]>
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 | 38 | int memblock_debug __initdata_memblock; |
1aadc056 | 39 | static 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 | ||
7bd0b0f0 TH |
82 | /** |
83 | * memblock_find_in_range_node - find free area in given range and node | |
84 | * @start: start of candidate range | |
85 | * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} | |
86 | * @size: size of free area to find | |
87 | * @align: alignment of free area to find | |
88 | * @nid: nid of the free area to find, %MAX_NUMNODES for any node | |
89 | * | |
90 | * Find @size free area aligned to @align in the specified range and node. | |
91 | * | |
92 | * RETURNS: | |
93 | * Found address on success, %0 on failure. | |
6ed311b2 | 94 | */ |
7bd0b0f0 TH |
95 | phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t start, |
96 | phys_addr_t end, phys_addr_t size, | |
97 | phys_addr_t align, int nid) | |
6ed311b2 | 98 | { |
7bd0b0f0 TH |
99 | phys_addr_t this_start, this_end, cand; |
100 | u64 i; | |
6ed311b2 | 101 | |
7bd0b0f0 TH |
102 | /* pump up @end */ |
103 | if (end == MEMBLOCK_ALLOC_ACCESSIBLE) | |
104 | end = memblock.current_limit; | |
f1af98c7 | 105 | |
5d53cb27 TH |
106 | /* avoid allocating the first page */ |
107 | start = max_t(phys_addr_t, start, PAGE_SIZE); | |
7bd0b0f0 | 108 | end = max(start, end); |
f1af98c7 | 109 | |
7bd0b0f0 TH |
110 | for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) { |
111 | this_start = clamp(this_start, start, end); | |
112 | this_end = clamp(this_end, start, end); | |
6ed311b2 | 113 | |
5d53cb27 TH |
114 | if (this_end < size) |
115 | continue; | |
116 | ||
7bd0b0f0 TH |
117 | cand = round_down(this_end - size, align); |
118 | if (cand >= this_start) | |
119 | return cand; | |
120 | } | |
1f5026a7 | 121 | return 0; |
6ed311b2 BH |
122 | } |
123 | ||
7bd0b0f0 TH |
124 | /** |
125 | * memblock_find_in_range - find free area in given range | |
126 | * @start: start of candidate range | |
127 | * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE} | |
128 | * @size: size of free area to find | |
129 | * @align: alignment of free area to find | |
130 | * | |
131 | * Find @size free area aligned to @align in the specified range. | |
132 | * | |
133 | * RETURNS: | |
134 | * Found address on success, %0 on failure. | |
fc769a8e | 135 | */ |
7bd0b0f0 TH |
136 | phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, |
137 | phys_addr_t end, phys_addr_t size, | |
138 | phys_addr_t align) | |
6ed311b2 | 139 | { |
7bd0b0f0 TH |
140 | return memblock_find_in_range_node(start, end, size, align, |
141 | MAX_NUMNODES); | |
6ed311b2 BH |
142 | } |
143 | ||
7950c407 YL |
144 | /* |
145 | * Free memblock.reserved.regions | |
146 | */ | |
147 | int __init_memblock memblock_free_reserved_regions(void) | |
148 | { | |
149 | if (memblock.reserved.regions == memblock_reserved_init_regions) | |
150 | return 0; | |
151 | ||
152 | return memblock_free(__pa(memblock.reserved.regions), | |
153 | sizeof(struct memblock_region) * memblock.reserved.max); | |
154 | } | |
155 | ||
156 | /* | |
157 | * Reserve memblock.reserved.regions | |
158 | */ | |
159 | int __init_memblock memblock_reserve_reserved_regions(void) | |
160 | { | |
161 | if (memblock.reserved.regions == memblock_reserved_init_regions) | |
162 | return 0; | |
163 | ||
164 | return memblock_reserve(__pa(memblock.reserved.regions), | |
165 | sizeof(struct memblock_region) * memblock.reserved.max); | |
166 | } | |
167 | ||
10d06439 | 168 | static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r) |
95f72d1e | 169 | { |
1440c4e2 | 170 | type->total_size -= type->regions[r].size; |
7c0caeb8 TH |
171 | memmove(&type->regions[r], &type->regions[r + 1], |
172 | (type->cnt - (r + 1)) * sizeof(type->regions[r])); | |
e3239ff9 | 173 | type->cnt--; |
95f72d1e | 174 | |
8f7a6605 BH |
175 | /* Special case for empty arrays */ |
176 | if (type->cnt == 0) { | |
1440c4e2 | 177 | WARN_ON(type->total_size != 0); |
8f7a6605 BH |
178 | type->cnt = 1; |
179 | type->regions[0].base = 0; | |
180 | type->regions[0].size = 0; | |
7c0caeb8 | 181 | memblock_set_region_node(&type->regions[0], MAX_NUMNODES); |
8f7a6605 | 182 | } |
95f72d1e YL |
183 | } |
184 | ||
10d06439 | 185 | static int __init_memblock memblock_double_array(struct memblock_type *type) |
142b45a7 BH |
186 | { |
187 | struct memblock_region *new_array, *old_array; | |
188 | phys_addr_t old_size, new_size, addr; | |
189 | int use_slab = slab_is_available(); | |
190 | ||
191 | /* We don't allow resizing until we know about the reserved regions | |
192 | * of memory that aren't suitable for allocation | |
193 | */ | |
194 | if (!memblock_can_resize) | |
195 | return -1; | |
196 | ||
142b45a7 BH |
197 | /* Calculate new doubled size */ |
198 | old_size = type->max * sizeof(struct memblock_region); | |
199 | new_size = old_size << 1; | |
200 | ||
201 | /* Try to find some space for it. | |
202 | * | |
203 | * WARNING: We assume that either slab_is_available() and we use it or | |
204 | * we use MEMBLOCK for allocations. That means that this is unsafe to use | |
205 | * when bootmem is currently active (unless bootmem itself is implemented | |
206 | * on top of MEMBLOCK which isn't the case yet) | |
207 | * | |
208 | * This should however not be an issue for now, as we currently only | |
209 | * call into MEMBLOCK while it's still active, or much later when slab is | |
210 | * active for memory hotplug operations | |
211 | */ | |
212 | if (use_slab) { | |
213 | new_array = kmalloc(new_size, GFP_KERNEL); | |
1f5026a7 | 214 | addr = new_array ? __pa(new_array) : 0; |
4e2f0775 | 215 | } else { |
fc769a8e | 216 | addr = memblock_find_in_range(0, MEMBLOCK_ALLOC_ACCESSIBLE, new_size, sizeof(phys_addr_t)); |
4e2f0775 GS |
217 | new_array = addr ? __va(addr) : 0; |
218 | } | |
1f5026a7 | 219 | if (!addr) { |
142b45a7 BH |
220 | pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n", |
221 | memblock_type_name(type), type->max, type->max * 2); | |
222 | return -1; | |
223 | } | |
142b45a7 | 224 | |
ea9e4376 YL |
225 | memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]", |
226 | memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1); | |
227 | ||
142b45a7 BH |
228 | /* Found space, we now need to move the array over before |
229 | * we add the reserved region since it may be our reserved | |
230 | * array itself that is full. | |
231 | */ | |
232 | memcpy(new_array, type->regions, old_size); | |
233 | memset(new_array + type->max, 0, old_size); | |
234 | old_array = type->regions; | |
235 | type->regions = new_array; | |
236 | type->max <<= 1; | |
237 | ||
238 | /* If we use SLAB that's it, we are done */ | |
239 | if (use_slab) | |
240 | return 0; | |
241 | ||
242 | /* Add the new reserved region now. Should not fail ! */ | |
9c8c27e2 | 243 | BUG_ON(memblock_reserve(addr, new_size)); |
142b45a7 BH |
244 | |
245 | /* If the array wasn't our static init one, then free it. We only do | |
246 | * that before SLAB is available as later on, we don't know whether | |
247 | * to use kfree or free_bootmem_pages(). Shouldn't be a big deal | |
248 | * anyways | |
249 | */ | |
250 | if (old_array != memblock_memory_init_regions && | |
251 | old_array != memblock_reserved_init_regions) | |
252 | memblock_free(__pa(old_array), old_size); | |
253 | ||
254 | return 0; | |
255 | } | |
256 | ||
784656f9 TH |
257 | /** |
258 | * memblock_merge_regions - merge neighboring compatible regions | |
259 | * @type: memblock type to scan | |
260 | * | |
261 | * Scan @type and merge neighboring compatible regions. | |
262 | */ | |
263 | static void __init_memblock memblock_merge_regions(struct memblock_type *type) | |
95f72d1e | 264 | { |
784656f9 | 265 | int i = 0; |
95f72d1e | 266 | |
784656f9 TH |
267 | /* cnt never goes below 1 */ |
268 | while (i < type->cnt - 1) { | |
269 | struct memblock_region *this = &type->regions[i]; | |
270 | struct memblock_region *next = &type->regions[i + 1]; | |
95f72d1e | 271 | |
7c0caeb8 TH |
272 | if (this->base + this->size != next->base || |
273 | memblock_get_region_node(this) != | |
274 | memblock_get_region_node(next)) { | |
784656f9 TH |
275 | BUG_ON(this->base + this->size > next->base); |
276 | i++; | |
277 | continue; | |
8f7a6605 BH |
278 | } |
279 | ||
784656f9 TH |
280 | this->size += next->size; |
281 | memmove(next, next + 1, (type->cnt - (i + 1)) * sizeof(*next)); | |
282 | type->cnt--; | |
95f72d1e | 283 | } |
784656f9 | 284 | } |
95f72d1e | 285 | |
784656f9 TH |
286 | /** |
287 | * memblock_insert_region - insert new memblock region | |
288 | * @type: memblock type to insert into | |
289 | * @idx: index for the insertion point | |
290 | * @base: base address of the new region | |
291 | * @size: size of the new region | |
292 | * | |
293 | * Insert new memblock region [@base,@base+@size) into @type at @idx. | |
294 | * @type must already have extra room to accomodate the new region. | |
295 | */ | |
296 | static void __init_memblock memblock_insert_region(struct memblock_type *type, | |
297 | int idx, phys_addr_t base, | |
7c0caeb8 | 298 | phys_addr_t size, int nid) |
784656f9 TH |
299 | { |
300 | struct memblock_region *rgn = &type->regions[idx]; | |
301 | ||
302 | BUG_ON(type->cnt >= type->max); | |
303 | memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn)); | |
304 | rgn->base = base; | |
305 | rgn->size = size; | |
7c0caeb8 | 306 | memblock_set_region_node(rgn, nid); |
784656f9 | 307 | type->cnt++; |
1440c4e2 | 308 | type->total_size += size; |
784656f9 TH |
309 | } |
310 | ||
311 | /** | |
312 | * memblock_add_region - add new memblock region | |
313 | * @type: memblock type to add new region into | |
314 | * @base: base address of the new region | |
315 | * @size: size of the new region | |
7fb0bc3f | 316 | * @nid: nid of the new region |
784656f9 TH |
317 | * |
318 | * Add new memblock region [@base,@base+@size) into @type. The new region | |
319 | * is allowed to overlap with existing ones - overlaps don't affect already | |
320 | * existing regions. @type is guaranteed to be minimal (all neighbouring | |
321 | * compatible regions are merged) after the addition. | |
322 | * | |
323 | * RETURNS: | |
324 | * 0 on success, -errno on failure. | |
325 | */ | |
581adcbe | 326 | static int __init_memblock memblock_add_region(struct memblock_type *type, |
7fb0bc3f | 327 | phys_addr_t base, phys_addr_t size, int nid) |
784656f9 TH |
328 | { |
329 | bool insert = false; | |
eb18f1b5 TH |
330 | phys_addr_t obase = base; |
331 | phys_addr_t end = base + memblock_cap_size(base, &size); | |
784656f9 TH |
332 | int i, nr_new; |
333 | ||
b3dc627c TH |
334 | if (!size) |
335 | return 0; | |
336 | ||
784656f9 TH |
337 | /* special case for empty array */ |
338 | if (type->regions[0].size == 0) { | |
1440c4e2 | 339 | WARN_ON(type->cnt != 1 || type->total_size); |
8f7a6605 BH |
340 | type->regions[0].base = base; |
341 | type->regions[0].size = size; | |
7fb0bc3f | 342 | memblock_set_region_node(&type->regions[0], nid); |
1440c4e2 | 343 | type->total_size = size; |
8f7a6605 | 344 | return 0; |
95f72d1e | 345 | } |
784656f9 TH |
346 | repeat: |
347 | /* | |
348 | * The following is executed twice. Once with %false @insert and | |
349 | * then with %true. The first counts the number of regions needed | |
350 | * to accomodate the new area. The second actually inserts them. | |
142b45a7 | 351 | */ |
784656f9 TH |
352 | base = obase; |
353 | nr_new = 0; | |
95f72d1e | 354 | |
784656f9 TH |
355 | for (i = 0; i < type->cnt; i++) { |
356 | struct memblock_region *rgn = &type->regions[i]; | |
357 | phys_addr_t rbase = rgn->base; | |
358 | phys_addr_t rend = rbase + rgn->size; | |
359 | ||
360 | if (rbase >= end) | |
95f72d1e | 361 | break; |
784656f9 TH |
362 | if (rend <= base) |
363 | continue; | |
364 | /* | |
365 | * @rgn overlaps. If it separates the lower part of new | |
366 | * area, insert that portion. | |
367 | */ | |
368 | if (rbase > base) { | |
369 | nr_new++; | |
370 | if (insert) | |
371 | memblock_insert_region(type, i++, base, | |
7fb0bc3f | 372 | rbase - base, nid); |
95f72d1e | 373 | } |
784656f9 TH |
374 | /* area below @rend is dealt with, forget about it */ |
375 | base = min(rend, end); | |
95f72d1e | 376 | } |
784656f9 TH |
377 | |
378 | /* insert the remaining portion */ | |
379 | if (base < end) { | |
380 | nr_new++; | |
381 | if (insert) | |
7fb0bc3f | 382 | memblock_insert_region(type, i, base, end - base, nid); |
95f72d1e | 383 | } |
95f72d1e | 384 | |
784656f9 TH |
385 | /* |
386 | * If this was the first round, resize array and repeat for actual | |
387 | * insertions; otherwise, merge and return. | |
142b45a7 | 388 | */ |
784656f9 TH |
389 | if (!insert) { |
390 | while (type->cnt + nr_new > type->max) | |
391 | if (memblock_double_array(type) < 0) | |
392 | return -ENOMEM; | |
393 | insert = true; | |
394 | goto repeat; | |
395 | } else { | |
396 | memblock_merge_regions(type); | |
397 | return 0; | |
142b45a7 | 398 | } |
95f72d1e YL |
399 | } |
400 | ||
7fb0bc3f TH |
401 | int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size, |
402 | int nid) | |
403 | { | |
404 | return memblock_add_region(&memblock.memory, base, size, nid); | |
405 | } | |
406 | ||
581adcbe | 407 | int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size) |
95f72d1e | 408 | { |
7fb0bc3f | 409 | return memblock_add_region(&memblock.memory, base, size, MAX_NUMNODES); |
95f72d1e YL |
410 | } |
411 | ||
6a9ceb31 TH |
412 | /** |
413 | * memblock_isolate_range - isolate given range into disjoint memblocks | |
414 | * @type: memblock type to isolate range for | |
415 | * @base: base of range to isolate | |
416 | * @size: size of range to isolate | |
417 | * @start_rgn: out parameter for the start of isolated region | |
418 | * @end_rgn: out parameter for the end of isolated region | |
419 | * | |
420 | * Walk @type and ensure that regions don't cross the boundaries defined by | |
421 | * [@base,@base+@size). Crossing regions are split at the boundaries, | |
422 | * which may create at most two more regions. The index of the first | |
423 | * region inside the range is returned in *@start_rgn and end in *@end_rgn. | |
424 | * | |
425 | * RETURNS: | |
426 | * 0 on success, -errno on failure. | |
427 | */ | |
428 | static int __init_memblock memblock_isolate_range(struct memblock_type *type, | |
429 | phys_addr_t base, phys_addr_t size, | |
430 | int *start_rgn, int *end_rgn) | |
431 | { | |
eb18f1b5 | 432 | phys_addr_t end = base + memblock_cap_size(base, &size); |
6a9ceb31 TH |
433 | int i; |
434 | ||
435 | *start_rgn = *end_rgn = 0; | |
436 | ||
b3dc627c TH |
437 | if (!size) |
438 | return 0; | |
439 | ||
6a9ceb31 TH |
440 | /* we'll create at most two more regions */ |
441 | while (type->cnt + 2 > type->max) | |
442 | if (memblock_double_array(type) < 0) | |
443 | return -ENOMEM; | |
444 | ||
445 | for (i = 0; i < type->cnt; i++) { | |
446 | struct memblock_region *rgn = &type->regions[i]; | |
447 | phys_addr_t rbase = rgn->base; | |
448 | phys_addr_t rend = rbase + rgn->size; | |
449 | ||
450 | if (rbase >= end) | |
451 | break; | |
452 | if (rend <= base) | |
453 | continue; | |
454 | ||
455 | if (rbase < base) { | |
456 | /* | |
457 | * @rgn intersects from below. Split and continue | |
458 | * to process the next region - the new top half. | |
459 | */ | |
460 | rgn->base = base; | |
1440c4e2 TH |
461 | rgn->size -= base - rbase; |
462 | type->total_size -= base - rbase; | |
6a9ceb31 | 463 | memblock_insert_region(type, i, rbase, base - rbase, |
71936180 | 464 | memblock_get_region_node(rgn)); |
6a9ceb31 TH |
465 | } else if (rend > end) { |
466 | /* | |
467 | * @rgn intersects from above. Split and redo the | |
468 | * current region - the new bottom half. | |
469 | */ | |
470 | rgn->base = end; | |
1440c4e2 TH |
471 | rgn->size -= end - rbase; |
472 | type->total_size -= end - rbase; | |
6a9ceb31 | 473 | memblock_insert_region(type, i--, rbase, end - rbase, |
71936180 | 474 | memblock_get_region_node(rgn)); |
6a9ceb31 TH |
475 | } else { |
476 | /* @rgn is fully contained, record it */ | |
477 | if (!*end_rgn) | |
478 | *start_rgn = i; | |
479 | *end_rgn = i + 1; | |
480 | } | |
481 | } | |
482 | ||
483 | return 0; | |
484 | } | |
6a9ceb31 | 485 | |
581adcbe TH |
486 | static int __init_memblock __memblock_remove(struct memblock_type *type, |
487 | phys_addr_t base, phys_addr_t size) | |
95f72d1e | 488 | { |
71936180 TH |
489 | int start_rgn, end_rgn; |
490 | int i, ret; | |
95f72d1e | 491 | |
71936180 TH |
492 | ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); |
493 | if (ret) | |
494 | return ret; | |
95f72d1e | 495 | |
71936180 TH |
496 | for (i = end_rgn - 1; i >= start_rgn; i--) |
497 | memblock_remove_region(type, i); | |
8f7a6605 | 498 | return 0; |
95f72d1e YL |
499 | } |
500 | ||
581adcbe | 501 | int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size) |
95f72d1e YL |
502 | { |
503 | return __memblock_remove(&memblock.memory, base, size); | |
504 | } | |
505 | ||
581adcbe | 506 | int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size) |
95f72d1e | 507 | { |
24aa0788 | 508 | memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n", |
a150439c PA |
509 | (unsigned long long)base, |
510 | (unsigned long long)base + size, | |
511 | (void *)_RET_IP_); | |
24aa0788 | 512 | |
95f72d1e YL |
513 | return __memblock_remove(&memblock.reserved, base, size); |
514 | } | |
515 | ||
581adcbe | 516 | int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size) |
95f72d1e | 517 | { |
e3239ff9 | 518 | struct memblock_type *_rgn = &memblock.reserved; |
95f72d1e | 519 | |
24aa0788 | 520 | memblock_dbg("memblock_reserve: [%#016llx-%#016llx] %pF\n", |
a150439c PA |
521 | (unsigned long long)base, |
522 | (unsigned long long)base + size, | |
523 | (void *)_RET_IP_); | |
95f72d1e | 524 | |
7fb0bc3f | 525 | return memblock_add_region(_rgn, base, size, MAX_NUMNODES); |
95f72d1e YL |
526 | } |
527 | ||
35fd0808 TH |
528 | /** |
529 | * __next_free_mem_range - next function for for_each_free_mem_range() | |
530 | * @idx: pointer to u64 loop variable | |
531 | * @nid: nid: node selector, %MAX_NUMNODES for all nodes | |
532 | * @p_start: ptr to phys_addr_t for start address of the range, can be %NULL | |
533 | * @p_end: ptr to phys_addr_t for end address of the range, can be %NULL | |
534 | * @p_nid: ptr to int for nid of the range, can be %NULL | |
535 | * | |
536 | * Find the first free area from *@idx which matches @nid, fill the out | |
537 | * parameters, and update *@idx for the next iteration. The lower 32bit of | |
538 | * *@idx contains index into memory region and the upper 32bit indexes the | |
539 | * areas before each reserved region. For example, if reserved regions | |
540 | * look like the following, | |
541 | * | |
542 | * 0:[0-16), 1:[32-48), 2:[128-130) | |
543 | * | |
544 | * The upper 32bit indexes the following regions. | |
545 | * | |
546 | * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX) | |
547 | * | |
548 | * As both region arrays are sorted, the function advances the two indices | |
549 | * in lockstep and returns each intersection. | |
550 | */ | |
551 | void __init_memblock __next_free_mem_range(u64 *idx, int nid, | |
552 | phys_addr_t *out_start, | |
553 | phys_addr_t *out_end, int *out_nid) | |
554 | { | |
555 | struct memblock_type *mem = &memblock.memory; | |
556 | struct memblock_type *rsv = &memblock.reserved; | |
557 | int mi = *idx & 0xffffffff; | |
558 | int ri = *idx >> 32; | |
559 | ||
560 | for ( ; mi < mem->cnt; mi++) { | |
561 | struct memblock_region *m = &mem->regions[mi]; | |
562 | phys_addr_t m_start = m->base; | |
563 | phys_addr_t m_end = m->base + m->size; | |
564 | ||
565 | /* only memory regions are associated with nodes, check it */ | |
566 | if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m)) | |
567 | continue; | |
568 | ||
569 | /* scan areas before each reservation for intersection */ | |
570 | for ( ; ri < rsv->cnt + 1; ri++) { | |
571 | struct memblock_region *r = &rsv->regions[ri]; | |
572 | phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0; | |
573 | phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX; | |
574 | ||
575 | /* if ri advanced past mi, break out to advance mi */ | |
576 | if (r_start >= m_end) | |
577 | break; | |
578 | /* if the two regions intersect, we're done */ | |
579 | if (m_start < r_end) { | |
580 | if (out_start) | |
581 | *out_start = max(m_start, r_start); | |
582 | if (out_end) | |
583 | *out_end = min(m_end, r_end); | |
584 | if (out_nid) | |
585 | *out_nid = memblock_get_region_node(m); | |
586 | /* | |
587 | * The region which ends first is advanced | |
588 | * for the next iteration. | |
589 | */ | |
590 | if (m_end <= r_end) | |
591 | mi++; | |
592 | else | |
593 | ri++; | |
594 | *idx = (u32)mi | (u64)ri << 32; | |
595 | return; | |
596 | } | |
597 | } | |
598 | } | |
599 | ||
600 | /* signal end of iteration */ | |
601 | *idx = ULLONG_MAX; | |
602 | } | |
603 | ||
7bd0b0f0 TH |
604 | /** |
605 | * __next_free_mem_range_rev - next function for for_each_free_mem_range_reverse() | |
606 | * @idx: pointer to u64 loop variable | |
607 | * @nid: nid: node selector, %MAX_NUMNODES for all nodes | |
608 | * @p_start: ptr to phys_addr_t for start address of the range, can be %NULL | |
609 | * @p_end: ptr to phys_addr_t for end address of the range, can be %NULL | |
610 | * @p_nid: ptr to int for nid of the range, can be %NULL | |
611 | * | |
612 | * Reverse of __next_free_mem_range(). | |
613 | */ | |
614 | void __init_memblock __next_free_mem_range_rev(u64 *idx, int nid, | |
615 | phys_addr_t *out_start, | |
616 | phys_addr_t *out_end, int *out_nid) | |
617 | { | |
618 | struct memblock_type *mem = &memblock.memory; | |
619 | struct memblock_type *rsv = &memblock.reserved; | |
620 | int mi = *idx & 0xffffffff; | |
621 | int ri = *idx >> 32; | |
622 | ||
623 | if (*idx == (u64)ULLONG_MAX) { | |
624 | mi = mem->cnt - 1; | |
625 | ri = rsv->cnt; | |
626 | } | |
627 | ||
628 | for ( ; mi >= 0; mi--) { | |
629 | struct memblock_region *m = &mem->regions[mi]; | |
630 | phys_addr_t m_start = m->base; | |
631 | phys_addr_t m_end = m->base + m->size; | |
632 | ||
633 | /* only memory regions are associated with nodes, check it */ | |
634 | if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m)) | |
635 | continue; | |
636 | ||
637 | /* scan areas before each reservation for intersection */ | |
638 | for ( ; ri >= 0; ri--) { | |
639 | struct memblock_region *r = &rsv->regions[ri]; | |
640 | phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0; | |
641 | phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX; | |
642 | ||
643 | /* if ri advanced past mi, break out to advance mi */ | |
644 | if (r_end <= m_start) | |
645 | break; | |
646 | /* if the two regions intersect, we're done */ | |
647 | if (m_end > r_start) { | |
648 | if (out_start) | |
649 | *out_start = max(m_start, r_start); | |
650 | if (out_end) | |
651 | *out_end = min(m_end, r_end); | |
652 | if (out_nid) | |
653 | *out_nid = memblock_get_region_node(m); | |
654 | ||
655 | if (m_start >= r_start) | |
656 | mi--; | |
657 | else | |
658 | ri--; | |
659 | *idx = (u32)mi | (u64)ri << 32; | |
660 | return; | |
661 | } | |
662 | } | |
663 | } | |
664 | ||
665 | *idx = ULLONG_MAX; | |
666 | } | |
667 | ||
7c0caeb8 TH |
668 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
669 | /* | |
670 | * Common iterator interface used to define for_each_mem_range(). | |
671 | */ | |
672 | void __init_memblock __next_mem_pfn_range(int *idx, int nid, | |
673 | unsigned long *out_start_pfn, | |
674 | unsigned long *out_end_pfn, int *out_nid) | |
675 | { | |
676 | struct memblock_type *type = &memblock.memory; | |
677 | struct memblock_region *r; | |
678 | ||
679 | while (++*idx < type->cnt) { | |
680 | r = &type->regions[*idx]; | |
681 | ||
682 | if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size)) | |
683 | continue; | |
684 | if (nid == MAX_NUMNODES || nid == r->nid) | |
685 | break; | |
686 | } | |
687 | if (*idx >= type->cnt) { | |
688 | *idx = -1; | |
689 | return; | |
690 | } | |
691 | ||
692 | if (out_start_pfn) | |
693 | *out_start_pfn = PFN_UP(r->base); | |
694 | if (out_end_pfn) | |
695 | *out_end_pfn = PFN_DOWN(r->base + r->size); | |
696 | if (out_nid) | |
697 | *out_nid = r->nid; | |
698 | } | |
699 | ||
700 | /** | |
701 | * memblock_set_node - set node ID on memblock regions | |
702 | * @base: base of area to set node ID for | |
703 | * @size: size of area to set node ID for | |
704 | * @nid: node ID to set | |
705 | * | |
706 | * Set the nid of memblock memory regions in [@base,@base+@size) to @nid. | |
707 | * Regions which cross the area boundaries are split as necessary. | |
708 | * | |
709 | * RETURNS: | |
710 | * 0 on success, -errno on failure. | |
711 | */ | |
712 | int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size, | |
713 | int nid) | |
714 | { | |
715 | struct memblock_type *type = &memblock.memory; | |
6a9ceb31 TH |
716 | int start_rgn, end_rgn; |
717 | int i, ret; | |
7c0caeb8 | 718 | |
6a9ceb31 TH |
719 | ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn); |
720 | if (ret) | |
721 | return ret; | |
7c0caeb8 | 722 | |
6a9ceb31 TH |
723 | for (i = start_rgn; i < end_rgn; i++) |
724 | type->regions[i].nid = nid; | |
7c0caeb8 TH |
725 | |
726 | memblock_merge_regions(type); | |
727 | return 0; | |
728 | } | |
729 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ | |
730 | ||
7bd0b0f0 TH |
731 | static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size, |
732 | phys_addr_t align, phys_addr_t max_addr, | |
733 | int nid) | |
95f72d1e | 734 | { |
6ed311b2 | 735 | phys_addr_t found; |
95f72d1e | 736 | |
847854f5 TH |
737 | /* align @size to avoid excessive fragmentation on reserved array */ |
738 | size = round_up(size, align); | |
739 | ||
7bd0b0f0 | 740 | found = memblock_find_in_range_node(0, max_addr, size, align, nid); |
9c8c27e2 | 741 | if (found && !memblock_reserve(found, size)) |
6ed311b2 | 742 | return found; |
95f72d1e | 743 | |
6ed311b2 | 744 | return 0; |
95f72d1e YL |
745 | } |
746 | ||
7bd0b0f0 TH |
747 | phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid) |
748 | { | |
749 | return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid); | |
750 | } | |
751 | ||
752 | phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) | |
753 | { | |
754 | return memblock_alloc_base_nid(size, align, max_addr, MAX_NUMNODES); | |
755 | } | |
756 | ||
6ed311b2 | 757 | phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr) |
95f72d1e | 758 | { |
6ed311b2 BH |
759 | phys_addr_t alloc; |
760 | ||
761 | alloc = __memblock_alloc_base(size, align, max_addr); | |
762 | ||
763 | if (alloc == 0) | |
764 | panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n", | |
765 | (unsigned long long) size, (unsigned long long) max_addr); | |
766 | ||
767 | return alloc; | |
95f72d1e YL |
768 | } |
769 | ||
6ed311b2 | 770 | phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align) |
95f72d1e | 771 | { |
6ed311b2 BH |
772 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
773 | } | |
95f72d1e | 774 | |
9d1e2492 BH |
775 | phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid) |
776 | { | |
777 | phys_addr_t res = memblock_alloc_nid(size, align, nid); | |
778 | ||
779 | if (res) | |
780 | return res; | |
15fb0972 | 781 | return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE); |
95f72d1e YL |
782 | } |
783 | ||
9d1e2492 BH |
784 | |
785 | /* | |
786 | * Remaining API functions | |
787 | */ | |
788 | ||
2898cc4c | 789 | phys_addr_t __init memblock_phys_mem_size(void) |
95f72d1e | 790 | { |
1440c4e2 | 791 | return memblock.memory.total_size; |
95f72d1e YL |
792 | } |
793 | ||
0a93ebef SR |
794 | /* lowest address */ |
795 | phys_addr_t __init_memblock memblock_start_of_DRAM(void) | |
796 | { | |
797 | return memblock.memory.regions[0].base; | |
798 | } | |
799 | ||
10d06439 | 800 | phys_addr_t __init_memblock memblock_end_of_DRAM(void) |
95f72d1e YL |
801 | { |
802 | int idx = memblock.memory.cnt - 1; | |
803 | ||
e3239ff9 | 804 | return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size); |
95f72d1e YL |
805 | } |
806 | ||
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 | 908 | pr_info("MEMBLOCK configuration:\n"); |
1440c4e2 TH |
909 | pr_info(" memory size = %#llx reserved size = %#llx\n", |
910 | (unsigned long long)memblock.memory.total_size, | |
911 | (unsigned long long)memblock.reserved.total_size); | |
6ed311b2 BH |
912 | |
913 | memblock_dump(&memblock.memory, "memory"); | |
914 | memblock_dump(&memblock.reserved, "reserved"); | |
915 | } | |
916 | ||
1aadc056 | 917 | void __init memblock_allow_resize(void) |
6ed311b2 | 918 | { |
142b45a7 | 919 | memblock_can_resize = 1; |
6ed311b2 BH |
920 | } |
921 | ||
6ed311b2 BH |
922 | static int __init early_memblock(char *p) |
923 | { | |
924 | if (p && strstr(p, "debug")) | |
925 | memblock_debug = 1; | |
926 | return 0; | |
927 | } | |
928 | early_param("memblock", early_memblock); | |
929 | ||
c378ddd5 | 930 | #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK) |
6d03b885 BH |
931 | |
932 | static int memblock_debug_show(struct seq_file *m, void *private) | |
933 | { | |
934 | struct memblock_type *type = m->private; | |
935 | struct memblock_region *reg; | |
936 | int i; | |
937 | ||
938 | for (i = 0; i < type->cnt; i++) { | |
939 | reg = &type->regions[i]; | |
940 | seq_printf(m, "%4d: ", i); | |
941 | if (sizeof(phys_addr_t) == 4) | |
942 | seq_printf(m, "0x%08lx..0x%08lx\n", | |
943 | (unsigned long)reg->base, | |
944 | (unsigned long)(reg->base + reg->size - 1)); | |
945 | else | |
946 | seq_printf(m, "0x%016llx..0x%016llx\n", | |
947 | (unsigned long long)reg->base, | |
948 | (unsigned long long)(reg->base + reg->size - 1)); | |
949 | ||
950 | } | |
951 | return 0; | |
952 | } | |
953 | ||
954 | static int memblock_debug_open(struct inode *inode, struct file *file) | |
955 | { | |
956 | return single_open(file, memblock_debug_show, inode->i_private); | |
957 | } | |
958 | ||
959 | static const struct file_operations memblock_debug_fops = { | |
960 | .open = memblock_debug_open, | |
961 | .read = seq_read, | |
962 | .llseek = seq_lseek, | |
963 | .release = single_release, | |
964 | }; | |
965 | ||
966 | static int __init memblock_init_debugfs(void) | |
967 | { | |
968 | struct dentry *root = debugfs_create_dir("memblock", NULL); | |
969 | if (!root) | |
970 | return -ENXIO; | |
971 | debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops); | |
972 | debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops); | |
973 | ||
974 | return 0; | |
975 | } | |
976 | __initcall(memblock_init_debugfs); | |
977 | ||
978 | #endif /* CONFIG_DEBUG_FS */ |