blob: 19699366134651ae917cb6c048dd342938f697b6 [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
Yinghai Lu3661ca62010-09-15 13:05:29 -0700100static phys_addr_t __init_memblock memblock_find_base(phys_addr_t size,
101 phys_addr_t align, phys_addr_t start, phys_addr_t end)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000102{
103 long i;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000104
105 BUG_ON(0 == size);
106
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000107 /* Pump up max_addr */
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000108 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
109 end = memblock.current_limit;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000110
111 /* We do a top-down search, this tends to limit memory
112 * fragmentation by keeping early boot allocs near the
113 * top of memory
114 */
115 for (i = memblock.memory.cnt - 1; i >= 0; i--) {
116 phys_addr_t memblockbase = memblock.memory.regions[i].base;
117 phys_addr_t memblocksize = memblock.memory.regions[i].size;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000118 phys_addr_t bottom, top, found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000119
120 if (memblocksize < size)
121 continue;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000122 if ((memblockbase + memblocksize) <= start)
123 break;
124 bottom = max(memblockbase, start);
125 top = min(memblockbase + memblocksize, end);
126 if (bottom >= top)
127 continue;
128 found = memblock_find_region(bottom, top, size, align);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200129 if (found)
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000130 return found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000131 }
Tejun Heo1f5026a2011-07-12 09:58:09 +0200132 return 0;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000133}
134
Yinghai Lu5303b682010-07-28 15:38:40 +1000135/*
136 * Find a free area with specified alignment in a specific range.
137 */
138u64 __init_memblock memblock_find_in_range(u64 start, u64 end, u64 size, u64 align)
139{
140 return memblock_find_base(size, align, start, end);
141}
142
Yinghai Lu7950c402010-08-25 13:39:14 -0700143/*
144 * Free memblock.reserved.regions
145 */
146int __init_memblock memblock_free_reserved_regions(void)
147{
148 if (memblock.reserved.regions == memblock_reserved_init_regions)
149 return 0;
150
151 return memblock_free(__pa(memblock.reserved.regions),
152 sizeof(struct memblock_region) * memblock.reserved.max);
153}
154
155/*
156 * Reserve memblock.reserved.regions
157 */
158int __init_memblock memblock_reserve_reserved_regions(void)
159{
160 if (memblock.reserved.regions == memblock_reserved_init_regions)
161 return 0;
162
163 return memblock_reserve(__pa(memblock.reserved.regions),
164 sizeof(struct memblock_region) * memblock.reserved.max);
165}
166
Yinghai Lu10d06432010-07-28 15:43:02 +1000167static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000168{
169 unsigned long i;
170
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000171 for (i = r; i < type->cnt - 1; i++) {
172 type->regions[i].base = type->regions[i + 1].base;
173 type->regions[i].size = type->regions[i + 1].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000174 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000175 type->cnt--;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000176
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700177 /* Special case for empty arrays */
178 if (type->cnt == 0) {
179 type->cnt = 1;
180 type->regions[0].base = 0;
181 type->regions[0].size = 0;
182 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000183}
184
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700185/* Defined below but needed now */
186static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size);
187
Yinghai Lu10d06432010-07-28 15:43:02 +1000188static int __init_memblock memblock_double_array(struct memblock_type *type)
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700189{
190 struct memblock_region *new_array, *old_array;
191 phys_addr_t old_size, new_size, addr;
192 int use_slab = slab_is_available();
193
194 /* We don't allow resizing until we know about the reserved regions
195 * of memory that aren't suitable for allocation
196 */
197 if (!memblock_can_resize)
198 return -1;
199
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700200 /* Calculate new doubled size */
201 old_size = type->max * sizeof(struct memblock_region);
202 new_size = old_size << 1;
203
204 /* Try to find some space for it.
205 *
206 * WARNING: We assume that either slab_is_available() and we use it or
207 * we use MEMBLOCK for allocations. That means that this is unsafe to use
208 * when bootmem is currently active (unless bootmem itself is implemented
209 * on top of MEMBLOCK which isn't the case yet)
210 *
211 * This should however not be an issue for now, as we currently only
212 * call into MEMBLOCK while it's still active, or much later when slab is
213 * active for memory hotplug operations
214 */
215 if (use_slab) {
216 new_array = kmalloc(new_size, GFP_KERNEL);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200217 addr = new_array ? __pa(new_array) : 0;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700218 } else
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000219 addr = memblock_find_base(new_size, sizeof(phys_addr_t), 0, MEMBLOCK_ALLOC_ACCESSIBLE);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200220 if (!addr) {
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700221 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
222 memblock_type_name(type), type->max, type->max * 2);
223 return -1;
224 }
225 new_array = __va(addr);
226
Yinghai Luea9e4372010-07-28 15:13:22 +1000227 memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]",
228 memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1);
229
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700230 /* Found space, we now need to move the array over before
231 * we add the reserved region since it may be our reserved
232 * array itself that is full.
233 */
234 memcpy(new_array, type->regions, old_size);
235 memset(new_array + type->max, 0, old_size);
236 old_array = type->regions;
237 type->regions = new_array;
238 type->max <<= 1;
239
240 /* If we use SLAB that's it, we are done */
241 if (use_slab)
242 return 0;
243
244 /* Add the new reserved region now. Should not fail ! */
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700245 BUG_ON(memblock_add_region(&memblock.reserved, addr, new_size));
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700246
247 /* If the array wasn't our static init one, then free it. We only do
248 * that before SLAB is available as later on, we don't know whether
249 * to use kfree or free_bootmem_pages(). Shouldn't be a big deal
250 * anyways
251 */
252 if (old_array != memblock_memory_init_regions &&
253 old_array != memblock_reserved_init_regions)
254 memblock_free(__pa(old_array), old_size);
255
256 return 0;
257}
258
Yinghai Lu10d06432010-07-28 15:43:02 +1000259extern int __init_memblock __weak memblock_memory_can_coalesce(phys_addr_t addr1, phys_addr_t size1,
Benjamin Herrenschmidtd2cd5632010-07-06 15:39:14 -0700260 phys_addr_t addr2, phys_addr_t size2)
261{
262 return 1;
263}
264
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700265static long __init_memblock memblock_add_region(struct memblock_type *type,
266 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000267{
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700268 phys_addr_t end = base + size;
269 int i, slot = -1;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000270
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700271 /* First try and coalesce this MEMBLOCK with others */
272 for (i = 0; i < type->cnt; i++) {
273 struct memblock_region *rgn = &type->regions[i];
274 phys_addr_t rend = rgn->base + rgn->size;
275
276 /* Exit if there's no possible hits */
277 if (rgn->base > end || rgn->size == 0)
278 break;
279
280 /* Check if we are fully enclosed within an existing
281 * block
282 */
283 if (rgn->base <= base && rend >= end)
284 return 0;
285
286 /* Check if we overlap or are adjacent with the bottom
287 * of a block.
288 */
289 if (base < rgn->base && end >= rgn->base) {
290 /* If we can't coalesce, create a new block */
291 if (!memblock_memory_can_coalesce(base, size,
292 rgn->base,
293 rgn->size)) {
294 /* Overlap & can't coalesce are mutually
295 * exclusive, if you do that, be prepared
296 * for trouble
297 */
298 WARN_ON(end != rgn->base);
299 goto new_block;
300 }
301 /* We extend the bottom of the block down to our
302 * base
303 */
304 rgn->base = base;
305 rgn->size = rend - base;
306
307 /* Return if we have nothing else to allocate
308 * (fully coalesced)
309 */
310 if (rend >= end)
311 return 0;
312
313 /* We continue processing from the end of the
314 * coalesced block.
315 */
316 base = rend;
317 size = end - base;
318 }
319
320 /* Now check if we overlap or are adjacent with the
321 * top of a block
322 */
323 if (base <= rend && end >= rend) {
324 /* If we can't coalesce, create a new block */
325 if (!memblock_memory_can_coalesce(rgn->base,
326 rgn->size,
327 base, size)) {
328 /* Overlap & can't coalesce are mutually
329 * exclusive, if you do that, be prepared
330 * for trouble
331 */
332 WARN_ON(rend != base);
333 goto new_block;
334 }
335 /* We adjust our base down to enclose the
336 * original block and destroy it. It will be
337 * part of our new allocation. Since we've
338 * freed an entry, we know we won't fail
339 * to allocate one later, so we won't risk
340 * losing the original block allocation.
341 */
342 size += (base - rgn->base);
343 base = rgn->base;
344 memblock_remove_region(type, i--);
345 }
346 }
347
348 /* If the array is empty, special case, replace the fake
349 * filler region and return
350 */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000351 if ((type->cnt == 1) && (type->regions[0].size == 0)) {
352 type->regions[0].base = base;
353 type->regions[0].size = size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000354 return 0;
355 }
356
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700357 new_block:
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700358 /* If we are out of space, we fail. It's too late to resize the array
359 * but then this shouldn't have happened in the first place.
360 */
361 if (WARN_ON(type->cnt >= type->max))
Yinghai Lu95f72d12010-07-12 14:36:09 +1000362 return -1;
363
364 /* Couldn't coalesce the MEMBLOCK, so add it to the sorted table. */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000365 for (i = type->cnt - 1; i >= 0; i--) {
366 if (base < type->regions[i].base) {
367 type->regions[i+1].base = type->regions[i].base;
368 type->regions[i+1].size = type->regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000369 } else {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000370 type->regions[i+1].base = base;
371 type->regions[i+1].size = size;
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700372 slot = i + 1;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000373 break;
374 }
375 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000376 if (base < type->regions[0].base) {
377 type->regions[0].base = base;
378 type->regions[0].size = size;
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700379 slot = 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000380 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000381 type->cnt++;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000382
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700383 /* The array is full ? Try to resize it. If that fails, we undo
384 * our allocation and return an error
385 */
386 if (type->cnt == type->max && memblock_double_array(type)) {
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700387 BUG_ON(slot < 0);
388 memblock_remove_region(type, slot);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700389 return -1;
390 }
391
Yinghai Lu95f72d12010-07-12 14:36:09 +1000392 return 0;
393}
394
Yinghai Lu10d06432010-07-28 15:43:02 +1000395long __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000396{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000397 return memblock_add_region(&memblock.memory, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000398
399}
400
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700401static long __init_memblock __memblock_remove(struct memblock_type *type,
402 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000403{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000404 phys_addr_t end = base + size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000405 int i;
406
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700407 /* Walk through the array for collisions */
408 for (i = 0; i < type->cnt; i++) {
409 struct memblock_region *rgn = &type->regions[i];
410 phys_addr_t rend = rgn->base + rgn->size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000411
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700412 /* Nothing more to do, exit */
413 if (rgn->base > end || rgn->size == 0)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000414 break;
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700415
416 /* If we fully enclose the block, drop it */
417 if (base <= rgn->base && end >= rend) {
418 memblock_remove_region(type, i--);
419 continue;
420 }
421
422 /* If we are fully enclosed within a block
423 * then we need to split it and we are done
424 */
425 if (base > rgn->base && end < rend) {
426 rgn->size = base - rgn->base;
427 if (!memblock_add_region(type, end, rend - end))
428 return 0;
429 /* Failure to split is bad, we at least
430 * restore the block before erroring
431 */
432 rgn->size = rend - rgn->base;
433 WARN_ON(1);
434 return -1;
435 }
436
437 /* Check if we need to trim the bottom of a block */
438 if (rgn->base < end && rend > end) {
439 rgn->size -= end - rgn->base;
440 rgn->base = end;
441 break;
442 }
443
444 /* And check if we need to trim the top of a block */
445 if (base < rend)
446 rgn->size -= rend - base;
447
Yinghai Lu95f72d12010-07-12 14:36:09 +1000448 }
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700449 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000450}
451
Yinghai Lu10d06432010-07-28 15:43:02 +1000452long __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000453{
454 return __memblock_remove(&memblock.memory, base, size);
455}
456
Yinghai Lu3661ca62010-09-15 13:05:29 -0700457long __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000458{
459 return __memblock_remove(&memblock.reserved, base, size);
460}
461
Yinghai Lu3661ca62010-09-15 13:05:29 -0700462long __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000463{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000464 struct memblock_type *_rgn = &memblock.reserved;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000465
466 BUG_ON(0 == size);
467
468 return memblock_add_region(_rgn, base, size);
469}
470
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000471phys_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 +1000472{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000473 phys_addr_t found;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000474
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000475 /* We align the size to limit fragmentation. Without this, a lot of
476 * small allocs quickly eat up the whole reserve array on sparc
477 */
Tejun Heo348968e2011-07-12 09:58:08 +0200478 size = round_up(size, align);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000479
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000480 found = memblock_find_base(size, align, 0, max_addr);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200481 if (found && !memblock_add_region(&memblock.reserved, found, size))
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000482 return found;
483
484 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000485}
486
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000487phys_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 +1000488{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000489 phys_addr_t alloc;
490
491 alloc = __memblock_alloc_base(size, align, max_addr);
492
493 if (alloc == 0)
494 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
495 (unsigned long long) size, (unsigned long long) max_addr);
496
497 return alloc;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000498}
499
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000500phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000501{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000502 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000503}
504
Yinghai Lu95f72d12010-07-12 14:36:09 +1000505
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000506/*
507 * Additional node-local allocators. Search for node memory is bottom up
508 * and walks memblock regions within that node bottom-up as well, but allocation
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700509 * within an memblock region is top-down. XXX I plan to fix that at some stage
510 *
511 * WARNING: Only available after early_node_map[] has been populated,
512 * on some architectures, that is after all the calls to add_active_range()
513 * have been done to populate it.
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000514 */
Yinghai Lu95f72d12010-07-12 14:36:09 +1000515
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000516phys_addr_t __weak __init memblock_nid_range(phys_addr_t start, phys_addr_t end, int *nid)
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700517{
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700518#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
519 /*
520 * This code originates from sparc which really wants use to walk by addresses
521 * and returns the nid. This is not very convenient for early_pfn_map[] users
522 * as the map isn't sorted yet, and it really wants to be walked by nid.
523 *
524 * For now, I implement the inefficient method below which walks the early
525 * map multiple times. Eventually we may want to use an ARCH config option
526 * to implement a completely different method for both case.
527 */
528 unsigned long start_pfn, end_pfn;
529 int i;
530
531 for (i = 0; i < MAX_NUMNODES; i++) {
532 get_pfn_range_for_nid(i, &start_pfn, &end_pfn);
533 if (start < PFN_PHYS(start_pfn) || start >= PFN_PHYS(end_pfn))
534 continue;
535 *nid = i;
536 return min(end, PFN_PHYS(end_pfn));
537 }
538#endif
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700539 *nid = 0;
540
541 return end;
542}
543
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000544static phys_addr_t __init memblock_alloc_nid_region(struct memblock_region *mp,
545 phys_addr_t size,
546 phys_addr_t align, int nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000547{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000548 phys_addr_t start, end;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000549
550 start = mp->base;
551 end = start + mp->size;
552
Tejun Heo348968e2011-07-12 09:58:08 +0200553 start = round_up(start, align);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000554 while (start < end) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000555 phys_addr_t this_end;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000556 int this_nid;
557
Benjamin Herrenschmidt35a1f0b2010-07-06 15:38:58 -0700558 this_end = memblock_nid_range(start, end, &this_nid);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000559 if (this_nid == nid) {
Benjamin Herrenschmidt3a9c2c82010-07-12 13:28:15 +1000560 phys_addr_t ret = memblock_find_region(start, this_end, size, align);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200561 if (ret &&
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700562 !memblock_add_region(&memblock.reserved, ret, size))
Yinghai Lu95f72d12010-07-12 14:36:09 +1000563 return ret;
564 }
565 start = this_end;
566 }
567
Tejun Heo1f5026a2011-07-12 09:58:09 +0200568 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000569}
570
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000571phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000572{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000573 struct memblock_type *mem = &memblock.memory;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000574 int i;
575
576 BUG_ON(0 == size);
577
Benjamin Herrenschmidt7f219c72010-07-12 14:24:57 +1000578 /* We align the size to limit fragmentation. Without this, a lot of
579 * small allocs quickly eat up the whole reserve array on sparc
580 */
Tejun Heo348968e2011-07-12 09:58:08 +0200581 size = round_up(size, align);
Benjamin Herrenschmidt7f219c72010-07-12 14:24:57 +1000582
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700583 /* We do a bottom-up search for a region with the right
584 * nid since that's easier considering how memblock_nid_range()
585 * works
586 */
Yinghai Lu95f72d12010-07-12 14:36:09 +1000587 for (i = 0; i < mem->cnt; i++) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000588 phys_addr_t ret = memblock_alloc_nid_region(&mem->regions[i],
Yinghai Lu95f72d12010-07-12 14:36:09 +1000589 size, align, nid);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200590 if (ret)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000591 return ret;
592 }
593
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700594 return 0;
595}
596
597phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
598{
599 phys_addr_t res = memblock_alloc_nid(size, align, nid);
600
601 if (res)
602 return res;
Tejun Heo15fb0972011-07-12 09:58:07 +0200603 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000604}
605
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700606
607/*
608 * Remaining API functions
609 */
610
Yinghai Lu95f72d12010-07-12 14:36:09 +1000611/* You must call memblock_analyze() before this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000612phys_addr_t __init memblock_phys_mem_size(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000613{
Benjamin Herrenschmidt4734b592010-07-28 14:31:29 +1000614 return memblock.memory_size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000615}
616
Yinghai Lu10d06432010-07-28 15:43:02 +1000617phys_addr_t __init_memblock memblock_end_of_DRAM(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000618{
619 int idx = memblock.memory.cnt - 1;
620
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000621 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000622}
623
624/* You must call memblock_analyze() after this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000625void __init memblock_enforce_memory_limit(phys_addr_t memory_limit)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000626{
627 unsigned long i;
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000628 phys_addr_t limit;
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000629 struct memblock_region *p;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000630
631 if (!memory_limit)
632 return;
633
634 /* Truncate the memblock regions to satisfy the memory limit. */
635 limit = memory_limit;
636 for (i = 0; i < memblock.memory.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000637 if (limit > memblock.memory.regions[i].size) {
638 limit -= memblock.memory.regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000639 continue;
640 }
641
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000642 memblock.memory.regions[i].size = limit;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000643 memblock.memory.cnt = i + 1;
644 break;
645 }
646
Yinghai Lu95f72d12010-07-12 14:36:09 +1000647 memory_limit = memblock_end_of_DRAM();
648
649 /* And truncate any reserves above the limit also. */
650 for (i = 0; i < memblock.reserved.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000651 p = &memblock.reserved.regions[i];
Yinghai Lu95f72d12010-07-12 14:36:09 +1000652
653 if (p->base > memory_limit)
654 p->size = 0;
655 else if ((p->base + p->size) > memory_limit)
656 p->size = memory_limit - p->base;
657
658 if (p->size == 0) {
659 memblock_remove_region(&memblock.reserved, i);
660 i--;
661 }
662 }
663}
664
Yinghai Lucd794812010-10-11 12:34:09 -0700665static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000666{
667 unsigned int left = 0, right = type->cnt;
668
669 do {
670 unsigned int mid = (right + left) / 2;
671
672 if (addr < type->regions[mid].base)
673 right = mid;
674 else if (addr >= (type->regions[mid].base +
675 type->regions[mid].size))
676 left = mid + 1;
677 else
678 return mid;
679 } while (left < right);
680 return -1;
681}
682
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000683int __init memblock_is_reserved(phys_addr_t addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000684{
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000685 return memblock_search(&memblock.reserved, addr) != -1;
686}
Yinghai Lu95f72d12010-07-12 14:36:09 +1000687
Yinghai Lu3661ca62010-09-15 13:05:29 -0700688int __init_memblock memblock_is_memory(phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000689{
690 return memblock_search(&memblock.memory, addr) != -1;
691}
692
Yinghai Lu3661ca62010-09-15 13:05:29 -0700693int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000694{
Tomi Valkeinenabb65272011-01-20 14:44:20 -0800695 int idx = memblock_search(&memblock.memory, base);
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000696
697 if (idx == -1)
698 return 0;
Tomi Valkeinenabb65272011-01-20 14:44:20 -0800699 return memblock.memory.regions[idx].base <= base &&
700 (memblock.memory.regions[idx].base +
701 memblock.memory.regions[idx].size) >= (base + size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000702}
703
Yinghai Lu10d06432010-07-28 15:43:02 +1000704int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000705{
Benjamin Herrenschmidtf1c2c192010-08-04 14:17:17 +1000706 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000707}
708
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700709
Yinghai Lu3661ca62010-09-15 13:05:29 -0700710void __init_memblock memblock_set_current_limit(phys_addr_t limit)
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700711{
712 memblock.current_limit = limit;
713}
714
Yinghai Lu10d06432010-07-28 15:43:02 +1000715static void __init_memblock memblock_dump(struct memblock_type *region, char *name)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000716{
717 unsigned long long base, size;
718 int i;
719
720 pr_info(" %s.cnt = 0x%lx\n", name, region->cnt);
721
722 for (i = 0; i < region->cnt; i++) {
723 base = region->regions[i].base;
724 size = region->regions[i].size;
725
Yinghai Luea9e4372010-07-28 15:13:22 +1000726 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes\n",
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000727 name, i, base, base + size - 1, size);
728 }
729}
730
Yinghai Lu10d06432010-07-28 15:43:02 +1000731void __init_memblock memblock_dump_all(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000732{
733 if (!memblock_debug)
734 return;
735
736 pr_info("MEMBLOCK configuration:\n");
737 pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size);
738
739 memblock_dump(&memblock.memory, "memory");
740 memblock_dump(&memblock.reserved, "reserved");
741}
742
743void __init memblock_analyze(void)
744{
745 int i;
746
747 /* Check marker in the unused last array entry */
748 WARN_ON(memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS].base
749 != (phys_addr_t)RED_INACTIVE);
750 WARN_ON(memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS].base
751 != (phys_addr_t)RED_INACTIVE);
752
753 memblock.memory_size = 0;
754
755 for (i = 0; i < memblock.memory.cnt; i++)
756 memblock.memory_size += memblock.memory.regions[i].size;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700757
758 /* We allow resizing from there */
759 memblock_can_resize = 1;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000760}
761
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700762void __init memblock_init(void)
763{
Jeremy Fitzhardinge236260b2010-10-06 15:52:29 -0700764 static int init_done __initdata = 0;
765
766 if (init_done)
767 return;
768 init_done = 1;
769
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700770 /* Hookup the initial arrays */
771 memblock.memory.regions = memblock_memory_init_regions;
772 memblock.memory.max = INIT_MEMBLOCK_REGIONS;
773 memblock.reserved.regions = memblock_reserved_init_regions;
774 memblock.reserved.max = INIT_MEMBLOCK_REGIONS;
775
776 /* Write a marker in the unused last array entry */
777 memblock.memory.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
778 memblock.reserved.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
779
780 /* Create a dummy zero size MEMBLOCK which will get coalesced away later.
781 * This simplifies the memblock_add() code below...
782 */
783 memblock.memory.regions[0].base = 0;
784 memblock.memory.regions[0].size = 0;
785 memblock.memory.cnt = 1;
786
787 /* Ditto. */
788 memblock.reserved.regions[0].base = 0;
789 memblock.reserved.regions[0].size = 0;
790 memblock.reserved.cnt = 1;
791
792 memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE;
793}
794
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000795static int __init early_memblock(char *p)
796{
797 if (p && strstr(p, "debug"))
798 memblock_debug = 1;
799 return 0;
800}
801early_param("memblock", early_memblock);
802
Yinghai Lu10d06432010-07-28 15:43:02 +1000803#if defined(CONFIG_DEBUG_FS) && !defined(ARCH_DISCARD_MEMBLOCK)
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -0700804
805static int memblock_debug_show(struct seq_file *m, void *private)
806{
807 struct memblock_type *type = m->private;
808 struct memblock_region *reg;
809 int i;
810
811 for (i = 0; i < type->cnt; i++) {
812 reg = &type->regions[i];
813 seq_printf(m, "%4d: ", i);
814 if (sizeof(phys_addr_t) == 4)
815 seq_printf(m, "0x%08lx..0x%08lx\n",
816 (unsigned long)reg->base,
817 (unsigned long)(reg->base + reg->size - 1));
818 else
819 seq_printf(m, "0x%016llx..0x%016llx\n",
820 (unsigned long long)reg->base,
821 (unsigned long long)(reg->base + reg->size - 1));
822
823 }
824 return 0;
825}
826
827static int memblock_debug_open(struct inode *inode, struct file *file)
828{
829 return single_open(file, memblock_debug_show, inode->i_private);
830}
831
832static const struct file_operations memblock_debug_fops = {
833 .open = memblock_debug_open,
834 .read = seq_read,
835 .llseek = seq_lseek,
836 .release = single_release,
837};
838
839static int __init memblock_init_debugfs(void)
840{
841 struct dentry *root = debugfs_create_dir("memblock", NULL);
842 if (!root)
843 return -ENXIO;
844 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
845 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
846
847 return 0;
848}
849__initcall(memblock_init_debugfs);
850
851#endif /* CONFIG_DEBUG_FS */