]>
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 | ||
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 | ||
142b45a7 BH |
178 | /* Defined below but needed now */ |
179 | static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size); | |
180 | ||
10d06439 | 181 | static int __init_memblock memblock_double_array(struct memblock_type *type) |
142b45a7 BH |
182 | { |
183 | struct memblock_region *new_array, *old_array; | |
184 | phys_addr_t old_size, new_size, addr; | |
185 | int use_slab = slab_is_available(); | |
186 | ||
187 | /* We don't allow resizing until we know about the reserved regions | |
188 | * of memory that aren't suitable for allocation | |
189 | */ | |
190 | if (!memblock_can_resize) | |
191 | return -1; | |
192 | ||
142b45a7 BH |
193 | /* Calculate new doubled size */ |
194 | old_size = type->max * sizeof(struct memblock_region); | |
195 | new_size = old_size << 1; | |
196 | ||
197 | /* Try to find some space for it. | |
198 | * | |
199 | * WARNING: We assume that either slab_is_available() and we use it or | |
200 | * we use MEMBLOCK for allocations. That means that this is unsafe to use | |
201 | * when bootmem is currently active (unless bootmem itself is implemented | |
202 | * on top of MEMBLOCK which isn't the case yet) | |
203 | * | |
204 | * This should however not be an issue for now, as we currently only | |
205 | * call into MEMBLOCK while it's still active, or much later when slab is | |
206 | * active for memory hotplug operations | |
207 | */ | |
208 | if (use_slab) { | |
209 | new_array = kmalloc(new_size, GFP_KERNEL); | |
1f5026a7 | 210 | addr = new_array ? __pa(new_array) : 0; |
142b45a7 | 211 | } else |
fc769a8e | 212 | addr = memblock_find_in_range(0, MEMBLOCK_ALLOC_ACCESSIBLE, new_size, sizeof(phys_addr_t)); |
1f5026a7 | 213 | if (!addr) { |
142b45a7 BH |
214 | pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", |
215 | memblock_type_name(type), type->max, type->max * 2); | |
216 | return -1; | |
217 | } | |
218 | new_array = __va(addr); | |
219 | ||
ea9e4376 YL |
220 | memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]", |
221 | memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1); | |
222 | ||
142b45a7 BH |
223 | /* Found space, we now need to move the array over before |
224 | * we add the reserved region since it may be our reserved | |
225 | * array itself that is full. | |
226 | */ | |
227 | memcpy(new_array, type->regions, old_size); | |
228 | memset(new_array + type->max, 0, old_size); | |
229 | old_array = type->regions; | |
230 | type->regions = new_array; | |
231 | type->max <<= 1; | |
232 | ||
233 | /* If we use SLAB that's it, we are done */ | |
234 | if (use_slab) | |
235 | return 0; | |
236 | ||
237 | /* Add the new reserved region now. Should not fail ! */ | |
8f7a6605 | 238 | BUG_ON(memblock_add_region(&memblock.reserved, addr, new_size)); |
142b45a7 BH |
239 | |
240 | /* If the array wasn't our static init one, then free it. We only do | |
241 | * that before SLAB is available as later on, we don't know whether | |
242 | * to use kfree or free_bootmem_pages(). Shouldn't be a big deal | |
243 | * anyways | |
244 | */ | |
245 | if (old_array != memblock_memory_init_regions && | |
246 | old_array != memblock_reserved_init_regions) | |
247 | memblock_free(__pa(old_array), old_size); | |
248 | ||
249 | return 0; | |
250 | } | |
251 | ||
784656f9 TH |
252 | /** |
253 | * memblock_merge_regions - merge neighboring compatible regions | |
254 | * @type: memblock type to scan | |
255 | * | |
256 | * Scan @type and merge neighboring compatible regions. | |
257 | */ | |
258 | static void __init_memblock memblock_merge_regions(struct memblock_type *type) | |
95f72d1e | 259 | { |
784656f9 | 260 | int i = 0; |
95f72d1e | 261 | |
784656f9 TH |
262 | /* cnt never goes below 1 */ |
263 | while (i < type->cnt - 1) { | |
264 | struct memblock_region *this = &type->regions[i]; | |
265 | struct memblock_region *next = &type->regions[i + 1]; | |
95f72d1e | 266 | |
7c0caeb8 TH |
267 | if (this->base + this->size != next->base || |
268 | memblock_get_region_node(this) != | |
269 | memblock_get_region_node(next)) { | |
784656f9 TH |
270 | BUG_ON(this->base + this->size > next->base); |
271 | i++; | |
272 | continue; | |
8f7a6605 BH |
273 | } |
274 | ||
784656f9 TH |
275 | this->size += next->size; |
276 | memmove(next, next + 1, (type->cnt - (i + 1)) * sizeof(*next)); | |
277 | type->cnt--; | |
95f72d1e | 278 | } |
784656f9 | 279 | } |
95f72d1e | 280 | |
784656f9 TH |
281 | /** |
282 | * memblock_insert_region - insert new memblock region | |
283 | * @type: memblock type to insert into | |
284 | * @idx: index for the insertion point | |
285 | * @base: base address of the new region | |
286 | * @size: size of the new region | |
287 | * | |
288 | * Insert new memblock region [@base,@base+@size) into @type at @idx. | |
289 | * @type must already have extra room to accomodate the new region. | |
290 | */ | |
291 | static void __init_memblock memblock_insert_region(struct memblock_type *type, | |
292 | int idx, phys_addr_t base, | |
7c0caeb8 | 293 | phys_addr_t size, int nid) |
784656f9 TH |
294 | { |
295 | struct memblock_region *rgn = &type->regions[idx]; | |
296 | ||
297 | BUG_ON(type->cnt >= type->max); | |
298 | memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); | |
299 | rgn->base = base; | |
300 | rgn->size = size; | |
7c0caeb8 | 301 | memblock_set_region_node(rgn, nid); |
784656f9 TH |
302 | type->cnt++; |
303 | } | |
304 | ||
305 | /** | |
306 | * memblock_add_region - add new memblock region | |
307 | * @type: memblock type to add new region into | |
308 | * @base: base address of the new region | |
309 | * @size: size of the new region | |
310 | * | |
311 | * Add new memblock region [@base,@base+@size) into @type. The new region | |
312 | * is allowed to overlap with existing ones - overlaps don't affect already | |
313 | * existing regions. @type is guaranteed to be minimal (all neighbouring | |
314 | * compatible regions are merged) after the addition. | |
315 | * | |
316 | * RETURNS: | |
317 | * 0 on success, -errno on failure. | |
318 | */ | |
319 | static long __init_memblock memblock_add_region(struct memblock_type *type, | |
320 | phys_addr_t base, phys_addr_t size) | |
321 | { | |
322 | bool insert = false; | |
323 | phys_addr_t obase = base, end = base + size; | |
324 | int i, nr_new; | |
325 | ||
326 | /* special case for empty array */ | |
327 | if (type->regions[0].size == 0) { | |
328 | WARN_ON(type->cnt != 1); | |
8f7a6605 BH |
329 | type->regions[0].base = base; |
330 | type->regions[0].size = size; | |
7c0caeb8 | 331 | memblock_set_region_node(&type->regions[0], MAX_NUMNODES); |
8f7a6605 | 332 | return 0; |
95f72d1e | 333 | } |
784656f9 TH |
334 | repeat: |
335 | /* | |
336 | * The following is executed twice. Once with %false @insert and | |
337 | * then with %true. The first counts the number of regions needed | |
338 | * to accomodate the new area. The second actually inserts them. | |
142b45a7 | 339 | */ |
784656f9 TH |
340 | base = obase; |
341 | nr_new = 0; | |
95f72d1e | 342 | |
784656f9 TH |
343 | for (i = 0; i < type->cnt; i++) { |
344 | struct memblock_region *rgn = &type->regions[i]; | |
345 | phys_addr_t rbase = rgn->base; | |
346 | phys_addr_t rend = rbase + rgn->size; | |
347 | ||
348 | if (rbase >= end) | |
95f72d1e | 349 | break; |
784656f9 TH |
350 | if (rend <= base) |
351 | continue; | |
352 | /* | |
353 | * @rgn overlaps. If it separates the lower part of new | |
354 | * area, insert that portion. | |
355 | */ | |
356 | if (rbase > base) { | |
357 | nr_new++; | |
358 | if (insert) | |
359 | memblock_insert_region(type, i++, base, | |
7c0caeb8 | 360 | rbase - base, MAX_NUMNODES); |
95f72d1e | 361 | } |
784656f9 TH |
362 | /* area below @rend is dealt with, forget about it */ |
363 | base = min(rend, end); | |
95f72d1e | 364 | } |
784656f9 TH |
365 | |
366 | /* insert the remaining portion */ | |
367 | if (base < end) { | |
368 | nr_new++; | |
369 | if (insert) | |
7c0caeb8 TH |
370 | memblock_insert_region(type, i, base, end - base, |
371 | MAX_NUMNODES); | |
95f72d1e | 372 | } |
95f72d1e | 373 | |
784656f9 TH |
374 | /* |
375 | * If this was the first round, resize array and repeat for actual | |
376 | * insertions; otherwise, merge and return. | |
142b45a7 | 377 | */ |
784656f9 TH |
378 | if (!insert) { |
379 | while (type->cnt + nr_new > type->max) | |
380 | if (memblock_double_array(type) < 0) | |
381 | return -ENOMEM; | |
382 | insert = true; | |
383 | goto repeat; | |
384 | } else { | |
385 | memblock_merge_regions(type); | |
386 | return 0; | |
142b45a7 | 387 | } |
95f72d1e YL |
388 | } |
389 | ||
10d06439 | 390 | long __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) |
95f72d1e | 391 | { |
e3239ff9 | 392 | return memblock_add_region(&memblock.memory, base, size); |
95f72d1e YL |
393 | } |
394 | ||
8f7a6605 BH |
395 | static long __init_memblock __memblock_remove(struct memblock_type *type, |
396 | phys_addr_t base, phys_addr_t size) | |
95f72d1e | 397 | { |
2898cc4c | 398 | phys_addr_t end = base + size; |
95f72d1e YL |
399 | int i; |
400 | ||
8f7a6605 BH |
401 | /* Walk through the array for collisions */ |
402 | for (i = 0; i < type->cnt; i++) { | |
403 | struct memblock_region *rgn = &type->regions[i]; | |
404 | phys_addr_t rend = rgn->base + rgn->size; | |
95f72d1e | 405 | |
8f7a6605 BH |
406 | /* Nothing more to do, exit */ |
407 | if (rgn->base > end || rgn->size == 0) | |
95f72d1e | 408 | break; |
95f72d1e | 409 | |
8f7a6605 BH |
410 | /* If we fully enclose the block, drop it */ |
411 | if (base <= rgn->base && end >= rend) { | |
412 | memblock_remove_region(type, i--); | |
413 | continue; | |
414 | } | |
95f72d1e | 415 | |
8f7a6605 BH |
416 | /* If we are fully enclosed within a block |
417 | * then we need to split it and we are done | |
418 | */ | |
419 | if (base > rgn->base && end < rend) { | |
420 | rgn->size = base - rgn->base; | |
421 | if (!memblock_add_region(type, end, rend - end)) | |
422 | return 0; | |
423 | /* Failure to split is bad, we at least | |
424 | * restore the block before erroring | |
425 | */ | |
426 | rgn->size = rend - rgn->base; | |
427 | WARN_ON(1); | |
428 | return -1; | |
429 | } | |
95f72d1e | 430 | |
8f7a6605 BH |
431 | /* Check if we need to trim the bottom of a block */ |
432 | if (rgn->base < end && rend > end) { | |
433 | rgn->size -= end - rgn->base; | |
434 | rgn->base = end; | |
435 | break; | |
436 | } | |
95f72d1e | 437 | |
8f7a6605 BH |
438 | /* And check if we need to trim the top of a block */ |
439 | if (base < rend) | |
440 | rgn->size -= rend - base; | |
95f72d1e | 441 | |
8f7a6605 BH |
442 | } |
443 | return 0; | |
95f72d1e YL |
444 | } |
445 | ||
10d06439 | 446 | long __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) |
95f72d1e YL |
447 | { |
448 | return __memblock_remove(&memblock.memory, base, size); | |
449 | } | |
450 | ||
3661ca66 | 451 | long __init_memblock memblock_free(phys_addr_t base, phys_addr_t size) |
95f72d1e | 452 | { |
24aa0788 | 453 | memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n", |
a150439c PA |
454 | (unsigned long long)base, |
455 | (unsigned long long)base + size, | |
456 | (void *)_RET_IP_); | |
24aa0788 | 457 | |
95f72d1e YL |
458 | return __memblock_remove(&memblock.reserved, base, size); |
459 | } | |
460 | ||
3661ca66 | 461 | long __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) |
95f72d1e | 462 | { |
e3239ff9 | 463 | struct memblock_type *_rgn = &memblock.reserved; |
95f72d1e | 464 | |
24aa0788 | 465 | memblock_dbg("memblock_reserve: [%#016llx-%#016llx] %pF\n", |
a150439c PA |
466 | (unsigned long long)base, |
467 | (unsigned long long)base + size, | |
468 | (void *)_RET_IP_); | |
95f72d1e YL |
469 | BUG_ON(0 == size); |
470 | ||
471 | return memblock_add_region(_rgn, base, size); | |
472 | } | |
473 | ||
35fd0808 TH |
474 | /** |
475 | * __next_free_mem_range - next function for for_each_free_mem_range() | |
476 | * @idx: pointer to u64 loop variable | |
477 | * @nid: nid: node selector, %MAX_NUMNODES for all nodes | |
478 | * @p_start: ptr to phys_addr_t for start address of the range, can be %NULL | |
479 | * @p_end: ptr to phys_addr_t for end address of the range, can be %NULL | |
480 | * @p_nid: ptr to int for nid of the range, can be %NULL | |
481 | * | |
482 | * Find the first free area from *@idx which matches @nid, fill the out | |
483 | * parameters, and update *@idx for the next iteration. The lower 32bit of | |
484 | * *@idx contains index into memory region and the upper 32bit indexes the | |
485 | * areas before each reserved region. For example, if reserved regions | |
486 | * look like the following, | |
487 | * | |
488 | * 0:[0-16), 1:[32-48), 2:[128-130) | |
489 | * | |
490 | * The upper 32bit indexes the following regions. | |
491 | * | |
492 | * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) | |
493 | * | |
494 | * As both region arrays are sorted, the function advances the two indices | |
495 | * in lockstep and returns each intersection. | |
496 | */ | |
497 | void __init_memblock __next_free_mem_range(u64 *idx, int nid, | |
498 | phys_addr_t *out_start, | |
499 | phys_addr_t *out_end, int *out_nid) | |
500 | { | |
501 | struct memblock_type *mem = &memblock.memory; | |
502 | struct memblock_type *rsv = &memblock.reserved; | |
503 | int mi = *idx & 0xffffffff; | |
504 | int ri = *idx >> 32; | |
505 | ||
506 | for ( ; mi < mem->cnt; mi++) { | |
507 | struct memblock_region *m = &mem->regions[mi]; | |
508 | phys_addr_t m_start = m->base; | |
509 | phys_addr_t m_end = m->base + m->size; | |
510 | ||
511 | /* only memory regions are associated with nodes, check it */ | |
512 | if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m)) | |
513 | continue; | |
514 | ||
515 | /* scan areas before each reservation for intersection */ | |
516 | for ( ; ri < rsv->cnt + 1; ri++) { | |
517 | struct memblock_region *r = &rsv->regions[ri]; | |
518 | phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0; | |
519 | phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX; | |
520 | ||
521 | /* if ri advanced past mi, break out to advance mi */ | |
522 | if (r_start >= m_end) | |
523 | break; | |
524 | /* if the two regions intersect, we're done */ | |
525 | if (m_start < r_end) { | |
526 | if (out_start) | |
527 | *out_start = max(m_start, r_start); | |
528 | if (out_end) | |
529 | *out_end = min(m_end, r_end); | |
530 | if (out_nid) | |
531 | *out_nid = memblock_get_region_node(m); | |
532 | /* | |
533 | * The region which ends first is advanced | |
534 | * for the next iteration. | |
535 | */ | |
536 | if (m_end <= r_end) | |
537 | mi++; | |
538 | else | |
539 | ri++; | |
540 | *idx = (u32)mi | (u64)ri << 32; | |
541 | return; | |
542 | } | |
543 | } | |
544 | } | |
545 | ||
546 | /* signal end of iteration */ | |
547 | *idx = ULLONG_MAX; | |
548 | } | |
549 | ||
7c0caeb8 TH |
550 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
551 | /* | |
552 | * Common iterator interface used to define for_each_mem_range(). | |
553 | */ | |
554 | void __init_memblock __next_mem_pfn_range(int *idx, int nid, | |
555 | unsigned long *out_start_pfn, | |
556 | unsigned long *out_end_pfn, int *out_nid) | |
557 | { | |
558 | struct memblock_type *type = &memblock.memory; | |
559 | struct memblock_region *r; | |
560 | ||
561 | while (++*idx < type->cnt) { | |
562 | r = &type->regions[*idx]; | |
563 | ||
564 | if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) | |
565 | continue; | |
566 | if (nid == MAX_NUMNODES || nid == r->nid) | |
567 | break; | |
568 | } | |
569 | if (*idx >= type->cnt) { | |
570 | *idx = -1; | |
571 | return; | |
572 | } | |
573 | ||
574 | if (out_start_pfn) | |
575 | *out_start_pfn = PFN_UP(r->base); | |
576 | if (out_end_pfn) | |
577 | *out_end_pfn = PFN_DOWN(r->base + r->size); | |
578 | if (out_nid) | |
579 | *out_nid = r->nid; | |
580 | } | |
581 | ||
582 | /** | |
583 | * memblock_set_node - set node ID on memblock regions | |
584 | * @base: base of area to set node ID for | |
585 | * @size: size of area to set node ID for | |
586 | * @nid: node ID to set | |
587 | * | |
588 | * Set the nid of memblock memory regions in [@base,@base+@size) to @nid. | |
589 | * Regions which cross the area boundaries are split as necessary. | |
590 | * | |
591 | * RETURNS: | |
592 | * 0 on success, -errno on failure. | |
593 | */ | |
594 | int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, | |
595 | int nid) | |
596 | { | |
597 | struct memblock_type *type = &memblock.memory; | |
598 | phys_addr_t end = base + size; | |
599 | int i; | |
600 | ||
601 | /* we'll create at most two more regions */ | |
602 | while (type->cnt + 2 > type->max) | |
603 | if (memblock_double_array(type) < 0) | |
604 | return -ENOMEM; | |
605 | ||
606 | for (i = 0; i < type->cnt; i++) { | |
607 | struct memblock_region *rgn = &type->regions[i]; | |
608 | phys_addr_t rbase = rgn->base; | |
609 | phys_addr_t rend = rbase + rgn->size; | |
610 | ||
611 | if (rbase >= end) | |
612 | break; | |
613 | if (rend <= base) | |
614 | continue; | |
615 | ||
616 | if (rbase < base) { | |
617 | /* | |
618 | * @rgn intersects from below. Split and continue | |
619 | * to process the next region - the new top half. | |
620 | */ | |
621 | rgn->base = base; | |
622 | rgn->size = rend - rgn->base; | |
623 | memblock_insert_region(type, i, rbase, base - rbase, | |
624 | rgn->nid); | |
625 | } else if (rend > end) { | |
626 | /* | |
627 | * @rgn intersects from above. Split and redo the | |
628 | * current region - the new bottom half. | |
629 | */ | |
630 | rgn->base = end; | |
631 | rgn->size = rend - rgn->base; | |
632 | memblock_insert_region(type, i--, rbase, end - rbase, | |
633 | rgn->nid); | |
634 | } else { | |
635 | /* @rgn is fully contained, set ->nid */ | |
636 | rgn->nid = nid; | |
637 | } | |
638 | } | |
639 | ||
640 | memblock_merge_regions(type); | |
641 | return 0; | |
642 | } | |
643 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
644 | ||
6ed311b2 | 645 | phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) |
95f72d1e | 646 | { |
6ed311b2 | 647 | phys_addr_t found; |
95f72d1e | 648 | |
6ed311b2 BH |
649 | /* We align the size to limit fragmentation. Without this, a lot of |
650 | * small allocs quickly eat up the whole reserve array on sparc | |
651 | */ | |
348968eb | 652 | size = round_up(size, align); |
95f72d1e | 653 | |
fc769a8e | 654 | found = memblock_find_in_range(0, max_addr, size, align); |
1f5026a7 | 655 | if (found && !memblock_add_region(&memblock.reserved, found, size)) |
6ed311b2 | 656 | return found; |
95f72d1e | 657 | |
6ed311b2 | 658 | return 0; |
95f72d1e YL |
659 | } |
660 | ||
6ed311b2 | 661 | phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) |
95f72d1e | 662 | { |
6ed311b2 BH |
663 | phys_addr_t alloc; |
664 | ||
665 | alloc = __memblock_alloc_base(size, align, max_addr); | |
666 | ||
667 | if (alloc == 0) | |
668 | panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n", | |
669 | (unsigned long long) size, (unsigned long long) max_addr); | |
670 | ||
671 | return alloc; | |
95f72d1e YL |
672 | } |
673 | ||
6ed311b2 | 674 | phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align) |
95f72d1e | 675 | { |
6ed311b2 BH |
676 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
677 | } | |
95f72d1e | 678 | |
95f72d1e | 679 | |
6ed311b2 | 680 | /* |
34e18455 | 681 | * Additional node-local top-down allocators. |
c196f76f BH |
682 | * |
683 | * WARNING: Only available after early_node_map[] has been populated, | |
684 | * on some architectures, that is after all the calls to add_active_range() | |
685 | * have been done to populate it. | |
6ed311b2 | 686 | */ |
95f72d1e | 687 | |
34e18455 TH |
688 | static phys_addr_t __init memblock_nid_range_rev(phys_addr_t start, |
689 | phys_addr_t end, int *nid) | |
c3f72b57 | 690 | { |
c196f76f | 691 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
c196f76f BH |
692 | unsigned long start_pfn, end_pfn; |
693 | int i; | |
694 | ||
b2fea988 | 695 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, nid) |
34e18455 TH |
696 | if (end > PFN_PHYS(start_pfn) && end <= PFN_PHYS(end_pfn)) |
697 | return max(start, PFN_PHYS(start_pfn)); | |
c196f76f | 698 | #endif |
c3f72b57 | 699 | *nid = 0; |
34e18455 | 700 | return start; |
c3f72b57 BH |
701 | } |
702 | ||
e6498040 TH |
703 | phys_addr_t __init memblock_find_in_range_node(phys_addr_t start, |
704 | phys_addr_t end, | |
2898cc4c BH |
705 | phys_addr_t size, |
706 | phys_addr_t align, int nid) | |
95f72d1e | 707 | { |
e6498040 TH |
708 | struct memblock_type *mem = &memblock.memory; |
709 | int i; | |
95f72d1e | 710 | |
e6498040 | 711 | BUG_ON(0 == size); |
95f72d1e | 712 | |
e6498040 TH |
713 | /* Pump up max_addr */ |
714 | if (end == MEMBLOCK_ALLOC_ACCESSIBLE) | |
715 | end = memblock.current_limit; | |
95f72d1e | 716 | |
e6498040 TH |
717 | for (i = mem->cnt - 1; i >= 0; i--) { |
718 | struct memblock_region *r = &mem->regions[i]; | |
719 | phys_addr_t base = max(start, r->base); | |
720 | phys_addr_t top = min(end, r->base + r->size); | |
721 | ||
722 | while (base < top) { | |
723 | phys_addr_t tbase, ret; | |
724 | int tnid; | |
725 | ||
726 | tbase = memblock_nid_range_rev(base, top, &tnid); | |
727 | if (nid == MAX_NUMNODES || tnid == nid) { | |
728 | ret = memblock_find_region(tbase, top, size, align); | |
729 | if (ret) | |
730 | return ret; | |
731 | } | |
732 | top = tbase; | |
95f72d1e | 733 | } |
95f72d1e | 734 | } |
e6498040 | 735 | |
1f5026a7 | 736 | return 0; |
95f72d1e YL |
737 | } |
738 | ||
2898cc4c | 739 | phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid) |
95f72d1e | 740 | { |
e6498040 | 741 | phys_addr_t found; |
95f72d1e | 742 | |
e6498040 TH |
743 | /* |
744 | * We align the size to limit fragmentation. Without this, a lot of | |
7f219c73 BH |
745 | * small allocs quickly eat up the whole reserve array on sparc |
746 | */ | |
348968eb | 747 | size = round_up(size, align); |
7f219c73 | 748 | |
e6498040 TH |
749 | found = memblock_find_in_range_node(0, MEMBLOCK_ALLOC_ACCESSIBLE, |
750 | size, align, nid); | |
751 | if (found && !memblock_add_region(&memblock.reserved, found, size)) | |
752 | return found; | |
95f72d1e | 753 | |
9d1e2492 BH |
754 | return 0; |
755 | } | |
756 | ||
757 | phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) | |
758 | { | |
759 | phys_addr_t res = memblock_alloc_nid(size, align, nid); | |
760 | ||
761 | if (res) | |
762 | return res; | |
15fb0972 | 763 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
95f72d1e YL |
764 | } |
765 | ||
9d1e2492 BH |
766 | |
767 | /* | |
768 | * Remaining API functions | |
769 | */ | |
770 | ||
95f72d1e | 771 | /* You must call memblock_analyze() before this. */ |
2898cc4c | 772 | phys_addr_t __init memblock_phys_mem_size(void) |
95f72d1e | 773 | { |
4734b594 | 774 | return memblock.memory_size; |
95f72d1e YL |
775 | } |
776 | ||
0a93ebef SR |
777 | /* lowest address */ |
778 | phys_addr_t __init_memblock memblock_start_of_DRAM(void) | |
779 | { | |
780 | return memblock.memory.regions[0].base; | |
781 | } | |
782 | ||
10d06439 | 783 | phys_addr_t __init_memblock memblock_end_of_DRAM(void) |
95f72d1e YL |
784 | { |
785 | int idx = memblock.memory.cnt - 1; | |
786 | ||
e3239ff9 | 787 | return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); |
95f72d1e YL |
788 | } |
789 | ||
790 | /* You must call memblock_analyze() after this. */ | |
2898cc4c | 791 | void __init memblock_enforce_memory_limit(phys_addr_t memory_limit) |
95f72d1e YL |
792 | { |
793 | unsigned long i; | |
2898cc4c | 794 | phys_addr_t limit; |
e3239ff9 | 795 | struct memblock_region *p; |
95f72d1e YL |
796 | |
797 | if (!memory_limit) | |
798 | return; | |
799 | ||
800 | /* Truncate the memblock regions to satisfy the memory limit. */ | |
801 | limit = memory_limit; | |
802 | for (i = 0; i < memblock.memory.cnt; i++) { | |
e3239ff9 BH |
803 | if (limit > memblock.memory.regions[i].size) { |
804 | limit -= memblock.memory.regions[i].size; | |
95f72d1e YL |
805 | continue; |
806 | } | |
807 | ||
e3239ff9 | 808 | memblock.memory.regions[i].size = limit; |
95f72d1e YL |
809 | memblock.memory.cnt = i + 1; |
810 | break; | |
811 | } | |
812 | ||
95f72d1e YL |
813 | memory_limit = memblock_end_of_DRAM(); |
814 | ||
815 | /* And truncate any reserves above the limit also. */ | |
816 | for (i = 0; i < memblock.reserved.cnt; i++) { | |
e3239ff9 | 817 | p = &memblock.reserved.regions[i]; |
95f72d1e YL |
818 | |
819 | if (p->base > memory_limit) | |
820 | p->size = 0; | |
821 | else if ((p->base + p->size) > memory_limit) | |
822 | p->size = memory_limit - p->base; | |
823 | ||
824 | if (p->size == 0) { | |
825 | memblock_remove_region(&memblock.reserved, i); | |
826 | i--; | |
827 | } | |
828 | } | |
829 | } | |
830 | ||
cd79481d | 831 | static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr) |
72d4b0b4 BH |
832 | { |
833 | unsigned int left = 0, right = type->cnt; | |
834 | ||
835 | do { | |
836 | unsigned int mid = (right + left) / 2; | |
837 | ||
838 | if (addr < type->regions[mid].base) | |
839 | right = mid; | |
840 | else if (addr >= (type->regions[mid].base + | |
841 | type->regions[mid].size)) | |
842 | left = mid + 1; | |
843 | else | |
844 | return mid; | |
845 | } while (left < right); | |
846 | return -1; | |
847 | } | |
848 | ||
2898cc4c | 849 | int __init memblock_is_reserved(phys_addr_t addr) |
95f72d1e | 850 | { |
72d4b0b4 BH |
851 | return memblock_search(&memblock.reserved, addr) != -1; |
852 | } | |
95f72d1e | 853 | |
3661ca66 | 854 | int __init_memblock memblock_is_memory(phys_addr_t addr) |
72d4b0b4 BH |
855 | { |
856 | return memblock_search(&memblock.memory, addr) != -1; | |
857 | } | |
858 | ||
3661ca66 | 859 | int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size) |
72d4b0b4 | 860 | { |
abb65272 | 861 | int idx = memblock_search(&memblock.memory, base); |
72d4b0b4 BH |
862 | |
863 | if (idx == -1) | |
864 | return 0; | |
abb65272 TV |
865 | return memblock.memory.regions[idx].base <= base && |
866 | (memblock.memory.regions[idx].base + | |
867 | memblock.memory.regions[idx].size) >= (base + size); | |
95f72d1e YL |
868 | } |
869 | ||
10d06439 | 870 | int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) |
95f72d1e | 871 | { |
f1c2c19c | 872 | return memblock_overlaps_region(&memblock.reserved, base, size) >= 0; |
95f72d1e YL |
873 | } |
874 | ||
e63075a3 | 875 | |
3661ca66 | 876 | void __init_memblock memblock_set_current_limit(phys_addr_t limit) |
e63075a3 BH |
877 | { |
878 | memblock.current_limit = limit; | |
879 | } | |
880 | ||
7c0caeb8 | 881 | static void __init_memblock memblock_dump(struct memblock_type *type, char *name) |
6ed311b2 BH |
882 | { |
883 | unsigned long long base, size; | |
884 | int i; | |
885 | ||
7c0caeb8 | 886 | pr_info(" %s.cnt = 0x%lx\n", name, type->cnt); |
6ed311b2 | 887 | |
7c0caeb8 TH |
888 | for (i = 0; i < type->cnt; i++) { |
889 | struct memblock_region *rgn = &type->regions[i]; | |
890 | char nid_buf[32] = ""; | |
891 | ||
892 | base = rgn->base; | |
893 | size = rgn->size; | |
894 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
895 | if (memblock_get_region_node(rgn) != MAX_NUMNODES) | |
896 | snprintf(nid_buf, sizeof(nid_buf), " on node %d", | |
897 | memblock_get_region_node(rgn)); | |
898 | #endif | |
899 | pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s\n", | |
900 | name, i, base, base + size - 1, size, nid_buf); | |
6ed311b2 BH |
901 | } |
902 | } | |
903 | ||
10d06439 | 904 | void __init_memblock memblock_dump_all(void) |
6ed311b2 BH |
905 | { |
906 | if (!memblock_debug) | |
907 | return; | |
908 | ||
909 | pr_info("MEMBLOCK configuration:\n"); | |
910 | pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size); | |
911 | ||
912 | memblock_dump(&memblock.memory, "memory"); | |
913 | memblock_dump(&memblock.reserved, "reserved"); | |
914 | } | |
915 | ||
916 | void __init memblock_analyze(void) | |
917 | { | |
918 | int i; | |
919 | ||
920 | /* Check marker in the unused last array entry */ | |
921 | WARN_ON(memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS].base | |
c9d8c3d0 | 922 | != MEMBLOCK_INACTIVE); |
6ed311b2 | 923 | WARN_ON(memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS].base |
c9d8c3d0 | 924 | != MEMBLOCK_INACTIVE); |
6ed311b2 BH |
925 | |
926 | memblock.memory_size = 0; | |
927 | ||
928 | for (i = 0; i < memblock.memory.cnt; i++) | |
929 | memblock.memory_size += memblock.memory.regions[i].size; | |
142b45a7 BH |
930 | |
931 | /* We allow resizing from there */ | |
932 | memblock_can_resize = 1; | |
6ed311b2 BH |
933 | } |
934 | ||
7590abe8 BH |
935 | void __init memblock_init(void) |
936 | { | |
236260b9 JF |
937 | static int init_done __initdata = 0; |
938 | ||
939 | if (init_done) | |
940 | return; | |
941 | init_done = 1; | |
942 | ||
7590abe8 BH |
943 | /* Hookup the initial arrays */ |
944 | memblock.memory.regions = memblock_memory_init_regions; | |
945 | memblock.memory.max = INIT_MEMBLOCK_REGIONS; | |
946 | memblock.reserved.regions = memblock_reserved_init_regions; | |
947 | memblock.reserved.max = INIT_MEMBLOCK_REGIONS; | |
948 | ||
949 | /* Write a marker in the unused last array entry */ | |
c9d8c3d0 AM |
950 | memblock.memory.regions[INIT_MEMBLOCK_REGIONS].base = MEMBLOCK_INACTIVE; |
951 | memblock.reserved.regions[INIT_MEMBLOCK_REGIONS].base = MEMBLOCK_INACTIVE; | |
7590abe8 BH |
952 | |
953 | /* Create a dummy zero size MEMBLOCK which will get coalesced away later. | |
954 | * This simplifies the memblock_add() code below... | |
955 | */ | |
956 | memblock.memory.regions[0].base = 0; | |
957 | memblock.memory.regions[0].size = 0; | |
7c0caeb8 | 958 | memblock_set_region_node(&memblock.memory.regions[0], MAX_NUMNODES); |
7590abe8 BH |
959 | memblock.memory.cnt = 1; |
960 | ||
961 | /* Ditto. */ | |
962 | memblock.reserved.regions[0].base = 0; | |
963 | memblock.reserved.regions[0].size = 0; | |
7c0caeb8 | 964 | memblock_set_region_node(&memblock.reserved.regions[0], MAX_NUMNODES); |
7590abe8 BH |
965 | memblock.reserved.cnt = 1; |
966 | ||
967 | memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE; | |
968 | } | |
969 | ||
6ed311b2 BH |
970 | static int __init early_memblock(char *p) |
971 | { | |
972 | if (p && strstr(p, "debug")) | |
973 | memblock_debug = 1; | |
974 | return 0; | |
975 | } | |
976 | early_param("memblock", early_memblock); | |
977 | ||
c378ddd5 | 978 | #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK) |
6d03b885 BH |
979 | |
980 | static int memblock_debug_show(struct seq_file *m, void *private) | |
981 | { | |
982 | struct memblock_type *type = m->private; | |
983 | struct memblock_region *reg; | |
984 | int i; | |
985 | ||
986 | for (i = 0; i < type->cnt; i++) { | |
987 | reg = &type->regions[i]; | |
988 | seq_printf(m, "%4d: ", i); | |
989 | if (sizeof(phys_addr_t) == 4) | |
990 | seq_printf(m, "0x%08lx..0x%08lx\n", | |
991 | (unsigned long)reg->base, | |
992 | (unsigned long)(reg->base + reg->size - 1)); | |
993 | else | |
994 | seq_printf(m, "0x%016llx..0x%016llx\n", | |
995 | (unsigned long long)reg->base, | |
996 | (unsigned long long)(reg->base + reg->size - 1)); | |
997 | ||
998 | } | |
999 | return 0; | |
1000 | } | |
1001 | ||
1002 | static int memblock_debug_open(struct inode *inode, struct file *file) | |
1003 | { | |
1004 | return single_open(file, memblock_debug_show, inode->i_private); | |
1005 | } | |
1006 | ||
1007 | static const struct file_operations memblock_debug_fops = { | |
1008 | .open = memblock_debug_open, | |
1009 | .read = seq_read, | |
1010 | .llseek = seq_lseek, | |
1011 | .release = single_release, | |
1012 | }; | |
1013 | ||
1014 | static int __init memblock_init_debugfs(void) | |
1015 | { | |
1016 | struct dentry *root = debugfs_create_dir("memblock", NULL); | |
1017 | if (!root) | |
1018 | return -ENXIO; | |
1019 | debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops); | |
1020 | debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops); | |
1021 | ||
1022 | return 0; | |
1023 | } | |
1024 | __initcall(memblock_init_debugfs); | |
1025 | ||
1026 | #endif /* CONFIG_DEBUG_FS */ |