blob: e815f4b758099cc37c42bb0a95ac401e1d3d75f4 [file] [log] [blame]
Yinghai Lu95f72d12010-07-12 14:36:09 +10001/*
2 * Procedures for maintaining information about logical memory blocks.
3 *
4 * Peter Bergner, IBM Corp. June 2001.
5 * Copyright (C) 2001 Peter Bergner.
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 */
12
13#include <linux/kernel.h>
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -070014#include <linux/slab.h>
Yinghai Lu95f72d12010-07-12 14:36:09 +100015#include <linux/init.h>
16#include <linux/bitops.h>
Benjamin Herrenschmidt449e8df2010-07-06 15:39:07 -070017#include <linux/poison.h>
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -070018#include <linux/pfn.h>
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -070019#include <linux/debugfs.h>
20#include <linux/seq_file.h>
Yinghai Lu95f72d12010-07-12 14:36:09 +100021#include <linux/memblock.h>
22
Yinghai Lu10d06432010-07-28 15:43:02 +100023struct memblock memblock __initdata_memblock;
Yinghai Lu95f72d12010-07-12 14:36:09 +100024
Yinghai Lu10d06432010-07-28 15:43:02 +100025int memblock_debug __initdata_memblock;
26int memblock_can_resize __initdata_memblock;
27static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS + 1] __initdata_memblock;
28static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS + 1] __initdata_memblock;
Yinghai Lu95f72d12010-07-12 14:36:09 +100029
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -070030/* inline so we don't get a warning when pr_debug is compiled out */
31static inline const char *memblock_type_name(struct memblock_type *type)
32{
33 if (type == &memblock.memory)
34 return "memory";
35 else if (type == &memblock.reserved)
36 return "reserved";
37 else
38 return "unknown";
39}
40
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100041/*
42 * Address comparison utilities
43 */
Yinghai Lu10d06432010-07-28 15:43:02 +100044static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +100045 phys_addr_t base2, phys_addr_t size2)
Yinghai Lu95f72d12010-07-12 14:36:09 +100046{
47 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
48}
49
Yinghai Lu10d06432010-07-28 15:43:02 +100050long __init_memblock memblock_overlaps_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100051{
52 unsigned long i;
53
54 for (i = 0; i < type->cnt; i++) {
55 phys_addr_t rgnbase = type->regions[i].base;
56 phys_addr_t rgnsize = type->regions[i].size;
57 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
58 break;
59 }
60
61 return (i < type->cnt) ? i : -1;
62}
63
64/*
65 * Find, allocate, deallocate or reserve unreserved regions. All allocations
66 * are top-down.
67 */
68
Yinghai Lucd794812010-10-11 12:34:09 -070069static phys_addr_t __init_memblock memblock_find_region(phys_addr_t start, phys_addr_t end,
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100070 phys_addr_t size, phys_addr_t align)
71{
72 phys_addr_t base, res_base;
73 long j;
74
Yinghai Luf1af98c2010-10-04 14:57:39 -070075 /* In case, huge size is requested */
76 if (end < size)
Tejun Heo1f5026a2011-07-12 09:58:09 +020077 return 0;
Yinghai Luf1af98c2010-10-04 14:57:39 -070078
Tejun Heo348968e2011-07-12 09:58:08 +020079 base = round_down(end - size, align);
Yinghai Luf1af98c2010-10-04 14:57:39 -070080
Benjamin Herrenschmidt25818f02010-07-28 15:25:10 +100081 /* Prevent allocations returning 0 as it's also used to
82 * indicate an allocation failure
83 */
84 if (start == 0)
85 start = PAGE_SIZE;
86
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100087 while (start <= base) {
88 j = memblock_overlaps_region(&memblock.reserved, base, size);
89 if (j < 0)
90 return base;
91 res_base = memblock.reserved.regions[j].base;
92 if (res_base < size)
93 break;
Tejun Heo348968e2011-07-12 09:58:08 +020094 base = round_down(res_base - size, align);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100095 }
96
Tejun Heo1f5026a2011-07-12 09:58:09 +020097 return 0;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100098}
99
Tejun Heofc769a82011-07-12 09:58:10 +0200100/*
101 * Find a free area with specified alignment in a specific range.
102 */
103phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, phys_addr_t end,
104 phys_addr_t size, phys_addr_t align)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000105{
106 long i;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000107
108 BUG_ON(0 == size);
109
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000110 /* Pump up max_addr */
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000111 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
112 end = memblock.current_limit;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000113
114 /* We do a top-down search, this tends to limit memory
115 * fragmentation by keeping early boot allocs near the
116 * top of memory
117 */
118 for (i = memblock.memory.cnt - 1; i >= 0; i--) {
119 phys_addr_t memblockbase = memblock.memory.regions[i].base;
120 phys_addr_t memblocksize = memblock.memory.regions[i].size;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000121 phys_addr_t bottom, top, found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000122
123 if (memblocksize < size)
124 continue;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000125 if ((memblockbase + memblocksize) <= start)
126 break;
127 bottom = max(memblockbase, start);
128 top = min(memblockbase + memblocksize, end);
129 if (bottom >= top)
130 continue;
131 found = memblock_find_region(bottom, top, size, align);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200132 if (found)
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000133 return found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000134 }
Tejun Heo1f5026a2011-07-12 09:58:09 +0200135 return 0;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000136}
137
Yinghai Lu5303b682010-07-28 15:38:40 +1000138/*
Yinghai Lu7950c402010-08-25 13:39:14 -0700139 * Free memblock.reserved.regions
140 */
141int __init_memblock memblock_free_reserved_regions(void)
142{
143 if (memblock.reserved.regions == memblock_reserved_init_regions)
144 return 0;
145
146 return memblock_free(__pa(memblock.reserved.regions),
147 sizeof(struct memblock_region) * memblock.reserved.max);
148}
149
150/*
151 * Reserve memblock.reserved.regions
152 */
153int __init_memblock memblock_reserve_reserved_regions(void)
154{
155 if (memblock.reserved.regions == memblock_reserved_init_regions)
156 return 0;
157
158 return memblock_reserve(__pa(memblock.reserved.regions),
159 sizeof(struct memblock_region) * memblock.reserved.max);
160}
161
Yinghai Lu10d06432010-07-28 15:43:02 +1000162static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000163{
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200164 memmove(&type->regions[r], &type->regions[r + 1],
165 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000166 type->cnt--;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000167
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700168 /* Special case for empty arrays */
169 if (type->cnt == 0) {
170 type->cnt = 1;
171 type->regions[0].base = 0;
172 type->regions[0].size = 0;
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200173 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700174 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000175}
176
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700177/* Defined below but needed now */
178static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size);
179
Yinghai Lu10d06432010-07-28 15:43:02 +1000180static int __init_memblock memblock_double_array(struct memblock_type *type)
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700181{
182 struct memblock_region *new_array, *old_array;
183 phys_addr_t old_size, new_size, addr;
184 int use_slab = slab_is_available();
185
186 /* We don't allow resizing until we know about the reserved regions
187 * of memory that aren't suitable for allocation
188 */
189 if (!memblock_can_resize)
190 return -1;
191
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700192 /* Calculate new doubled size */
193 old_size = type->max * sizeof(struct memblock_region);
194 new_size = old_size << 1;
195
196 /* Try to find some space for it.
197 *
198 * WARNING: We assume that either slab_is_available() and we use it or
199 * we use MEMBLOCK for allocations. That means that this is unsafe to use
200 * when bootmem is currently active (unless bootmem itself is implemented
201 * on top of MEMBLOCK which isn't the case yet)
202 *
203 * This should however not be an issue for now, as we currently only
204 * call into MEMBLOCK while it's still active, or much later when slab is
205 * active for memory hotplug operations
206 */
207 if (use_slab) {
208 new_array = kmalloc(new_size, GFP_KERNEL);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200209 addr = new_array ? __pa(new_array) : 0;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700210 } else
Tejun Heofc769a82011-07-12 09:58:10 +0200211 addr = memblock_find_in_range(0, MEMBLOCK_ALLOC_ACCESSIBLE, new_size, sizeof(phys_addr_t));
Tejun Heo1f5026a2011-07-12 09:58:09 +0200212 if (!addr) {
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700213 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
214 memblock_type_name(type), type->max, type->max * 2);
215 return -1;
216 }
217 new_array = __va(addr);
218
Yinghai Luea9e4372010-07-28 15:13:22 +1000219 memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]",
220 memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1);
221
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700222 /* Found space, we now need to move the array over before
223 * we add the reserved region since it may be our reserved
224 * array itself that is full.
225 */
226 memcpy(new_array, type->regions, old_size);
227 memset(new_array + type->max, 0, old_size);
228 old_array = type->regions;
229 type->regions = new_array;
230 type->max <<= 1;
231
232 /* If we use SLAB that's it, we are done */
233 if (use_slab)
234 return 0;
235
236 /* Add the new reserved region now. Should not fail ! */
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700237 BUG_ON(memblock_add_region(&memblock.reserved, addr, new_size));
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700238
239 /* If the array wasn't our static init one, then free it. We only do
240 * that before SLAB is available as later on, we don't know whether
241 * to use kfree or free_bootmem_pages(). Shouldn't be a big deal
242 * anyways
243 */
244 if (old_array != memblock_memory_init_regions &&
245 old_array != memblock_reserved_init_regions)
246 memblock_free(__pa(old_array), old_size);
247
248 return 0;
249}
250
Tejun Heo784656f2011-07-12 11:15:55 +0200251/**
252 * memblock_merge_regions - merge neighboring compatible regions
253 * @type: memblock type to scan
254 *
255 * Scan @type and merge neighboring compatible regions.
256 */
257static void __init_memblock memblock_merge_regions(struct memblock_type *type)
258{
259 int i = 0;
260
261 /* cnt never goes below 1 */
262 while (i < type->cnt - 1) {
263 struct memblock_region *this = &type->regions[i];
264 struct memblock_region *next = &type->regions[i + 1];
265
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200266 if (this->base + this->size != next->base ||
267 memblock_get_region_node(this) !=
268 memblock_get_region_node(next)) {
Tejun Heo784656f2011-07-12 11:15:55 +0200269 BUG_ON(this->base + this->size > next->base);
270 i++;
271 continue;
272 }
273
274 this->size += next->size;
275 memmove(next, next + 1, (type->cnt - (i + 1)) * sizeof(*next));
276 type->cnt--;
277 }
278}
279
280/**
281 * memblock_insert_region - insert new memblock region
282 * @type: memblock type to insert into
283 * @idx: index for the insertion point
284 * @base: base address of the new region
285 * @size: size of the new region
286 *
287 * Insert new memblock region [@base,@base+@size) into @type at @idx.
288 * @type must already have extra room to accomodate the new region.
289 */
290static void __init_memblock memblock_insert_region(struct memblock_type *type,
291 int idx, phys_addr_t base,
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200292 phys_addr_t size, int nid)
Tejun Heo784656f2011-07-12 11:15:55 +0200293{
294 struct memblock_region *rgn = &type->regions[idx];
295
296 BUG_ON(type->cnt >= type->max);
297 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
298 rgn->base = base;
299 rgn->size = size;
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200300 memblock_set_region_node(rgn, nid);
Tejun Heo784656f2011-07-12 11:15:55 +0200301 type->cnt++;
302}
303
304/**
305 * memblock_add_region - add new memblock region
306 * @type: memblock type to add new region into
307 * @base: base address of the new region
308 * @size: size of the new region
309 *
310 * Add new memblock region [@base,@base+@size) into @type. The new region
311 * is allowed to overlap with existing ones - overlaps don't affect already
312 * existing regions. @type is guaranteed to be minimal (all neighbouring
313 * compatible regions are merged) after the addition.
314 *
315 * RETURNS:
316 * 0 on success, -errno on failure.
317 */
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700318static long __init_memblock memblock_add_region(struct memblock_type *type,
319 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000320{
Tejun Heo784656f2011-07-12 11:15:55 +0200321 bool insert = false;
322 phys_addr_t obase = base, end = base + size;
323 int i, nr_new;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000324
Tejun Heo784656f2011-07-12 11:15:55 +0200325 /* special case for empty array */
326 if (type->regions[0].size == 0) {
327 WARN_ON(type->cnt != 1);
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000328 type->regions[0].base = base;
329 type->regions[0].size = size;
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200330 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000331 return 0;
332 }
Tejun Heo784656f2011-07-12 11:15:55 +0200333repeat:
334 /*
335 * The following is executed twice. Once with %false @insert and
336 * then with %true. The first counts the number of regions needed
337 * to accomodate the new area. The second actually inserts them.
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700338 */
Tejun Heo784656f2011-07-12 11:15:55 +0200339 base = obase;
340 nr_new = 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000341
Tejun Heo784656f2011-07-12 11:15:55 +0200342 for (i = 0; i < type->cnt; i++) {
343 struct memblock_region *rgn = &type->regions[i];
344 phys_addr_t rbase = rgn->base;
345 phys_addr_t rend = rbase + rgn->size;
346
347 if (rbase >= end)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000348 break;
Tejun Heo784656f2011-07-12 11:15:55 +0200349 if (rend <= base)
350 continue;
351 /*
352 * @rgn overlaps. If it separates the lower part of new
353 * area, insert that portion.
354 */
355 if (rbase > base) {
356 nr_new++;
357 if (insert)
358 memblock_insert_region(type, i++, base,
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200359 rbase - base, MAX_NUMNODES);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000360 }
Tejun Heo784656f2011-07-12 11:15:55 +0200361 /* area below @rend is dealt with, forget about it */
362 base = min(rend, end);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000363 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000364
Tejun Heo784656f2011-07-12 11:15:55 +0200365 /* insert the remaining portion */
366 if (base < end) {
367 nr_new++;
368 if (insert)
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200369 memblock_insert_region(type, i, base, end - base,
370 MAX_NUMNODES);
Tejun Heo784656f2011-07-12 11:15:55 +0200371 }
372
373 /*
374 * If this was the first round, resize array and repeat for actual
375 * insertions; otherwise, merge and return.
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700376 */
Tejun Heo784656f2011-07-12 11:15:55 +0200377 if (!insert) {
378 while (type->cnt + nr_new > type->max)
379 if (memblock_double_array(type) < 0)
380 return -ENOMEM;
381 insert = true;
382 goto repeat;
383 } else {
384 memblock_merge_regions(type);
385 return 0;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700386 }
Yinghai Lu95f72d12010-07-12 14:36:09 +1000387}
388
Yinghai Lu10d06432010-07-28 15:43:02 +1000389long __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000390{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000391 return memblock_add_region(&memblock.memory, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000392}
393
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700394static long __init_memblock __memblock_remove(struct memblock_type *type,
395 phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000396{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000397 phys_addr_t end = base + size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000398 int i;
399
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700400 /* Walk through the array for collisions */
401 for (i = 0; i < type->cnt; i++) {
402 struct memblock_region *rgn = &type->regions[i];
403 phys_addr_t rend = rgn->base + rgn->size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000404
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700405 /* Nothing more to do, exit */
406 if (rgn->base > end || rgn->size == 0)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000407 break;
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700408
409 /* If we fully enclose the block, drop it */
410 if (base <= rgn->base && end >= rend) {
411 memblock_remove_region(type, i--);
412 continue;
413 }
414
415 /* If we are fully enclosed within a block
416 * then we need to split it and we are done
417 */
418 if (base > rgn->base && end < rend) {
419 rgn->size = base - rgn->base;
420 if (!memblock_add_region(type, end, rend - end))
421 return 0;
422 /* Failure to split is bad, we at least
423 * restore the block before erroring
424 */
425 rgn->size = rend - rgn->base;
426 WARN_ON(1);
427 return -1;
428 }
429
430 /* Check if we need to trim the bottom of a block */
431 if (rgn->base < end && rend > end) {
432 rgn->size -= end - rgn->base;
433 rgn->base = end;
434 break;
435 }
436
437 /* And check if we need to trim the top of a block */
438 if (base < rend)
439 rgn->size -= rend - base;
440
Yinghai Lu95f72d12010-07-12 14:36:09 +1000441 }
Benjamin Herrenschmidt8f7a6602011-03-22 16:33:43 -0700442 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000443}
444
Yinghai Lu10d06432010-07-28 15:43:02 +1000445long __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000446{
447 return __memblock_remove(&memblock.memory, base, size);
448}
449
Yinghai Lu3661ca62010-09-15 13:05:29 -0700450long __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000451{
452 return __memblock_remove(&memblock.reserved, base, size);
453}
454
Yinghai Lu3661ca62010-09-15 13:05:29 -0700455long __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000456{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000457 struct memblock_type *_rgn = &memblock.reserved;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000458
459 BUG_ON(0 == size);
460
461 return memblock_add_region(_rgn, base, size);
462}
463
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200464#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
465/*
466 * Common iterator interface used to define for_each_mem_range().
467 */
468void __init_memblock __next_mem_pfn_range(int *idx, int nid,
469 unsigned long *out_start_pfn,
470 unsigned long *out_end_pfn, int *out_nid)
471{
472 struct memblock_type *type = &memblock.memory;
473 struct memblock_region *r;
474
475 while (++*idx < type->cnt) {
476 r = &type->regions[*idx];
477
478 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
479 continue;
480 if (nid == MAX_NUMNODES || nid == r->nid)
481 break;
482 }
483 if (*idx >= type->cnt) {
484 *idx = -1;
485 return;
486 }
487
488 if (out_start_pfn)
489 *out_start_pfn = PFN_UP(r->base);
490 if (out_end_pfn)
491 *out_end_pfn = PFN_DOWN(r->base + r->size);
492 if (out_nid)
493 *out_nid = r->nid;
494}
495
496/**
497 * memblock_set_node - set node ID on memblock regions
498 * @base: base of area to set node ID for
499 * @size: size of area to set node ID for
500 * @nid: node ID to set
501 *
502 * Set the nid of memblock memory regions in [@base,@base+@size) to @nid.
503 * Regions which cross the area boundaries are split as necessary.
504 *
505 * RETURNS:
506 * 0 on success, -errno on failure.
507 */
508int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
509 int nid)
510{
511 struct memblock_type *type = &memblock.memory;
512 phys_addr_t end = base + size;
513 int i;
514
515 /* we'll create at most two more regions */
516 while (type->cnt + 2 > type->max)
517 if (memblock_double_array(type) < 0)
518 return -ENOMEM;
519
520 for (i = 0; i < type->cnt; i++) {
521 struct memblock_region *rgn = &type->regions[i];
522 phys_addr_t rbase = rgn->base;
523 phys_addr_t rend = rbase + rgn->size;
524
525 if (rbase >= end)
526 break;
527 if (rend <= base)
528 continue;
529
530 if (rbase < base) {
531 /*
532 * @rgn intersects from below. Split and continue
533 * to process the next region - the new top half.
534 */
535 rgn->base = base;
536 rgn->size = rend - rgn->base;
537 memblock_insert_region(type, i, rbase, base - rbase,
538 rgn->nid);
539 } else if (rend > end) {
540 /*
541 * @rgn intersects from above. Split and redo the
542 * current region - the new bottom half.
543 */
544 rgn->base = end;
545 rgn->size = rend - rgn->base;
546 memblock_insert_region(type, i--, rbase, end - rbase,
547 rgn->nid);
548 } else {
549 /* @rgn is fully contained, set ->nid */
550 rgn->nid = nid;
551 }
552 }
553
554 memblock_merge_regions(type);
555 return 0;
556}
557#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
558
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000559phys_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 +1000560{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000561 phys_addr_t found;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000562
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000563 /* We align the size to limit fragmentation. Without this, a lot of
564 * small allocs quickly eat up the whole reserve array on sparc
565 */
Tejun Heo348968e2011-07-12 09:58:08 +0200566 size = round_up(size, align);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000567
Tejun Heofc769a82011-07-12 09:58:10 +0200568 found = memblock_find_in_range(0, max_addr, size, align);
Tejun Heo1f5026a2011-07-12 09:58:09 +0200569 if (found && !memblock_add_region(&memblock.reserved, found, size))
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000570 return found;
571
572 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000573}
574
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000575phys_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 +1000576{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000577 phys_addr_t alloc;
578
579 alloc = __memblock_alloc_base(size, align, max_addr);
580
581 if (alloc == 0)
582 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
583 (unsigned long long) size, (unsigned long long) max_addr);
584
585 return alloc;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000586}
587
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000588phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000589{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000590 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000591}
592
Yinghai Lu95f72d12010-07-12 14:36:09 +1000593
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000594/*
Tejun Heo34e18452011-07-12 10:46:33 +0200595 * Additional node-local top-down allocators.
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700596 *
597 * WARNING: Only available after early_node_map[] has been populated,
598 * on some architectures, that is after all the calls to add_active_range()
599 * have been done to populate it.
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000600 */
Yinghai Lu95f72d12010-07-12 14:36:09 +1000601
Tejun Heo34e18452011-07-12 10:46:33 +0200602static phys_addr_t __init memblock_nid_range_rev(phys_addr_t start,
603 phys_addr_t end, int *nid)
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700604{
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700605#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700606 unsigned long start_pfn, end_pfn;
607 int i;
608
Tejun Heob2fea982011-07-12 10:46:31 +0200609 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, nid)
Tejun Heo34e18452011-07-12 10:46:33 +0200610 if (end > PFN_PHYS(start_pfn) && end <= PFN_PHYS(end_pfn))
611 return max(start, PFN_PHYS(start_pfn));
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700612#endif
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700613 *nid = 0;
Tejun Heo34e18452011-07-12 10:46:33 +0200614 return start;
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700615}
616
Tejun Heoe6498042011-07-12 10:46:34 +0200617phys_addr_t __init memblock_find_in_range_node(phys_addr_t start,
618 phys_addr_t end,
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000619 phys_addr_t size,
620 phys_addr_t align, int nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000621{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000622 struct memblock_type *mem = &memblock.memory;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000623 int i;
624
625 BUG_ON(0 == size);
626
Tejun Heoe6498042011-07-12 10:46:34 +0200627 /* Pump up max_addr */
628 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
629 end = memblock.current_limit;
630
631 for (i = mem->cnt - 1; i >= 0; i--) {
632 struct memblock_region *r = &mem->regions[i];
633 phys_addr_t base = max(start, r->base);
634 phys_addr_t top = min(end, r->base + r->size);
635
636 while (base < top) {
637 phys_addr_t tbase, ret;
638 int tnid;
639
640 tbase = memblock_nid_range_rev(base, top, &tnid);
641 if (nid == MAX_NUMNODES || tnid == nid) {
642 ret = memblock_find_region(tbase, top, size, align);
643 if (ret)
644 return ret;
645 }
646 top = tbase;
647 }
648 }
649
650 return 0;
651}
652
653phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
654{
655 phys_addr_t found;
656
657 /*
658 * We align the size to limit fragmentation. Without this, a lot of
Benjamin Herrenschmidt7f219c72010-07-12 14:24:57 +1000659 * small allocs quickly eat up the whole reserve array on sparc
660 */
Tejun Heo348968e2011-07-12 09:58:08 +0200661 size = round_up(size, align);
Benjamin Herrenschmidt7f219c72010-07-12 14:24:57 +1000662
Tejun Heoe6498042011-07-12 10:46:34 +0200663 found = memblock_find_in_range_node(0, MEMBLOCK_ALLOC_ACCESSIBLE,
664 size, align, nid);
665 if (found && !memblock_add_region(&memblock.reserved, found, size))
666 return found;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000667
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700668 return 0;
669}
670
671phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
672{
673 phys_addr_t res = memblock_alloc_nid(size, align, nid);
674
675 if (res)
676 return res;
Tejun Heo15fb0972011-07-12 09:58:07 +0200677 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000678}
679
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700680
681/*
682 * Remaining API functions
683 */
684
Yinghai Lu95f72d12010-07-12 14:36:09 +1000685/* You must call memblock_analyze() before this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000686phys_addr_t __init memblock_phys_mem_size(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000687{
Benjamin Herrenschmidt4734b592010-07-28 14:31:29 +1000688 return memblock.memory_size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000689}
690
Yinghai Lu10d06432010-07-28 15:43:02 +1000691phys_addr_t __init_memblock memblock_end_of_DRAM(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000692{
693 int idx = memblock.memory.cnt - 1;
694
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000695 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000696}
697
698/* You must call memblock_analyze() after this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000699void __init memblock_enforce_memory_limit(phys_addr_t memory_limit)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000700{
701 unsigned long i;
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000702 phys_addr_t limit;
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000703 struct memblock_region *p;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000704
705 if (!memory_limit)
706 return;
707
708 /* Truncate the memblock regions to satisfy the memory limit. */
709 limit = memory_limit;
710 for (i = 0; i < memblock.memory.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000711 if (limit > memblock.memory.regions[i].size) {
712 limit -= memblock.memory.regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000713 continue;
714 }
715
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000716 memblock.memory.regions[i].size = limit;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000717 memblock.memory.cnt = i + 1;
718 break;
719 }
720
Yinghai Lu95f72d12010-07-12 14:36:09 +1000721 memory_limit = memblock_end_of_DRAM();
722
723 /* And truncate any reserves above the limit also. */
724 for (i = 0; i < memblock.reserved.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000725 p = &memblock.reserved.regions[i];
Yinghai Lu95f72d12010-07-12 14:36:09 +1000726
727 if (p->base > memory_limit)
728 p->size = 0;
729 else if ((p->base + p->size) > memory_limit)
730 p->size = memory_limit - p->base;
731
732 if (p->size == 0) {
733 memblock_remove_region(&memblock.reserved, i);
734 i--;
735 }
736 }
737}
738
Yinghai Lucd794812010-10-11 12:34:09 -0700739static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000740{
741 unsigned int left = 0, right = type->cnt;
742
743 do {
744 unsigned int mid = (right + left) / 2;
745
746 if (addr < type->regions[mid].base)
747 right = mid;
748 else if (addr >= (type->regions[mid].base +
749 type->regions[mid].size))
750 left = mid + 1;
751 else
752 return mid;
753 } while (left < right);
754 return -1;
755}
756
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000757int __init memblock_is_reserved(phys_addr_t addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000758{
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000759 return memblock_search(&memblock.reserved, addr) != -1;
760}
Yinghai Lu95f72d12010-07-12 14:36:09 +1000761
Yinghai Lu3661ca62010-09-15 13:05:29 -0700762int __init_memblock memblock_is_memory(phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000763{
764 return memblock_search(&memblock.memory, addr) != -1;
765}
766
Yinghai Lu3661ca62010-09-15 13:05:29 -0700767int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000768{
Tomi Valkeinenabb65272011-01-20 14:44:20 -0800769 int idx = memblock_search(&memblock.memory, base);
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000770
771 if (idx == -1)
772 return 0;
Tomi Valkeinenabb65272011-01-20 14:44:20 -0800773 return memblock.memory.regions[idx].base <= base &&
774 (memblock.memory.regions[idx].base +
775 memblock.memory.regions[idx].size) >= (base + size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000776}
777
Yinghai Lu10d06432010-07-28 15:43:02 +1000778int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000779{
Benjamin Herrenschmidtf1c2c192010-08-04 14:17:17 +1000780 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000781}
782
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700783
Yinghai Lu3661ca62010-09-15 13:05:29 -0700784void __init_memblock memblock_set_current_limit(phys_addr_t limit)
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700785{
786 memblock.current_limit = limit;
787}
788
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200789static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000790{
791 unsigned long long base, size;
792 int i;
793
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200794 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000795
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200796 for (i = 0; i < type->cnt; i++) {
797 struct memblock_region *rgn = &type->regions[i];
798 char nid_buf[32] = "";
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000799
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200800 base = rgn->base;
801 size = rgn->size;
802#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
803 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
804 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
805 memblock_get_region_node(rgn));
806#endif
807 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s\n",
808 name, i, base, base + size - 1, size, nid_buf);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000809 }
810}
811
Yinghai Lu10d06432010-07-28 15:43:02 +1000812void __init_memblock memblock_dump_all(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000813{
814 if (!memblock_debug)
815 return;
816
817 pr_info("MEMBLOCK configuration:\n");
818 pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size);
819
820 memblock_dump(&memblock.memory, "memory");
821 memblock_dump(&memblock.reserved, "reserved");
822}
823
824void __init memblock_analyze(void)
825{
826 int i;
827
828 /* Check marker in the unused last array entry */
829 WARN_ON(memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS].base
830 != (phys_addr_t)RED_INACTIVE);
831 WARN_ON(memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS].base
832 != (phys_addr_t)RED_INACTIVE);
833
834 memblock.memory_size = 0;
835
836 for (i = 0; i < memblock.memory.cnt; i++)
837 memblock.memory_size += memblock.memory.regions[i].size;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700838
839 /* We allow resizing from there */
840 memblock_can_resize = 1;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000841}
842
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700843void __init memblock_init(void)
844{
Jeremy Fitzhardinge236260b2010-10-06 15:52:29 -0700845 static int init_done __initdata = 0;
846
847 if (init_done)
848 return;
849 init_done = 1;
850
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700851 /* Hookup the initial arrays */
852 memblock.memory.regions = memblock_memory_init_regions;
853 memblock.memory.max = INIT_MEMBLOCK_REGIONS;
854 memblock.reserved.regions = memblock_reserved_init_regions;
855 memblock.reserved.max = INIT_MEMBLOCK_REGIONS;
856
857 /* Write a marker in the unused last array entry */
858 memblock.memory.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
859 memblock.reserved.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
860
861 /* Create a dummy zero size MEMBLOCK which will get coalesced away later.
862 * This simplifies the memblock_add() code below...
863 */
864 memblock.memory.regions[0].base = 0;
865 memblock.memory.regions[0].size = 0;
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200866 memblock_set_region_node(&memblock.memory.regions[0], MAX_NUMNODES);
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700867 memblock.memory.cnt = 1;
868
869 /* Ditto. */
870 memblock.reserved.regions[0].base = 0;
871 memblock.reserved.regions[0].size = 0;
Tejun Heo7c0caeb2011-07-14 11:43:42 +0200872 memblock_set_region_node(&memblock.reserved.regions[0], MAX_NUMNODES);
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700873 memblock.reserved.cnt = 1;
874
875 memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE;
876}
877
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000878static int __init early_memblock(char *p)
879{
880 if (p && strstr(p, "debug"))
881 memblock_debug = 1;
882 return 0;
883}
884early_param("memblock", early_memblock);
885
Yinghai Lu10d06432010-07-28 15:43:02 +1000886#if defined(CONFIG_DEBUG_FS) && !defined(ARCH_DISCARD_MEMBLOCK)
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -0700887
888static int memblock_debug_show(struct seq_file *m, void *private)
889{
890 struct memblock_type *type = m->private;
891 struct memblock_region *reg;
892 int i;
893
894 for (i = 0; i < type->cnt; i++) {
895 reg = &type->regions[i];
896 seq_printf(m, "%4d: ", i);
897 if (sizeof(phys_addr_t) == 4)
898 seq_printf(m, "0x%08lx..0x%08lx\n",
899 (unsigned long)reg->base,
900 (unsigned long)(reg->base + reg->size - 1));
901 else
902 seq_printf(m, "0x%016llx..0x%016llx\n",
903 (unsigned long long)reg->base,
904 (unsigned long long)(reg->base + reg->size - 1));
905
906 }
907 return 0;
908}
909
910static int memblock_debug_open(struct inode *inode, struct file *file)
911{
912 return single_open(file, memblock_debug_show, inode->i_private);
913}
914
915static const struct file_operations memblock_debug_fops = {
916 .open = memblock_debug_open,
917 .read = seq_read,
918 .llseek = seq_lseek,
919 .release = single_release,
920};
921
922static int __init memblock_init_debugfs(void)
923{
924 struct dentry *root = debugfs_create_dir("memblock", NULL);
925 if (!root)
926 return -ENXIO;
927 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
928 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
929
930 return 0;
931}
932__initcall(memblock_init_debugfs);
933
934#endif /* CONFIG_DEBUG_FS */