blob: cb520df2a414a7bfd3fecf41d0116d44a466272a [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 Lu95f72d12010-07-12 14:36:09 +100023struct memblock memblock;
24
Yinghai Lu5e63cf42010-07-28 15:07:21 +100025int memblock_debug;
26int memblock_can_resize;
Benjamin Herrenschmidtbf23c512010-07-06 15:39:06 -070027static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS + 1];
28static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS + 1];
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
45static phys_addr_t 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
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100050static phys_addr_t 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
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +100055static unsigned long memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
56 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
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +100061static long memblock_addrs_adjacent(phys_addr_t base1, phys_addr_t size1,
62 phys_addr_t base2, phys_addr_t size2)
Yinghai Lu95f72d12010-07-12 14:36:09 +100063{
64 if (base2 == base1 + size1)
65 return 1;
66 else if (base1 == base2 + size2)
67 return -1;
68
69 return 0;
70}
71
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +100072static long memblock_regions_adjacent(struct memblock_type *type,
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +100073 unsigned long r1, unsigned long r2)
Yinghai Lu95f72d12010-07-12 14:36:09 +100074{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +100075 phys_addr_t base1 = type->regions[r1].base;
76 phys_addr_t size1 = type->regions[r1].size;
77 phys_addr_t base2 = type->regions[r2].base;
78 phys_addr_t size2 = type->regions[r2].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +100079
80 return memblock_addrs_adjacent(base1, size1, base2, size2);
81}
82
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100083long memblock_overlaps_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
84{
85 unsigned long i;
86
87 for (i = 0; i < type->cnt; i++) {
88 phys_addr_t rgnbase = type->regions[i].base;
89 phys_addr_t rgnsize = type->regions[i].size;
90 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
91 break;
92 }
93
94 return (i < type->cnt) ? i : -1;
95}
96
97/*
98 * Find, allocate, deallocate or reserve unreserved regions. All allocations
99 * are top-down.
100 */
101
102static phys_addr_t __init memblock_find_region(phys_addr_t start, phys_addr_t end,
103 phys_addr_t size, phys_addr_t align)
104{
105 phys_addr_t base, res_base;
106 long j;
107
Benjamin Herrenschmidt25818f02010-07-28 15:25:10 +1000108 /* Prevent allocations returning 0 as it's also used to
109 * indicate an allocation failure
110 */
111 if (start == 0)
112 start = PAGE_SIZE;
113
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000114 base = memblock_align_down((end - size), align);
115 while (start <= base) {
116 j = memblock_overlaps_region(&memblock.reserved, base, size);
117 if (j < 0)
118 return base;
119 res_base = memblock.reserved.regions[j].base;
120 if (res_base < size)
121 break;
122 base = memblock_align_down(res_base - size, align);
123 }
124
125 return MEMBLOCK_ERROR;
126}
127
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000128static phys_addr_t __init memblock_find_base(phys_addr_t size, phys_addr_t align,
129 phys_addr_t start, phys_addr_t end)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000130{
131 long i;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000132
133 BUG_ON(0 == size);
134
135 size = memblock_align_up(size, align);
136
137 /* Pump up max_addr */
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000138 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
139 end = memblock.current_limit;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000140
141 /* We do a top-down search, this tends to limit memory
142 * fragmentation by keeping early boot allocs near the
143 * top of memory
144 */
145 for (i = memblock.memory.cnt - 1; i >= 0; i--) {
146 phys_addr_t memblockbase = memblock.memory.regions[i].base;
147 phys_addr_t memblocksize = memblock.memory.regions[i].size;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000148 phys_addr_t bottom, top, found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000149
150 if (memblocksize < size)
151 continue;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000152 if ((memblockbase + memblocksize) <= start)
153 break;
154 bottom = max(memblockbase, start);
155 top = min(memblockbase + memblocksize, end);
156 if (bottom >= top)
157 continue;
158 found = memblock_find_region(bottom, top, size, align);
159 if (found != MEMBLOCK_ERROR)
160 return found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000161 }
162 return MEMBLOCK_ERROR;
163}
164
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000165static void memblock_remove_region(struct memblock_type *type, unsigned long r)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000166{
167 unsigned long i;
168
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000169 for (i = r; i < type->cnt - 1; i++) {
170 type->regions[i].base = type->regions[i + 1].base;
171 type->regions[i].size = type->regions[i + 1].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000172 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000173 type->cnt--;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000174}
175
176/* Assumption: base addr of region 1 < base addr of region 2 */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000177static void memblock_coalesce_regions(struct memblock_type *type,
Yinghai Lu95f72d12010-07-12 14:36:09 +1000178 unsigned long r1, unsigned long r2)
179{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000180 type->regions[r1].size += type->regions[r2].size;
181 memblock_remove_region(type, r2);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000182}
183
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700184/* Defined below but needed now */
185static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size);
186
187static int memblock_double_array(struct memblock_type *type)
188{
189 struct memblock_region *new_array, *old_array;
190 phys_addr_t old_size, new_size, addr;
191 int use_slab = slab_is_available();
192
193 /* We don't allow resizing until we know about the reserved regions
194 * of memory that aren't suitable for allocation
195 */
196 if (!memblock_can_resize)
197 return -1;
198
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700199 /* Calculate new doubled size */
200 old_size = type->max * sizeof(struct memblock_region);
201 new_size = old_size << 1;
202
203 /* Try to find some space for it.
204 *
205 * WARNING: We assume that either slab_is_available() and we use it or
206 * we use MEMBLOCK for allocations. That means that this is unsafe to use
207 * when bootmem is currently active (unless bootmem itself is implemented
208 * on top of MEMBLOCK which isn't the case yet)
209 *
210 * This should however not be an issue for now, as we currently only
211 * call into MEMBLOCK while it's still active, or much later when slab is
212 * active for memory hotplug operations
213 */
214 if (use_slab) {
215 new_array = kmalloc(new_size, GFP_KERNEL);
216 addr = new_array == NULL ? MEMBLOCK_ERROR : __pa(new_array);
217 } else
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000218 addr = memblock_find_base(new_size, sizeof(phys_addr_t), 0, MEMBLOCK_ALLOC_ACCESSIBLE);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700219 if (addr == MEMBLOCK_ERROR) {
220 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
221 memblock_type_name(type), type->max, type->max * 2);
222 return -1;
223 }
224 new_array = __va(addr);
225
Yinghai Luea9e4372010-07-28 15:13:22 +1000226 memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]",
227 memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1);
228
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700229 /* Found space, we now need to move the array over before
230 * we add the reserved region since it may be our reserved
231 * array itself that is full.
232 */
233 memcpy(new_array, type->regions, old_size);
234 memset(new_array + type->max, 0, old_size);
235 old_array = type->regions;
236 type->regions = new_array;
237 type->max <<= 1;
238
239 /* If we use SLAB that's it, we are done */
240 if (use_slab)
241 return 0;
242
243 /* Add the new reserved region now. Should not fail ! */
244 BUG_ON(memblock_add_region(&memblock.reserved, addr, new_size) < 0);
245
246 /* If the array wasn't our static init one, then free it. We only do
247 * that before SLAB is available as later on, we don't know whether
248 * to use kfree or free_bootmem_pages(). Shouldn't be a big deal
249 * anyways
250 */
251 if (old_array != memblock_memory_init_regions &&
252 old_array != memblock_reserved_init_regions)
253 memblock_free(__pa(old_array), old_size);
254
255 return 0;
256}
257
Benjamin Herrenschmidtd2cd5632010-07-06 15:39:14 -0700258extern int __weak memblock_memory_can_coalesce(phys_addr_t addr1, phys_addr_t size1,
259 phys_addr_t addr2, phys_addr_t size2)
260{
261 return 1;
262}
263
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000264static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000265{
266 unsigned long coalesced = 0;
267 long adjacent, i;
268
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000269 if ((type->cnt == 1) && (type->regions[0].size == 0)) {
270 type->regions[0].base = base;
271 type->regions[0].size = size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000272 return 0;
273 }
274
275 /* First try and coalesce this MEMBLOCK with another. */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000276 for (i = 0; i < type->cnt; i++) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000277 phys_addr_t rgnbase = type->regions[i].base;
278 phys_addr_t rgnsize = type->regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000279
280 if ((rgnbase == base) && (rgnsize == size))
281 /* Already have this region, so we're done */
282 return 0;
283
284 adjacent = memblock_addrs_adjacent(base, size, rgnbase, rgnsize);
Benjamin Herrenschmidtd2cd5632010-07-06 15:39:14 -0700285 /* Check if arch allows coalescing */
286 if (adjacent != 0 && type == &memblock.memory &&
287 !memblock_memory_can_coalesce(base, size, rgnbase, rgnsize))
288 break;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000289 if (adjacent > 0) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000290 type->regions[i].base -= size;
291 type->regions[i].size += size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000292 coalesced++;
293 break;
294 } else if (adjacent < 0) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000295 type->regions[i].size += size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000296 coalesced++;
297 break;
298 }
299 }
300
Benjamin Herrenschmidtd2cd5632010-07-06 15:39:14 -0700301 /* If we plugged a hole, we may want to also coalesce with the
302 * next region
303 */
304 if ((i < type->cnt - 1) && memblock_regions_adjacent(type, i, i+1) &&
305 ((type != &memblock.memory || memblock_memory_can_coalesce(type->regions[i].base,
306 type->regions[i].size,
307 type->regions[i+1].base,
308 type->regions[i+1].size)))) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000309 memblock_coalesce_regions(type, i, i+1);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000310 coalesced++;
311 }
312
313 if (coalesced)
314 return coalesced;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700315
316 /* If we are out of space, we fail. It's too late to resize the array
317 * but then this shouldn't have happened in the first place.
318 */
319 if (WARN_ON(type->cnt >= type->max))
Yinghai Lu95f72d12010-07-12 14:36:09 +1000320 return -1;
321
322 /* Couldn't coalesce the MEMBLOCK, so add it to the sorted table. */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000323 for (i = type->cnt - 1; i >= 0; i--) {
324 if (base < type->regions[i].base) {
325 type->regions[i+1].base = type->regions[i].base;
326 type->regions[i+1].size = type->regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000327 } else {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000328 type->regions[i+1].base = base;
329 type->regions[i+1].size = size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000330 break;
331 }
332 }
333
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000334 if (base < type->regions[0].base) {
335 type->regions[0].base = base;
336 type->regions[0].size = size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000337 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000338 type->cnt++;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000339
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700340 /* The array is full ? Try to resize it. If that fails, we undo
341 * our allocation and return an error
342 */
343 if (type->cnt == type->max && memblock_double_array(type)) {
344 type->cnt--;
345 return -1;
346 }
347
Yinghai Lu95f72d12010-07-12 14:36:09 +1000348 return 0;
349}
350
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000351long memblock_add(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000352{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000353 return memblock_add_region(&memblock.memory, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000354
355}
356
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000357static long __memblock_remove(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000358{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000359 phys_addr_t rgnbegin, rgnend;
360 phys_addr_t end = base + size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000361 int i;
362
363 rgnbegin = rgnend = 0; /* supress gcc warnings */
364
365 /* Find the region where (base, size) belongs to */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000366 for (i=0; i < type->cnt; i++) {
367 rgnbegin = type->regions[i].base;
368 rgnend = rgnbegin + type->regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000369
370 if ((rgnbegin <= base) && (end <= rgnend))
371 break;
372 }
373
374 /* Didn't find the region */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000375 if (i == type->cnt)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000376 return -1;
377
378 /* Check to see if we are removing entire region */
379 if ((rgnbegin == base) && (rgnend == end)) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000380 memblock_remove_region(type, i);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000381 return 0;
382 }
383
384 /* Check to see if region is matching at the front */
385 if (rgnbegin == base) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000386 type->regions[i].base = end;
387 type->regions[i].size -= size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000388 return 0;
389 }
390
391 /* Check to see if the region is matching at the end */
392 if (rgnend == end) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000393 type->regions[i].size -= size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000394 return 0;
395 }
396
397 /*
398 * We need to split the entry - adjust the current one to the
399 * beginging of the hole and add the region after hole.
400 */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000401 type->regions[i].size = base - type->regions[i].base;
402 return memblock_add_region(type, end, rgnend - end);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000403}
404
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000405long memblock_remove(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000406{
407 return __memblock_remove(&memblock.memory, base, size);
408}
409
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000410long __init memblock_free(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000411{
412 return __memblock_remove(&memblock.reserved, base, size);
413}
414
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000415long __init memblock_reserve(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000416{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000417 struct memblock_type *_rgn = &memblock.reserved;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000418
419 BUG_ON(0 == size);
420
421 return memblock_add_region(_rgn, base, size);
422}
423
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000424phys_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 +1000425{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000426 phys_addr_t found;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000427
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000428 /* We align the size to limit fragmentation. Without this, a lot of
429 * small allocs quickly eat up the whole reserve array on sparc
430 */
431 size = memblock_align_up(size, align);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000432
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000433 found = memblock_find_base(size, align, 0, max_addr);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000434 if (found != MEMBLOCK_ERROR &&
435 memblock_add_region(&memblock.reserved, found, size) >= 0)
436 return found;
437
438 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000439}
440
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000441phys_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 +1000442{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000443 phys_addr_t alloc;
444
445 alloc = __memblock_alloc_base(size, align, max_addr);
446
447 if (alloc == 0)
448 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
449 (unsigned long long) size, (unsigned long long) max_addr);
450
451 return alloc;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000452}
453
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000454phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000455{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000456 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000457}
458
Yinghai Lu95f72d12010-07-12 14:36:09 +1000459
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000460/*
461 * Additional node-local allocators. Search for node memory is bottom up
462 * and walks memblock regions within that node bottom-up as well, but allocation
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700463 * within an memblock region is top-down. XXX I plan to fix that at some stage
464 *
465 * WARNING: Only available after early_node_map[] has been populated,
466 * on some architectures, that is after all the calls to add_active_range()
467 * have been done to populate it.
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000468 */
Yinghai Lu95f72d12010-07-12 14:36:09 +1000469
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000470phys_addr_t __weak __init memblock_nid_range(phys_addr_t start, phys_addr_t end, int *nid)
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700471{
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700472#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
473 /*
474 * This code originates from sparc which really wants use to walk by addresses
475 * and returns the nid. This is not very convenient for early_pfn_map[] users
476 * as the map isn't sorted yet, and it really wants to be walked by nid.
477 *
478 * For now, I implement the inefficient method below which walks the early
479 * map multiple times. Eventually we may want to use an ARCH config option
480 * to implement a completely different method for both case.
481 */
482 unsigned long start_pfn, end_pfn;
483 int i;
484
485 for (i = 0; i < MAX_NUMNODES; i++) {
486 get_pfn_range_for_nid(i, &start_pfn, &end_pfn);
487 if (start < PFN_PHYS(start_pfn) || start >= PFN_PHYS(end_pfn))
488 continue;
489 *nid = i;
490 return min(end, PFN_PHYS(end_pfn));
491 }
492#endif
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700493 *nid = 0;
494
495 return end;
496}
497
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000498static phys_addr_t __init memblock_alloc_nid_region(struct memblock_region *mp,
499 phys_addr_t size,
500 phys_addr_t align, int nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000501{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000502 phys_addr_t start, end;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000503
504 start = mp->base;
505 end = start + mp->size;
506
507 start = memblock_align_up(start, align);
508 while (start < end) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000509 phys_addr_t this_end;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000510 int this_nid;
511
Benjamin Herrenschmidt35a1f0b2010-07-06 15:38:58 -0700512 this_end = memblock_nid_range(start, end, &this_nid);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000513 if (this_nid == nid) {
Benjamin Herrenschmidt3a9c2c82010-07-12 13:28:15 +1000514 phys_addr_t ret = memblock_find_region(start, this_end, size, align);
Benjamin Herrenschmidt4d629f92010-07-06 15:39:09 -0700515 if (ret != MEMBLOCK_ERROR &&
Benjamin Herrenschmidt3a9c2c82010-07-12 13:28:15 +1000516 memblock_add_region(&memblock.reserved, ret, size) >= 0)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000517 return ret;
518 }
519 start = this_end;
520 }
521
Benjamin Herrenschmidt4d629f92010-07-06 15:39:09 -0700522 return MEMBLOCK_ERROR;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000523}
524
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000525phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000526{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000527 struct memblock_type *mem = &memblock.memory;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000528 int i;
529
530 BUG_ON(0 == size);
531
Benjamin Herrenschmidt7f219c72010-07-12 14:24:57 +1000532 /* We align the size to limit fragmentation. Without this, a lot of
533 * small allocs quickly eat up the whole reserve array on sparc
534 */
535 size = memblock_align_up(size, align);
536
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700537 /* We do a bottom-up search for a region with the right
538 * nid since that's easier considering how memblock_nid_range()
539 * works
540 */
Yinghai Lu95f72d12010-07-12 14:36:09 +1000541 for (i = 0; i < mem->cnt; i++) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000542 phys_addr_t ret = memblock_alloc_nid_region(&mem->regions[i],
Yinghai Lu95f72d12010-07-12 14:36:09 +1000543 size, align, nid);
Benjamin Herrenschmidt4d629f92010-07-06 15:39:09 -0700544 if (ret != MEMBLOCK_ERROR)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000545 return ret;
546 }
547
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700548 return 0;
549}
550
551phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
552{
553 phys_addr_t res = memblock_alloc_nid(size, align, nid);
554
555 if (res)
556 return res;
Benjamin Herrenschmidt918fe8d2010-07-06 15:39:18 -0700557 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000558}
559
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700560
561/*
562 * Remaining API functions
563 */
564
Yinghai Lu95f72d12010-07-12 14:36:09 +1000565/* You must call memblock_analyze() before this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000566phys_addr_t __init memblock_phys_mem_size(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000567{
Benjamin Herrenschmidt4734b592010-07-28 14:31:29 +1000568 return memblock.memory_size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000569}
570
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000571phys_addr_t memblock_end_of_DRAM(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000572{
573 int idx = memblock.memory.cnt - 1;
574
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000575 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000576}
577
578/* You must call memblock_analyze() after this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000579void __init memblock_enforce_memory_limit(phys_addr_t memory_limit)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000580{
581 unsigned long i;
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000582 phys_addr_t limit;
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000583 struct memblock_region *p;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000584
585 if (!memory_limit)
586 return;
587
588 /* Truncate the memblock regions to satisfy the memory limit. */
589 limit = memory_limit;
590 for (i = 0; i < memblock.memory.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000591 if (limit > memblock.memory.regions[i].size) {
592 limit -= memblock.memory.regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000593 continue;
594 }
595
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000596 memblock.memory.regions[i].size = limit;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000597 memblock.memory.cnt = i + 1;
598 break;
599 }
600
Yinghai Lu95f72d12010-07-12 14:36:09 +1000601 memory_limit = memblock_end_of_DRAM();
602
603 /* And truncate any reserves above the limit also. */
604 for (i = 0; i < memblock.reserved.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000605 p = &memblock.reserved.regions[i];
Yinghai Lu95f72d12010-07-12 14:36:09 +1000606
607 if (p->base > memory_limit)
608 p->size = 0;
609 else if ((p->base + p->size) > memory_limit)
610 p->size = memory_limit - p->base;
611
612 if (p->size == 0) {
613 memblock_remove_region(&memblock.reserved, i);
614 i--;
615 }
616 }
617}
618
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000619static int memblock_search(struct memblock_type *type, phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000620{
621 unsigned int left = 0, right = type->cnt;
622
623 do {
624 unsigned int mid = (right + left) / 2;
625
626 if (addr < type->regions[mid].base)
627 right = mid;
628 else if (addr >= (type->regions[mid].base +
629 type->regions[mid].size))
630 left = mid + 1;
631 else
632 return mid;
633 } while (left < right);
634 return -1;
635}
636
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000637int __init memblock_is_reserved(phys_addr_t addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000638{
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000639 return memblock_search(&memblock.reserved, addr) != -1;
640}
Yinghai Lu95f72d12010-07-12 14:36:09 +1000641
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000642int memblock_is_memory(phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000643{
644 return memblock_search(&memblock.memory, addr) != -1;
645}
646
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000647int memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000648{
649 int idx = memblock_search(&memblock.reserved, base);
650
651 if (idx == -1)
652 return 0;
653 return memblock.reserved.regions[idx].base <= base &&
654 (memblock.reserved.regions[idx].base +
655 memblock.reserved.regions[idx].size) >= (base + size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000656}
657
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000658int memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000659{
Benjamin Herrenschmidtf1c2c192010-08-04 14:17:17 +1000660 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000661}
662
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700663
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000664void __init memblock_set_current_limit(phys_addr_t limit)
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700665{
666 memblock.current_limit = limit;
667}
668
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000669static void memblock_dump(struct memblock_type *region, char *name)
670{
671 unsigned long long base, size;
672 int i;
673
674 pr_info(" %s.cnt = 0x%lx\n", name, region->cnt);
675
676 for (i = 0; i < region->cnt; i++) {
677 base = region->regions[i].base;
678 size = region->regions[i].size;
679
Yinghai Luea9e4372010-07-28 15:13:22 +1000680 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes\n",
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000681 name, i, base, base + size - 1, size);
682 }
683}
684
685void memblock_dump_all(void)
686{
687 if (!memblock_debug)
688 return;
689
690 pr_info("MEMBLOCK configuration:\n");
691 pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size);
692
693 memblock_dump(&memblock.memory, "memory");
694 memblock_dump(&memblock.reserved, "reserved");
695}
696
697void __init memblock_analyze(void)
698{
699 int i;
700
701 /* Check marker in the unused last array entry */
702 WARN_ON(memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS].base
703 != (phys_addr_t)RED_INACTIVE);
704 WARN_ON(memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS].base
705 != (phys_addr_t)RED_INACTIVE);
706
707 memblock.memory_size = 0;
708
709 for (i = 0; i < memblock.memory.cnt; i++)
710 memblock.memory_size += memblock.memory.regions[i].size;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700711
712 /* We allow resizing from there */
713 memblock_can_resize = 1;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000714}
715
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700716void __init memblock_init(void)
717{
718 /* Hookup the initial arrays */
719 memblock.memory.regions = memblock_memory_init_regions;
720 memblock.memory.max = INIT_MEMBLOCK_REGIONS;
721 memblock.reserved.regions = memblock_reserved_init_regions;
722 memblock.reserved.max = INIT_MEMBLOCK_REGIONS;
723
724 /* Write a marker in the unused last array entry */
725 memblock.memory.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
726 memblock.reserved.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
727
728 /* Create a dummy zero size MEMBLOCK which will get coalesced away later.
729 * This simplifies the memblock_add() code below...
730 */
731 memblock.memory.regions[0].base = 0;
732 memblock.memory.regions[0].size = 0;
733 memblock.memory.cnt = 1;
734
735 /* Ditto. */
736 memblock.reserved.regions[0].base = 0;
737 memblock.reserved.regions[0].size = 0;
738 memblock.reserved.cnt = 1;
739
740 memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE;
741}
742
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000743static int __init early_memblock(char *p)
744{
745 if (p && strstr(p, "debug"))
746 memblock_debug = 1;
747 return 0;
748}
749early_param("memblock", early_memblock);
750
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -0700751#ifdef CONFIG_DEBUG_FS
752
753static int memblock_debug_show(struct seq_file *m, void *private)
754{
755 struct memblock_type *type = m->private;
756 struct memblock_region *reg;
757 int i;
758
759 for (i = 0; i < type->cnt; i++) {
760 reg = &type->regions[i];
761 seq_printf(m, "%4d: ", i);
762 if (sizeof(phys_addr_t) == 4)
763 seq_printf(m, "0x%08lx..0x%08lx\n",
764 (unsigned long)reg->base,
765 (unsigned long)(reg->base + reg->size - 1));
766 else
767 seq_printf(m, "0x%016llx..0x%016llx\n",
768 (unsigned long long)reg->base,
769 (unsigned long long)(reg->base + reg->size - 1));
770
771 }
772 return 0;
773}
774
775static int memblock_debug_open(struct inode *inode, struct file *file)
776{
777 return single_open(file, memblock_debug_show, inode->i_private);
778}
779
780static const struct file_operations memblock_debug_fops = {
781 .open = memblock_debug_open,
782 .read = seq_read,
783 .llseek = seq_lseek,
784 .release = single_release,
785};
786
787static int __init memblock_init_debugfs(void)
788{
789 struct dentry *root = debugfs_create_dir("memblock", NULL);
790 if (!root)
791 return -ENXIO;
792 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
793 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
794
795 return 0;
796}
797__initcall(memblock_init_debugfs);
798
799#endif /* CONFIG_DEBUG_FS */