blob: 84bec4969ed5d3a27d9c6035da6e39ebf1dfdd18 [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
H Hartley Sweeten2d7d3eb2011-10-31 17:09:15 -070061static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
62 phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100063{
64 unsigned long i;
65
66 for (i = 0; i < type->cnt; i++) {
67 phys_addr_t rgnbase = type->regions[i].base;
68 phys_addr_t rgnsize = type->regions[i].size;
69 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
70 break;
71 }
72
73 return (i < type->cnt) ? i : -1;
74}
75
76/*
77 * Find, allocate, deallocate or reserve unreserved regions. All allocations
78 * are top-down.
79 */
80
Yinghai Lucd794812010-10-11 12:34:09 -070081static phys_addr_t __init_memblock memblock_find_region(phys_addr_t start, phys_addr_t end,
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100082 phys_addr_t size, phys_addr_t align)
83{
84 phys_addr_t base, res_base;
85 long j;
86
Yinghai Luf1af98c2010-10-04 14:57:39 -070087 /* In case, huge size is requested */
88 if (end < size)
89 return MEMBLOCK_ERROR;
90
91 base = memblock_align_down((end - size), align);
92
Benjamin Herrenschmidt25818f02010-07-28 15:25:10 +100093 /* Prevent allocations returning 0 as it's also used to
94 * indicate an allocation failure
95 */
96 if (start == 0)
97 start = PAGE_SIZE;
98
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100099 while (start <= base) {
100 j = memblock_overlaps_region(&memblock.reserved, base, size);
101 if (j < 0)
102 return base;
103 res_base = memblock.reserved.regions[j].base;
104 if (res_base < size)
105 break;
106 base = memblock_align_down(res_base - size, align);
107 }
108
109 return MEMBLOCK_ERROR;
110}
111
Yinghai Lu3661ca62010-09-15 13:05:29 -0700112static phys_addr_t __init_memblock memblock_find_base(phys_addr_t size,
113 phys_addr_t align, phys_addr_t start, phys_addr_t end)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000114{
115 long i;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000116
117 BUG_ON(0 == size);
118
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000119 /* Pump up max_addr */
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000120 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
121 end = memblock.current_limit;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000122
123 /* We do a top-down search, this tends to limit memory
124 * fragmentation by keeping early boot allocs near the
125 * top of memory
126 */
127 for (i = memblock.memory.cnt - 1; i >= 0; i--) {
128 phys_addr_t memblockbase = memblock.memory.regions[i].base;
129 phys_addr_t memblocksize = memblock.memory.regions[i].size;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000130 phys_addr_t bottom, top, found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000131
132 if (memblocksize < size)
133 continue;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000134 if ((memblockbase + memblocksize) <= start)
135 break;
136 bottom = max(memblockbase, start);
137 top = min(memblockbase + memblocksize, end);
138 if (bottom >= top)
139 continue;
140 found = memblock_find_region(bottom, top, size, align);
141 if (found != MEMBLOCK_ERROR)
142 return found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000143 }
144 return MEMBLOCK_ERROR;
145}
146
Yinghai Lu5303b682010-07-28 15:38:40 +1000147/*
148 * Find a free area with specified alignment in a specific range.
149 */
150u64 __init_memblock memblock_find_in_range(u64 start, u64 end, u64 size, u64 align)
151{
152 return memblock_find_base(size, align, start, end);
153}
154
Yinghai Lu7950c402010-08-25 13:39:14 -0700155/*
156 * Free memblock.reserved.regions
157 */
158int __init_memblock memblock_free_reserved_regions(void)
159{
160 if (memblock.reserved.regions == memblock_reserved_init_regions)
161 return 0;
162
163 return memblock_free(__pa(memblock.reserved.regions),
164 sizeof(struct memblock_region) * memblock.reserved.max);
165}
166
167/*
168 * Reserve memblock.reserved.regions
169 */
170int __init_memblock memblock_reserve_reserved_regions(void)
171{
172 if (memblock.reserved.regions == memblock_reserved_init_regions)
173 return 0;
174
175 return memblock_reserve(__pa(memblock.reserved.regions),
176 sizeof(struct memblock_region) * memblock.reserved.max);
177}
178
Yinghai Lu10d06432010-07-28 15:43:02 +1000179static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000180{
181 unsigned long i;
182
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000183 for (i = r; i < type->cnt - 1; i++) {
184 type->regions[i].base = type->regions[i + 1].base;
185 type->regions[i].size = type->regions[i + 1].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000186 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000187 type->cnt--;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000188
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700189 /* Special case for empty arrays */
190 if (type->cnt == 0) {
191 type->cnt = 1;
192 type->regions[0].base = 0;
193 type->regions[0].size = 0;
194 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000195}
196
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700197/* Defined below but needed now */
198static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size);
199
Yinghai Lu10d06432010-07-28 15:43:02 +1000200static int __init_memblock memblock_double_array(struct memblock_type *type)
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700201{
202 struct memblock_region *new_array, *old_array;
203 phys_addr_t old_size, new_size, addr;
204 int use_slab = slab_is_available();
205
206 /* We don't allow resizing until we know about the reserved regions
207 * of memory that aren't suitable for allocation
208 */
209 if (!memblock_can_resize)
210 return -1;
211
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700212 /* Calculate new doubled size */
213 old_size = type->max * sizeof(struct memblock_region);
214 new_size = old_size << 1;
215
216 /* Try to find some space for it.
217 *
218 * WARNING: We assume that either slab_is_available() and we use it or
219 * we use MEMBLOCK for allocations. That means that this is unsafe to use
220 * when bootmem is currently active (unless bootmem itself is implemented
221 * on top of MEMBLOCK which isn't the case yet)
222 *
223 * This should however not be an issue for now, as we currently only
224 * call into MEMBLOCK while it's still active, or much later when slab is
225 * active for memory hotplug operations
226 */
227 if (use_slab) {
228 new_array = kmalloc(new_size, GFP_KERNEL);
229 addr = new_array == NULL ? MEMBLOCK_ERROR : __pa(new_array);
230 } else
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000231 addr = memblock_find_base(new_size, sizeof(phys_addr_t), 0, MEMBLOCK_ALLOC_ACCESSIBLE);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700232 if (addr == MEMBLOCK_ERROR) {
233 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
234 memblock_type_name(type), type->max, type->max * 2);
235 return -1;
236 }
237 new_array = __va(addr);
238
Yinghai Luea9e4372010-07-28 15:13:22 +1000239 memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]",
240 memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1);
241
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700242 /* Found space, we now need to move the array over before
243 * we add the reserved region since it may be our reserved
244 * array itself that is full.
245 */
246 memcpy(new_array, type->regions, old_size);
247 memset(new_array + type->max, 0, old_size);
248 old_array = type->regions;
249 type->regions = new_array;
250 type->max <<= 1;
251
252 /* If we use SLAB that's it, we are done */
253 if (use_slab)
254 return 0;
255
256 /* Add the new reserved region now. Should not fail ! */
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700257 BUG_ON(memblock_add_region(&memblock.reserved, addr, new_size));
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700258
259 /* If the array wasn't our static init one, then free it. We only do
260 * that before SLAB is available as later on, we don't know whether
261 * to use kfree or free_bootmem_pages(). Shouldn't be a big deal
262 * anyways
263 */
264 if (old_array != memblock_memory_init_regions &&
265 old_array != memblock_reserved_init_regions)
266 memblock_free(__pa(old_array), old_size);
267
268 return 0;
269}
270
Jonghwan Choid1f0ece2011-10-31 17:08:42 -0700271int __init_memblock __weak memblock_memory_can_coalesce(phys_addr_t addr1, phys_addr_t size1,
Benjamin Herrenschmidtd2cd5632010-07-06 15:39:14 -0700272 phys_addr_t addr2, phys_addr_t size2)
273{
274 return 1;
275}
276
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700277static long __init_memblock memblock_add_region(struct memblock_type *type,
278 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000279{
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700280 phys_addr_t end = base + size;
281 int i, slot = -1;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000282
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700283 /* First try and coalesce this MEMBLOCK with others */
284 for (i = 0; i < type->cnt; i++) {
285 struct memblock_region *rgn = &type->regions[i];
286 phys_addr_t rend = rgn->base + rgn->size;
287
288 /* Exit if there's no possible hits */
289 if (rgn->base > end || rgn->size == 0)
290 break;
291
292 /* Check if we are fully enclosed within an existing
293 * block
294 */
295 if (rgn->base <= base && rend >= end)
296 return 0;
297
298 /* Check if we overlap or are adjacent with the bottom
299 * of a block.
300 */
301 if (base < rgn->base && end >= rgn->base) {
302 /* If we can't coalesce, create a new block */
303 if (!memblock_memory_can_coalesce(base, size,
304 rgn->base,
305 rgn->size)) {
306 /* Overlap & can't coalesce are mutually
307 * exclusive, if you do that, be prepared
308 * for trouble
309 */
310 WARN_ON(end != rgn->base);
311 goto new_block;
312 }
313 /* We extend the bottom of the block down to our
314 * base
315 */
316 rgn->base = base;
317 rgn->size = rend - base;
318
319 /* Return if we have nothing else to allocate
320 * (fully coalesced)
321 */
322 if (rend >= end)
323 return 0;
324
325 /* We continue processing from the end of the
326 * coalesced block.
327 */
328 base = rend;
329 size = end - base;
330 }
331
332 /* Now check if we overlap or are adjacent with the
333 * top of a block
334 */
335 if (base <= rend && end >= rend) {
336 /* If we can't coalesce, create a new block */
337 if (!memblock_memory_can_coalesce(rgn->base,
338 rgn->size,
339 base, size)) {
340 /* Overlap & can't coalesce are mutually
341 * exclusive, if you do that, be prepared
342 * for trouble
343 */
344 WARN_ON(rend != base);
345 goto new_block;
346 }
347 /* We adjust our base down to enclose the
348 * original block and destroy it. It will be
349 * part of our new allocation. Since we've
350 * freed an entry, we know we won't fail
351 * to allocate one later, so we won't risk
352 * losing the original block allocation.
353 */
354 size += (base - rgn->base);
355 base = rgn->base;
356 memblock_remove_region(type, i--);
357 }
358 }
359
360 /* If the array is empty, special case, replace the fake
361 * filler region and return
362 */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000363 if ((type->cnt == 1) && (type->regions[0].size == 0)) {
364 type->regions[0].base = base;
365 type->regions[0].size = size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000366 return 0;
367 }
368
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700369 new_block:
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700370 /* If we are out of space, we fail. It's too late to resize the array
371 * but then this shouldn't have happened in the first place.
372 */
373 if (WARN_ON(type->cnt >= type->max))
Yinghai Lu95f72d12010-07-12 14:36:09 +1000374 return -1;
375
376 /* Couldn't coalesce the MEMBLOCK, so add it to the sorted table. */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000377 for (i = type->cnt - 1; i >= 0; i--) {
378 if (base < type->regions[i].base) {
379 type->regions[i+1].base = type->regions[i].base;
380 type->regions[i+1].size = type->regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000381 } else {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000382 type->regions[i+1].base = base;
383 type->regions[i+1].size = size;
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700384 slot = i + 1;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000385 break;
386 }
387 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000388 if (base < type->regions[0].base) {
389 type->regions[0].base = base;
390 type->regions[0].size = size;
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700391 slot = 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000392 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000393 type->cnt++;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000394
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700395 /* The array is full ? Try to resize it. If that fails, we undo
396 * our allocation and return an error
397 */
398 if (type->cnt == type->max && memblock_double_array(type)) {
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700399 BUG_ON(slot < 0);
400 memblock_remove_region(type, slot);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700401 return -1;
402 }
403
Yinghai Lu95f72d12010-07-12 14:36:09 +1000404 return 0;
405}
406
Yinghai Lu10d06432010-07-28 15:43:02 +1000407long __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000408{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000409 return memblock_add_region(&memblock.memory, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000410
411}
412
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700413static long __init_memblock __memblock_remove(struct memblock_type *type,
414 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000415{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000416 phys_addr_t end = base + size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000417 int i;
418
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700419 /* Walk through the array for collisions */
420 for (i = 0; i < type->cnt; i++) {
421 struct memblock_region *rgn = &type->regions[i];
422 phys_addr_t rend = rgn->base + rgn->size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000423
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700424 /* Nothing more to do, exit */
425 if (rgn->base > end || rgn->size == 0)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000426 break;
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700427
428 /* If we fully enclose the block, drop it */
429 if (base <= rgn->base && end >= rend) {
430 memblock_remove_region(type, i--);
431 continue;
432 }
433
434 /* If we are fully enclosed within a block
435 * then we need to split it and we are done
436 */
437 if (base > rgn->base && end < rend) {
438 rgn->size = base - rgn->base;
439 if (!memblock_add_region(type, end, rend - end))
440 return 0;
441 /* Failure to split is bad, we at least
442 * restore the block before erroring
443 */
444 rgn->size = rend - rgn->base;
445 WARN_ON(1);
446 return -1;
447 }
448
449 /* Check if we need to trim the bottom of a block */
450 if (rgn->base < end && rend > end) {
451 rgn->size -= end - rgn->base;
452 rgn->base = end;
453 break;
454 }
455
456 /* And check if we need to trim the top of a block */
457 if (base < rend)
458 rgn->size -= rend - base;
459
Yinghai Lu95f72d12010-07-12 14:36:09 +1000460 }
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700461 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000462}
463
Yinghai Lu10d06432010-07-28 15:43:02 +1000464long __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000465{
466 return __memblock_remove(&memblock.memory, base, size);
467}
468
Yinghai Lu3661ca62010-09-15 13:05:29 -0700469long __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000470{
471 return __memblock_remove(&memblock.reserved, base, size);
472}
473
Yinghai Lu3661ca62010-09-15 13:05:29 -0700474long __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000475{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000476 struct memblock_type *_rgn = &memblock.reserved;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000477
478 BUG_ON(0 == size);
479
480 return memblock_add_region(_rgn, base, size);
481}
482
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000483phys_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 +1000484{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000485 phys_addr_t found;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000486
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000487 /* We align the size to limit fragmentation. Without this, a lot of
488 * small allocs quickly eat up the whole reserve array on sparc
489 */
490 size = memblock_align_up(size, align);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000491
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000492 found = memblock_find_base(size, align, 0, max_addr);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000493 if (found != MEMBLOCK_ERROR &&
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700494 !memblock_add_region(&memblock.reserved, found, size))
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000495 return found;
496
497 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000498}
499
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000500phys_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 +1000501{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000502 phys_addr_t alloc;
503
504 alloc = __memblock_alloc_base(size, align, max_addr);
505
506 if (alloc == 0)
507 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
508 (unsigned long long) size, (unsigned long long) max_addr);
509
510 return alloc;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000511}
512
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000513phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000514{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000515 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000516}
517
Yinghai Lu95f72d12010-07-12 14:36:09 +1000518
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000519/*
520 * Additional node-local allocators. Search for node memory is bottom up
521 * and walks memblock regions within that node bottom-up as well, but allocation
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700522 * within an memblock region is top-down. XXX I plan to fix that at some stage
523 *
524 * WARNING: Only available after early_node_map[] has been populated,
525 * on some architectures, that is after all the calls to add_active_range()
526 * have been done to populate it.
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000527 */
Yinghai Lu95f72d12010-07-12 14:36:09 +1000528
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000529phys_addr_t __weak __init memblock_nid_range(phys_addr_t start, phys_addr_t end, int *nid)
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700530{
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700531#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
532 /*
533 * This code originates from sparc which really wants use to walk by addresses
534 * and returns the nid. This is not very convenient for early_pfn_map[] users
535 * as the map isn't sorted yet, and it really wants to be walked by nid.
536 *
537 * For now, I implement the inefficient method below which walks the early
538 * map multiple times. Eventually we may want to use an ARCH config option
539 * to implement a completely different method for both case.
540 */
541 unsigned long start_pfn, end_pfn;
542 int i;
543
544 for (i = 0; i < MAX_NUMNODES; i++) {
545 get_pfn_range_for_nid(i, &start_pfn, &end_pfn);
546 if (start < PFN_PHYS(start_pfn) || start >= PFN_PHYS(end_pfn))
547 continue;
548 *nid = i;
549 return min(end, PFN_PHYS(end_pfn));
550 }
551#endif
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700552 *nid = 0;
553
554 return end;
555}
556
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000557static phys_addr_t __init memblock_alloc_nid_region(struct memblock_region *mp,
558 phys_addr_t size,
559 phys_addr_t align, int nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000560{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000561 phys_addr_t start, end;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000562
563 start = mp->base;
564 end = start + mp->size;
565
566 start = memblock_align_up(start, align);
567 while (start < end) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000568 phys_addr_t this_end;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000569 int this_nid;
570
Benjamin Herrenschmidt35a1f0b2010-07-06 15:38:58 -0700571 this_end = memblock_nid_range(start, end, &this_nid);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000572 if (this_nid == nid) {
Benjamin Herrenschmidt3a9c2c82010-07-12 13:28:15 +1000573 phys_addr_t ret = memblock_find_region(start, this_end, size, align);
Benjamin Herrenschmidt4d629f92010-07-06 15:39:09 -0700574 if (ret != MEMBLOCK_ERROR &&
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700575 !memblock_add_region(&memblock.reserved, ret, size))
Yinghai Lu95f72d12010-07-12 14:36:09 +1000576 return ret;
577 }
578 start = this_end;
579 }
580
Benjamin Herrenschmidt4d629f92010-07-06 15:39:09 -0700581 return MEMBLOCK_ERROR;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000582}
583
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000584phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000585{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000586 struct memblock_type *mem = &memblock.memory;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000587 int i;
588
589 BUG_ON(0 == size);
590
Benjamin Herrenschmidt7f219c72010-07-12 14:24:57 +1000591 /* We align the size to limit fragmentation. Without this, a lot of
592 * small allocs quickly eat up the whole reserve array on sparc
593 */
594 size = memblock_align_up(size, align);
595
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700596 /* We do a bottom-up search for a region with the right
597 * nid since that's easier considering how memblock_nid_range()
598 * works
599 */
Yinghai Lu95f72d12010-07-12 14:36:09 +1000600 for (i = 0; i < mem->cnt; i++) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000601 phys_addr_t ret = memblock_alloc_nid_region(&mem->regions[i],
Yinghai Lu95f72d12010-07-12 14:36:09 +1000602 size, align, nid);
Benjamin Herrenschmidt4d629f92010-07-06 15:39:09 -0700603 if (ret != MEMBLOCK_ERROR)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000604 return ret;
605 }
606
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700607 return 0;
608}
609
610phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
611{
612 phys_addr_t res = memblock_alloc_nid(size, align, nid);
613
614 if (res)
615 return res;
Benjamin Herrenschmidt918fe8d2010-07-06 15:39:18 -0700616 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000617}
618
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700619
620/*
621 * Remaining API functions
622 */
623
Yinghai Lu95f72d12010-07-12 14:36:09 +1000624/* You must call memblock_analyze() before this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000625phys_addr_t __init memblock_phys_mem_size(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000626{
Benjamin Herrenschmidt4734b592010-07-28 14:31:29 +1000627 return memblock.memory_size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000628}
629
Sam Ravnborg0a93ebe2011-10-31 17:08:16 -0700630/* lowest address */
631phys_addr_t __init_memblock memblock_start_of_DRAM(void)
632{
633 return memblock.memory.regions[0].base;
634}
635
Yinghai Lu10d06432010-07-28 15:43:02 +1000636phys_addr_t __init_memblock memblock_end_of_DRAM(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000637{
638 int idx = memblock.memory.cnt - 1;
639
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000640 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000641}
642
643/* You must call memblock_analyze() after this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000644void __init memblock_enforce_memory_limit(phys_addr_t memory_limit)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000645{
646 unsigned long i;
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000647 phys_addr_t limit;
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000648 struct memblock_region *p;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000649
650 if (!memory_limit)
651 return;
652
653 /* Truncate the memblock regions to satisfy the memory limit. */
654 limit = memory_limit;
655 for (i = 0; i < memblock.memory.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000656 if (limit > memblock.memory.regions[i].size) {
657 limit -= memblock.memory.regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000658 continue;
659 }
660
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000661 memblock.memory.regions[i].size = limit;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000662 memblock.memory.cnt = i + 1;
663 break;
664 }
665
Yinghai Lu95f72d12010-07-12 14:36:09 +1000666 memory_limit = memblock_end_of_DRAM();
667
668 /* And truncate any reserves above the limit also. */
669 for (i = 0; i < memblock.reserved.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000670 p = &memblock.reserved.regions[i];
Yinghai Lu95f72d12010-07-12 14:36:09 +1000671
672 if (p->base > memory_limit)
673 p->size = 0;
674 else if ((p->base + p->size) > memory_limit)
675 p->size = memory_limit - p->base;
676
677 if (p->size == 0) {
678 memblock_remove_region(&memblock.reserved, i);
679 i--;
680 }
681 }
682}
683
Yinghai Lucd794812010-10-11 12:34:09 -0700684static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000685{
686 unsigned int left = 0, right = type->cnt;
687
688 do {
689 unsigned int mid = (right + left) / 2;
690
691 if (addr < type->regions[mid].base)
692 right = mid;
693 else if (addr >= (type->regions[mid].base +
694 type->regions[mid].size))
695 left = mid + 1;
696 else
697 return mid;
698 } while (left < right);
699 return -1;
700}
701
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000702int __init memblock_is_reserved(phys_addr_t addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000703{
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000704 return memblock_search(&memblock.reserved, addr) != -1;
705}
Yinghai Lu95f72d12010-07-12 14:36:09 +1000706
Yinghai Lu3661ca62010-09-15 13:05:29 -0700707int __init_memblock memblock_is_memory(phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000708{
709 return memblock_search(&memblock.memory, addr) != -1;
710}
711
Yinghai Lu3661ca62010-09-15 13:05:29 -0700712int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000713{
Tomi Valkeinenabb65272011-01-20 14:44:20 -0800714 int idx = memblock_search(&memblock.memory, base);
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000715
716 if (idx == -1)
717 return 0;
Tomi Valkeinenabb65272011-01-20 14:44:20 -0800718 return memblock.memory.regions[idx].base <= base &&
719 (memblock.memory.regions[idx].base +
720 memblock.memory.regions[idx].size) >= (base + size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000721}
722
Yinghai Lu10d06432010-07-28 15:43:02 +1000723int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000724{
Benjamin Herrenschmidtf1c2c192010-08-04 14:17:17 +1000725 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000726}
727
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700728
Yinghai Lu3661ca62010-09-15 13:05:29 -0700729void __init_memblock memblock_set_current_limit(phys_addr_t limit)
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700730{
731 memblock.current_limit = limit;
732}
733
Yinghai Lu10d06432010-07-28 15:43:02 +1000734static void __init_memblock memblock_dump(struct memblock_type *region, char *name)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000735{
736 unsigned long long base, size;
737 int i;
738
739 pr_info(" %s.cnt = 0x%lx\n", name, region->cnt);
740
741 for (i = 0; i < region->cnt; i++) {
742 base = region->regions[i].base;
743 size = region->regions[i].size;
744
Yinghai Luea9e4372010-07-28 15:13:22 +1000745 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes\n",
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000746 name, i, base, base + size - 1, size);
747 }
748}
749
Yinghai Lu10d06432010-07-28 15:43:02 +1000750void __init_memblock memblock_dump_all(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000751{
752 if (!memblock_debug)
753 return;
754
755 pr_info("MEMBLOCK configuration:\n");
756 pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size);
757
758 memblock_dump(&memblock.memory, "memory");
759 memblock_dump(&memblock.reserved, "reserved");
760}
761
762void __init memblock_analyze(void)
763{
764 int i;
765
766 /* Check marker in the unused last array entry */
767 WARN_ON(memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS].base
Andrew Mortonc9d8c3d2011-07-25 17:12:18 -0700768 != MEMBLOCK_INACTIVE);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000769 WARN_ON(memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS].base
Andrew Mortonc9d8c3d2011-07-25 17:12:18 -0700770 != MEMBLOCK_INACTIVE);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000771
772 memblock.memory_size = 0;
773
774 for (i = 0; i < memblock.memory.cnt; i++)
775 memblock.memory_size += memblock.memory.regions[i].size;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700776
777 /* We allow resizing from there */
778 memblock_can_resize = 1;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000779}
780
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700781void __init memblock_init(void)
782{
Jeremy Fitzhardinge236260b2010-10-06 15:52:29 -0700783 static int init_done __initdata = 0;
784
785 if (init_done)
786 return;
787 init_done = 1;
788
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700789 /* Hookup the initial arrays */
790 memblock.memory.regions = memblock_memory_init_regions;
791 memblock.memory.max = INIT_MEMBLOCK_REGIONS;
792 memblock.reserved.regions = memblock_reserved_init_regions;
793 memblock.reserved.max = INIT_MEMBLOCK_REGIONS;
794
795 /* Write a marker in the unused last array entry */
Andrew Mortonc9d8c3d2011-07-25 17:12:18 -0700796 memblock.memory.regions[INIT_MEMBLOCK_REGIONS].base = MEMBLOCK_INACTIVE;
797 memblock.reserved.regions[INIT_MEMBLOCK_REGIONS].base = MEMBLOCK_INACTIVE;
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700798
799 /* Create a dummy zero size MEMBLOCK which will get coalesced away later.
800 * This simplifies the memblock_add() code below...
801 */
802 memblock.memory.regions[0].base = 0;
803 memblock.memory.regions[0].size = 0;
804 memblock.memory.cnt = 1;
805
806 /* Ditto. */
807 memblock.reserved.regions[0].base = 0;
808 memblock.reserved.regions[0].size = 0;
809 memblock.reserved.cnt = 1;
810
811 memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE;
812}
813
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000814static int __init early_memblock(char *p)
815{
816 if (p && strstr(p, "debug"))
817 memblock_debug = 1;
818 return 0;
819}
820early_param("memblock", early_memblock);
821
Yinghai Lu10d06432010-07-28 15:43:02 +1000822#if defined(CONFIG_DEBUG_FS) && !defined(ARCH_DISCARD_MEMBLOCK)
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -0700823
824static int memblock_debug_show(struct seq_file *m, void *private)
825{
826 struct memblock_type *type = m->private;
827 struct memblock_region *reg;
828 int i;
829
830 for (i = 0; i < type->cnt; i++) {
831 reg = &type->regions[i];
832 seq_printf(m, "%4d: ", i);
833 if (sizeof(phys_addr_t) == 4)
834 seq_printf(m, "0x%08lx..0x%08lx\n",
835 (unsigned long)reg->base,
836 (unsigned long)(reg->base + reg->size - 1));
837 else
838 seq_printf(m, "0x%016llx..0x%016llx\n",
839 (unsigned long long)reg->base,
840 (unsigned long long)(reg->base + reg->size - 1));
841
842 }
843 return 0;
844}
845
846static int memblock_debug_open(struct inode *inode, struct file *file)
847{
848 return single_open(file, memblock_debug_show, inode->i_private);
849}
850
851static const struct file_operations memblock_debug_fops = {
852 .open = memblock_debug_open,
853 .read = seq_read,
854 .llseek = seq_lseek,
855 .release = single_release,
856};
857
858static int __init memblock_init_debugfs(void)
859{
860 struct dentry *root = debugfs_create_dir("memblock", NULL);
861 if (!root)
862 return -ENXIO;
863 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
864 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
865
866 return 0;
867}
868__initcall(memblock_init_debugfs);
869
870#endif /* CONFIG_DEBUG_FS */