Commit | Line | Data |
---|---|---|
95f72d1e YL |
1 | /* |
2 | * Procedures for maintaining information about logical memory blocks. | |
3 | * | |
4 | * Peter Bergner, IBM Corp. June 2001. | |
5 | * Copyright (C) 2001 Peter Bergner. | |
6 | * | |
7 | * This program is free software; you can redistribute it and/or | |
8 | * modify it under the terms of the GNU General Public License | |
9 | * as published by the Free Software Foundation; either version | |
10 | * 2 of the License, or (at your option) any later version. | |
11 | */ | |
12 | ||
13 | #include <linux/kernel.h> | |
142b45a7 | 14 | #include <linux/slab.h> |
95f72d1e YL |
15 | #include <linux/init.h> |
16 | #include <linux/bitops.h> | |
449e8df3 | 17 | #include <linux/poison.h> |
c196f76f | 18 | #include <linux/pfn.h> |
6d03b885 BH |
19 | #include <linux/debugfs.h> |
20 | #include <linux/seq_file.h> | |
95f72d1e YL |
21 | #include <linux/memblock.h> |
22 | ||
10d06439 | 23 | struct memblock memblock __initdata_memblock; |
95f72d1e | 24 | |
10d06439 YL |
25 | int memblock_debug __initdata_memblock; |
26 | int memblock_can_resize __initdata_memblock; | |
27 | static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS + 1] __initdata_memblock; | |
28 | static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS + 1] __initdata_memblock; | |
95f72d1e | 29 | |
142b45a7 BH |
30 | /* inline so we don't get a warning when pr_debug is compiled out */ |
31 | static inline const char *memblock_type_name(struct memblock_type *type) | |
32 | { | |
33 | if (type == &memblock.memory) | |
34 | return "memory"; | |
35 | else if (type == &memblock.reserved) | |
36 | return "reserved"; | |
37 | else | |
38 | return "unknown"; | |
39 | } | |
40 | ||
6ed311b2 BH |
41 | /* |
42 | * Address comparison utilities | |
43 | */ | |
10d06439 | 44 | static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, |
2898cc4c | 45 | phys_addr_t base2, phys_addr_t size2) |
95f72d1e YL |
46 | { |
47 | return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); | |
48 | } | |
49 | ||
10d06439 | 50 | long __init_memblock memblock_overlaps_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size) |
6ed311b2 BH |
51 | { |
52 | unsigned long i; | |
53 | ||
54 | for (i = 0; i < type->cnt; i++) { | |
55 | phys_addr_t rgnbase = type->regions[i].base; | |
56 | phys_addr_t rgnsize = type->regions[i].size; | |
57 | if (memblock_addrs_overlap(base, size, rgnbase, rgnsize)) | |
58 | break; | |
59 | } | |
60 | ||
61 | return (i < type->cnt) ? i : -1; | |
62 | } | |
63 | ||
64 | /* | |
65 | * Find, allocate, deallocate or reserve unreserved regions. All allocations | |
66 | * are top-down. | |
67 | */ | |
68 | ||
cd79481d | 69 | static phys_addr_t __init_memblock memblock_find_region(phys_addr_t start, phys_addr_t end, |
6ed311b2 BH |
70 | phys_addr_t size, phys_addr_t align) |
71 | { | |
72 | phys_addr_t base, res_base; | |
73 | long j; | |
74 | ||
f1af98c7 YL |
75 | /* In case, huge size is requested */ |
76 | if (end < size) | |
77 | return MEMBLOCK_ERROR; | |
78 | ||
348968eb | 79 | base = round_down(end - size, align); |
f1af98c7 | 80 | |
25818f0f BH |
81 | /* Prevent allocations returning 0 as it's also used to |
82 | * indicate an allocation failure | |
83 | */ | |
84 | if (start == 0) | |
85 | start = PAGE_SIZE; | |
86 | ||
6ed311b2 BH |
87 | while (start <= base) { |
88 | j = memblock_overlaps_region(&memblock.reserved, base, size); | |
89 | if (j < 0) | |
90 | return base; | |
91 | res_base = memblock.reserved.regions[j].base; | |
92 | if (res_base < size) | |
93 | break; | |
348968eb | 94 | base = round_down(res_base - size, align); |
6ed311b2 BH |
95 | } |
96 | ||
97 | return MEMBLOCK_ERROR; | |
98 | } | |
99 | ||
3661ca66 YL |
100 | static phys_addr_t __init_memblock memblock_find_base(phys_addr_t size, |
101 | phys_addr_t align, phys_addr_t start, phys_addr_t end) | |
6ed311b2 BH |
102 | { |
103 | long i; | |
6ed311b2 BH |
104 | |
105 | BUG_ON(0 == size); | |
106 | ||
6ed311b2 | 107 | /* Pump up max_addr */ |
fef501d4 BH |
108 | if (end == MEMBLOCK_ALLOC_ACCESSIBLE) |
109 | end = memblock.current_limit; | |
6ed311b2 BH |
110 | |
111 | /* We do a top-down search, this tends to limit memory | |
112 | * fragmentation by keeping early boot allocs near the | |
113 | * top of memory | |
114 | */ | |
115 | for (i = memblock.memory.cnt - 1; i >= 0; i--) { | |
116 | phys_addr_t memblockbase = memblock.memory.regions[i].base; | |
117 | phys_addr_t memblocksize = memblock.memory.regions[i].size; | |
fef501d4 | 118 | phys_addr_t bottom, top, found; |
6ed311b2 BH |
119 | |
120 | if (memblocksize < size) | |
121 | continue; | |
fef501d4 BH |
122 | if ((memblockbase + memblocksize) <= start) |
123 | break; | |
124 | bottom = max(memblockbase, start); | |
125 | top = min(memblockbase + memblocksize, end); | |
126 | if (bottom >= top) | |
127 | continue; | |
128 | found = memblock_find_region(bottom, top, size, align); | |
129 | if (found != MEMBLOCK_ERROR) | |
130 | return found; | |
6ed311b2 BH |
131 | } |
132 | return MEMBLOCK_ERROR; | |
133 | } | |
134 | ||
5303b68f YL |
135 | /* |
136 | * Find a free area with specified alignment in a specific range. | |
137 | */ | |
138 | u64 __init_memblock memblock_find_in_range(u64 start, u64 end, u64 size, u64 align) | |
139 | { | |
140 | return memblock_find_base(size, align, start, end); | |
141 | } | |
142 | ||
7950c407 YL |
143 | /* |
144 | * Free memblock.reserved.regions | |
145 | */ | |
146 | int __init_memblock memblock_free_reserved_regions(void) | |
147 | { | |
148 | if (memblock.reserved.regions == memblock_reserved_init_regions) | |
149 | return 0; | |
150 | ||
151 | return memblock_free(__pa(memblock.reserved.regions), | |
152 | sizeof(struct memblock_region) * memblock.reserved.max); | |
153 | } | |
154 | ||
155 | /* | |
156 | * Reserve memblock.reserved.regions | |
157 | */ | |
158 | int __init_memblock memblock_reserve_reserved_regions(void) | |
159 | { | |
160 | if (memblock.reserved.regions == memblock_reserved_init_regions) | |
161 | return 0; | |
162 | ||
163 | return memblock_reserve(__pa(memblock.reserved.regions), | |
164 | sizeof(struct memblock_region) * memblock.reserved.max); | |
165 | } | |
166 | ||
10d06439 | 167 | static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) |
95f72d1e YL |
168 | { |
169 | unsigned long i; | |
170 | ||
e3239ff9 BH |
171 | for (i = r; i < type->cnt - 1; i++) { |
172 | type->regions[i].base = type->regions[i + 1].base; | |
173 | type->regions[i].size = type->regions[i + 1].size; | |
95f72d1e | 174 | } |
e3239ff9 | 175 | type->cnt--; |
95f72d1e | 176 | |
8f7a6605 BH |
177 | /* Special case for empty arrays */ |
178 | if (type->cnt == 0) { | |
179 | type->cnt = 1; | |
180 | type->regions[0].base = 0; | |
181 | type->regions[0].size = 0; | |
182 | } | |
95f72d1e YL |
183 | } |
184 | ||
142b45a7 BH |
185 | /* Defined below but needed now */ |
186 | static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size); | |
187 | ||
10d06439 | 188 | static int __init_memblock memblock_double_array(struct memblock_type *type) |
142b45a7 BH |
189 | { |
190 | struct memblock_region *new_array, *old_array; | |
191 | phys_addr_t old_size, new_size, addr; | |
192 | int use_slab = slab_is_available(); | |
193 | ||
194 | /* We don't allow resizing until we know about the reserved regions | |
195 | * of memory that aren't suitable for allocation | |
196 | */ | |
197 | if (!memblock_can_resize) | |
198 | return -1; | |
199 | ||
142b45a7 BH |
200 | /* Calculate new doubled size */ |
201 | old_size = type->max * sizeof(struct memblock_region); | |
202 | new_size = old_size << 1; | |
203 | ||
204 | /* Try to find some space for it. | |
205 | * | |
206 | * WARNING: We assume that either slab_is_available() and we use it or | |
207 | * we use MEMBLOCK for allocations. That means that this is unsafe to use | |
208 | * when bootmem is currently active (unless bootmem itself is implemented | |
209 | * on top of MEMBLOCK which isn't the case yet) | |
210 | * | |
211 | * This should however not be an issue for now, as we currently only | |
212 | * call into MEMBLOCK while it's still active, or much later when slab is | |
213 | * active for memory hotplug operations | |
214 | */ | |
215 | if (use_slab) { | |
216 | new_array = kmalloc(new_size, GFP_KERNEL); | |
217 | addr = new_array == NULL ? MEMBLOCK_ERROR : __pa(new_array); | |
218 | } else | |
fef501d4 | 219 | addr = memblock_find_base(new_size, sizeof(phys_addr_t), 0, MEMBLOCK_ALLOC_ACCESSIBLE); |
142b45a7 BH |
220 | if (addr == MEMBLOCK_ERROR) { |
221 | pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", | |
222 | memblock_type_name(type), type->max, type->max * 2); | |
223 | return -1; | |
224 | } | |
225 | new_array = __va(addr); | |
226 | ||
ea9e4376 YL |
227 | memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]", |
228 | memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1); | |
229 | ||
142b45a7 BH |
230 | /* Found space, we now need to move the array over before |
231 | * we add the reserved region since it may be our reserved | |
232 | * array itself that is full. | |
233 | */ | |
234 | memcpy(new_array, type->regions, old_size); | |
235 | memset(new_array + type->max, 0, old_size); | |
236 | old_array = type->regions; | |
237 | type->regions = new_array; | |
238 | type->max <<= 1; | |
239 | ||
240 | /* If we use SLAB that's it, we are done */ | |
241 | if (use_slab) | |
242 | return 0; | |
243 | ||
244 | /* Add the new reserved region now. Should not fail ! */ | |
8f7a6605 | 245 | BUG_ON(memblock_add_region(&memblock.reserved, addr, new_size)); |
142b45a7 BH |
246 | |
247 | /* If the array wasn't our static init one, then free it. We only do | |
248 | * that before SLAB is available as later on, we don't know whether | |
249 | * to use kfree or free_bootmem_pages(). Shouldn't be a big deal | |
250 | * anyways | |
251 | */ | |
252 | if (old_array != memblock_memory_init_regions && | |
253 | old_array != memblock_reserved_init_regions) | |
254 | memblock_free(__pa(old_array), old_size); | |
255 | ||
256 | return 0; | |
257 | } | |
258 | ||
10d06439 | 259 | extern int __init_memblock __weak memblock_memory_can_coalesce(phys_addr_t addr1, phys_addr_t size1, |
d2cd563b BH |
260 | phys_addr_t addr2, phys_addr_t size2) |
261 | { | |
262 | return 1; | |
263 | } | |
264 | ||
8f7a6605 BH |
265 | static long __init_memblock memblock_add_region(struct memblock_type *type, |
266 | phys_addr_t base, phys_addr_t size) | |
95f72d1e | 267 | { |
8f7a6605 BH |
268 | phys_addr_t end = base + size; |
269 | int i, slot = -1; | |
95f72d1e | 270 | |
8f7a6605 | 271 | /* First try and coalesce this MEMBLOCK with others */ |
e3239ff9 | 272 | for (i = 0; i < type->cnt; i++) { |
8f7a6605 BH |
273 | struct memblock_region *rgn = &type->regions[i]; |
274 | phys_addr_t rend = rgn->base + rgn->size; | |
275 | ||
276 | /* Exit if there's no possible hits */ | |
277 | if (rgn->base > end || rgn->size == 0) | |
278 | break; | |
95f72d1e | 279 | |
8f7a6605 BH |
280 | /* Check if we are fully enclosed within an existing |
281 | * block | |
282 | */ | |
283 | if (rgn->base <= base && rend >= end) | |
95f72d1e YL |
284 | return 0; |
285 | ||
8f7a6605 BH |
286 | /* Check if we overlap or are adjacent with the bottom |
287 | * of a block. | |
288 | */ | |
289 | if (base < rgn->base && end >= rgn->base) { | |
290 | /* If we can't coalesce, create a new block */ | |
291 | if (!memblock_memory_can_coalesce(base, size, | |
292 | rgn->base, | |
293 | rgn->size)) { | |
294 | /* Overlap & can't coalesce are mutually | |
295 | * exclusive, if you do that, be prepared | |
296 | * for trouble | |
297 | */ | |
298 | WARN_ON(end != rgn->base); | |
299 | goto new_block; | |
300 | } | |
301 | /* We extend the bottom of the block down to our | |
302 | * base | |
303 | */ | |
304 | rgn->base = base; | |
305 | rgn->size = rend - base; | |
306 | ||
307 | /* Return if we have nothing else to allocate | |
308 | * (fully coalesced) | |
309 | */ | |
310 | if (rend >= end) | |
311 | return 0; | |
312 | ||
313 | /* We continue processing from the end of the | |
314 | * coalesced block. | |
315 | */ | |
316 | base = rend; | |
317 | size = end - base; | |
318 | } | |
319 | ||
320 | /* Now check if we overlap or are adjacent with the | |
321 | * top of a block | |
322 | */ | |
323 | if (base <= rend && end >= rend) { | |
324 | /* If we can't coalesce, create a new block */ | |
325 | if (!memblock_memory_can_coalesce(rgn->base, | |
326 | rgn->size, | |
327 | base, size)) { | |
328 | /* Overlap & can't coalesce are mutually | |
329 | * exclusive, if you do that, be prepared | |
330 | * for trouble | |
331 | */ | |
332 | WARN_ON(rend != base); | |
333 | goto new_block; | |
334 | } | |
335 | /* We adjust our base down to enclose the | |
336 | * original block and destroy it. It will be | |
337 | * part of our new allocation. Since we've | |
338 | * freed an entry, we know we won't fail | |
339 | * to allocate one later, so we won't risk | |
340 | * losing the original block allocation. | |
341 | */ | |
342 | size += (base - rgn->base); | |
343 | base = rgn->base; | |
344 | memblock_remove_region(type, i--); | |
95f72d1e YL |
345 | } |
346 | } | |
347 | ||
8f7a6605 BH |
348 | /* If the array is empty, special case, replace the fake |
349 | * filler region and return | |
d2cd563b | 350 | */ |
8f7a6605 BH |
351 | if ((type->cnt == 1) && (type->regions[0].size == 0)) { |
352 | type->regions[0].base = base; | |
353 | type->regions[0].size = size; | |
354 | return 0; | |
95f72d1e YL |
355 | } |
356 | ||
8f7a6605 | 357 | new_block: |
142b45a7 BH |
358 | /* If we are out of space, we fail. It's too late to resize the array |
359 | * but then this shouldn't have happened in the first place. | |
360 | */ | |
361 | if (WARN_ON(type->cnt >= type->max)) | |
95f72d1e YL |
362 | return -1; |
363 | ||
364 | /* Couldn't coalesce the MEMBLOCK, so add it to the sorted table. */ | |
e3239ff9 BH |
365 | for (i = type->cnt - 1; i >= 0; i--) { |
366 | if (base < type->regions[i].base) { | |
367 | type->regions[i+1].base = type->regions[i].base; | |
368 | type->regions[i+1].size = type->regions[i].size; | |
95f72d1e | 369 | } else { |
e3239ff9 BH |
370 | type->regions[i+1].base = base; |
371 | type->regions[i+1].size = size; | |
8f7a6605 | 372 | slot = i + 1; |
95f72d1e YL |
373 | break; |
374 | } | |
375 | } | |
e3239ff9 BH |
376 | if (base < type->regions[0].base) { |
377 | type->regions[0].base = base; | |
378 | type->regions[0].size = size; | |
8f7a6605 | 379 | slot = 0; |
95f72d1e | 380 | } |
e3239ff9 | 381 | type->cnt++; |
95f72d1e | 382 | |
142b45a7 BH |
383 | /* The array is full ? Try to resize it. If that fails, we undo |
384 | * our allocation and return an error | |
385 | */ | |
386 | if (type->cnt == type->max && memblock_double_array(type)) { | |
8f7a6605 BH |
387 | BUG_ON(slot < 0); |
388 | memblock_remove_region(type, slot); | |
142b45a7 BH |
389 | return -1; |
390 | } | |
391 | ||
95f72d1e YL |
392 | return 0; |
393 | } | |
394 | ||
10d06439 | 395 | long __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) |
95f72d1e | 396 | { |
e3239ff9 | 397 | return memblock_add_region(&memblock.memory, base, size); |
95f72d1e YL |
398 | |
399 | } | |
400 | ||
8f7a6605 BH |
401 | static long __init_memblock __memblock_remove(struct memblock_type *type, |
402 | phys_addr_t base, phys_addr_t size) | |
95f72d1e | 403 | { |
2898cc4c | 404 | phys_addr_t end = base + size; |
95f72d1e YL |
405 | int i; |
406 | ||
8f7a6605 BH |
407 | /* Walk through the array for collisions */ |
408 | for (i = 0; i < type->cnt; i++) { | |
409 | struct memblock_region *rgn = &type->regions[i]; | |
410 | phys_addr_t rend = rgn->base + rgn->size; | |
95f72d1e | 411 | |
8f7a6605 BH |
412 | /* Nothing more to do, exit */ |
413 | if (rgn->base > end || rgn->size == 0) | |
95f72d1e | 414 | break; |
95f72d1e | 415 | |
8f7a6605 BH |
416 | /* If we fully enclose the block, drop it */ |
417 | if (base <= rgn->base && end >= rend) { | |
418 | memblock_remove_region(type, i--); | |
419 | continue; | |
420 | } | |
95f72d1e | 421 | |
8f7a6605 BH |
422 | /* If we are fully enclosed within a block |
423 | * then we need to split it and we are done | |
424 | */ | |
425 | if (base > rgn->base && end < rend) { | |
426 | rgn->size = base - rgn->base; | |
427 | if (!memblock_add_region(type, end, rend - end)) | |
428 | return 0; | |
429 | /* Failure to split is bad, we at least | |
430 | * restore the block before erroring | |
431 | */ | |
432 | rgn->size = rend - rgn->base; | |
433 | WARN_ON(1); | |
434 | return -1; | |
435 | } | |
95f72d1e | 436 | |
8f7a6605 BH |
437 | /* Check if we need to trim the bottom of a block */ |
438 | if (rgn->base < end && rend > end) { | |
439 | rgn->size -= end - rgn->base; | |
440 | rgn->base = end; | |
441 | break; | |
442 | } | |
95f72d1e | 443 | |
8f7a6605 BH |
444 | /* And check if we need to trim the top of a block */ |
445 | if (base < rend) | |
446 | rgn->size -= rend - base; | |
95f72d1e | 447 | |
8f7a6605 BH |
448 | } |
449 | return 0; | |
95f72d1e YL |
450 | } |
451 | ||
10d06439 | 452 | long __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) |
95f72d1e YL |
453 | { |
454 | return __memblock_remove(&memblock.memory, base, size); | |
455 | } | |
456 | ||
3661ca66 | 457 | long __init_memblock memblock_free(phys_addr_t base, phys_addr_t size) |
95f72d1e YL |
458 | { |
459 | return __memblock_remove(&memblock.reserved, base, size); | |
460 | } | |
461 | ||
3661ca66 | 462 | long __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) |
95f72d1e | 463 | { |
e3239ff9 | 464 | struct memblock_type *_rgn = &memblock.reserved; |
95f72d1e YL |
465 | |
466 | BUG_ON(0 == size); | |
467 | ||
468 | return memblock_add_region(_rgn, base, size); | |
469 | } | |
470 | ||
6ed311b2 | 471 | phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) |
95f72d1e | 472 | { |
6ed311b2 | 473 | phys_addr_t found; |
95f72d1e | 474 | |
6ed311b2 BH |
475 | /* We align the size to limit fragmentation. Without this, a lot of |
476 | * small allocs quickly eat up the whole reserve array on sparc | |
477 | */ | |
348968eb | 478 | size = round_up(size, align); |
95f72d1e | 479 | |
fef501d4 | 480 | found = memblock_find_base(size, align, 0, max_addr); |
6ed311b2 | 481 | if (found != MEMBLOCK_ERROR && |
8f7a6605 | 482 | !memblock_add_region(&memblock.reserved, found, size)) |
6ed311b2 | 483 | return found; |
95f72d1e | 484 | |
6ed311b2 | 485 | return 0; |
95f72d1e YL |
486 | } |
487 | ||
6ed311b2 | 488 | phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) |
95f72d1e | 489 | { |
6ed311b2 BH |
490 | phys_addr_t alloc; |
491 | ||
492 | alloc = __memblock_alloc_base(size, align, max_addr); | |
493 | ||
494 | if (alloc == 0) | |
495 | panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n", | |
496 | (unsigned long long) size, (unsigned long long) max_addr); | |
497 | ||
498 | return alloc; | |
95f72d1e YL |
499 | } |
500 | ||
6ed311b2 | 501 | phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align) |
95f72d1e | 502 | { |
6ed311b2 BH |
503 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
504 | } | |
95f72d1e | 505 | |
95f72d1e | 506 | |
6ed311b2 BH |
507 | /* |
508 | * Additional node-local allocators. Search for node memory is bottom up | |
509 | * and walks memblock regions within that node bottom-up as well, but allocation | |
c196f76f BH |
510 | * within an memblock region is top-down. XXX I plan to fix that at some stage |
511 | * | |
512 | * WARNING: Only available after early_node_map[] has been populated, | |
513 | * on some architectures, that is after all the calls to add_active_range() | |
514 | * have been done to populate it. | |
6ed311b2 | 515 | */ |
95f72d1e | 516 | |
2898cc4c | 517 | phys_addr_t __weak __init memblock_nid_range(phys_addr_t start, phys_addr_t end, int *nid) |
c3f72b57 | 518 | { |
c196f76f BH |
519 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
520 | /* | |
521 | * This code originates from sparc which really wants use to walk by addresses | |
522 | * and returns the nid. This is not very convenient for early_pfn_map[] users | |
523 | * as the map isn't sorted yet, and it really wants to be walked by nid. | |
524 | * | |
525 | * For now, I implement the inefficient method below which walks the early | |
526 | * map multiple times. Eventually we may want to use an ARCH config option | |
527 | * to implement a completely different method for both case. | |
528 | */ | |
529 | unsigned long start_pfn, end_pfn; | |
530 | int i; | |
531 | ||
532 | for (i = 0; i < MAX_NUMNODES; i++) { | |
533 | get_pfn_range_for_nid(i, &start_pfn, &end_pfn); | |
534 | if (start < PFN_PHYS(start_pfn) || start >= PFN_PHYS(end_pfn)) | |
535 | continue; | |
536 | *nid = i; | |
537 | return min(end, PFN_PHYS(end_pfn)); | |
538 | } | |
539 | #endif | |
c3f72b57 BH |
540 | *nid = 0; |
541 | ||
542 | return end; | |
543 | } | |
544 | ||
2898cc4c BH |
545 | static phys_addr_t __init memblock_alloc_nid_region(struct memblock_region *mp, |
546 | phys_addr_t size, | |
547 | phys_addr_t align, int nid) | |
95f72d1e | 548 | { |
2898cc4c | 549 | phys_addr_t start, end; |
95f72d1e YL |
550 | |
551 | start = mp->base; | |
552 | end = start + mp->size; | |
553 | ||
348968eb | 554 | start = round_up(start, align); |
95f72d1e | 555 | while (start < end) { |
2898cc4c | 556 | phys_addr_t this_end; |
95f72d1e YL |
557 | int this_nid; |
558 | ||
35a1f0bd | 559 | this_end = memblock_nid_range(start, end, &this_nid); |
95f72d1e | 560 | if (this_nid == nid) { |
3a9c2c81 | 561 | phys_addr_t ret = memblock_find_region(start, this_end, size, align); |
4d629f9a | 562 | if (ret != MEMBLOCK_ERROR && |
8f7a6605 | 563 | !memblock_add_region(&memblock.reserved, ret, size)) |
95f72d1e YL |
564 | return ret; |
565 | } | |
566 | start = this_end; | |
567 | } | |
568 | ||
4d629f9a | 569 | return MEMBLOCK_ERROR; |
95f72d1e YL |
570 | } |
571 | ||
2898cc4c | 572 | phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid) |
95f72d1e | 573 | { |
e3239ff9 | 574 | struct memblock_type *mem = &memblock.memory; |
95f72d1e YL |
575 | int i; |
576 | ||
577 | BUG_ON(0 == size); | |
578 | ||
7f219c73 BH |
579 | /* We align the size to limit fragmentation. Without this, a lot of |
580 | * small allocs quickly eat up the whole reserve array on sparc | |
581 | */ | |
348968eb | 582 | size = round_up(size, align); |
7f219c73 | 583 | |
c3f72b57 BH |
584 | /* We do a bottom-up search for a region with the right |
585 | * nid since that's easier considering how memblock_nid_range() | |
586 | * works | |
587 | */ | |
95f72d1e | 588 | for (i = 0; i < mem->cnt; i++) { |
2898cc4c | 589 | phys_addr_t ret = memblock_alloc_nid_region(&mem->regions[i], |
95f72d1e | 590 | size, align, nid); |
4d629f9a | 591 | if (ret != MEMBLOCK_ERROR) |
95f72d1e YL |
592 | return ret; |
593 | } | |
594 | ||
9d1e2492 BH |
595 | return 0; |
596 | } | |
597 | ||
598 | phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) | |
599 | { | |
600 | phys_addr_t res = memblock_alloc_nid(size, align, nid); | |
601 | ||
602 | if (res) | |
603 | return res; | |
15fb0972 | 604 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
95f72d1e YL |
605 | } |
606 | ||
9d1e2492 BH |
607 | |
608 | /* | |
609 | * Remaining API functions | |
610 | */ | |
611 | ||
95f72d1e | 612 | /* You must call memblock_analyze() before this. */ |
2898cc4c | 613 | phys_addr_t __init memblock_phys_mem_size(void) |
95f72d1e | 614 | { |
4734b594 | 615 | return memblock.memory_size; |
95f72d1e YL |
616 | } |
617 | ||
10d06439 | 618 | phys_addr_t __init_memblock memblock_end_of_DRAM(void) |
95f72d1e YL |
619 | { |
620 | int idx = memblock.memory.cnt - 1; | |
621 | ||
e3239ff9 | 622 | return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); |
95f72d1e YL |
623 | } |
624 | ||
625 | /* You must call memblock_analyze() after this. */ | |
2898cc4c | 626 | void __init memblock_enforce_memory_limit(phys_addr_t memory_limit) |
95f72d1e YL |
627 | { |
628 | unsigned long i; | |
2898cc4c | 629 | phys_addr_t limit; |
e3239ff9 | 630 | struct memblock_region *p; |
95f72d1e YL |
631 | |
632 | if (!memory_limit) | |
633 | return; | |
634 | ||
635 | /* Truncate the memblock regions to satisfy the memory limit. */ | |
636 | limit = memory_limit; | |
637 | for (i = 0; i < memblock.memory.cnt; i++) { | |
e3239ff9 BH |
638 | if (limit > memblock.memory.regions[i].size) { |
639 | limit -= memblock.memory.regions[i].size; | |
95f72d1e YL |
640 | continue; |
641 | } | |
642 | ||
e3239ff9 | 643 | memblock.memory.regions[i].size = limit; |
95f72d1e YL |
644 | memblock.memory.cnt = i + 1; |
645 | break; | |
646 | } | |
647 | ||
95f72d1e YL |
648 | memory_limit = memblock_end_of_DRAM(); |
649 | ||
650 | /* And truncate any reserves above the limit also. */ | |
651 | for (i = 0; i < memblock.reserved.cnt; i++) { | |
e3239ff9 | 652 | p = &memblock.reserved.regions[i]; |
95f72d1e YL |
653 | |
654 | if (p->base > memory_limit) | |
655 | p->size = 0; | |
656 | else if ((p->base + p->size) > memory_limit) | |
657 | p->size = memory_limit - p->base; | |
658 | ||
659 | if (p->size == 0) { | |
660 | memblock_remove_region(&memblock.reserved, i); | |
661 | i--; | |
662 | } | |
663 | } | |
664 | } | |
665 | ||
cd79481d | 666 | static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) |
72d4b0b4 BH |
667 | { |
668 | unsigned int left = 0, right = type->cnt; | |
669 | ||
670 | do { | |
671 | unsigned int mid = (right + left) / 2; | |
672 | ||
673 | if (addr < type->regions[mid].base) | |
674 | right = mid; | |
675 | else if (addr >= (type->regions[mid].base + | |
676 | type->regions[mid].size)) | |
677 | left = mid + 1; | |
678 | else | |
679 | return mid; | |
680 | } while (left < right); | |
681 | return -1; | |
682 | } | |
683 | ||
2898cc4c | 684 | int __init memblock_is_reserved(phys_addr_t addr) |
95f72d1e | 685 | { |
72d4b0b4 BH |
686 | return memblock_search(&memblock.reserved, addr) != -1; |
687 | } | |
95f72d1e | 688 | |
3661ca66 | 689 | int __init_memblock memblock_is_memory(phys_addr_t addr) |
72d4b0b4 BH |
690 | { |
691 | return memblock_search(&memblock.memory, addr) != -1; | |
692 | } | |
693 | ||
3661ca66 | 694 | int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) |
72d4b0b4 | 695 | { |
abb65272 | 696 | int idx = memblock_search(&memblock.memory, base); |
72d4b0b4 BH |
697 | |
698 | if (idx == -1) | |
699 | return 0; | |
abb65272 TV |
700 | return memblock.memory.regions[idx].base <= base && |
701 | (memblock.memory.regions[idx].base + | |
702 | memblock.memory.regions[idx].size) >= (base + size); | |
95f72d1e YL |
703 | } |
704 | ||
10d06439 | 705 | int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) |
95f72d1e | 706 | { |
f1c2c19c | 707 | return memblock_overlaps_region(&memblock.reserved, base, size) >= 0; |
95f72d1e YL |
708 | } |
709 | ||
e63075a3 | 710 | |
3661ca66 | 711 | void __init_memblock memblock_set_current_limit(phys_addr_t limit) |
e63075a3 BH |
712 | { |
713 | memblock.current_limit = limit; | |
714 | } | |
715 | ||
10d06439 | 716 | static void __init_memblock memblock_dump(struct memblock_type *region, char *name) |
6ed311b2 BH |
717 | { |
718 | unsigned long long base, size; | |
719 | int i; | |
720 | ||
721 | pr_info(" %s.cnt = 0x%lx\n", name, region->cnt); | |
722 | ||
723 | for (i = 0; i < region->cnt; i++) { | |
724 | base = region->regions[i].base; | |
725 | size = region->regions[i].size; | |
726 | ||
ea9e4376 | 727 | pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes\n", |
6ed311b2 BH |
728 | name, i, base, base + size - 1, size); |
729 | } | |
730 | } | |
731 | ||
10d06439 | 732 | void __init_memblock memblock_dump_all(void) |
6ed311b2 BH |
733 | { |
734 | if (!memblock_debug) | |
735 | return; | |
736 | ||
737 | pr_info("MEMBLOCK configuration:\n"); | |
738 | pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size); | |
739 | ||
740 | memblock_dump(&memblock.memory, "memory"); | |
741 | memblock_dump(&memblock.reserved, "reserved"); | |
742 | } | |
743 | ||
744 | void __init memblock_analyze(void) | |
745 | { | |
746 | int i; | |
747 | ||
748 | /* Check marker in the unused last array entry */ | |
749 | WARN_ON(memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS].base | |
750 | != (phys_addr_t)RED_INACTIVE); | |
751 | WARN_ON(memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS].base | |
752 | != (phys_addr_t)RED_INACTIVE); | |
753 | ||
754 | memblock.memory_size = 0; | |
755 | ||
756 | for (i = 0; i < memblock.memory.cnt; i++) | |
757 | memblock.memory_size += memblock.memory.regions[i].size; | |
142b45a7 BH |
758 | |
759 | /* We allow resizing from there */ | |
760 | memblock_can_resize = 1; | |
6ed311b2 BH |
761 | } |
762 | ||
7590abe8 BH |
763 | void __init memblock_init(void) |
764 | { | |
236260b9 JF |
765 | static int init_done __initdata = 0; |
766 | ||
767 | if (init_done) | |
768 | return; | |
769 | init_done = 1; | |
770 | ||
7590abe8 BH |
771 | /* Hookup the initial arrays */ |
772 | memblock.memory.regions = memblock_memory_init_regions; | |
773 | memblock.memory.max = INIT_MEMBLOCK_REGIONS; | |
774 | memblock.reserved.regions = memblock_reserved_init_regions; | |
775 | memblock.reserved.max = INIT_MEMBLOCK_REGIONS; | |
776 | ||
777 | /* Write a marker in the unused last array entry */ | |
778 | memblock.memory.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE; | |
779 | memblock.reserved.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE; | |
780 | ||
781 | /* Create a dummy zero size MEMBLOCK which will get coalesced away later. | |
782 | * This simplifies the memblock_add() code below... | |
783 | */ | |
784 | memblock.memory.regions[0].base = 0; | |
785 | memblock.memory.regions[0].size = 0; | |
786 | memblock.memory.cnt = 1; | |
787 | ||
788 | /* Ditto. */ | |
789 | memblock.reserved.regions[0].base = 0; | |
790 | memblock.reserved.regions[0].size = 0; | |
791 | memblock.reserved.cnt = 1; | |
792 | ||
793 | memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE; | |
794 | } | |
795 | ||
6ed311b2 BH |
796 | static int __init early_memblock(char *p) |
797 | { | |
798 | if (p && strstr(p, "debug")) | |
799 | memblock_debug = 1; | |
800 | return 0; | |
801 | } | |
802 | early_param("memblock", early_memblock); | |
803 | ||
10d06439 | 804 | #if defined(CONFIG_DEBUG_FS) && !defined(ARCH_DISCARD_MEMBLOCK) |
6d03b885 BH |
805 | |
806 | static int memblock_debug_show(struct seq_file *m, void *private) | |
807 | { | |
808 | struct memblock_type *type = m->private; | |
809 | struct memblock_region *reg; | |
810 | int i; | |
811 | ||
812 | for (i = 0; i < type->cnt; i++) { | |
813 | reg = &type->regions[i]; | |
814 | seq_printf(m, "%4d: ", i); | |
815 | if (sizeof(phys_addr_t) == 4) | |
816 | seq_printf(m, "0x%08lx..0x%08lx\n", | |
817 | (unsigned long)reg->base, | |
818 | (unsigned long)(reg->base + reg->size - 1)); | |
819 | else | |
820 | seq_printf(m, "0x%016llx..0x%016llx\n", | |
821 | (unsigned long long)reg->base, | |
822 | (unsigned long long)(reg->base + reg->size - 1)); | |
823 | ||
824 | } | |
825 | return 0; | |
826 | } | |
827 | ||
828 | static int memblock_debug_open(struct inode *inode, struct file *file) | |
829 | { | |
830 | return single_open(file, memblock_debug_show, inode->i_private); | |
831 | } | |
832 | ||
833 | static const struct file_operations memblock_debug_fops = { | |
834 | .open = memblock_debug_open, | |
835 | .read = seq_read, | |
836 | .llseek = seq_lseek, | |
837 | .release = single_release, | |
838 | }; | |
839 | ||
840 | static int __init memblock_init_debugfs(void) | |
841 | { | |
842 | struct dentry *root = debugfs_create_dir("memblock", NULL); | |
843 | if (!root) | |
844 | return -ENXIO; | |
845 | debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops); | |
846 | debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops); | |
847 | ||
848 | return 0; | |
849 | } | |
850 | __initcall(memblock_init_debugfs); | |
851 | ||
852 | #endif /* CONFIG_DEBUG_FS */ |