
660 | unsigned long goal, unsigned long limit) | |
661 | { | |
441c7e0a PE |
662 | if (WARN_ON_ONCE(slab_is_available())) |
663 | return kzalloc(size, GFP_NOWAIT); | |
664 | ||
d0c4f570 | 665 | #ifdef CONFIG_HAVE_ARCH_BOOTMEM |
433f13a7 JP |
666 | { |
667 | bootmem_data_t *p_bdata; | |
668 | ||
669 | p_bdata = bootmem_arch_preferred_node(bdata, size, align, | |
670 | goal, limit); | |
671 | if (p_bdata) | |
672 | return alloc_bootmem_core(p_bdata, size, align, | |
673 | goal, limit); | |
674 | } | |
d0c4f570 TH |
675 | #endif |
676 | return NULL; | |
677 | } | |
08677214 | 678 | #endif |
d0c4f570 | 679 | |
0f3caba2 JW |
680 | static void * __init ___alloc_bootmem_nopanic(unsigned long size, |
681 | unsigned long align, | |
682 | unsigned long goal, | |
683 | unsigned long limit) | |
684 | { | |
08677214 YL |
685 | #ifdef CONFIG_NO_BOOTMEM |
686 | void *ptr; | |
687 | ||
688 | if (WARN_ON_ONCE(slab_is_available())) | |
689 | return kzalloc(size, GFP_NOWAIT); | |
690 | ||
691 | restart: | |
692 | ||
693 | ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit); | |
694 | ||
695 | if (ptr) | |
696 | return ptr; | |
697 | ||
698 | if (goal != 0) { | |
699 | goal = 0; | |
700 | goto restart; | |
701 | } | |
702 | ||
703 | return NULL; | |
704 | #else | |
0f3caba2 | 705 | bootmem_data_t *bdata; |
d0c4f570 | 706 | void *region; |
0f3caba2 JW |
707 | |
708 | restart: | |
d0c4f570 TH |
709 | region = alloc_arch_preferred_bootmem(NULL, size, align, goal, limit); |
710 | if (region) | |
711 | return region; | |
0f3caba2 | 712 | |
d0c4f570 | 713 | list_for_each_entry(bdata, &bdata_list, list) { |
0f3caba2 JW |
714 | if (goal && bdata->node_low_pfn <= PFN_DOWN(goal)) |
715 | continue; | |
3560e249 | 716 | if (limit && bdata->node_min_pfn >= PFN_DOWN(limit)) |
0f3caba2 JW |
717 | break; |
718 | ||
719 | region = alloc_bootmem_core(bdata, size, align, goal, limit); | |
720 | if (region) | |
721 | return region; | |
722 | } | |
723 | ||
5f2809e6 JW |
724 | if (goal) { |
725 | goal = 0; | |
0f3caba2 | 726 | goto restart; |
5f2809e6 | 727 | } |
2e5237da | 728 | |
5f2809e6 | 729 | return NULL; |
08677214 | 730 | #endif |
1da177e4 LT |
731 | } |
732 | ||
a66fd7da JW |
733 | /** |
734 | * __alloc_bootmem_nopanic - allocate boot memory without panicking | |
735 | * @size: size of the request in bytes | |
736 | * @align: alignment of the region | |
737 | * @goal: preferred starting address of the region | |
738 | * | |
739 | * The goal is dropped if it can not be satisfied and the allocation will | |
740 | * fall back to memory below @goal. | |
741 | * | |
742 | * Allocation may happen on any node in the system. | |
743 | * | |
744 | * Returns NULL on failure. | |
745 | */ | |
bb0923a6 | 746 | void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, |
0f3caba2 | 747 | unsigned long goal) |
1da177e4 | 748 | { |
08677214 YL |
749 | unsigned long limit = 0; |
750 | ||
751 | #ifdef CONFIG_NO_BOOTMEM | |
752 | limit = -1UL; | |
753 | #endif | |
754 | ||
755 | return ___alloc_bootmem_nopanic(size, align, goal, limit); | |
0f3caba2 | 756 | } |
1da177e4 | 757 | |
0f3caba2 JW |
758 | static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, |
759 | unsigned long goal, unsigned long limit) | |
760 | { | |
761 | void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit); | |
762 | ||
763 | if (mem) | |
764 | return mem; | |
765 | /* | |
766 | * Whoops, we cannot satisfy the allocation request. | |
767 | */ | |
768 | printk(KERN_ALERT "bootmem alloc of %lu bytes failed!\n", size); | |
769 | panic("Out of memory"); | |
a8062231 AK |
770 | return NULL; |
771 | } | |
1da177e4 | 772 | |
a66fd7da JW |
773 | /** |
774 | * __alloc_bootmem - allocate boot memory | |
775 | * @size: size of the request in bytes | |
776 | * @align: alignment of the region | |
777 | * @goal: preferred starting address of the region | |
778 | * | |
779 | * The goal is dropped if it can not be satisfied and the allocation will | |
780 | * fall back to memory below @goal. | |
781 | * | |
782 | * Allocation may happen on any node in the system. | |
783 | * | |
784 | * The function panics if the request can not be satisfied. | |
785 | */ | |
bb0923a6 FBH |
786 | void * __init __alloc_bootmem(unsigned long size, unsigned long align, |
787 | unsigned long goal) | |
a8062231 | 788 | { |
08677214 YL |
789 | unsigned long limit = 0; |
790 | ||
791 | #ifdef CONFIG_NO_BOOTMEM | |
792 | limit = -1UL; | |
793 | #endif | |
794 | ||
795 | return ___alloc_bootmem(size, align, goal, limit); | |
1da177e4 LT |
796 | } |
797 | ||
08677214 | 798 | #ifndef CONFIG_NO_BOOTMEM |
4cc278b7 JW |
799 | static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata, |
800 | unsigned long size, unsigned long align, | |
801 | unsigned long goal, unsigned long limit) | |
802 | { | |
803 | void *ptr; | |
804 | ||
d0c4f570 TH |
805 | ptr = alloc_arch_preferred_bootmem(bdata, size, align, goal, limit); |
806 | if (ptr) | |
807 | return ptr; | |
808 | ||
4cc278b7 JW |
809 | ptr = alloc_bootmem_core(bdata, size, align, goal, limit); |
810 | if (ptr) | |
811 | return ptr; | |
812 | ||
813 | return ___alloc_bootmem(size, align, goal, limit); | |
814 | } | |
08677214 | 815 | #endif |
4cc278b7 | 816 | |
a66fd7da JW |
817 | /** |
818 | * __alloc_bootmem_node - allocate boot memory from a specific node | |
819 | * @pgdat: node to allocate from | |
820 | * @size: size of the request in bytes | |
821 | * @align: alignment of the region | |
822 | * @goal: preferred starting address of the region | |
823 | * | |
824 | * The goal is dropped if it can not be satisfied and the allocation will | |
825 | * fall back to memory below @goal. | |
826 | * | |
827 | * Allocation may fall back to any node in the system if the specified node | |
828 | * can not hold the requested memory. | |
829 | * | |
830 | * The function panics if the request can not be satisfied. | |
831 | */ | |
bb0923a6 FBH |
832 | void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size, |
833 | unsigned long align, unsigned long goal) | |
1da177e4 | 834 | { |
c91c4773 PE |
835 | if (WARN_ON_ONCE(slab_is_available())) |
836 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
837 | ||
08677214 YL |
838 | #ifdef CONFIG_NO_BOOTMEM |
839 | return __alloc_memory_core_early(pgdat->node_id, size, align, | |
840 | goal, -1ULL); | |
841 | #else | |
4cc278b7 | 842 | return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0); |
08677214 YL |
843 | #endif |
844 | } | |
845 | ||
846 | void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size, | |
847 | unsigned long align, unsigned long goal) | |
848 | { | |
849 | #ifdef MAX_DMA32_PFN | |
850 | unsigned long end_pfn; | |
851 | ||
852 | if (WARN_ON_ONCE(slab_is_available())) | |
853 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
854 | ||
855 | /* update goal according ...MAX_DMA32_PFN */ | |
856 | end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages; | |
857 | ||
858 | if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) && | |
859 | (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) { | |
860 | void *ptr; | |
861 | unsigned long new_goal; | |
862 | ||
863 | new_goal = MAX_DMA32_PFN << PAGE_SHIFT; | |
864 | #ifdef CONFIG_NO_BOOTMEM | |
865 | ptr = __alloc_memory_core_early(pgdat->node_id, size, align, | |
866 | new_goal, -1ULL); | |
867 | #else | |
868 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, | |
869 | new_goal, 0); | |
870 | #endif | |
871 | if (ptr) | |
872 | return ptr; | |
873 | } | |
874 | #endif | |
875 | ||
876 | return __alloc_bootmem_node(pgdat, size, align, goal); | |
877 | ||
1da177e4 LT |
878 | } |
879 | ||
e70260aa | 880 | #ifdef CONFIG_SPARSEMEM |
a66fd7da JW |
881 | /** |
882 | * alloc_bootmem_section - allocate boot memory from a specific section | |
883 | * @size: size of the request in bytes | |
884 | * @section_nr: sparse map section to allocate from | |
885 | * | |
886 | * Return NULL on failure. | |
887 | */ | |
e70260aa YG |
888 | void * __init alloc_bootmem_section(unsigned long size, |
889 | unsigned long section_nr) | |
890 | { | |
08677214 YL |
891 | #ifdef CONFIG_NO_BOOTMEM |
892 | unsigned long pfn, goal, limit; | |
893 | ||
894 | pfn = section_nr_to_pfn(section_nr); | |
895 | goal = pfn << PAGE_SHIFT; | |
896 | limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT; | |
897 | ||
898 | return __alloc_memory_core_early(early_pfn_to_nid(pfn), size, | |
899 | SMP_CACHE_BYTES, goal, limit); | |
900 | #else | |
75a56cfe JW |
901 | bootmem_data_t *bdata; |
902 | unsigned long pfn, goal, limit; | |
e70260aa YG |
903 | |
904 | pfn = section_nr_to_pfn(section_nr); | |
75a56cfe JW |
905 | goal = pfn << PAGE_SHIFT; |
906 | limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT; | |
907 | bdata = &bootmem_node_data[early_pfn_to_nid(pfn)]; | |
e70260aa | 908 | |
75a56cfe | 909 | return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit); |
08677214 | 910 | #endif |
e70260aa YG |
911 | } |
912 | #endif | |
913 | ||
b54bbf7b AK |
914 | void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size, |
915 | unsigned long align, unsigned long goal) | |
916 | { | |
917 | void *ptr; | |
918 | ||
c91c4773 PE |
919 | if (WARN_ON_ONCE(slab_is_available())) |
920 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
921 | ||
08677214 YL |
922 | #ifdef CONFIG_NO_BOOTMEM |
923 | ptr = __alloc_memory_core_early(pgdat->node_id, size, align, | |
924 | goal, -1ULL); | |
925 | #else | |
d0c4f570 TH |
926 | ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0); |
927 | if (ptr) | |
928 | return ptr; | |
929 | ||
b54bbf7b | 930 | ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0); |
08677214 | 931 | #endif |
b54bbf7b AK |
932 | if (ptr) |
933 | return ptr; | |
934 | ||
935 | return __alloc_bootmem_nopanic(size, align, goal); | |
936 | } | |
937 | ||
dfd54cbc HC |
938 | #ifndef ARCH_LOW_ADDRESS_LIMIT |
939 | #define ARCH_LOW_ADDRESS_LIMIT 0xffffffffUL | |
940 | #endif | |
008857c1 | 941 | |
a66fd7da JW |
942 | /** |
943 | * __alloc_bootmem_low - allocate low boot memory | |
944 | * @size: size of the request in bytes | |
945 | * @align: alignment of the region | |
946 | * @goal: preferred starting address of the region | |
947 | * | |
948 | * The goal is dropped if it can not be satisfied and the allocation will | |
949 | * fall back to memory below @goal. | |
950 | * | |
951 | * Allocation may happen on any node in the system. | |
952 | * | |
953 | * The function panics if the request can not be satisfied. | |
954 | */ | |
bb0923a6 FBH |
955 | void * __init __alloc_bootmem_low(unsigned long size, unsigned long align, |
956 | unsigned long goal) | |
008857c1 | 957 | { |
0f3caba2 | 958 | return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT); |
008857c1 RT |
959 | } |
960 | ||
a66fd7da JW |
961 | /** |
962 | * __alloc_bootmem_low_node - allocate low boot memory from a specific node | |
963 | * @pgdat: node to allocate from | |
964 | * @size: size of the request in bytes | |
965 | * @align: alignment of the region | |
966 | * @goal: preferred starting address of the region | |
967 | * | |
968 | * The goal is dropped if it can not be satisfied and the allocation will | |
969 | * fall back to memory below @goal. | |
970 | * | |
971 | * Allocation may fall back to any node in the system if the specified node | |
972 | * can not hold the requested memory. | |
973 | * | |
974 | * The function panics if the request can not be satisfied. | |
975 | */ | |
008857c1 RT |
976 | void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size, |
977 | unsigned long align, unsigned long goal) | |
978 | { | |
c91c4773 PE |
979 | if (WARN_ON_ONCE(slab_is_available())) |
980 | return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); | |
981 | ||
08677214 YL |
982 | #ifdef CONFIG_NO_BOOTMEM |
983 | return __alloc_memory_core_early(pgdat->node_id, size, align, | |
984 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
985 | #else | |
4cc278b7 JW |
986 | return ___alloc_bootmem_node(pgdat->bdata, size, align, |
987 | goal, ARCH_LOW_ADDRESS_LIMIT); | |
08677214 | 988 | #endif |
008857c1 | 989 | } |