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; | |
c5a1cb28 TH |
27 | static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock; |
28 | static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __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 | ||
2d7d3eb2 HS |
50 | static long __init_memblock memblock_overlaps_region(struct memblock_type *type, |
51 | phys_addr_t base, phys_addr_t size) | |
6ed311b2 BH |
52 | { |
53 | unsigned long i; | |
54 | ||
55 | for (i = 0; i < type->cnt; i++) { | |
56 | phys_addr_t rgnbase = type->regions[i].base; | |
57 | phys_addr_t rgnsize = type->regions[i].size; | |
58 | if (memblock_addrs_overlap(base, size, rgnbase, rgnsize)) | |
59 | break; | |
60 | } | |
61 | ||
62 | return (i < type->cnt) ? i : -1; | |
63 | } | |
64 | ||
65 | /* | |
66 | * Find, allocate, deallocate or reserve unreserved regions. All allocations | |
67 | * are top-down. | |
68 | */ | |
69 | ||
cd79481d | 70 | static phys_addr_t __init_memblock memblock_find_region(phys_addr_t start, phys_addr_t end, |
6ed311b2 BH |
71 | phys_addr_t size, phys_addr_t align) |
72 | { | |
73 | phys_addr_t base, res_base; | |
74 | long j; | |
75 | ||
f1af98c7 YL |
76 | /* In case, huge size is requested */ |
77 | if (end < size) | |
1f5026a7 | 78 | return 0; |
f1af98c7 | 79 | |
348968eb | 80 | base = round_down(end - size, align); |
f1af98c7 | 81 | |
25818f0f BH |
82 | /* Prevent allocations returning 0 as it's also used to |
83 | * indicate an allocation failure | |
84 | */ | |
85 | if (start == 0) | |
86 | start = PAGE_SIZE; | |
87 | ||
6ed311b2 BH |
88 | while (start <= base) { |
89 | j = memblock_overlaps_region(&memblock.reserved, base, size); | |
90 | if (j < 0) | |
91 | return base; | |
92 | res_base = memblock.reserved.regions[j].base; | |
93 | if (res_base < size) | |
94 | break; | |
348968eb | 95 | base = round_down(res_base - size, align); |
6ed311b2 BH |
96 | } |
97 | ||
1f5026a7 | 98 | return 0; |
6ed311b2 BH |
99 | } |
100 | ||
fc769a8e TH |
101 | /* |
102 | * Find a free area with specified alignment in a specific range. | |
103 | */ | |
104 | phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, phys_addr_t end, | |
105 | phys_addr_t size, phys_addr_t align) | |
6ed311b2 BH |
106 | { |
107 | long i; | |
6ed311b2 BH |
108 | |
109 | BUG_ON(0 == size); | |
110 | ||
6ed311b2 | 111 | /* Pump up max_addr */ |
fef501d4 BH |
112 | if (end == MEMBLOCK_ALLOC_ACCESSIBLE) |
113 | end = memblock.current_limit; | |
6ed311b2 BH |
114 | |
115 | /* We do a top-down search, this tends to limit memory | |
116 | * fragmentation by keeping early boot allocs near the | |
117 | * top of memory | |
118 | */ | |
119 | for (i = memblock.memory.cnt - 1; i >= 0; i--) { | |
120 | phys_addr_t memblockbase = memblock.memory.regions[i].base; | |
121 | phys_addr_t memblocksize = memblock.memory.regions[i].size; | |
fef501d4 | 122 | phys_addr_t bottom, top, found; |
6ed311b2 BH |
123 | |
124 | if (memblocksize < size) | |
125 | continue; | |
fef501d4 BH |
126 | if ((memblockbase + memblocksize) <= start) |
127 | break; | |
128 | bottom = max(memblockbase, start); | |
129 | top = min(memblockbase + memblocksize, end); | |
130 | if (bottom >= top) | |
131 | continue; | |
132 | found = memblock_find_region(bottom, top, size, align); | |
1f5026a7 | 133 | if (found) |
fef501d4 | 134 | return found; |
6ed311b2 | 135 | } |
1f5026a7 | 136 | return 0; |
6ed311b2 BH |
137 | } |
138 | ||
7950c407 YL |
139 | /* |
140 | * Free memblock.reserved.regions | |
141 | */ | |
142 | int __init_memblock memblock_free_reserved_regions(void) | |
143 | { | |
144 | if (memblock.reserved.regions == memblock_reserved_init_regions) | |
145 | return 0; | |
146 | ||
147 | return memblock_free(__pa(memblock.reserved.regions), | |
148 | sizeof(struct memblock_region) * memblock.reserved.max); | |
149 | } | |
150 | ||
151 | /* | |
152 | * Reserve memblock.reserved.regions | |
153 | */ | |
154 | int __init_memblock memblock_reserve_reserved_regions(void) | |
155 | { | |
156 | if (memblock.reserved.regions == memblock_reserved_init_regions) | |
157 | return 0; | |
158 | ||
159 | return memblock_reserve(__pa(memblock.reserved.regions), | |
160 | sizeof(struct memblock_region) * memblock.reserved.max); | |
161 | } | |
162 | ||
10d06439 | 163 | static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) |
95f72d1e | 164 | { |
7c0caeb8 TH |
165 | memmove(&type->regions[r], &type->regions[r + 1], |
166 | (type->cnt - (r + 1)) * sizeof(type->regions[r])); | |
e3239ff9 | 167 | type->cnt--; |
95f72d1e | 168 | |
8f7a6605 BH |
169 | /* Special case for empty arrays */ |
170 | if (type->cnt == 0) { | |
171 | type->cnt = 1; | |
172 | type->regions[0].base = 0; | |
173 | type->regions[0].size = 0; | |
7c0caeb8 | 174 | memblock_set_region_node(&type->regions[0], MAX_NUMNODES); |
8f7a6605 | 175 | } |
95f72d1e YL |
176 | } |
177 | ||
10d06439 | 178 | static int __init_memblock memblock_double_array(struct memblock_type *type) |
142b45a7 BH |
179 | { |
180 | struct memblock_region *new_array, *old_array; | |
181 | phys_addr_t old_size, new_size, addr; | |
182 | int use_slab = slab_is_available(); | |
183 | ||
184 | /* We don't allow resizing until we know about the reserved regions | |
185 | * of memory that aren't suitable for allocation | |
186 | */ | |
187 | if (!memblock_can_resize) | |
188 | return -1; | |
189 | ||
142b45a7 BH |
190 | /* Calculate new doubled size */ |
191 | old_size = type->max * sizeof(struct memblock_region); | |
192 | new_size = old_size << 1; | |
193 | ||
194 | /* Try to find some space for it. | |
195 | * | |
196 | * WARNING: We assume that either slab_is_available() and we use it or | |
197 | * we use MEMBLOCK for allocations. That means that this is unsafe to use | |
198 | * when bootmem is currently active (unless bootmem itself is implemented | |
199 | * on top of MEMBLOCK which isn't the case yet) | |
200 | * | |
201 | * This should however not be an issue for now, as we currently only | |
202 | * call into MEMBLOCK while it's still active, or much later when slab is | |
203 | * active for memory hotplug operations | |
204 | */ | |
205 | if (use_slab) { | |
206 | new_array = kmalloc(new_size, GFP_KERNEL); | |
1f5026a7 | 207 | addr = new_array ? __pa(new_array) : 0; |
142b45a7 | 208 | } else |
fc769a8e | 209 | addr = memblock_find_in_range(0, MEMBLOCK_ALLOC_ACCESSIBLE, new_size, sizeof(phys_addr_t)); |
1f5026a7 | 210 | if (!addr) { |
142b45a7 BH |
211 | pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", |
212 | memblock_type_name(type), type->max, type->max * 2); | |
213 | return -1; | |
214 | } | |
215 | new_array = __va(addr); | |
216 | ||
ea9e4376 YL |
217 | memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]", |
218 | memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1); | |
219 | ||
142b45a7 BH |
220 | /* Found space, we now need to move the array over before |
221 | * we add the reserved region since it may be our reserved | |
222 | * array itself that is full. | |
223 | */ | |
224 | memcpy(new_array, type->regions, old_size); | |
225 | memset(new_array + type->max, 0, old_size); | |
226 | old_array = type->regions; | |
227 | type->regions = new_array; | |
228 | type->max <<= 1; | |
229 | ||
230 | /* If we use SLAB that's it, we are done */ | |
231 | if (use_slab) | |
232 | return 0; | |
233 | ||
234 | /* Add the new reserved region now. Should not fail ! */ | |
9c8c27e2 | 235 | BUG_ON(memblock_reserve(addr, new_size)); |
142b45a7 BH |
236 | |
237 | /* If the array wasn't our static init one, then free it. We only do | |
238 | * that before SLAB is available as later on, we don't know whether | |
239 | * to use kfree or free_bootmem_pages(). Shouldn't be a big deal | |
240 | * anyways | |
241 | */ | |
242 | if (old_array != memblock_memory_init_regions && | |
243 | old_array != memblock_reserved_init_regions) | |
244 | memblock_free(__pa(old_array), old_size); | |
245 | ||
246 | return 0; | |
247 | } | |
248 | ||
784656f9 TH |
249 | /** |
250 | * memblock_merge_regions - merge neighboring compatible regions | |
251 | * @type: memblock type to scan | |
252 | * | |
253 | * Scan @type and merge neighboring compatible regions. | |
254 | */ | |
255 | static void __init_memblock memblock_merge_regions(struct memblock_type *type) | |
95f72d1e | 256 | { |
784656f9 | 257 | int i = 0; |
95f72d1e | 258 | |
784656f9 TH |
259 | /* cnt never goes below 1 */ |
260 | while (i < type->cnt - 1) { | |
261 | struct memblock_region *this = &type->regions[i]; | |
262 | struct memblock_region *next = &type->regions[i + 1]; | |
95f72d1e | 263 | |
7c0caeb8 TH |
264 | if (this->base + this->size != next->base || |
265 | memblock_get_region_node(this) != | |
266 | memblock_get_region_node(next)) { | |
784656f9 TH |
267 | BUG_ON(this->base + this->size > next->base); |
268 | i++; | |
269 | continue; | |
8f7a6605 BH |
270 | } |
271 | ||
784656f9 TH |
272 | this->size += next->size; |
273 | memmove(next, next + 1, (type->cnt - (i + 1)) * sizeof(*next)); | |
274 | type->cnt--; | |
95f72d1e | 275 | } |
784656f9 | 276 | } |
95f72d1e | 277 | |
784656f9 TH |
278 | /** |
279 | * memblock_insert_region - insert new memblock region | |
280 | * @type: memblock type to insert into | |
281 | * @idx: index for the insertion point | |
282 | * @base: base address of the new region | |
283 | * @size: size of the new region | |
284 | * | |
285 | * Insert new memblock region [@base,@base+@size) into @type at @idx. | |
286 | * @type must already have extra room to accomodate the new region. | |
287 | */ | |
288 | static void __init_memblock memblock_insert_region(struct memblock_type *type, | |
289 | int idx, phys_addr_t base, | |
7c0caeb8 | 290 | phys_addr_t size, int nid) |
784656f9 TH |
291 | { |
292 | struct memblock_region *rgn = &type->regions[idx]; | |
293 | ||
294 | BUG_ON(type->cnt >= type->max); | |
295 | memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); | |
296 | rgn->base = base; | |
297 | rgn->size = size; | |
7c0caeb8 | 298 | memblock_set_region_node(rgn, nid); |
784656f9 TH |
299 | type->cnt++; |
300 | } | |
301 | ||
302 | /** | |
303 | * memblock_add_region - add new memblock region | |
304 | * @type: memblock type to add new region into | |
305 | * @base: base address of the new region | |
306 | * @size: size of the new region | |
307 | * | |
308 | * Add new memblock region [@base,@base+@size) into @type. The new region | |
309 | * is allowed to overlap with existing ones - overlaps don't affect already | |
310 | * existing regions. @type is guaranteed to be minimal (all neighbouring | |
311 | * compatible regions are merged) after the addition. | |
312 | * | |
313 | * RETURNS: | |
314 | * 0 on success, -errno on failure. | |
315 | */ | |
581adcbe TH |
316 | static int __init_memblock memblock_add_region(struct memblock_type *type, |
317 | phys_addr_t base, phys_addr_t size) | |
784656f9 TH |
318 | { |
319 | bool insert = false; | |
320 | phys_addr_t obase = base, end = base + size; | |
321 | int i, nr_new; | |
322 | ||
323 | /* special case for empty array */ | |
324 | if (type->regions[0].size == 0) { | |
325 | WARN_ON(type->cnt != 1); | |
8f7a6605 BH |
326 | type->regions[0].base = base; |
327 | type->regions[0].size = size; | |
7c0caeb8 | 328 | memblock_set_region_node(&type->regions[0], MAX_NUMNODES); |
8f7a6605 | 329 | return 0; |
95f72d1e | 330 | } |
784656f9 TH |
331 | repeat: |
332 | /* | |
333 | * The following is executed twice. Once with %false @insert and | |
334 | * then with %true. The first counts the number of regions needed | |
335 | * to accomodate the new area. The second actually inserts them. | |
142b45a7 | 336 | */ |
784656f9 TH |
337 | base = obase; |
338 | nr_new = 0; | |
95f72d1e | 339 | |
784656f9 TH |
340 | for (i = 0; i < type->cnt; i++) { |
341 | struct memblock_region *rgn = &type->regions[i]; | |
342 | phys_addr_t rbase = rgn->base; | |
343 | phys_addr_t rend = rbase + rgn->size; | |
344 | ||
345 | if (rbase >= end) | |
95f72d1e | 346 | break; |
784656f9 TH |
347 | if (rend <= base) |
348 | continue; | |
349 | /* | |
350 | * @rgn overlaps. If it separates the lower part of new | |
351 | * area, insert that portion. | |
352 | */ | |
353 | if (rbase > base) { | |
354 | nr_new++; | |
355 | if (insert) | |
356 | memblock_insert_region(type, i++, base, | |
7c0caeb8 | 357 | rbase - base, MAX_NUMNODES); |
95f72d1e | 358 | } |
784656f9 TH |
359 | /* area below @rend is dealt with, forget about it */ |
360 | base = min(rend, end); | |
95f72d1e | 361 | } |
784656f9 TH |
362 | |
363 | /* insert the remaining portion */ | |
364 | if (base < end) { | |
365 | nr_new++; | |
366 | if (insert) | |
7c0caeb8 TH |
367 | memblock_insert_region(type, i, base, end - base, |
368 | MAX_NUMNODES); | |
95f72d1e | 369 | } |
95f72d1e | 370 | |
784656f9 TH |
371 | /* |
372 | * If this was the first round, resize array and repeat for actual | |
373 | * insertions; otherwise, merge and return. | |
142b45a7 | 374 | */ |
784656f9 TH |
375 | if (!insert) { |
376 | while (type->cnt + nr_new > type->max) | |
377 | if (memblock_double_array(type) < 0) | |
378 | return -ENOMEM; | |
379 | insert = true; | |
380 | goto repeat; | |
381 | } else { | |
382 | memblock_merge_regions(type); | |
383 | return 0; | |
142b45a7 | 384 | } |
95f72d1e YL |
385 | } |
386 | ||
581adcbe | 387 | int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) |
95f72d1e | 388 | { |
e3239ff9 | 389 | return memblock_add_region(&memblock.memory, base, size); |
95f72d1e YL |
390 | } |
391 | ||
581adcbe TH |
392 | static int __init_memblock __memblock_remove(struct memblock_type *type, |
393 | phys_addr_t base, phys_addr_t size) | |
95f72d1e | 394 | { |
2898cc4c | 395 | phys_addr_t end = base + size; |
95f72d1e YL |
396 | int i; |
397 | ||
8f7a6605 BH |
398 | /* Walk through the array for collisions */ |
399 | for (i = 0; i < type->cnt; i++) { | |
400 | struct memblock_region *rgn = &type->regions[i]; | |
401 | phys_addr_t rend = rgn->base + rgn->size; | |
95f72d1e | 402 | |
8f7a6605 BH |
403 | /* Nothing more to do, exit */ |
404 | if (rgn->base > end || rgn->size == 0) | |
95f72d1e | 405 | break; |
95f72d1e | 406 | |
8f7a6605 BH |
407 | /* If we fully enclose the block, drop it */ |
408 | if (base <= rgn->base && end >= rend) { | |
409 | memblock_remove_region(type, i--); | |
410 | continue; | |
411 | } | |
95f72d1e | 412 | |
8f7a6605 BH |
413 | /* If we are fully enclosed within a block |
414 | * then we need to split it and we are done | |
415 | */ | |
416 | if (base > rgn->base && end < rend) { | |
417 | rgn->size = base - rgn->base; | |
418 | if (!memblock_add_region(type, end, rend - end)) | |
419 | return 0; | |
420 | /* Failure to split is bad, we at least | |
421 | * restore the block before erroring | |
422 | */ | |
423 | rgn->size = rend - rgn->base; | |
424 | WARN_ON(1); | |
425 | return -1; | |
426 | } | |
95f72d1e | 427 | |
8f7a6605 BH |
428 | /* Check if we need to trim the bottom of a block */ |
429 | if (rgn->base < end && rend > end) { | |
430 | rgn->size -= end - rgn->base; | |
431 | rgn->base = end; | |
432 | break; | |
433 | } | |
95f72d1e | 434 | |
8f7a6605 BH |
435 | /* And check if we need to trim the top of a block */ |
436 | if (base < rend) | |
437 | rgn->size -= rend - base; | |
95f72d1e | 438 | |
8f7a6605 BH |
439 | } |
440 | return 0; | |
95f72d1e YL |
441 | } |
442 | ||
581adcbe | 443 | int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) |
95f72d1e YL |
444 | { |
445 | return __memblock_remove(&memblock.memory, base, size); | |
446 | } | |
447 | ||
581adcbe | 448 | int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size) |
95f72d1e | 449 | { |
24aa0788 | 450 | memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n", |
a150439c PA |
451 | (unsigned long long)base, |
452 | (unsigned long long)base + size, | |
453 | (void *)_RET_IP_); | |
24aa0788 | 454 | |
95f72d1e YL |
455 | return __memblock_remove(&memblock.reserved, base, size); |
456 | } | |
457 | ||
581adcbe | 458 | int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) |
95f72d1e | 459 | { |
e3239ff9 | 460 | struct memblock_type *_rgn = &memblock.reserved; |
95f72d1e | 461 | |
24aa0788 | 462 | memblock_dbg("memblock_reserve: [%#016llx-%#016llx] %pF\n", |
a150439c PA |
463 | (unsigned long long)base, |
464 | (unsigned long long)base + size, | |
465 | (void *)_RET_IP_); | |
95f72d1e YL |
466 | BUG_ON(0 == size); |
467 | ||
468 | return memblock_add_region(_rgn, base, size); | |
469 | } | |
470 | ||
35fd0808 TH |
471 | /** |
472 | * __next_free_mem_range - next function for for_each_free_mem_range() | |
473 | * @idx: pointer to u64 loop variable | |
474 | * @nid: nid: node selector, %MAX_NUMNODES for all nodes | |
475 | * @p_start: ptr to phys_addr_t for start address of the range, can be %NULL | |
476 | * @p_end: ptr to phys_addr_t for end address of the range, can be %NULL | |
477 | * @p_nid: ptr to int for nid of the range, can be %NULL | |
478 | * | |
479 | * Find the first free area from *@idx which matches @nid, fill the out | |
480 | * parameters, and update *@idx for the next iteration. The lower 32bit of | |
481 | * *@idx contains index into memory region and the upper 32bit indexes the | |
482 | * areas before each reserved region. For example, if reserved regions | |
483 | * look like the following, | |
484 | * | |
485 | * 0:[0-16), 1:[32-48), 2:[128-130) | |
486 | * | |
487 | * The upper 32bit indexes the following regions. | |
488 | * | |
489 | * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) | |
490 | * | |
491 | * As both region arrays are sorted, the function advances the two indices | |
492 | * in lockstep and returns each intersection. | |
493 | */ | |
494 | void __init_memblock __next_free_mem_range(u64 *idx, int nid, | |
495 | phys_addr_t *out_start, | |
496 | phys_addr_t *out_end, int *out_nid) | |
497 | { | |
498 | struct memblock_type *mem = &memblock.memory; | |
499 | struct memblock_type *rsv = &memblock.reserved; | |
500 | int mi = *idx & 0xffffffff; | |
501 | int ri = *idx >> 32; | |
502 | ||
503 | for ( ; mi < mem->cnt; mi++) { | |
504 | struct memblock_region *m = &mem->regions[mi]; | |
505 | phys_addr_t m_start = m->base; | |
506 | phys_addr_t m_end = m->base + m->size; | |
507 | ||
508 | /* only memory regions are associated with nodes, check it */ | |
509 | if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m)) | |
510 | continue; | |
511 | ||
512 | /* scan areas before each reservation for intersection */ | |
513 | for ( ; ri < rsv->cnt + 1; ri++) { | |
514 | struct memblock_region *r = &rsv->regions[ri]; | |
515 | phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0; | |
516 | phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX; | |
517 | ||
518 | /* if ri advanced past mi, break out to advance mi */ | |
519 | if (r_start >= m_end) | |
520 | break; | |
521 | /* if the two regions intersect, we're done */ | |
522 | if (m_start < r_end) { | |
523 | if (out_start) | |
524 | *out_start = max(m_start, r_start); | |
525 | if (out_end) | |
526 | *out_end = min(m_end, r_end); | |
527 | if (out_nid) | |
528 | *out_nid = memblock_get_region_node(m); | |
529 | /* | |
530 | * The region which ends first is advanced | |
531 | * for the next iteration. | |
532 | */ | |
533 | if (m_end <= r_end) | |
534 | mi++; | |
535 | else | |
536 | ri++; | |
537 | *idx = (u32)mi | (u64)ri << 32; | |
538 | return; | |
539 | } | |
540 | } | |
541 | } | |
542 | ||
543 | /* signal end of iteration */ | |
544 | *idx = ULLONG_MAX; | |
545 | } | |
546 | ||
7c0caeb8 TH |
547 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
548 | /* | |
549 | * Common iterator interface used to define for_each_mem_range(). | |
550 | */ | |
551 | void __init_memblock __next_mem_pfn_range(int *idx, int nid, | |
552 | unsigned long *out_start_pfn, | |
553 | unsigned long *out_end_pfn, int *out_nid) | |
554 | { | |
555 | struct memblock_type *type = &memblock.memory; | |
556 | struct memblock_region *r; | |
557 | ||
558 | while (++*idx < type->cnt) { | |
559 | r = &type->regions[*idx]; | |
560 | ||
561 | if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) | |
562 | continue; | |
563 | if (nid == MAX_NUMNODES || nid == r->nid) | |
564 | break; | |
565 | } | |
566 | if (*idx >= type->cnt) { | |
567 | *idx = -1; | |
568 | return; | |
569 | } | |
570 | ||
571 | if (out_start_pfn) | |
572 | *out_start_pfn = PFN_UP(r->base); | |
573 | if (out_end_pfn) | |
574 | *out_end_pfn = PFN_DOWN(r->base + r->size); | |
575 | if (out_nid) | |
576 | *out_nid = r->nid; | |
577 | } | |
578 | ||
579 | /** | |
580 | * memblock_set_node - set node ID on memblock regions | |
581 | * @base: base of area to set node ID for | |
582 | * @size: size of area to set node ID for | |
583 | * @nid: node ID to set | |
584 | * | |
585 | * Set the nid of memblock memory regions in [@base,@base+@size) to @nid. | |
586 | * Regions which cross the area boundaries are split as necessary. | |
587 | * | |
588 | * RETURNS: | |
589 | * 0 on success, -errno on failure. | |
590 | */ | |
591 | int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, | |
592 | int nid) | |
593 | { | |
594 | struct memblock_type *type = &memblock.memory; | |
595 | phys_addr_t end = base + size; | |
596 | int i; | |
597 | ||
598 | /* we'll create at most two more regions */ | |
599 | while (type->cnt + 2 > type->max) | |
600 | if (memblock_double_array(type) < 0) | |
601 | return -ENOMEM; | |
602 | ||
603 | for (i = 0; i < type->cnt; i++) { | |
604 | struct memblock_region *rgn = &type->regions[i]; | |
605 | phys_addr_t rbase = rgn->base; | |
606 | phys_addr_t rend = rbase + rgn->size; | |
607 | ||
608 | if (rbase >= end) | |
609 | break; | |
610 | if (rend <= base) | |
611 | continue; | |
612 | ||
613 | if (rbase < base) { | |
614 | /* | |
615 | * @rgn intersects from below. Split and continue | |
616 | * to process the next region - the new top half. | |
617 | */ | |
618 | rgn->base = base; | |
619 | rgn->size = rend - rgn->base; | |
620 | memblock_insert_region(type, i, rbase, base - rbase, | |
621 | rgn->nid); | |
622 | } else if (rend > end) { | |
623 | /* | |
624 | * @rgn intersects from above. Split and redo the | |
625 | * current region - the new bottom half. | |
626 | */ | |
627 | rgn->base = end; | |
628 | rgn->size = rend - rgn->base; | |
629 | memblock_insert_region(type, i--, rbase, end - rbase, | |
630 | rgn->nid); | |
631 | } else { | |
632 | /* @rgn is fully contained, set ->nid */ | |
633 | rgn->nid = nid; | |
634 | } | |
635 | } | |
636 | ||
637 | memblock_merge_regions(type); | |
638 | return 0; | |
639 | } | |
640 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
641 | ||
6ed311b2 | 642 | phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) |
95f72d1e | 643 | { |
6ed311b2 | 644 | phys_addr_t found; |
95f72d1e | 645 | |
6ed311b2 BH |
646 | /* We align the size to limit fragmentation. Without this, a lot of |
647 | * small allocs quickly eat up the whole reserve array on sparc | |
648 | */ | |
348968eb | 649 | size = round_up(size, align); |
95f72d1e | 650 | |
fc769a8e | 651 | found = memblock_find_in_range(0, max_addr, size, align); |
9c8c27e2 | 652 | if (found && !memblock_reserve(found, size)) |
6ed311b2 | 653 | return found; |
95f72d1e | 654 | |
6ed311b2 | 655 | return 0; |
95f72d1e YL |
656 | } |
657 | ||
6ed311b2 | 658 | phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) |
95f72d1e | 659 | { |
6ed311b2 BH |
660 | phys_addr_t alloc; |
661 | ||
662 | alloc = __memblock_alloc_base(size, align, max_addr); | |
663 | ||
664 | if (alloc == 0) | |
665 | panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n", | |
666 | (unsigned long long) size, (unsigned long long) max_addr); | |
667 | ||
668 | return alloc; | |
95f72d1e YL |
669 | } |
670 | ||
6ed311b2 | 671 | phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align) |
95f72d1e | 672 | { |
6ed311b2 BH |
673 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
674 | } | |
95f72d1e | 675 | |
95f72d1e | 676 | |
6ed311b2 | 677 | /* |
34e18455 | 678 | * Additional node-local top-down allocators. |
c196f76f BH |
679 | * |
680 | * WARNING: Only available after early_node_map[] has been populated, | |
681 | * on some architectures, that is after all the calls to add_active_range() | |
682 | * have been done to populate it. | |
6ed311b2 | 683 | */ |
95f72d1e | 684 | |
34e18455 TH |
685 | static phys_addr_t __init memblock_nid_range_rev(phys_addr_t start, |
686 | phys_addr_t end, int *nid) | |
c3f72b57 | 687 | { |
c196f76f | 688 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
c196f76f BH |
689 | unsigned long start_pfn, end_pfn; |
690 | int i; | |
691 | ||
b2fea988 | 692 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, nid) |
34e18455 TH |
693 | if (end > PFN_PHYS(start_pfn) && end <= PFN_PHYS(end_pfn)) |
694 | return max(start, PFN_PHYS(start_pfn)); | |
c196f76f | 695 | #endif |
c3f72b57 | 696 | *nid = 0; |
34e18455 | 697 | return start; |
c3f72b57 BH |
698 | } |
699 | ||
e6498040 TH |
700 | phys_addr_t __init memblock_find_in_range_node(phys_addr_t start, |
701 | phys_addr_t end, | |
2898cc4c BH |
702 | phys_addr_t size, |
703 | phys_addr_t align, int nid) | |
95f72d1e | 704 | { |
e6498040 TH |
705 | struct memblock_type *mem = &memblock.memory; |
706 | int i; | |
95f72d1e | 707 | |
e6498040 | 708 | BUG_ON(0 == size); |
95f72d1e | 709 | |
e6498040 TH |
710 | /* Pump up max_addr */ |
711 | if (end == MEMBLOCK_ALLOC_ACCESSIBLE) | |
712 | end = memblock.current_limit; | |
95f72d1e | 713 | |
e6498040 TH |
714 | for (i = mem->cnt - 1; i >= 0; i--) { |
715 | struct memblock_region *r = &mem->regions[i]; | |
716 | phys_addr_t base = max(start, r->base); | |
717 | phys_addr_t top = min(end, r->base + r->size); | |
718 | ||
719 | while (base < top) { | |
720 | phys_addr_t tbase, ret; | |
721 | int tnid; | |
722 | ||
723 | tbase = memblock_nid_range_rev(base, top, &tnid); | |
724 | if (nid == MAX_NUMNODES || tnid == nid) { | |
725 | ret = memblock_find_region(tbase, top, size, align); | |
726 | if (ret) | |
727 | return ret; | |
728 | } | |
729 | top = tbase; | |
95f72d1e | 730 | } |
95f72d1e | 731 | } |
e6498040 | 732 | |
1f5026a7 | 733 | return 0; |
95f72d1e YL |
734 | } |
735 | ||
2898cc4c | 736 | phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid) |
95f72d1e | 737 | { |
e6498040 | 738 | phys_addr_t found; |
95f72d1e | 739 | |
e6498040 TH |
740 | /* |
741 | * We align the size to limit fragmentation. Without this, a lot of | |
7f219c73 BH |
742 | * small allocs quickly eat up the whole reserve array on sparc |
743 | */ | |
348968eb | 744 | size = round_up(size, align); |
7f219c73 | 745 | |
e6498040 TH |
746 | found = memblock_find_in_range_node(0, MEMBLOCK_ALLOC_ACCESSIBLE, |
747 | size, align, nid); | |
9c8c27e2 | 748 | if (found && !memblock_reserve(found, size)) |
e6498040 | 749 | return found; |
95f72d1e | 750 | |
9d1e2492 BH |
751 | return 0; |
752 | } | |
753 | ||
754 | phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) | |
755 | { | |
756 | phys_addr_t res = memblock_alloc_nid(size, align, nid); | |
757 | ||
758 | if (res) | |
759 | return res; | |
15fb0972 | 760 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
95f72d1e YL |
761 | } |
762 | ||
9d1e2492 BH |
763 | |
764 | /* | |
765 | * Remaining API functions | |
766 | */ | |
767 | ||
95f72d1e | 768 | /* You must call memblock_analyze() before this. */ |
2898cc4c | 769 | phys_addr_t __init memblock_phys_mem_size(void) |
95f72d1e | 770 | { |
4734b594 | 771 | return memblock.memory_size; |
95f72d1e YL |
772 | } |
773 | ||
0a93ebef SR |
774 | /* lowest address */ |
775 | phys_addr_t __init_memblock memblock_start_of_DRAM(void) | |
776 | { | |
777 | return memblock.memory.regions[0].base; | |
778 | } | |
779 | ||
10d06439 | 780 | phys_addr_t __init_memblock memblock_end_of_DRAM(void) |
95f72d1e YL |
781 | { |
782 | int idx = memblock.memory.cnt - 1; | |
783 | ||
e3239ff9 | 784 | return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); |
95f72d1e YL |
785 | } |
786 | ||
787 | /* You must call memblock_analyze() after this. */ | |
2898cc4c | 788 | void __init memblock_enforce_memory_limit(phys_addr_t memory_limit) |
95f72d1e YL |
789 | { |
790 | unsigned long i; | |
2898cc4c | 791 | phys_addr_t limit; |
e3239ff9 | 792 | struct memblock_region *p; |
95f72d1e YL |
793 | |
794 | if (!memory_limit) | |
795 | return; | |
796 | ||
797 | /* Truncate the memblock regions to satisfy the memory limit. */ | |
798 | limit = memory_limit; | |
799 | for (i = 0; i < memblock.memory.cnt; i++) { | |
e3239ff9 BH |
800 | if (limit > memblock.memory.regions[i].size) { |
801 | limit -= memblock.memory.regions[i].size; | |
95f72d1e YL |
802 | continue; |
803 | } | |
804 | ||
e3239ff9 | 805 | memblock.memory.regions[i].size = limit; |
95f72d1e YL |
806 | memblock.memory.cnt = i + 1; |
807 | break; | |
808 | } | |
809 | ||
95f72d1e YL |
810 | memory_limit = memblock_end_of_DRAM(); |
811 | ||
812 | /* And truncate any reserves above the limit also. */ | |
813 | for (i = 0; i < memblock.reserved.cnt; i++) { | |
e3239ff9 | 814 | p = &memblock.reserved.regions[i]; |
95f72d1e YL |
815 | |
816 | if (p->base > memory_limit) | |
817 | p->size = 0; | |
818 | else if ((p->base + p->size) > memory_limit) | |
819 | p->size = memory_limit - p->base; | |
820 | ||
821 | if (p->size == 0) { | |
822 | memblock_remove_region(&memblock.reserved, i); | |
823 | i--; | |
824 | } | |
825 | } | |
826 | } | |
827 | ||
cd79481d | 828 | static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) |
72d4b0b4 BH |
829 | { |
830 | unsigned int left = 0, right = type->cnt; | |
831 | ||
832 | do { | |
833 | unsigned int mid = (right + left) / 2; | |
834 | ||
835 | if (addr < type->regions[mid].base) | |
836 | right = mid; | |
837 | else if (addr >= (type->regions[mid].base + | |
838 | type->regions[mid].size)) | |
839 | left = mid + 1; | |
840 | else | |
841 | return mid; | |
842 | } while (left < right); | |
843 | return -1; | |
844 | } | |
845 | ||
2898cc4c | 846 | int __init memblock_is_reserved(phys_addr_t addr) |
95f72d1e | 847 | { |
72d4b0b4 BH |
848 | return memblock_search(&memblock.reserved, addr) != -1; |
849 | } | |
95f72d1e | 850 | |
3661ca66 | 851 | int __init_memblock memblock_is_memory(phys_addr_t addr) |
72d4b0b4 BH |
852 | { |
853 | return memblock_search(&memblock.memory, addr) != -1; | |
854 | } | |
855 | ||
3661ca66 | 856 | int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) |
72d4b0b4 | 857 | { |
abb65272 | 858 | int idx = memblock_search(&memblock.memory, base); |
72d4b0b4 BH |
859 | |
860 | if (idx == -1) | |
861 | return 0; | |
abb65272 TV |
862 | return memblock.memory.regions[idx].base <= base && |
863 | (memblock.memory.regions[idx].base + | |
864 | memblock.memory.regions[idx].size) >= (base + size); | |
95f72d1e YL |
865 | } |
866 | ||
10d06439 | 867 | int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) |
95f72d1e | 868 | { |
f1c2c19c | 869 | return memblock_overlaps_region(&memblock.reserved, base, size) >= 0; |
95f72d1e YL |
870 | } |
871 | ||
e63075a3 | 872 | |
3661ca66 | 873 | void __init_memblock memblock_set_current_limit(phys_addr_t limit) |
e63075a3 BH |
874 | { |
875 | memblock.current_limit = limit; | |
876 | } | |
877 | ||
7c0caeb8 | 878 | static void __init_memblock memblock_dump(struct memblock_type *type, char *name) |
6ed311b2 BH |
879 | { |
880 | unsigned long long base, size; | |
881 | int i; | |
882 | ||
7c0caeb8 | 883 | pr_info(" %s.cnt = 0x%lx\n", name, type->cnt); |
6ed311b2 | 884 | |
7c0caeb8 TH |
885 | for (i = 0; i < type->cnt; i++) { |
886 | struct memblock_region *rgn = &type->regions[i]; | |
887 | char nid_buf[32] = ""; | |
888 | ||
889 | base = rgn->base; | |
890 | size = rgn->size; | |
891 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
892 | if (memblock_get_region_node(rgn) != MAX_NUMNODES) | |
893 | snprintf(nid_buf, sizeof(nid_buf), " on node %d", | |
894 | memblock_get_region_node(rgn)); | |
895 | #endif | |
896 | pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s\n", | |
897 | name, i, base, base + size - 1, size, nid_buf); | |
6ed311b2 BH |
898 | } |
899 | } | |
900 | ||
4ff7b82f | 901 | void __init_memblock __memblock_dump_all(void) |
6ed311b2 | 902 | { |
6ed311b2 BH |
903 | pr_info("MEMBLOCK configuration:\n"); |
904 | pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size); | |
905 | ||
906 | memblock_dump(&memblock.memory, "memory"); | |
907 | memblock_dump(&memblock.reserved, "reserved"); | |
908 | } | |
909 | ||
910 | void __init memblock_analyze(void) | |
911 | { | |
912 | int i; | |
913 | ||
6ed311b2 BH |
914 | memblock.memory_size = 0; |
915 | ||
916 | for (i = 0; i < memblock.memory.cnt; i++) | |
917 | memblock.memory_size += memblock.memory.regions[i].size; | |
142b45a7 BH |
918 | |
919 | /* We allow resizing from there */ | |
920 | memblock_can_resize = 1; | |
6ed311b2 BH |
921 | } |
922 | ||
7590abe8 BH |
923 | void __init memblock_init(void) |
924 | { | |
236260b9 JF |
925 | static int init_done __initdata = 0; |
926 | ||
927 | if (init_done) | |
928 | return; | |
929 | init_done = 1; | |
930 | ||
7590abe8 BH |
931 | /* Hookup the initial arrays */ |
932 | memblock.memory.regions = memblock_memory_init_regions; | |
933 | memblock.memory.max = INIT_MEMBLOCK_REGIONS; | |
934 | memblock.reserved.regions = memblock_reserved_init_regions; | |
935 | memblock.reserved.max = INIT_MEMBLOCK_REGIONS; | |
936 | ||
7590abe8 BH |
937 | /* Create a dummy zero size MEMBLOCK which will get coalesced away later. |
938 | * This simplifies the memblock_add() code below... | |
939 | */ | |
940 | memblock.memory.regions[0].base = 0; | |
941 | memblock.memory.regions[0].size = 0; | |
7c0caeb8 | 942 | memblock_set_region_node(&memblock.memory.regions[0], MAX_NUMNODES); |
7590abe8 BH |
943 | memblock.memory.cnt = 1; |
944 | ||
945 | /* Ditto. */ | |
946 | memblock.reserved.regions[0].base = 0; | |
947 | memblock.reserved.regions[0].size = 0; | |
7c0caeb8 | 948 | memblock_set_region_node(&memblock.reserved.regions[0], MAX_NUMNODES); |
7590abe8 BH |
949 | memblock.reserved.cnt = 1; |
950 | ||
951 | memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE; | |
952 | } | |
953 | ||
6ed311b2 BH |
954 | static int __init early_memblock(char *p) |
955 | { | |
956 | if (p && strstr(p, "debug")) | |
957 | memblock_debug = 1; | |
958 | return 0; | |
959 | } | |
960 | early_param("memblock", early_memblock); | |
961 | ||
c378ddd5 | 962 | #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK) |
6d03b885 BH |
963 | |
964 | static int memblock_debug_show(struct seq_file *m, void *private) | |
965 | { | |
966 | struct memblock_type *type = m->private; | |
967 | struct memblock_region *reg; | |
968 | int i; | |
969 | ||
970 | for (i = 0; i < type->cnt; i++) { | |
971 | reg = &type->regions[i]; | |
972 | seq_printf(m, "%4d: ", i); | |
973 | if (sizeof(phys_addr_t) == 4) | |
974 | seq_printf(m, "0x%08lx..0x%08lx\n", | |
975 | (unsigned long)reg->base, | |
976 | (unsigned long)(reg->base + reg->size - 1)); | |
977 | else | |
978 | seq_printf(m, "0x%016llx..0x%016llx\n", | |
979 | (unsigned long long)reg->base, | |
980 | (unsigned long long)(reg->base + reg->size - 1)); | |
981 | ||
982 | } | |
983 | return 0; | |
984 | } | |
985 | ||
986 | static int memblock_debug_open(struct inode *inode, struct file *file) | |
987 | { | |
988 | return single_open(file, memblock_debug_show, inode->i_private); | |
989 | } | |
990 | ||
991 | static const struct file_operations memblock_debug_fops = { | |
992 | .open = memblock_debug_open, | |
993 | .read = seq_read, | |
994 | .llseek = seq_lseek, | |
995 | .release = single_release, | |
996 | }; | |
997 | ||
998 | static int __init memblock_init_debugfs(void) | |
999 | { | |
1000 | struct dentry *root = debugfs_create_dir("memblock", NULL); | |
1001 | if (!root) | |
1002 | return -ENXIO; | |
1003 | debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops); | |
1004 | debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops); | |
1005 | ||
1006 | return 0; | |
1007 | } | |
1008 | __initcall(memblock_init_debugfs); | |
1009 | ||
1010 | #endif /* CONFIG_DEBUG_FS */ |