]>
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 | ||
fe091c20 TH |
23 | static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock; |
24 | static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock; | |
25 | ||
26 | struct memblock memblock __initdata_memblock = { | |
27 | .memory.regions = memblock_memory_init_regions, | |
28 | .memory.cnt = 1, /* empty dummy entry */ | |
29 | .memory.max = INIT_MEMBLOCK_REGIONS, | |
30 | ||
31 | .reserved.regions = memblock_reserved_init_regions, | |
32 | .reserved.cnt = 1, /* empty dummy entry */ | |
33 | .reserved.max = INIT_MEMBLOCK_REGIONS, | |
34 | ||
35 | .current_limit = MEMBLOCK_ALLOC_ANYWHERE, | |
36 | }; | |
95f72d1e | 37 | |
10d06439 YL |
38 | int memblock_debug __initdata_memblock; |
39 | int memblock_can_resize __initdata_memblock; | |
95f72d1e | 40 | |
142b45a7 BH |
41 | /* inline so we don't get a warning when pr_debug is compiled out */ |
42 | static inline const char *memblock_type_name(struct memblock_type *type) | |
43 | { | |
44 | if (type == &memblock.memory) | |
45 | return "memory"; | |
46 | else if (type == &memblock.reserved) | |
47 | return "reserved"; | |
48 | else | |
49 | return "unknown"; | |
50 | } | |
51 | ||
eb18f1b5 TH |
52 | /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */ |
53 | static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size) | |
54 | { | |
55 | return *size = min(*size, (phys_addr_t)ULLONG_MAX - base); | |
56 | } | |
57 | ||
6ed311b2 BH |
58 | /* |
59 | * Address comparison utilities | |
60 | */ | |
10d06439 | 61 | static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1, |
2898cc4c | 62 | phys_addr_t base2, phys_addr_t size2) |
95f72d1e YL |
63 | { |
64 | return ((base1 < (base2 + size2)) && (base2 < (base1 + size1))); | |
65 | } | |
66 | ||
2d7d3eb2 HS |
67 | static long __init_memblock memblock_overlaps_region(struct memblock_type *type, |
68 | phys_addr_t base, phys_addr_t size) | |
6ed311b2 BH |
69 | { |
70 | unsigned long i; | |
71 | ||
72 | for (i = 0; i < type->cnt; i++) { | |
73 | phys_addr_t rgnbase = type->regions[i].base; | |
74 | phys_addr_t rgnsize = type->regions[i].size; | |
75 | if (memblock_addrs_overlap(base, size, rgnbase, rgnsize)) | |
76 | break; | |
77 | } | |
78 | ||
79 | return (i < type->cnt) ? i : -1; | |
80 | } | |
81 | ||
82 | /* | |
83 | * Find, allocate, deallocate or reserve unreserved regions. All allocations | |
84 | * are top-down. | |
85 | */ | |
86 | ||
cd79481d | 87 | static phys_addr_t __init_memblock memblock_find_region(phys_addr_t start, phys_addr_t end, |
6ed311b2 BH |
88 | phys_addr_t size, phys_addr_t align) |
89 | { | |
90 | phys_addr_t base, res_base; | |
91 | long j; | |
92 | ||
f1af98c7 YL |
93 | /* In case, huge size is requested */ |
94 | if (end < size) | |
1f5026a7 | 95 | return 0; |
f1af98c7 | 96 | |
348968eb | 97 | base = round_down(end - size, align); |
f1af98c7 | 98 | |
25818f0f BH |
99 | /* Prevent allocations returning 0 as it's also used to |
100 | * indicate an allocation failure | |
101 | */ | |
102 | if (start == 0) | |
103 | start = PAGE_SIZE; | |
104 | ||
6ed311b2 BH |
105 | while (start <= base) { |
106 | j = memblock_overlaps_region(&memblock.reserved, base, size); | |
107 | if (j < 0) | |
108 | return base; | |
109 | res_base = memblock.reserved.regions[j].base; | |
110 | if (res_base < size) | |
111 | break; | |
348968eb | 112 | base = round_down(res_base - size, align); |
6ed311b2 BH |
113 | } |
114 | ||
1f5026a7 | 115 | return 0; |
6ed311b2 BH |
116 | } |
117 | ||
fc769a8e TH |
118 | /* |
119 | * Find a free area with specified alignment in a specific range. | |
120 | */ | |
121 | phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, phys_addr_t end, | |
122 | phys_addr_t size, phys_addr_t align) | |
6ed311b2 BH |
123 | { |
124 | long i; | |
6ed311b2 BH |
125 | |
126 | BUG_ON(0 == size); | |
127 | ||
6ed311b2 | 128 | /* Pump up max_addr */ |
fef501d4 BH |
129 | if (end == MEMBLOCK_ALLOC_ACCESSIBLE) |
130 | end = memblock.current_limit; | |
6ed311b2 BH |
131 | |
132 | /* We do a top-down search, this tends to limit memory | |
133 | * fragmentation by keeping early boot allocs near the | |
134 | * top of memory | |
135 | */ | |
136 | for (i = memblock.memory.cnt - 1; i >= 0; i--) { | |
137 | phys_addr_t memblockbase = memblock.memory.regions[i].base; | |
138 | phys_addr_t memblocksize = memblock.memory.regions[i].size; | |
fef501d4 | 139 | phys_addr_t bottom, top, found; |
6ed311b2 BH |
140 | |
141 | if (memblocksize < size) | |
142 | continue; | |
fef501d4 BH |
143 | if ((memblockbase + memblocksize) <= start) |
144 | break; | |
145 | bottom = max(memblockbase, start); | |
146 | top = min(memblockbase + memblocksize, end); | |
147 | if (bottom >= top) | |
148 | continue; | |
149 | found = memblock_find_region(bottom, top, size, align); | |
1f5026a7 | 150 | if (found) |
fef501d4 | 151 | return found; |
6ed311b2 | 152 | } |
1f5026a7 | 153 | return 0; |
6ed311b2 BH |
154 | } |
155 | ||
7950c407 YL |
156 | /* |
157 | * Free memblock.reserved.regions | |
158 | */ | |
159 | int __init_memblock memblock_free_reserved_regions(void) | |
160 | { | |
161 | if (memblock.reserved.regions == memblock_reserved_init_regions) | |
162 | return 0; | |
163 | ||
164 | return memblock_free(__pa(memblock.reserved.regions), | |
165 | sizeof(struct memblock_region) * memblock.reserved.max); | |
166 | } | |
167 | ||
168 | /* | |
169 | * Reserve memblock.reserved.regions | |
170 | */ | |
171 | int __init_memblock memblock_reserve_reserved_regions(void) | |
172 | { | |
173 | if (memblock.reserved.regions == memblock_reserved_init_regions) | |
174 | return 0; | |
175 | ||
176 | return memblock_reserve(__pa(memblock.reserved.regions), | |
177 | sizeof(struct memblock_region) * memblock.reserved.max); | |
178 | } | |
179 | ||
10d06439 | 180 | static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) |
95f72d1e | 181 | { |
7c0caeb8 TH |
182 | memmove(&type->regions[r], &type->regions[r + 1], |
183 | (type->cnt - (r + 1)) * sizeof(type->regions[r])); | |
e3239ff9 | 184 | type->cnt--; |
95f72d1e | 185 | |
8f7a6605 BH |
186 | /* Special case for empty arrays */ |
187 | if (type->cnt == 0) { | |
188 | type->cnt = 1; | |
189 | type->regions[0].base = 0; | |
190 | type->regions[0].size = 0; | |
7c0caeb8 | 191 | memblock_set_region_node(&type->regions[0], MAX_NUMNODES); |
8f7a6605 | 192 | } |
95f72d1e YL |
193 | } |
194 | ||
10d06439 | 195 | static int __init_memblock memblock_double_array(struct memblock_type *type) |
142b45a7 BH |
196 | { |
197 | struct memblock_region *new_array, *old_array; | |
198 | phys_addr_t old_size, new_size, addr; | |
199 | int use_slab = slab_is_available(); | |
200 | ||
201 | /* We don't allow resizing until we know about the reserved regions | |
202 | * of memory that aren't suitable for allocation | |
203 | */ | |
204 | if (!memblock_can_resize) | |
205 | return -1; | |
206 | ||
142b45a7 BH |
207 | /* Calculate new doubled size */ |
208 | old_size = type->max * sizeof(struct memblock_region); | |
209 | new_size = old_size << 1; | |
210 | ||
211 | /* Try to find some space for it. | |
212 | * | |
213 | * WARNING: We assume that either slab_is_available() and we use it or | |
214 | * we use MEMBLOCK for allocations. That means that this is unsafe to use | |
215 | * when bootmem is currently active (unless bootmem itself is implemented | |
216 | * on top of MEMBLOCK which isn't the case yet) | |
217 | * | |
218 | * This should however not be an issue for now, as we currently only | |
219 | * call into MEMBLOCK while it's still active, or much later when slab is | |
220 | * active for memory hotplug operations | |
221 | */ | |
222 | if (use_slab) { | |
223 | new_array = kmalloc(new_size, GFP_KERNEL); | |
1f5026a7 | 224 | addr = new_array ? __pa(new_array) : 0; |
142b45a7 | 225 | } else |
fc769a8e | 226 | addr = memblock_find_in_range(0, MEMBLOCK_ALLOC_ACCESSIBLE, new_size, sizeof(phys_addr_t)); |
1f5026a7 | 227 | if (!addr) { |
142b45a7 BH |
228 | pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", |
229 | memblock_type_name(type), type->max, type->max * 2); | |
230 | return -1; | |
231 | } | |
232 | new_array = __va(addr); | |
233 | ||
ea9e4376 YL |
234 | memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]", |
235 | memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1); | |
236 | ||
142b45a7 BH |
237 | /* Found space, we now need to move the array over before |
238 | * we add the reserved region since it may be our reserved | |
239 | * array itself that is full. | |
240 | */ | |
241 | memcpy(new_array, type->regions, old_size); | |
242 | memset(new_array + type->max, 0, old_size); | |
243 | old_array = type->regions; | |
244 | type->regions = new_array; | |
245 | type->max <<= 1; | |
246 | ||
247 | /* If we use SLAB that's it, we are done */ | |
248 | if (use_slab) | |
249 | return 0; | |
250 | ||
251 | /* Add the new reserved region now. Should not fail ! */ | |
9c8c27e2 | 252 | BUG_ON(memblock_reserve(addr, new_size)); |
142b45a7 BH |
253 | |
254 | /* If the array wasn't our static init one, then free it. We only do | |
255 | * that before SLAB is available as later on, we don't know whether | |
256 | * to use kfree or free_bootmem_pages(). Shouldn't be a big deal | |
257 | * anyways | |
258 | */ | |
259 | if (old_array != memblock_memory_init_regions && | |
260 | old_array != memblock_reserved_init_regions) | |
261 | memblock_free(__pa(old_array), old_size); | |
262 | ||
263 | return 0; | |
264 | } | |
265 | ||
784656f9 TH |
266 | /** |
267 | * memblock_merge_regions - merge neighboring compatible regions | |
268 | * @type: memblock type to scan | |
269 | * | |
270 | * Scan @type and merge neighboring compatible regions. | |
271 | */ | |
272 | static void __init_memblock memblock_merge_regions(struct memblock_type *type) | |
95f72d1e | 273 | { |
784656f9 | 274 | int i = 0; |
95f72d1e | 275 | |
784656f9 TH |
276 | /* cnt never goes below 1 */ |
277 | while (i < type->cnt - 1) { | |
278 | struct memblock_region *this = &type->regions[i]; | |
279 | struct memblock_region *next = &type->regions[i + 1]; | |
95f72d1e | 280 | |
7c0caeb8 TH |
281 | if (this->base + this->size != next->base || |
282 | memblock_get_region_node(this) != | |
283 | memblock_get_region_node(next)) { | |
784656f9 TH |
284 | BUG_ON(this->base + this->size > next->base); |
285 | i++; | |
286 | continue; | |
8f7a6605 BH |
287 | } |
288 | ||
784656f9 TH |
289 | this->size += next->size; |
290 | memmove(next, next + 1, (type->cnt - (i + 1)) * sizeof(*next)); | |
291 | type->cnt--; | |
95f72d1e | 292 | } |
784656f9 | 293 | } |
95f72d1e | 294 | |
784656f9 TH |
295 | /** |
296 | * memblock_insert_region - insert new memblock region | |
297 | * @type: memblock type to insert into | |
298 | * @idx: index for the insertion point | |
299 | * @base: base address of the new region | |
300 | * @size: size of the new region | |
301 | * | |
302 | * Insert new memblock region [@base,@base+@size) into @type at @idx. | |
303 | * @type must already have extra room to accomodate the new region. | |
304 | */ | |
305 | static void __init_memblock memblock_insert_region(struct memblock_type *type, | |
306 | int idx, phys_addr_t base, | |
7c0caeb8 | 307 | phys_addr_t size, int nid) |
784656f9 TH |
308 | { |
309 | struct memblock_region *rgn = &type->regions[idx]; | |
310 | ||
311 | BUG_ON(type->cnt >= type->max); | |
312 | memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); | |
313 | rgn->base = base; | |
314 | rgn->size = size; | |
7c0caeb8 | 315 | memblock_set_region_node(rgn, nid); |
784656f9 TH |
316 | type->cnt++; |
317 | } | |
318 | ||
319 | /** | |
320 | * memblock_add_region - add new memblock region | |
321 | * @type: memblock type to add new region into | |
322 | * @base: base address of the new region | |
323 | * @size: size of the new region | |
324 | * | |
325 | * Add new memblock region [@base,@base+@size) into @type. The new region | |
326 | * is allowed to overlap with existing ones - overlaps don't affect already | |
327 | * existing regions. @type is guaranteed to be minimal (all neighbouring | |
328 | * compatible regions are merged) after the addition. | |
329 | * | |
330 | * RETURNS: | |
331 | * 0 on success, -errno on failure. | |
332 | */ | |
581adcbe TH |
333 | static int __init_memblock memblock_add_region(struct memblock_type *type, |
334 | phys_addr_t base, phys_addr_t size) | |
784656f9 TH |
335 | { |
336 | bool insert = false; | |
eb18f1b5 TH |
337 | phys_addr_t obase = base; |
338 | phys_addr_t end = base + memblock_cap_size(base, &size); | |
784656f9 TH |
339 | int i, nr_new; |
340 | ||
341 | /* special case for empty array */ | |
342 | if (type->regions[0].size == 0) { | |
343 | WARN_ON(type->cnt != 1); | |
8f7a6605 BH |
344 | type->regions[0].base = base; |
345 | type->regions[0].size = size; | |
7c0caeb8 | 346 | memblock_set_region_node(&type->regions[0], MAX_NUMNODES); |
8f7a6605 | 347 | return 0; |
95f72d1e | 348 | } |
784656f9 TH |
349 | repeat: |
350 | /* | |
351 | * The following is executed twice. Once with %false @insert and | |
352 | * then with %true. The first counts the number of regions needed | |
353 | * to accomodate the new area. The second actually inserts them. | |
142b45a7 | 354 | */ |
784656f9 TH |
355 | base = obase; |
356 | nr_new = 0; | |
95f72d1e | 357 | |
784656f9 TH |
358 | for (i = 0; i < type->cnt; i++) { |
359 | struct memblock_region *rgn = &type->regions[i]; | |
360 | phys_addr_t rbase = rgn->base; | |
361 | phys_addr_t rend = rbase + rgn->size; | |
362 | ||
363 | if (rbase >= end) | |
95f72d1e | 364 | break; |
784656f9 TH |
365 | if (rend <= base) |
366 | continue; | |
367 | /* | |
368 | * @rgn overlaps. If it separates the lower part of new | |
369 | * area, insert that portion. | |
370 | */ | |
371 | if (rbase > base) { | |
372 | nr_new++; | |
373 | if (insert) | |
374 | memblock_insert_region(type, i++, base, | |
7c0caeb8 | 375 | rbase - base, MAX_NUMNODES); |
95f72d1e | 376 | } |
784656f9 TH |
377 | /* area below @rend is dealt with, forget about it */ |
378 | base = min(rend, end); | |
95f72d1e | 379 | } |
784656f9 TH |
380 | |
381 | /* insert the remaining portion */ | |
382 | if (base < end) { | |
383 | nr_new++; | |
384 | if (insert) | |
7c0caeb8 TH |
385 | memblock_insert_region(type, i, base, end - base, |
386 | MAX_NUMNODES); | |
95f72d1e | 387 | } |
95f72d1e | 388 | |
784656f9 TH |
389 | /* |
390 | * If this was the first round, resize array and repeat for actual | |
391 | * insertions; otherwise, merge and return. | |
142b45a7 | 392 | */ |
784656f9 TH |
393 | if (!insert) { |
394 | while (type->cnt + nr_new > type->max) | |
395 | if (memblock_double_array(type) < 0) | |
396 | return -ENOMEM; | |
397 | insert = true; | |
398 | goto repeat; | |
399 | } else { | |
400 | memblock_merge_regions(type); | |
401 | return 0; | |
142b45a7 | 402 | } |
95f72d1e YL |
403 | } |
404 | ||
581adcbe | 405 | int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) |
95f72d1e | 406 | { |
e3239ff9 | 407 | return memblock_add_region(&memblock.memory, base, size); |
95f72d1e YL |
408 | } |
409 | ||
6a9ceb31 TH |
410 | /** |
411 | * memblock_isolate_range - isolate given range into disjoint memblocks | |
412 | * @type: memblock type to isolate range for | |
413 | * @base: base of range to isolate | |
414 | * @size: size of range to isolate | |
415 | * @start_rgn: out parameter for the start of isolated region | |
416 | * @end_rgn: out parameter for the end of isolated region | |
417 | * | |
418 | * Walk @type and ensure that regions don't cross the boundaries defined by | |
419 | * [@base,@base+@size). Crossing regions are split at the boundaries, | |
420 | * which may create at most two more regions. The index of the first | |
421 | * region inside the range is returned in *@start_rgn and end in *@end_rgn. | |
422 | * | |
423 | * RETURNS: | |
424 | * 0 on success, -errno on failure. | |
425 | */ | |
426 | static int __init_memblock memblock_isolate_range(struct memblock_type *type, | |
427 | phys_addr_t base, phys_addr_t size, | |
428 | int *start_rgn, int *end_rgn) | |
429 | { | |
eb18f1b5 | 430 | phys_addr_t end = base + memblock_cap_size(base, &size); |
6a9ceb31 TH |
431 | int i; |
432 | ||
433 | *start_rgn = *end_rgn = 0; | |
434 | ||
435 | /* we'll create at most two more regions */ | |
436 | while (type->cnt + 2 > type->max) | |
437 | if (memblock_double_array(type) < 0) | |
438 | return -ENOMEM; | |
439 | ||
440 | for (i = 0; i < type->cnt; i++) { | |
441 | struct memblock_region *rgn = &type->regions[i]; | |
442 | phys_addr_t rbase = rgn->base; | |
443 | phys_addr_t rend = rbase + rgn->size; | |
444 | ||
445 | if (rbase >= end) | |
446 | break; | |
447 | if (rend <= base) | |
448 | continue; | |
449 | ||
450 | if (rbase < base) { | |
451 | /* | |
452 | * @rgn intersects from below. Split and continue | |
453 | * to process the next region - the new top half. | |
454 | */ | |
455 | rgn->base = base; | |
456 | rgn->size = rend - rgn->base; | |
457 | memblock_insert_region(type, i, rbase, base - rbase, | |
71936180 | 458 | memblock_get_region_node(rgn)); |
6a9ceb31 TH |
459 | } else if (rend > end) { |
460 | /* | |
461 | * @rgn intersects from above. Split and redo the | |
462 | * current region - the new bottom half. | |
463 | */ | |
464 | rgn->base = end; | |
465 | rgn->size = rend - rgn->base; | |
466 | memblock_insert_region(type, i--, rbase, end - rbase, | |
71936180 | 467 | memblock_get_region_node(rgn)); |
6a9ceb31 TH |
468 | } else { |
469 | /* @rgn is fully contained, record it */ | |
470 | if (!*end_rgn) | |
471 | *start_rgn = i; | |
472 | *end_rgn = i + 1; | |
473 | } | |
474 | } | |
475 | ||
476 | return 0; | |
477 | } | |
6a9ceb31 | 478 | |
581adcbe TH |
479 | static int __init_memblock __memblock_remove(struct memblock_type *type, |
480 | phys_addr_t base, phys_addr_t size) | |
95f72d1e | 481 | { |
71936180 TH |
482 | int start_rgn, end_rgn; |
483 | int i, ret; | |
95f72d1e | 484 | |
71936180 TH |
485 | ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); |
486 | if (ret) | |
487 | return ret; | |
95f72d1e | 488 | |
71936180 TH |
489 | for (i = end_rgn - 1; i >= start_rgn; i--) |
490 | memblock_remove_region(type, i); | |
8f7a6605 | 491 | return 0; |
95f72d1e YL |
492 | } |
493 | ||
581adcbe | 494 | int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) |
95f72d1e YL |
495 | { |
496 | return __memblock_remove(&memblock.memory, base, size); | |
497 | } | |
498 | ||
581adcbe | 499 | int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size) |
95f72d1e | 500 | { |
24aa0788 | 501 | memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n", |
a150439c PA |
502 | (unsigned long long)base, |
503 | (unsigned long long)base + size, | |
504 | (void *)_RET_IP_); | |
24aa0788 | 505 | |
95f72d1e YL |
506 | return __memblock_remove(&memblock.reserved, base, size); |
507 | } | |
508 | ||
581adcbe | 509 | int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) |
95f72d1e | 510 | { |
e3239ff9 | 511 | struct memblock_type *_rgn = &memblock.reserved; |
95f72d1e | 512 | |
24aa0788 | 513 | memblock_dbg("memblock_reserve: [%#016llx-%#016llx] %pF\n", |
a150439c PA |
514 | (unsigned long long)base, |
515 | (unsigned long long)base + size, | |
516 | (void *)_RET_IP_); | |
95f72d1e YL |
517 | BUG_ON(0 == size); |
518 | ||
519 | return memblock_add_region(_rgn, base, size); | |
520 | } | |
521 | ||
35fd0808 TH |
522 | /** |
523 | * __next_free_mem_range - next function for for_each_free_mem_range() | |
524 | * @idx: pointer to u64 loop variable | |
525 | * @nid: nid: node selector, %MAX_NUMNODES for all nodes | |
526 | * @p_start: ptr to phys_addr_t for start address of the range, can be %NULL | |
527 | * @p_end: ptr to phys_addr_t for end address of the range, can be %NULL | |
528 | * @p_nid: ptr to int for nid of the range, can be %NULL | |
529 | * | |
530 | * Find the first free area from *@idx which matches @nid, fill the out | |
531 | * parameters, and update *@idx for the next iteration. The lower 32bit of | |
532 | * *@idx contains index into memory region and the upper 32bit indexes the | |
533 | * areas before each reserved region. For example, if reserved regions | |
534 | * look like the following, | |
535 | * | |
536 | * 0:[0-16), 1:[32-48), 2:[128-130) | |
537 | * | |
538 | * The upper 32bit indexes the following regions. | |
539 | * | |
540 | * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) | |
541 | * | |
542 | * As both region arrays are sorted, the function advances the two indices | |
543 | * in lockstep and returns each intersection. | |
544 | */ | |
545 | void __init_memblock __next_free_mem_range(u64 *idx, int nid, | |
546 | phys_addr_t *out_start, | |
547 | phys_addr_t *out_end, int *out_nid) | |
548 | { | |
549 | struct memblock_type *mem = &memblock.memory; | |
550 | struct memblock_type *rsv = &memblock.reserved; | |
551 | int mi = *idx & 0xffffffff; | |
552 | int ri = *idx >> 32; | |
553 | ||
554 | for ( ; mi < mem->cnt; mi++) { | |
555 | struct memblock_region *m = &mem->regions[mi]; | |
556 | phys_addr_t m_start = m->base; | |
557 | phys_addr_t m_end = m->base + m->size; | |
558 | ||
559 | /* only memory regions are associated with nodes, check it */ | |
560 | if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m)) | |
561 | continue; | |
562 | ||
563 | /* scan areas before each reservation for intersection */ | |
564 | for ( ; ri < rsv->cnt + 1; ri++) { | |
565 | struct memblock_region *r = &rsv->regions[ri]; | |
566 | phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0; | |
567 | phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX; | |
568 | ||
569 | /* if ri advanced past mi, break out to advance mi */ | |
570 | if (r_start >= m_end) | |
571 | break; | |
572 | /* if the two regions intersect, we're done */ | |
573 | if (m_start < r_end) { | |
574 | if (out_start) | |
575 | *out_start = max(m_start, r_start); | |
576 | if (out_end) | |
577 | *out_end = min(m_end, r_end); | |
578 | if (out_nid) | |
579 | *out_nid = memblock_get_region_node(m); | |
580 | /* | |
581 | * The region which ends first is advanced | |
582 | * for the next iteration. | |
583 | */ | |
584 | if (m_end <= r_end) | |
585 | mi++; | |
586 | else | |
587 | ri++; | |
588 | *idx = (u32)mi | (u64)ri << 32; | |
589 | return; | |
590 | } | |
591 | } | |
592 | } | |
593 | ||
594 | /* signal end of iteration */ | |
595 | *idx = ULLONG_MAX; | |
596 | } | |
597 | ||
7c0caeb8 TH |
598 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
599 | /* | |
600 | * Common iterator interface used to define for_each_mem_range(). | |
601 | */ | |
602 | void __init_memblock __next_mem_pfn_range(int *idx, int nid, | |
603 | unsigned long *out_start_pfn, | |
604 | unsigned long *out_end_pfn, int *out_nid) | |
605 | { | |
606 | struct memblock_type *type = &memblock.memory; | |
607 | struct memblock_region *r; | |
608 | ||
609 | while (++*idx < type->cnt) { | |
610 | r = &type->regions[*idx]; | |
611 | ||
612 | if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) | |
613 | continue; | |
614 | if (nid == MAX_NUMNODES || nid == r->nid) | |
615 | break; | |
616 | } | |
617 | if (*idx >= type->cnt) { | |
618 | *idx = -1; | |
619 | return; | |
620 | } | |
621 | ||
622 | if (out_start_pfn) | |
623 | *out_start_pfn = PFN_UP(r->base); | |
624 | if (out_end_pfn) | |
625 | *out_end_pfn = PFN_DOWN(r->base + r->size); | |
626 | if (out_nid) | |
627 | *out_nid = r->nid; | |
628 | } | |
629 | ||
630 | /** | |
631 | * memblock_set_node - set node ID on memblock regions | |
632 | * @base: base of area to set node ID for | |
633 | * @size: size of area to set node ID for | |
634 | * @nid: node ID to set | |
635 | * | |
636 | * Set the nid of memblock memory regions in [@base,@base+@size) to @nid. | |
637 | * Regions which cross the area boundaries are split as necessary. | |
638 | * | |
639 | * RETURNS: | |
640 | * 0 on success, -errno on failure. | |
641 | */ | |
642 | int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, | |
643 | int nid) | |
644 | { | |
645 | struct memblock_type *type = &memblock.memory; | |
6a9ceb31 TH |
646 | int start_rgn, end_rgn; |
647 | int i, ret; | |
7c0caeb8 | 648 | |
6a9ceb31 TH |
649 | ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); |
650 | if (ret) | |
651 | return ret; | |
7c0caeb8 | 652 | |
6a9ceb31 TH |
653 | for (i = start_rgn; i < end_rgn; i++) |
654 | type->regions[i].nid = nid; | |
7c0caeb8 TH |
655 | |
656 | memblock_merge_regions(type); | |
657 | return 0; | |
658 | } | |
659 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
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 | 663 | phys_addr_t found; |
95f72d1e | 664 | |
6ed311b2 BH |
665 | /* We align the size to limit fragmentation. Without this, a lot of |
666 | * small allocs quickly eat up the whole reserve array on sparc | |
667 | */ | |
348968eb | 668 | size = round_up(size, align); |
95f72d1e | 669 | |
fc769a8e | 670 | found = memblock_find_in_range(0, max_addr, size, align); |
9c8c27e2 | 671 | if (found && !memblock_reserve(found, size)) |
6ed311b2 | 672 | return found; |
95f72d1e | 673 | |
6ed311b2 | 674 | return 0; |
95f72d1e YL |
675 | } |
676 | ||
6ed311b2 | 677 | phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) |
95f72d1e | 678 | { |
6ed311b2 BH |
679 | phys_addr_t alloc; |
680 | ||
681 | alloc = __memblock_alloc_base(size, align, max_addr); | |
682 | ||
683 | if (alloc == 0) | |
684 | panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n", | |
685 | (unsigned long long) size, (unsigned long long) max_addr); | |
686 | ||
687 | return alloc; | |
95f72d1e YL |
688 | } |
689 | ||
6ed311b2 | 690 | phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align) |
95f72d1e | 691 | { |
6ed311b2 BH |
692 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
693 | } | |
95f72d1e | 694 | |
95f72d1e | 695 | |
6ed311b2 | 696 | /* |
34e18455 | 697 | * Additional node-local top-down allocators. |
c196f76f BH |
698 | * |
699 | * WARNING: Only available after early_node_map[] has been populated, | |
700 | * on some architectures, that is after all the calls to add_active_range() | |
701 | * have been done to populate it. | |
6ed311b2 | 702 | */ |
95f72d1e | 703 | |
34e18455 TH |
704 | static phys_addr_t __init memblock_nid_range_rev(phys_addr_t start, |
705 | phys_addr_t end, int *nid) | |
c3f72b57 | 706 | { |
c196f76f | 707 | #ifdef CONFIG_ARCH_POPULATES_NODE_MAP |
c196f76f BH |
708 | unsigned long start_pfn, end_pfn; |
709 | int i; | |
710 | ||
b2fea988 | 711 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, nid) |
34e18455 TH |
712 | if (end > PFN_PHYS(start_pfn) && end <= PFN_PHYS(end_pfn)) |
713 | return max(start, PFN_PHYS(start_pfn)); | |
c196f76f | 714 | #endif |
c3f72b57 | 715 | *nid = 0; |
34e18455 | 716 | return start; |
c3f72b57 BH |
717 | } |
718 | ||
e6498040 TH |
719 | phys_addr_t __init memblock_find_in_range_node(phys_addr_t start, |
720 | phys_addr_t end, | |
2898cc4c BH |
721 | phys_addr_t size, |
722 | phys_addr_t align, int nid) | |
95f72d1e | 723 | { |
e6498040 TH |
724 | struct memblock_type *mem = &memblock.memory; |
725 | int i; | |
95f72d1e | 726 | |
e6498040 | 727 | BUG_ON(0 == size); |
95f72d1e | 728 | |
e6498040 TH |
729 | /* Pump up max_addr */ |
730 | if (end == MEMBLOCK_ALLOC_ACCESSIBLE) | |
731 | end = memblock.current_limit; | |
95f72d1e | 732 | |
e6498040 TH |
733 | for (i = mem->cnt - 1; i >= 0; i--) { |
734 | struct memblock_region *r = &mem->regions[i]; | |
735 | phys_addr_t base = max(start, r->base); | |
736 | phys_addr_t top = min(end, r->base + r->size); | |
737 | ||
738 | while (base < top) { | |
739 | phys_addr_t tbase, ret; | |
740 | int tnid; | |
741 | ||
742 | tbase = memblock_nid_range_rev(base, top, &tnid); | |
743 | if (nid == MAX_NUMNODES || tnid == nid) { | |
744 | ret = memblock_find_region(tbase, top, size, align); | |
745 | if (ret) | |
746 | return ret; | |
747 | } | |
748 | top = tbase; | |
95f72d1e | 749 | } |
95f72d1e | 750 | } |
e6498040 | 751 | |
1f5026a7 | 752 | return 0; |
95f72d1e YL |
753 | } |
754 | ||
2898cc4c | 755 | phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid) |
95f72d1e | 756 | { |
e6498040 | 757 | phys_addr_t found; |
95f72d1e | 758 | |
e6498040 TH |
759 | /* |
760 | * We align the size to limit fragmentation. Without this, a lot of | |
7f219c73 BH |
761 | * small allocs quickly eat up the whole reserve array on sparc |
762 | */ | |
348968eb | 763 | size = round_up(size, align); |
7f219c73 | 764 | |
e6498040 TH |
765 | found = memblock_find_in_range_node(0, MEMBLOCK_ALLOC_ACCESSIBLE, |
766 | size, align, nid); | |
9c8c27e2 | 767 | if (found && !memblock_reserve(found, size)) |
e6498040 | 768 | return found; |
95f72d1e | 769 | |
9d1e2492 BH |
770 | return 0; |
771 | } | |
772 | ||
773 | phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) | |
774 | { | |
775 | phys_addr_t res = memblock_alloc_nid(size, align, nid); | |
776 | ||
777 | if (res) | |
778 | return res; | |
15fb0972 | 779 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
95f72d1e YL |
780 | } |
781 | ||
9d1e2492 BH |
782 | |
783 | /* | |
784 | * Remaining API functions | |
785 | */ | |
786 | ||
95f72d1e | 787 | /* You must call memblock_analyze() before this. */ |
2898cc4c | 788 | phys_addr_t __init memblock_phys_mem_size(void) |
95f72d1e | 789 | { |
4734b594 | 790 | return memblock.memory_size; |
95f72d1e YL |
791 | } |
792 | ||
0a93ebef SR |
793 | /* lowest address */ |
794 | phys_addr_t __init_memblock memblock_start_of_DRAM(void) | |
795 | { | |
796 | return memblock.memory.regions[0].base; | |
797 | } | |
798 | ||
10d06439 | 799 | phys_addr_t __init_memblock memblock_end_of_DRAM(void) |
95f72d1e YL |
800 | { |
801 | int idx = memblock.memory.cnt - 1; | |
802 | ||
e3239ff9 | 803 | return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); |
95f72d1e YL |
804 | } |
805 | ||
806 | /* You must call memblock_analyze() after this. */ | |
c0ce8fef | 807 | void __init memblock_enforce_memory_limit(phys_addr_t limit) |
95f72d1e YL |
808 | { |
809 | unsigned long i; | |
c0ce8fef | 810 | phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX; |
95f72d1e | 811 | |
c0ce8fef | 812 | if (!limit) |
95f72d1e YL |
813 | return; |
814 | ||
c0ce8fef | 815 | /* find out max address */ |
95f72d1e | 816 | for (i = 0; i < memblock.memory.cnt; i++) { |
c0ce8fef | 817 | struct memblock_region *r = &memblock.memory.regions[i]; |
95f72d1e | 818 | |
c0ce8fef TH |
819 | if (limit <= r->size) { |
820 | max_addr = r->base + limit; | |
821 | break; | |
95f72d1e | 822 | } |
c0ce8fef | 823 | limit -= r->size; |
95f72d1e | 824 | } |
c0ce8fef TH |
825 | |
826 | /* truncate both memory and reserved regions */ | |
827 | __memblock_remove(&memblock.memory, max_addr, (phys_addr_t)ULLONG_MAX); | |
828 | __memblock_remove(&memblock.reserved, max_addr, (phys_addr_t)ULLONG_MAX); | |
95f72d1e YL |
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); |
eb18f1b5 | 862 | phys_addr_t end = base + memblock_cap_size(base, &size); |
72d4b0b4 BH |
863 | |
864 | if (idx == -1) | |
865 | return 0; | |
abb65272 TV |
866 | return memblock.memory.regions[idx].base <= base && |
867 | (memblock.memory.regions[idx].base + | |
eb18f1b5 | 868 | memblock.memory.regions[idx].size) >= end; |
95f72d1e YL |
869 | } |
870 | ||
10d06439 | 871 | int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size) |
95f72d1e | 872 | { |
eb18f1b5 | 873 | memblock_cap_size(base, &size); |
f1c2c19c | 874 | return memblock_overlaps_region(&memblock.reserved, base, size) >= 0; |
95f72d1e YL |
875 | } |
876 | ||
e63075a3 | 877 | |
3661ca66 | 878 | void __init_memblock memblock_set_current_limit(phys_addr_t limit) |
e63075a3 BH |
879 | { |
880 | memblock.current_limit = limit; | |
881 | } | |
882 | ||
7c0caeb8 | 883 | static void __init_memblock memblock_dump(struct memblock_type *type, char *name) |
6ed311b2 BH |
884 | { |
885 | unsigned long long base, size; | |
886 | int i; | |
887 | ||
7c0caeb8 | 888 | pr_info(" %s.cnt = 0x%lx\n", name, type->cnt); |
6ed311b2 | 889 | |
7c0caeb8 TH |
890 | for (i = 0; i < type->cnt; i++) { |
891 | struct memblock_region *rgn = &type->regions[i]; | |
892 | char nid_buf[32] = ""; | |
893 | ||
894 | base = rgn->base; | |
895 | size = rgn->size; | |
896 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP | |
897 | if (memblock_get_region_node(rgn) != MAX_NUMNODES) | |
898 | snprintf(nid_buf, sizeof(nid_buf), " on node %d", | |
899 | memblock_get_region_node(rgn)); | |
900 | #endif | |
901 | pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s\n", | |
902 | name, i, base, base + size - 1, size, nid_buf); | |
6ed311b2 BH |
903 | } |
904 | } | |
905 | ||
4ff7b82f | 906 | void __init_memblock __memblock_dump_all(void) |
6ed311b2 | 907 | { |
6ed311b2 BH |
908 | pr_info("MEMBLOCK configuration:\n"); |
909 | pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size); | |
910 | ||
911 | memblock_dump(&memblock.memory, "memory"); | |
912 | memblock_dump(&memblock.reserved, "reserved"); | |
913 | } | |
914 | ||
915 | void __init memblock_analyze(void) | |
916 | { | |
917 | int i; | |
918 | ||
6ed311b2 BH |
919 | memblock.memory_size = 0; |
920 | ||
921 | for (i = 0; i < memblock.memory.cnt; i++) | |
922 | memblock.memory_size += memblock.memory.regions[i].size; | |
142b45a7 BH |
923 | |
924 | /* We allow resizing from there */ | |
925 | memblock_can_resize = 1; | |
6ed311b2 BH |
926 | } |
927 | ||
6ed311b2 BH |
928 | static int __init early_memblock(char *p) |
929 | { | |
930 | if (p && strstr(p, "debug")) | |
931 | memblock_debug = 1; | |
932 | return 0; | |
933 | } | |
934 | early_param("memblock", early_memblock); | |
935 | ||
c378ddd5 | 936 | #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK) |
6d03b885 BH |
937 | |
938 | static int memblock_debug_show(struct seq_file *m, void *private) | |
939 | { | |
940 | struct memblock_type *type = m->private; | |
941 | struct memblock_region *reg; | |
942 | int i; | |
943 | ||
944 | for (i = 0; i < type->cnt; i++) { | |
945 | reg = &type->regions[i]; | |
946 | seq_printf(m, "%4d: ", i); | |
947 | if (sizeof(phys_addr_t) == 4) | |
948 | seq_printf(m, "0x%08lx..0x%08lx\n", | |
949 | (unsigned long)reg->base, | |
950 | (unsigned long)(reg->base + reg->size - 1)); | |
951 | else | |
952 | seq_printf(m, "0x%016llx..0x%016llx\n", | |
953 | (unsigned long long)reg->base, | |
954 | (unsigned long long)(reg->base + reg->size - 1)); | |
955 | ||
956 | } | |
957 | return 0; | |
958 | } | |
959 | ||
960 | static int memblock_debug_open(struct inode *inode, struct file *file) | |
961 | { | |
962 | return single_open(file, memblock_debug_show, inode->i_private); | |
963 | } | |
964 | ||
965 | static const struct file_operations memblock_debug_fops = { | |
966 | .open = memblock_debug_open, | |
967 | .read = seq_read, | |
968 | .llseek = seq_lseek, | |
969 | .release = single_release, | |
970 | }; | |
971 | ||
972 | static int __init memblock_init_debugfs(void) | |
973 | { | |
974 | struct dentry *root = debugfs_create_dir("memblock", NULL); | |
975 | if (!root) | |
976 | return -ENXIO; | |
977 | debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops); | |
978 | debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops); | |
979 | ||
980 | return 0; | |
981 | } | |
982 | __initcall(memblock_init_debugfs); | |
983 | ||
984 | #endif /* CONFIG_DEBUG_FS */ |