blob: a0562d1a6ad426e1b21d43882b1f5f187330edfa [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 */
44
Yinghai Lu10d06432010-07-28 15:43:02 +100045static phys_addr_t __init_memblock memblock_align_down(phys_addr_t addr, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +100046{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100047 return addr & ~(size - 1);
Yinghai Lu95f72d12010-07-12 14:36:09 +100048}
49
Yinghai Lu10d06432010-07-28 15:43:02 +100050static phys_addr_t __init_memblock memblock_align_up(phys_addr_t addr, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +100051{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100052 return (addr + (size - 1)) & ~(size - 1);
Yinghai Lu95f72d12010-07-12 14:36:09 +100053}
54
Yinghai Lu10d06432010-07-28 15:43:02 +100055static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +100056 phys_addr_t base2, phys_addr_t size2)
Yinghai Lu95f72d12010-07-12 14:36:09 +100057{
58 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
59}
60
Yinghai Lu10d06432010-07-28 15:43:02 +100061long __init_memblock memblock_overlaps_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100062{
63 unsigned long i;
64
65 for (i = 0; i < type->cnt; i++) {
66 phys_addr_t rgnbase = type->regions[i].base;
67 phys_addr_t rgnsize = type->regions[i].size;
68 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
69 break;
70 }
71
72 return (i < type->cnt) ? i : -1;
73}
74
75/*
76 * Find, allocate, deallocate or reserve unreserved regions. All allocations
77 * are top-down.
78 */
79
Yinghai Lucd794812010-10-11 12:34:09 -070080static phys_addr_t __init_memblock memblock_find_region(phys_addr_t start, phys_addr_t end,
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100081 phys_addr_t size, phys_addr_t align)
82{
83 phys_addr_t base, res_base;
84 long j;
85
Yinghai Luf1af98c2010-10-04 14:57:39 -070086 /* In case, huge size is requested */
87 if (end < size)
88 return MEMBLOCK_ERROR;
89
90 base = memblock_align_down((end - size), align);
91
Benjamin Herrenschmidt25818f02010-07-28 15:25:10 +100092 /* Prevent allocations returning 0 as it's also used to
93 * indicate an allocation failure
94 */
95 if (start == 0)
96 start = PAGE_SIZE;
97
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100098 while (start <= base) {
99 j = memblock_overlaps_region(&memblock.reserved, base, size);
100 if (j < 0)
101 return base;
102 res_base = memblock.reserved.regions[j].base;
103 if (res_base < size)
104 break;
105 base = memblock_align_down(res_base - size, align);
106 }
107
108 return MEMBLOCK_ERROR;
109}
110
Yinghai Lu3661ca62010-09-15 13:05:29 -0700111static phys_addr_t __init_memblock memblock_find_base(phys_addr_t size,
112 phys_addr_t align, phys_addr_t start, phys_addr_t end)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000113{
114 long i;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000115
116 BUG_ON(0 == size);
117
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000118 /* Pump up max_addr */
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000119 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
120 end = memblock.current_limit;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000121
122 /* We do a top-down search, this tends to limit memory
123 * fragmentation by keeping early boot allocs near the
124 * top of memory
125 */
126 for (i = memblock.memory.cnt - 1; i >= 0; i--) {
127 phys_addr_t memblockbase = memblock.memory.regions[i].base;
128 phys_addr_t memblocksize = memblock.memory.regions[i].size;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000129 phys_addr_t bottom, top, found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000130
131 if (memblocksize < size)
132 continue;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000133 if ((memblockbase + memblocksize) <= start)
134 break;
135 bottom = max(memblockbase, start);
136 top = min(memblockbase + memblocksize, end);
137 if (bottom >= top)
138 continue;
139 found = memblock_find_region(bottom, top, size, align);
140 if (found != MEMBLOCK_ERROR)
141 return found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000142 }
143 return MEMBLOCK_ERROR;
144}
145
Yinghai Lu5303b682010-07-28 15:38:40 +1000146/*
147 * Find a free area with specified alignment in a specific range.
148 */
149u64 __init_memblock memblock_find_in_range(u64 start, u64 end, u64 size, u64 align)
150{
151 return memblock_find_base(size, align, start, end);
152}
153
Yinghai Lu7950c402010-08-25 13:39:14 -0700154/*
155 * Free memblock.reserved.regions
156 */
157int __init_memblock memblock_free_reserved_regions(void)
158{
159 if (memblock.reserved.regions == memblock_reserved_init_regions)
160 return 0;
161
162 return memblock_free(__pa(memblock.reserved.regions),
163 sizeof(struct memblock_region) * memblock.reserved.max);
164}
165
166/*
167 * Reserve memblock.reserved.regions
168 */
169int __init_memblock memblock_reserve_reserved_regions(void)
170{
171 if (memblock.reserved.regions == memblock_reserved_init_regions)
172 return 0;
173
174 return memblock_reserve(__pa(memblock.reserved.regions),
175 sizeof(struct memblock_region) * memblock.reserved.max);
176}
177
Yinghai Lu10d06432010-07-28 15:43:02 +1000178static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000179{
180 unsigned long i;
181
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000182 for (i = r; i < type->cnt - 1; i++) {
183 type->regions[i].base = type->regions[i + 1].base;
184 type->regions[i].size = type->regions[i + 1].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000185 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000186 type->cnt--;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000187
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700188 /* Special case for empty arrays */
189 if (type->cnt == 0) {
190 type->cnt = 1;
191 type->regions[0].base = 0;
192 type->regions[0].size = 0;
193 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000194}
195
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700196/* Defined below but needed now */
197static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size);
198
Yinghai Lu10d06432010-07-28 15:43:02 +1000199static int __init_memblock memblock_double_array(struct memblock_type *type)
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700200{
201 struct memblock_region *new_array, *old_array;
202 phys_addr_t old_size, new_size, addr;
203 int use_slab = slab_is_available();
204
205 /* We don't allow resizing until we know about the reserved regions
206 * of memory that aren't suitable for allocation
207 */
208 if (!memblock_can_resize)
209 return -1;
210
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700211 /* Calculate new doubled size */
212 old_size = type->max * sizeof(struct memblock_region);
213 new_size = old_size << 1;
214
215 /* Try to find some space for it.
216 *
217 * WARNING: We assume that either slab_is_available() and we use it or
218 * we use MEMBLOCK for allocations. That means that this is unsafe to use
219 * when bootmem is currently active (unless bootmem itself is implemented
220 * on top of MEMBLOCK which isn't the case yet)
221 *
222 * This should however not be an issue for now, as we currently only
223 * call into MEMBLOCK while it's still active, or much later when slab is
224 * active for memory hotplug operations
225 */
226 if (use_slab) {
227 new_array = kmalloc(new_size, GFP_KERNEL);
228 addr = new_array == NULL ? MEMBLOCK_ERROR : __pa(new_array);
229 } else
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000230 addr = memblock_find_base(new_size, sizeof(phys_addr_t), 0, MEMBLOCK_ALLOC_ACCESSIBLE);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700231 if (addr == MEMBLOCK_ERROR) {
232 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
233 memblock_type_name(type), type->max, type->max * 2);
234 return -1;
235 }
236 new_array = __va(addr);
237
Yinghai Luea9e4372010-07-28 15:13:22 +1000238 memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]",
239 memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1);
240
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700241 /* Found space, we now need to move the array over before
242 * we add the reserved region since it may be our reserved
243 * array itself that is full.
244 */
245 memcpy(new_array, type->regions, old_size);
246 memset(new_array + type->max, 0, old_size);
247 old_array = type->regions;
248 type->regions = new_array;
249 type->max <<= 1;
250
251 /* If we use SLAB that's it, we are done */
252 if (use_slab)
253 return 0;
254
255 /* Add the new reserved region now. Should not fail ! */
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700256 BUG_ON(memblock_add_region(&memblock.reserved, addr, new_size));
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700257
258 /* If the array wasn't our static init one, then free it. We only do
259 * that before SLAB is available as later on, we don't know whether
260 * to use kfree or free_bootmem_pages(). Shouldn't be a big deal
261 * anyways
262 */
263 if (old_array != memblock_memory_init_regions &&
264 old_array != memblock_reserved_init_regions)
265 memblock_free(__pa(old_array), old_size);
266
267 return 0;
268}
269
Yinghai Lu10d06432010-07-28 15:43:02 +1000270extern int __init_memblock __weak memblock_memory_can_coalesce(phys_addr_t addr1, phys_addr_t size1,
Benjamin Herrenschmidtd2cd5632010-07-06 15:39:14 -0700271 phys_addr_t addr2, phys_addr_t size2)
272{
273 return 1;
274}
275
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700276static long __init_memblock memblock_add_region(struct memblock_type *type,
277 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000278{
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700279 phys_addr_t end = base + size;
280 int i, slot = -1;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000281
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700282 /* First try and coalesce this MEMBLOCK with others */
283 for (i = 0; i < type->cnt; i++) {
284 struct memblock_region *rgn = &type->regions[i];
285 phys_addr_t rend = rgn->base + rgn->size;
286
287 /* Exit if there's no possible hits */
288 if (rgn->base > end || rgn->size == 0)
289 break;
290
291 /* Check if we are fully enclosed within an existing
292 * block
293 */
294 if (rgn->base <= base && rend >= end)
295 return 0;
296
297 /* Check if we overlap or are adjacent with the bottom
298 * of a block.
299 */
300 if (base < rgn->base && end >= rgn->base) {
301 /* If we can't coalesce, create a new block */
302 if (!memblock_memory_can_coalesce(base, size,
303 rgn->base,
304 rgn->size)) {
305 /* Overlap & can't coalesce are mutually
306 * exclusive, if you do that, be prepared
307 * for trouble
308 */
309 WARN_ON(end != rgn->base);
310 goto new_block;
311 }
312 /* We extend the bottom of the block down to our
313 * base
314 */
315 rgn->base = base;
316 rgn->size = rend - base;
317
318 /* Return if we have nothing else to allocate
319 * (fully coalesced)
320 */
321 if (rend >= end)
322 return 0;
323
324 /* We continue processing from the end of the
325 * coalesced block.
326 */
327 base = rend;
328 size = end - base;
329 }
330
331 /* Now check if we overlap or are adjacent with the
332 * top of a block
333 */
334 if (base <= rend && end >= rend) {
335 /* If we can't coalesce, create a new block */
336 if (!memblock_memory_can_coalesce(rgn->base,
337 rgn->size,
338 base, size)) {
339 /* Overlap & can't coalesce are mutually
340 * exclusive, if you do that, be prepared
341 * for trouble
342 */
343 WARN_ON(rend != base);
344 goto new_block;
345 }
346 /* We adjust our base down to enclose the
347 * original block and destroy it. It will be
348 * part of our new allocation. Since we've
349 * freed an entry, we know we won't fail
350 * to allocate one later, so we won't risk
351 * losing the original block allocation.
352 */
353 size += (base - rgn->base);
354 base = rgn->base;
355 memblock_remove_region(type, i--);
356 }
357 }
358
359 /* If the array is empty, special case, replace the fake
360 * filler region and return
361 */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000362 if ((type->cnt == 1) && (type->regions[0].size == 0)) {
363 type->regions[0].base = base;
364 type->regions[0].size = size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000365 return 0;
366 }
367
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700368 new_block:
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700369 /* If we are out of space, we fail. It's too late to resize the array
370 * but then this shouldn't have happened in the first place.
371 */
372 if (WARN_ON(type->cnt >= type->max))
Yinghai Lu95f72d12010-07-12 14:36:09 +1000373 return -1;
374
375 /* Couldn't coalesce the MEMBLOCK, so add it to the sorted table. */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000376 for (i = type->cnt - 1; i >= 0; i--) {
377 if (base < type->regions[i].base) {
378 type->regions[i+1].base = type->regions[i].base;
379 type->regions[i+1].size = type->regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000380 } else {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000381 type->regions[i+1].base = base;
382 type->regions[i+1].size = size;
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700383 slot = i + 1;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000384 break;
385 }
386 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000387 if (base < type->regions[0].base) {
388 type->regions[0].base = base;
389 type->regions[0].size = size;
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700390 slot = 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000391 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000392 type->cnt++;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000393
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700394 /* The array is full ? Try to resize it. If that fails, we undo
395 * our allocation and return an error
396 */
397 if (type->cnt == type->max && memblock_double_array(type)) {
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700398 BUG_ON(slot < 0);
399 memblock_remove_region(type, slot);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700400 return -1;
401 }
402
Yinghai Lu95f72d12010-07-12 14:36:09 +1000403 return 0;
404}
405
Yinghai Lu10d06432010-07-28 15:43:02 +1000406long __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000407{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000408 return memblock_add_region(&memblock.memory, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000409
410}
411
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700412static long __init_memblock __memblock_remove(struct memblock_type *type,
413 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000414{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000415 phys_addr_t end = base + size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000416 int i;
417
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700418 /* Walk through the array for collisions */
419 for (i = 0; i < type->cnt; i++) {
420 struct memblock_region *rgn = &type->regions[i];
421 phys_addr_t rend = rgn->base + rgn->size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000422
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700423 /* Nothing more to do, exit */
424 if (rgn->base > end || rgn->size == 0)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000425 break;
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700426
427 /* If we fully enclose the block, drop it */
428 if (base <= rgn->base && end >= rend) {
429 memblock_remove_region(type, i--);
430 continue;
431 }
432
433 /* If we are fully enclosed within a block
434 * then we need to split it and we are done
435 */
436 if (base > rgn->base && end < rend) {
437 rgn->size = base - rgn->base;
438 if (!memblock_add_region(type, end, rend - end))
439 return 0;
440 /* Failure to split is bad, we at least
441 * restore the block before erroring
442 */
443 rgn->size = rend - rgn->base;
444 WARN_ON(1);
445 return -1;
446 }
447
448 /* Check if we need to trim the bottom of a block */
449 if (rgn->base < end && rend > end) {
450 rgn->size -= end - rgn->base;
451 rgn->base = end;
452 break;
453 }
454
455 /* And check if we need to trim the top of a block */
456 if (base < rend)
457 rgn->size -= rend - base;
458
Yinghai Lu95f72d12010-07-12 14:36:09 +1000459 }
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700460 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000461}
462
Yinghai Lu10d06432010-07-28 15:43:02 +1000463long __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000464{
465 return __memblock_remove(&memblock.memory, base, size);
466}
467
Yinghai Lu3661ca62010-09-15 13:05:29 -0700468long __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000469{
470 return __memblock_remove(&memblock.reserved, base, size);
471}
472
Yinghai Lu3661ca62010-09-15 13:05:29 -0700473long __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000474{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000475 struct memblock_type *_rgn = &memblock.reserved;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000476
477 BUG_ON(0 == size);
478
479 return memblock_add_region(_rgn, base, size);
480}
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 found;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000485
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000486 /* We align the size to limit fragmentation. Without this, a lot of
487 * small allocs quickly eat up the whole reserve array on sparc
488 */
489 size = memblock_align_up(size, align);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000490
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000491 found = memblock_find_base(size, align, 0, max_addr);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000492 if (found != MEMBLOCK_ERROR &&
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700493 !memblock_add_region(&memblock.reserved, found, size))
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000494 return found;
495
496 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000497}
498
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000499phys_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 +1000500{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000501 phys_addr_t alloc;
502
503 alloc = __memblock_alloc_base(size, align, max_addr);
504
505 if (alloc == 0)
506 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
507 (unsigned long long) size, (unsigned long long) max_addr);
508
509 return alloc;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000510}
511
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000512phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000513{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000514 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000515}
516
Yinghai Lu95f72d12010-07-12 14:36:09 +1000517
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000518/*
519 * Additional node-local allocators. Search for node memory is bottom up
520 * and walks memblock regions within that node bottom-up as well, but allocation
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700521 * within an memblock region is top-down. XXX I plan to fix that at some stage
522 *
523 * WARNING: Only available after early_node_map[] has been populated,
524 * on some architectures, that is after all the calls to add_active_range()
525 * have been done to populate it.
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000526 */
Yinghai Lu95f72d12010-07-12 14:36:09 +1000527
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000528phys_addr_t __weak __init memblock_nid_range(phys_addr_t start, phys_addr_t end, int *nid)
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700529{
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700530#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
531 /*
532 * This code originates from sparc which really wants use to walk by addresses
533 * and returns the nid. This is not very convenient for early_pfn_map[] users
534 * as the map isn't sorted yet, and it really wants to be walked by nid.
535 *
536 * For now, I implement the inefficient method below which walks the early
537 * map multiple times. Eventually we may want to use an ARCH config option
538 * to implement a completely different method for both case.
539 */
540 unsigned long start_pfn, end_pfn;
541 int i;
542
543 for (i = 0; i < MAX_NUMNODES; i++) {
544 get_pfn_range_for_nid(i, &start_pfn, &end_pfn);
545 if (start < PFN_PHYS(start_pfn) || start >= PFN_PHYS(end_pfn))
546 continue;
547 *nid = i;
548 return min(end, PFN_PHYS(end_pfn));
549 }
550#endif
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700551 *nid = 0;
552
553 return end;
554}
555
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000556static phys_addr_t __init memblock_alloc_nid_region(struct memblock_region *mp,
557 phys_addr_t size,
558 phys_addr_t align, int nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000559{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000560 phys_addr_t start, end;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000561
562 start = mp->base;
563 end = start + mp->size;
564
565 start = memblock_align_up(start, align);
566 while (start < end) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000567 phys_addr_t this_end;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000568 int this_nid;
569
Benjamin Herrenschmidt35a1f0b2010-07-06 15:38:58 -0700570 this_end = memblock_nid_range(start, end, &this_nid);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000571 if (this_nid == nid) {
Benjamin Herrenschmidt3a9c2c82010-07-12 13:28:15 +1000572 phys_addr_t ret = memblock_find_region(start, this_end, size, align);
Benjamin Herrenschmidt4d629f92010-07-06 15:39:09 -0700573 if (ret != MEMBLOCK_ERROR &&
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700574 !memblock_add_region(&memblock.reserved, ret, size))
Yinghai Lu95f72d12010-07-12 14:36:09 +1000575 return ret;
576 }
577 start = this_end;
578 }
579
Benjamin Herrenschmidt4d629f92010-07-06 15:39:09 -0700580 return MEMBLOCK_ERROR;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000581}
582
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000583phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000584{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000585 struct memblock_type *mem = &memblock.memory;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000586 int i;
587
588 BUG_ON(0 == size);
589
Benjamin Herrenschmidt7f219c72010-07-12 14:24:57 +1000590 /* We align the size to limit fragmentation. Without this, a lot of
591 * small allocs quickly eat up the whole reserve array on sparc
592 */
593 size = memblock_align_up(size, align);
594
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700595 /* We do a bottom-up search for a region with the right
596 * nid since that's easier considering how memblock_nid_range()
597 * works
598 */
Yinghai Lu95f72d12010-07-12 14:36:09 +1000599 for (i = 0; i < mem->cnt; i++) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000600 phys_addr_t ret = memblock_alloc_nid_region(&mem->regions[i],
Yinghai Lu95f72d12010-07-12 14:36:09 +1000601 size, align, nid);
Benjamin Herrenschmidt4d629f92010-07-06 15:39:09 -0700602 if (ret != MEMBLOCK_ERROR)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000603 return ret;
604 }
605
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700606 return 0;
607}
608
609phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
610{
611 phys_addr_t res = memblock_alloc_nid(size, align, nid);
612
613 if (res)
614 return res;
Benjamin Herrenschmidt918fe8d2010-07-06 15:39:18 -0700615 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000616}
617
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700618
619/*
620 * Remaining API functions
621 */
622
Yinghai Lu95f72d12010-07-12 14:36:09 +1000623/* You must call memblock_analyze() before this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000624phys_addr_t __init memblock_phys_mem_size(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000625{
Benjamin Herrenschmidt4734b592010-07-28 14:31:29 +1000626 return memblock.memory_size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000627}
628
Yinghai Lu10d06432010-07-28 15:43:02 +1000629phys_addr_t __init_memblock memblock_end_of_DRAM(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000630{
631 int idx = memblock.memory.cnt - 1;
632
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000633 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000634}
635
636/* You must call memblock_analyze() after this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000637void __init memblock_enforce_memory_limit(phys_addr_t memory_limit)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000638{
639 unsigned long i;
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000640 phys_addr_t limit;
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000641 struct memblock_region *p;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000642
643 if (!memory_limit)
644 return;
645
646 /* Truncate the memblock regions to satisfy the memory limit. */
647 limit = memory_limit;
648 for (i = 0; i < memblock.memory.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000649 if (limit > memblock.memory.regions[i].size) {
650 limit -= memblock.memory.regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000651 continue;
652 }
653
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000654 memblock.memory.regions[i].size = limit;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000655 memblock.memory.cnt = i + 1;
656 break;
657 }
658
Yinghai Lu95f72d12010-07-12 14:36:09 +1000659 memory_limit = memblock_end_of_DRAM();
660
661 /* And truncate any reserves above the limit also. */
662 for (i = 0; i < memblock.reserved.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000663 p = &memblock.reserved.regions[i];
Yinghai Lu95f72d12010-07-12 14:36:09 +1000664
665 if (p->base > memory_limit)
666 p->size = 0;
667 else if ((p->base + p->size) > memory_limit)
668 p->size = memory_limit - p->base;
669
670 if (p->size == 0) {
671 memblock_remove_region(&memblock.reserved, i);
672 i--;
673 }
674 }
675}
676
Yinghai Lucd794812010-10-11 12:34:09 -0700677static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000678{
679 unsigned int left = 0, right = type->cnt;
680
681 do {
682 unsigned int mid = (right + left) / 2;
683
684 if (addr < type->regions[mid].base)
685 right = mid;
686 else if (addr >= (type->regions[mid].base +
687 type->regions[mid].size))
688 left = mid + 1;
689 else
690 return mid;
691 } while (left < right);
692 return -1;
693}
694
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000695int __init memblock_is_reserved(phys_addr_t addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000696{
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000697 return memblock_search(&memblock.reserved, addr) != -1;
698}
Yinghai Lu95f72d12010-07-12 14:36:09 +1000699
Yinghai Lu3661ca62010-09-15 13:05:29 -0700700int __init_memblock memblock_is_memory(phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000701{
702 return memblock_search(&memblock.memory, addr) != -1;
703}
704
Yinghai Lu3661ca62010-09-15 13:05:29 -0700705int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000706{
Tomi Valkeinenabb65272011-01-20 14:44:20 -0800707 int idx = memblock_search(&memblock.memory, base);
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000708
709 if (idx == -1)
710 return 0;
Tomi Valkeinenabb65272011-01-20 14:44:20 -0800711 return memblock.memory.regions[idx].base <= base &&
712 (memblock.memory.regions[idx].base +
713 memblock.memory.regions[idx].size) >= (base + size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000714}
715
Yinghai Lu10d06432010-07-28 15:43:02 +1000716int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000717{
Benjamin Herrenschmidtf1c2c192010-08-04 14:17:17 +1000718 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000719}
720
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700721
Yinghai Lu3661ca62010-09-15 13:05:29 -0700722void __init_memblock memblock_set_current_limit(phys_addr_t limit)
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700723{
724 memblock.current_limit = limit;
725}
726
Yinghai Lu10d06432010-07-28 15:43:02 +1000727static void __init_memblock memblock_dump(struct memblock_type *region, char *name)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000728{
729 unsigned long long base, size;
730 int i;
731
732 pr_info(" %s.cnt = 0x%lx\n", name, region->cnt);
733
734 for (i = 0; i < region->cnt; i++) {
735 base = region->regions[i].base;
736 size = region->regions[i].size;
737
Yinghai Luea9e4372010-07-28 15:13:22 +1000738 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes\n",
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000739 name, i, base, base + size - 1, size);
740 }
741}
742
Yinghai Lu10d06432010-07-28 15:43:02 +1000743void __init_memblock memblock_dump_all(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000744{
745 if (!memblock_debug)
746 return;
747
748 pr_info("MEMBLOCK configuration:\n");
749 pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size);
750
751 memblock_dump(&memblock.memory, "memory");
752 memblock_dump(&memblock.reserved, "reserved");
753}
754
755void __init memblock_analyze(void)
756{
757 int i;
758
759 /* Check marker in the unused last array entry */
760 WARN_ON(memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS].base
761 != (phys_addr_t)RED_INACTIVE);
762 WARN_ON(memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS].base
763 != (phys_addr_t)RED_INACTIVE);
764
765 memblock.memory_size = 0;
766
767 for (i = 0; i < memblock.memory.cnt; i++)
768 memblock.memory_size += memblock.memory.regions[i].size;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700769
770 /* We allow resizing from there */
771 memblock_can_resize = 1;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000772}
773
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700774void __init memblock_init(void)
775{
Jeremy Fitzhardinge236260b2010-10-06 15:52:29 -0700776 static int init_done __initdata = 0;
777
778 if (init_done)
779 return;
780 init_done = 1;
781
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700782 /* Hookup the initial arrays */
783 memblock.memory.regions = memblock_memory_init_regions;
784 memblock.memory.max = INIT_MEMBLOCK_REGIONS;
785 memblock.reserved.regions = memblock_reserved_init_regions;
786 memblock.reserved.max = INIT_MEMBLOCK_REGIONS;
787
788 /* Write a marker in the unused last array entry */
789 memblock.memory.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
790 memblock.reserved.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
791
792 /* Create a dummy zero size MEMBLOCK which will get coalesced away later.
793 * This simplifies the memblock_add() code below...
794 */
795 memblock.memory.regions[0].base = 0;
796 memblock.memory.regions[0].size = 0;
797 memblock.memory.cnt = 1;
798
799 /* Ditto. */
800 memblock.reserved.regions[0].base = 0;
801 memblock.reserved.regions[0].size = 0;
802 memblock.reserved.cnt = 1;
803
804 memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE;
805}
806
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000807static int __init early_memblock(char *p)
808{
809 if (p && strstr(p, "debug"))
810 memblock_debug = 1;
811 return 0;
812}
813early_param("memblock", early_memblock);
814
Yinghai Lu10d06432010-07-28 15:43:02 +1000815#if defined(CONFIG_DEBUG_FS) && !defined(ARCH_DISCARD_MEMBLOCK)
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -0700816
817static int memblock_debug_show(struct seq_file *m, void *private)
818{
819 struct memblock_type *type = m->private;
820 struct memblock_region *reg;
821 int i;
822
823 for (i = 0; i < type->cnt; i++) {
824 reg = &type->regions[i];
825 seq_printf(m, "%4d: ", i);
826 if (sizeof(phys_addr_t) == 4)
827 seq_printf(m, "0x%08lx..0x%08lx\n",
828 (unsigned long)reg->base,
829 (unsigned long)(reg->base + reg->size - 1));
830 else
831 seq_printf(m, "0x%016llx..0x%016llx\n",
832 (unsigned long long)reg->base,
833 (unsigned long long)(reg->base + reg->size - 1));
834
835 }
836 return 0;
837}
838
839static int memblock_debug_open(struct inode *inode, struct file *file)
840{
841 return single_open(file, memblock_debug_show, inode->i_private);
842}
843
844static const struct file_operations memblock_debug_fops = {
845 .open = memblock_debug_open,
846 .read = seq_read,
847 .llseek = seq_lseek,
848 .release = single_release,
849};
850
851static int __init memblock_init_debugfs(void)
852{
853 struct dentry *root = debugfs_create_dir("memblock", NULL);
854 if (!root)
855 return -ENXIO;
856 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
857 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
858
859 return 0;
860}
861__initcall(memblock_init_debugfs);
862
863#endif /* CONFIG_DEBUG_FS */