blob: 0f9626f01b5e6cc8d7e286cc8e00cb6fc38857ed [file] [log] [blame]
Yinghai Lu95f72d12010-07-12 14:36:09 +10001/*
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>
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -070014#include <linux/slab.h>
Yinghai Lu95f72d12010-07-12 14:36:09 +100015#include <linux/init.h>
16#include <linux/bitops.h>
Benjamin Herrenschmidt449e8df2010-07-06 15:39:07 -070017#include <linux/poison.h>
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -070018#include <linux/pfn.h>
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -070019#include <linux/debugfs.h>
20#include <linux/seq_file.h>
Yinghai Lu95f72d12010-07-12 14:36:09 +100021#include <linux/memblock.h>
22
Yinghai Lu10d06432010-07-28 15:43:02 +100023struct memblock memblock __initdata_memblock;
Yinghai Lu95f72d12010-07-12 14:36:09 +100024
Yinghai Lu10d06432010-07-28 15:43:02 +100025int memblock_debug __initdata_memblock;
26int memblock_can_resize __initdata_memblock;
27static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS + 1] __initdata_memblock;
28static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS + 1] __initdata_memblock;
Yinghai Lu95f72d12010-07-12 14:36:09 +100029
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -070030/* inline so we don't get a warning when pr_debug is compiled out */
31static 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
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100041/*
42 * Address comparison utilities
43 */
Yinghai Lu10d06432010-07-28 15:43:02 +100044static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +100045 phys_addr_t base2, phys_addr_t size2)
Yinghai Lu95f72d12010-07-12 14:36:09 +100046{
47 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
48}
49
Yinghai Lu10d06432010-07-28 15:43:02 +100050long __init_memblock memblock_overlaps_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100051{
52 unsigned long i;
53
54 for (i = 0; i < type->cnt; i++) {
55 phys_addr_t rgnbase = type->regions[i].base;
56 phys_addr_t rgnsize = type->regions[i].size;
57 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
58 break;
59 }
60
61 return (i < type->cnt) ? i : -1;
62}
63
64/*
65 * Find, allocate, deallocate or reserve unreserved regions. All allocations
66 * are top-down.
67 */
68
Yinghai Lucd794812010-10-11 12:34:09 -070069static phys_addr_t __init_memblock memblock_find_region(phys_addr_t start, phys_addr_t end,
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100070 phys_addr_t size, phys_addr_t align)
71{
72 phys_addr_t base, res_base;
73 long j;
74
Yinghai Luf1af98c2010-10-04 14:57:39 -070075 /* In case, huge size is requested */
76 if (end < size)
Tejun Heo1f5026a2011-07-12 09:58:09 +020077 return 0;
Yinghai Luf1af98c2010-10-04 14:57:39 -070078
Tejun Heo348968e2011-07-12 09:58:08 +020079 base = round_down(end - size, align);
Yinghai Luf1af98c2010-10-04 14:57:39 -070080
Benjamin Herrenschmidt25818f02010-07-28 15:25:10 +100081 /* Prevent allocations returning 0 as it's also used to
82 * indicate an allocation failure
83 */
84 if (start == 0)
85 start = PAGE_SIZE;
86
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100087 while (start <= base) {
88 j = memblock_overlaps_region(&memblock.reserved, base, size);
89 if (j < 0)
90 return base;
91 res_base = memblock.reserved.regions[j].base;
92 if (res_base < size)
93 break;
Tejun Heo348968e2011-07-12 09:58:08 +020094 base = round_down(res_base - size, align);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100095 }
96
Tejun Heo1f5026a2011-07-12 09:58:09 +020097 return 0;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100098}
99
Tejun Heofc769a82011-07-12 09:58:10 +0200100/*
101 * Find a free area with specified alignment in a specific range.
102 */
103phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, phys_addr_t end,
104 phys_addr_t size, phys_addr_t align)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000105{
106 long i;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000107
108 BUG_ON(0 == size);
109
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000110 /* Pump up max_addr */
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000111 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
112 end = memblock.current_limit;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000113
114 /* We do a top-down search, this tends to limit memory
115 * fragmentation by keeping early boot allocs near the
116 * top of memory
117 */
118 for (i = memblock.memory.cnt - 1; i >= 0; i--) {
119 phys_addr_t memblockbase = memblock.memory.regions[i].base;
120 phys_addr_t memblocksize = memblock.memory.regions[i].size;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000121 phys_addr_t bottom, top, found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000122
123 if (memblocksize < size)
124 continue;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000125 if ((memblockbase + memblocksize) <= start)
126 break;
127 bottom = max(memblockbase, start);
128 top = min(memblockbase + memblocksize, end);
129 if (bottom >= top)
130 continue;
131 found = memblock_find_region(bottom, top, size, align);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200132 if (found)
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000133 return found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000134 }
Tejun Heo1f5026a2011-07-12 09:58:09 +0200135 return 0;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000136}
137
Yinghai Lu5303b682010-07-28 15:38:40 +1000138/*
Yinghai Lu7950c402010-08-25 13:39:14 -0700139 * Free memblock.reserved.regions
140 */
141int __init_memblock memblock_free_reserved_regions(void)
142{
143 if (memblock.reserved.regions == memblock_reserved_init_regions)
144 return 0;
145
146 return memblock_free(__pa(memblock.reserved.regions),
147 sizeof(struct memblock_region) * memblock.reserved.max);
148}
149
150/*
151 * Reserve memblock.reserved.regions
152 */
153int __init_memblock memblock_reserve_reserved_regions(void)
154{
155 if (memblock.reserved.regions == memblock_reserved_init_regions)
156 return 0;
157
158 return memblock_reserve(__pa(memblock.reserved.regions),
159 sizeof(struct memblock_region) * memblock.reserved.max);
160}
161
Yinghai Lu10d06432010-07-28 15:43:02 +1000162static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000163{
164 unsigned long i;
165
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000166 for (i = r; i < type->cnt - 1; i++) {
167 type->regions[i].base = type->regions[i + 1].base;
168 type->regions[i].size = type->regions[i + 1].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000169 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000170 type->cnt--;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000171
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700172 /* Special case for empty arrays */
173 if (type->cnt == 0) {
174 type->cnt = 1;
175 type->regions[0].base = 0;
176 type->regions[0].size = 0;
177 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000178}
179
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700180/* Defined below but needed now */
181static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size);
182
Yinghai Lu10d06432010-07-28 15:43:02 +1000183static int __init_memblock memblock_double_array(struct memblock_type *type)
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700184{
185 struct memblock_region *new_array, *old_array;
186 phys_addr_t old_size, new_size, addr;
187 int use_slab = slab_is_available();
188
189 /* We don't allow resizing until we know about the reserved regions
190 * of memory that aren't suitable for allocation
191 */
192 if (!memblock_can_resize)
193 return -1;
194
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700195 /* Calculate new doubled size */
196 old_size = type->max * sizeof(struct memblock_region);
197 new_size = old_size << 1;
198
199 /* Try to find some space for it.
200 *
201 * WARNING: We assume that either slab_is_available() and we use it or
202 * we use MEMBLOCK for allocations. That means that this is unsafe to use
203 * when bootmem is currently active (unless bootmem itself is implemented
204 * on top of MEMBLOCK which isn't the case yet)
205 *
206 * This should however not be an issue for now, as we currently only
207 * call into MEMBLOCK while it's still active, or much later when slab is
208 * active for memory hotplug operations
209 */
210 if (use_slab) {
211 new_array = kmalloc(new_size, GFP_KERNEL);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200212 addr = new_array ? __pa(new_array) : 0;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700213 } else
Tejun Heofc769a82011-07-12 09:58:10 +0200214 addr = memblock_find_in_range(0, MEMBLOCK_ALLOC_ACCESSIBLE, new_size, sizeof(phys_addr_t));
Tejun Heo1f5026a2011-07-12 09:58:09 +0200215 if (!addr) {
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700216 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
217 memblock_type_name(type), type->max, type->max * 2);
218 return -1;
219 }
220 new_array = __va(addr);
221
Yinghai Luea9e4372010-07-28 15:13:22 +1000222 memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]",
223 memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1);
224
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700225 /* Found space, we now need to move the array over before
226 * we add the reserved region since it may be our reserved
227 * array itself that is full.
228 */
229 memcpy(new_array, type->regions, old_size);
230 memset(new_array + type->max, 0, old_size);
231 old_array = type->regions;
232 type->regions = new_array;
233 type->max <<= 1;
234
235 /* If we use SLAB that's it, we are done */
236 if (use_slab)
237 return 0;
238
239 /* Add the new reserved region now. Should not fail ! */
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700240 BUG_ON(memblock_add_region(&memblock.reserved, addr, new_size));
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700241
242 /* If the array wasn't our static init one, then free it. We only do
243 * that before SLAB is available as later on, we don't know whether
244 * to use kfree or free_bootmem_pages(). Shouldn't be a big deal
245 * anyways
246 */
247 if (old_array != memblock_memory_init_regions &&
248 old_array != memblock_reserved_init_regions)
249 memblock_free(__pa(old_array), old_size);
250
251 return 0;
252}
253
Yinghai Lu10d06432010-07-28 15:43:02 +1000254extern int __init_memblock __weak memblock_memory_can_coalesce(phys_addr_t addr1, phys_addr_t size1,
Benjamin Herrenschmidtd2cd5632010-07-06 15:39:14 -0700255 phys_addr_t addr2, phys_addr_t size2)
256{
257 return 1;
258}
259
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700260static long __init_memblock memblock_add_region(struct memblock_type *type,
261 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000262{
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700263 phys_addr_t end = base + size;
264 int i, slot = -1;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000265
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700266 /* First try and coalesce this MEMBLOCK with others */
267 for (i = 0; i < type->cnt; i++) {
268 struct memblock_region *rgn = &type->regions[i];
269 phys_addr_t rend = rgn->base + rgn->size;
270
271 /* Exit if there's no possible hits */
272 if (rgn->base > end || rgn->size == 0)
273 break;
274
275 /* Check if we are fully enclosed within an existing
276 * block
277 */
278 if (rgn->base <= base && rend >= end)
279 return 0;
280
281 /* Check if we overlap or are adjacent with the bottom
282 * of a block.
283 */
284 if (base < rgn->base && end >= rgn->base) {
285 /* If we can't coalesce, create a new block */
286 if (!memblock_memory_can_coalesce(base, size,
287 rgn->base,
288 rgn->size)) {
289 /* Overlap & can't coalesce are mutually
290 * exclusive, if you do that, be prepared
291 * for trouble
292 */
293 WARN_ON(end != rgn->base);
294 goto new_block;
295 }
296 /* We extend the bottom of the block down to our
297 * base
298 */
299 rgn->base = base;
300 rgn->size = rend - base;
301
302 /* Return if we have nothing else to allocate
303 * (fully coalesced)
304 */
305 if (rend >= end)
306 return 0;
307
308 /* We continue processing from the end of the
309 * coalesced block.
310 */
311 base = rend;
312 size = end - base;
313 }
314
315 /* Now check if we overlap or are adjacent with the
316 * top of a block
317 */
318 if (base <= rend && end >= rend) {
319 /* If we can't coalesce, create a new block */
320 if (!memblock_memory_can_coalesce(rgn->base,
321 rgn->size,
322 base, size)) {
323 /* Overlap & can't coalesce are mutually
324 * exclusive, if you do that, be prepared
325 * for trouble
326 */
327 WARN_ON(rend != base);
328 goto new_block;
329 }
330 /* We adjust our base down to enclose the
331 * original block and destroy it. It will be
332 * part of our new allocation. Since we've
333 * freed an entry, we know we won't fail
334 * to allocate one later, so we won't risk
335 * losing the original block allocation.
336 */
337 size += (base - rgn->base);
338 base = rgn->base;
339 memblock_remove_region(type, i--);
340 }
341 }
342
343 /* If the array is empty, special case, replace the fake
344 * filler region and return
345 */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000346 if ((type->cnt == 1) && (type->regions[0].size == 0)) {
347 type->regions[0].base = base;
348 type->regions[0].size = size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000349 return 0;
350 }
351
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700352 new_block:
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700353 /* If we are out of space, we fail. It's too late to resize the array
354 * but then this shouldn't have happened in the first place.
355 */
356 if (WARN_ON(type->cnt >= type->max))
Yinghai Lu95f72d12010-07-12 14:36:09 +1000357 return -1;
358
359 /* Couldn't coalesce the MEMBLOCK, so add it to the sorted table. */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000360 for (i = type->cnt - 1; i >= 0; i--) {
361 if (base < type->regions[i].base) {
362 type->regions[i+1].base = type->regions[i].base;
363 type->regions[i+1].size = type->regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000364 } else {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000365 type->regions[i+1].base = base;
366 type->regions[i+1].size = size;
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700367 slot = i + 1;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000368 break;
369 }
370 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000371 if (base < type->regions[0].base) {
372 type->regions[0].base = base;
373 type->regions[0].size = size;
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700374 slot = 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000375 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000376 type->cnt++;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000377
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700378 /* The array is full ? Try to resize it. If that fails, we undo
379 * our allocation and return an error
380 */
381 if (type->cnt == type->max && memblock_double_array(type)) {
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700382 BUG_ON(slot < 0);
383 memblock_remove_region(type, slot);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700384 return -1;
385 }
386
Yinghai Lu95f72d12010-07-12 14:36:09 +1000387 return 0;
388}
389
Yinghai Lu10d06432010-07-28 15:43:02 +1000390long __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000391{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000392 return memblock_add_region(&memblock.memory, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000393
394}
395
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700396static long __init_memblock __memblock_remove(struct memblock_type *type,
397 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000398{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000399 phys_addr_t end = base + size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000400 int i;
401
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700402 /* Walk through the array for collisions */
403 for (i = 0; i < type->cnt; i++) {
404 struct memblock_region *rgn = &type->regions[i];
405 phys_addr_t rend = rgn->base + rgn->size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000406
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700407 /* Nothing more to do, exit */
408 if (rgn->base > end || rgn->size == 0)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000409 break;
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700410
411 /* If we fully enclose the block, drop it */
412 if (base <= rgn->base && end >= rend) {
413 memblock_remove_region(type, i--);
414 continue;
415 }
416
417 /* If we are fully enclosed within a block
418 * then we need to split it and we are done
419 */
420 if (base > rgn->base && end < rend) {
421 rgn->size = base - rgn->base;
422 if (!memblock_add_region(type, end, rend - end))
423 return 0;
424 /* Failure to split is bad, we at least
425 * restore the block before erroring
426 */
427 rgn->size = rend - rgn->base;
428 WARN_ON(1);
429 return -1;
430 }
431
432 /* Check if we need to trim the bottom of a block */
433 if (rgn->base < end && rend > end) {
434 rgn->size -= end - rgn->base;
435 rgn->base = end;
436 break;
437 }
438
439 /* And check if we need to trim the top of a block */
440 if (base < rend)
441 rgn->size -= rend - base;
442
Yinghai Lu95f72d12010-07-12 14:36:09 +1000443 }
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700444 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000445}
446
Yinghai Lu10d06432010-07-28 15:43:02 +1000447long __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000448{
449 return __memblock_remove(&memblock.memory, base, size);
450}
451
Yinghai Lu3661ca62010-09-15 13:05:29 -0700452long __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000453{
454 return __memblock_remove(&memblock.reserved, base, size);
455}
456
Yinghai Lu3661ca62010-09-15 13:05:29 -0700457long __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000458{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000459 struct memblock_type *_rgn = &memblock.reserved;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000460
461 BUG_ON(0 == size);
462
463 return memblock_add_region(_rgn, base, size);
464}
465
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000466phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000467{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000468 phys_addr_t found;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000469
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000470 /* We align the size to limit fragmentation. Without this, a lot of
471 * small allocs quickly eat up the whole reserve array on sparc
472 */
Tejun Heo348968e2011-07-12 09:58:08 +0200473 size = round_up(size, align);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000474
Tejun Heofc769a82011-07-12 09:58:10 +0200475 found = memblock_find_in_range(0, max_addr, size, align);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200476 if (found && !memblock_add_region(&memblock.reserved, found, size))
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000477 return found;
478
479 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000480}
481
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000482phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000483{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000484 phys_addr_t alloc;
485
486 alloc = __memblock_alloc_base(size, align, max_addr);
487
488 if (alloc == 0)
489 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
490 (unsigned long long) size, (unsigned long long) max_addr);
491
492 return alloc;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000493}
494
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000495phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000496{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000497 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000498}
499
Yinghai Lu95f72d12010-07-12 14:36:09 +1000500
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000501/*
502 * Additional node-local allocators. Search for node memory is bottom up
503 * and walks memblock regions within that node bottom-up as well, but allocation
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700504 * within an memblock region is top-down. XXX I plan to fix that at some stage
505 *
506 * WARNING: Only available after early_node_map[] has been populated,
507 * on some architectures, that is after all the calls to add_active_range()
508 * have been done to populate it.
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000509 */
Yinghai Lu95f72d12010-07-12 14:36:09 +1000510
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000511phys_addr_t __weak __init memblock_nid_range(phys_addr_t start, phys_addr_t end, int *nid)
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700512{
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700513#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
514 /*
515 * This code originates from sparc which really wants use to walk by addresses
516 * and returns the nid. This is not very convenient for early_pfn_map[] users
517 * as the map isn't sorted yet, and it really wants to be walked by nid.
518 *
519 * For now, I implement the inefficient method below which walks the early
520 * map multiple times. Eventually we may want to use an ARCH config option
521 * to implement a completely different method for both case.
522 */
523 unsigned long start_pfn, end_pfn;
524 int i;
525
526 for (i = 0; i < MAX_NUMNODES; i++) {
527 get_pfn_range_for_nid(i, &start_pfn, &end_pfn);
528 if (start < PFN_PHYS(start_pfn) || start >= PFN_PHYS(end_pfn))
529 continue;
530 *nid = i;
531 return min(end, PFN_PHYS(end_pfn));
532 }
533#endif
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700534 *nid = 0;
535
536 return end;
537}
538
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000539static phys_addr_t __init memblock_alloc_nid_region(struct memblock_region *mp,
540 phys_addr_t size,
541 phys_addr_t align, int nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000542{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000543 phys_addr_t start, end;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000544
545 start = mp->base;
546 end = start + mp->size;
547
Tejun Heo348968e2011-07-12 09:58:08 +0200548 start = round_up(start, align);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000549 while (start < end) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000550 phys_addr_t this_end;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000551 int this_nid;
552
Benjamin Herrenschmidt35a1f0b2010-07-06 15:38:58 -0700553 this_end = memblock_nid_range(start, end, &this_nid);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000554 if (this_nid == nid) {
Benjamin Herrenschmidt3a9c2c82010-07-12 13:28:15 +1000555 phys_addr_t ret = memblock_find_region(start, this_end, size, align);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200556 if (ret &&
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700557 !memblock_add_region(&memblock.reserved, ret, size))
Yinghai Lu95f72d12010-07-12 14:36:09 +1000558 return ret;
559 }
560 start = this_end;
561 }
562
Tejun Heo1f5026a2011-07-12 09:58:09 +0200563 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000564}
565
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000566phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000567{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000568 struct memblock_type *mem = &memblock.memory;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000569 int i;
570
571 BUG_ON(0 == size);
572
Benjamin Herrenschmidt7f219c72010-07-12 14:24:57 +1000573 /* We align the size to limit fragmentation. Without this, a lot of
574 * small allocs quickly eat up the whole reserve array on sparc
575 */
Tejun Heo348968e2011-07-12 09:58:08 +0200576 size = round_up(size, align);
Benjamin Herrenschmidt7f219c72010-07-12 14:24:57 +1000577
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700578 /* We do a bottom-up search for a region with the right
579 * nid since that's easier considering how memblock_nid_range()
580 * works
581 */
Yinghai Lu95f72d12010-07-12 14:36:09 +1000582 for (i = 0; i < mem->cnt; i++) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000583 phys_addr_t ret = memblock_alloc_nid_region(&mem->regions[i],
Yinghai Lu95f72d12010-07-12 14:36:09 +1000584 size, align, nid);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200585 if (ret)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000586 return ret;
587 }
588
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700589 return 0;
590}
591
592phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
593{
594 phys_addr_t res = memblock_alloc_nid(size, align, nid);
595
596 if (res)
597 return res;
Tejun Heo15fb0972011-07-12 09:58:07 +0200598 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000599}
600
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700601
602/*
603 * Remaining API functions
604 */
605
Yinghai Lu95f72d12010-07-12 14:36:09 +1000606/* You must call memblock_analyze() before this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000607phys_addr_t __init memblock_phys_mem_size(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000608{
Benjamin Herrenschmidt4734b592010-07-28 14:31:29 +1000609 return memblock.memory_size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000610}
611
Yinghai Lu10d06432010-07-28 15:43:02 +1000612phys_addr_t __init_memblock memblock_end_of_DRAM(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000613{
614 int idx = memblock.memory.cnt - 1;
615
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000616 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000617}
618
619/* You must call memblock_analyze() after this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000620void __init memblock_enforce_memory_limit(phys_addr_t memory_limit)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000621{
622 unsigned long i;
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000623 phys_addr_t limit;
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000624 struct memblock_region *p;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000625
626 if (!memory_limit)
627 return;
628
629 /* Truncate the memblock regions to satisfy the memory limit. */
630 limit = memory_limit;
631 for (i = 0; i < memblock.memory.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000632 if (limit > memblock.memory.regions[i].size) {
633 limit -= memblock.memory.regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000634 continue;
635 }
636
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000637 memblock.memory.regions[i].size = limit;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000638 memblock.memory.cnt = i + 1;
639 break;
640 }
641
Yinghai Lu95f72d12010-07-12 14:36:09 +1000642 memory_limit = memblock_end_of_DRAM();
643
644 /* And truncate any reserves above the limit also. */
645 for (i = 0; i < memblock.reserved.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000646 p = &memblock.reserved.regions[i];
Yinghai Lu95f72d12010-07-12 14:36:09 +1000647
648 if (p->base > memory_limit)
649 p->size = 0;
650 else if ((p->base + p->size) > memory_limit)
651 p->size = memory_limit - p->base;
652
653 if (p->size == 0) {
654 memblock_remove_region(&memblock.reserved, i);
655 i--;
656 }
657 }
658}
659
Yinghai Lucd794812010-10-11 12:34:09 -0700660static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000661{
662 unsigned int left = 0, right = type->cnt;
663
664 do {
665 unsigned int mid = (right + left) / 2;
666
667 if (addr < type->regions[mid].base)
668 right = mid;
669 else if (addr >= (type->regions[mid].base +
670 type->regions[mid].size))
671 left = mid + 1;
672 else
673 return mid;
674 } while (left < right);
675 return -1;
676}
677
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000678int __init memblock_is_reserved(phys_addr_t addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000679{
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000680 return memblock_search(&memblock.reserved, addr) != -1;
681}
Yinghai Lu95f72d12010-07-12 14:36:09 +1000682
Yinghai Lu3661ca62010-09-15 13:05:29 -0700683int __init_memblock memblock_is_memory(phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000684{
685 return memblock_search(&memblock.memory, addr) != -1;
686}
687
Yinghai Lu3661ca62010-09-15 13:05:29 -0700688int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000689{
Tomi Valkeinenabb65272011-01-20 14:44:20 -0800690 int idx = memblock_search(&memblock.memory, base);
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000691
692 if (idx == -1)
693 return 0;
Tomi Valkeinenabb65272011-01-20 14:44:20 -0800694 return memblock.memory.regions[idx].base <= base &&
695 (memblock.memory.regions[idx].base +
696 memblock.memory.regions[idx].size) >= (base + size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000697}
698
Yinghai Lu10d06432010-07-28 15:43:02 +1000699int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000700{
Benjamin Herrenschmidtf1c2c192010-08-04 14:17:17 +1000701 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000702}
703
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700704
Yinghai Lu3661ca62010-09-15 13:05:29 -0700705void __init_memblock memblock_set_current_limit(phys_addr_t limit)
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700706{
707 memblock.current_limit = limit;
708}
709
Yinghai Lu10d06432010-07-28 15:43:02 +1000710static void __init_memblock memblock_dump(struct memblock_type *region, char *name)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000711{
712 unsigned long long base, size;
713 int i;
714
715 pr_info(" %s.cnt = 0x%lx\n", name, region->cnt);
716
717 for (i = 0; i < region->cnt; i++) {
718 base = region->regions[i].base;
719 size = region->regions[i].size;
720
Yinghai Luea9e4372010-07-28 15:13:22 +1000721 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes\n",
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000722 name, i, base, base + size - 1, size);
723 }
724}
725
Yinghai Lu10d06432010-07-28 15:43:02 +1000726void __init_memblock memblock_dump_all(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000727{
728 if (!memblock_debug)
729 return;
730
731 pr_info("MEMBLOCK configuration:\n");
732 pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size);
733
734 memblock_dump(&memblock.memory, "memory");
735 memblock_dump(&memblock.reserved, "reserved");
736}
737
738void __init memblock_analyze(void)
739{
740 int i;
741
742 /* Check marker in the unused last array entry */
743 WARN_ON(memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS].base
744 != (phys_addr_t)RED_INACTIVE);
745 WARN_ON(memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS].base
746 != (phys_addr_t)RED_INACTIVE);
747
748 memblock.memory_size = 0;
749
750 for (i = 0; i < memblock.memory.cnt; i++)
751 memblock.memory_size += memblock.memory.regions[i].size;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700752
753 /* We allow resizing from there */
754 memblock_can_resize = 1;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000755}
756
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700757void __init memblock_init(void)
758{
Jeremy Fitzhardinge236260b2010-10-06 15:52:29 -0700759 static int init_done __initdata = 0;
760
761 if (init_done)
762 return;
763 init_done = 1;
764
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700765 /* Hookup the initial arrays */
766 memblock.memory.regions = memblock_memory_init_regions;
767 memblock.memory.max = INIT_MEMBLOCK_REGIONS;
768 memblock.reserved.regions = memblock_reserved_init_regions;
769 memblock.reserved.max = INIT_MEMBLOCK_REGIONS;
770
771 /* Write a marker in the unused last array entry */
772 memblock.memory.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
773 memblock.reserved.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
774
775 /* Create a dummy zero size MEMBLOCK which will get coalesced away later.
776 * This simplifies the memblock_add() code below...
777 */
778 memblock.memory.regions[0].base = 0;
779 memblock.memory.regions[0].size = 0;
780 memblock.memory.cnt = 1;
781
782 /* Ditto. */
783 memblock.reserved.regions[0].base = 0;
784 memblock.reserved.regions[0].size = 0;
785 memblock.reserved.cnt = 1;
786
787 memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE;
788}
789
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000790static int __init early_memblock(char *p)
791{
792 if (p && strstr(p, "debug"))
793 memblock_debug = 1;
794 return 0;
795}
796early_param("memblock", early_memblock);
797
Yinghai Lu10d06432010-07-28 15:43:02 +1000798#if defined(CONFIG_DEBUG_FS) && !defined(ARCH_DISCARD_MEMBLOCK)
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -0700799
800static int memblock_debug_show(struct seq_file *m, void *private)
801{
802 struct memblock_type *type = m->private;
803 struct memblock_region *reg;
804 int i;
805
806 for (i = 0; i < type->cnt; i++) {
807 reg = &type->regions[i];
808 seq_printf(m, "%4d: ", i);
809 if (sizeof(phys_addr_t) == 4)
810 seq_printf(m, "0x%08lx..0x%08lx\n",
811 (unsigned long)reg->base,
812 (unsigned long)(reg->base + reg->size - 1));
813 else
814 seq_printf(m, "0x%016llx..0x%016llx\n",
815 (unsigned long long)reg->base,
816 (unsigned long long)(reg->base + reg->size - 1));
817
818 }
819 return 0;
820}
821
822static int memblock_debug_open(struct inode *inode, struct file *file)
823{
824 return single_open(file, memblock_debug_show, inode->i_private);
825}
826
827static const struct file_operations memblock_debug_fops = {
828 .open = memblock_debug_open,
829 .read = seq_read,
830 .llseek = seq_lseek,
831 .release = single_release,
832};
833
834static int __init memblock_init_debugfs(void)
835{
836 struct dentry *root = debugfs_create_dir("memblock", NULL);
837 if (!root)
838 return -ENXIO;
839 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
840 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
841
842 return 0;
843}
844__initcall(memblock_init_debugfs);
845
846#endif /* CONFIG_DEBUG_FS */