blob: ae8b06c828c7fdeac2548ea4d85ecf94b52f257c [file] [log] [blame]
Yinghai Lu95f72d12010-07-12 14:36:09 +10001/*
2 * Procedures for maintaining information about logical memory blocks.
3 *
4 * Peter Bergner, IBM Corp. June 2001.
5 * Copyright (C) 2001 Peter Bergner.
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 */
12
13#include <linux/kernel.h>
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -070014#include <linux/slab.h>
Yinghai Lu95f72d12010-07-12 14:36:09 +100015#include <linux/init.h>
16#include <linux/bitops.h>
Benjamin Herrenschmidt449e8df2010-07-06 15:39:07 -070017#include <linux/poison.h>
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -070018#include <linux/pfn.h>
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -070019#include <linux/debugfs.h>
20#include <linux/seq_file.h>
Yinghai Lu95f72d12010-07-12 14:36:09 +100021#include <linux/memblock.h>
22
Yinghai Lu10d06432010-07-28 15:43:02 +100023struct memblock memblock __initdata_memblock;
Yinghai Lu95f72d12010-07-12 14:36:09 +100024
Yinghai Lu10d06432010-07-28 15:43:02 +100025int memblock_debug __initdata_memblock;
26int memblock_can_resize __initdata_memblock;
27static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS + 1] __initdata_memblock;
28static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS + 1] __initdata_memblock;
Yinghai Lu95f72d12010-07-12 14:36:09 +100029
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -070030/* inline so we don't get a warning when pr_debug is compiled out */
31static inline const char *memblock_type_name(struct memblock_type *type)
32{
33 if (type == &memblock.memory)
34 return "memory";
35 else if (type == &memblock.reserved)
36 return "reserved";
37 else
38 return "unknown";
39}
40
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100041/*
42 * Address comparison utilities
43 */
44
Yinghai Lu10d06432010-07-28 15:43:02 +100045static phys_addr_t __init_memblock memblock_align_down(phys_addr_t addr, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +100046{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100047 return addr & ~(size - 1);
Yinghai Lu95f72d12010-07-12 14:36:09 +100048}
49
Yinghai Lu10d06432010-07-28 15:43:02 +100050static phys_addr_t __init_memblock memblock_align_up(phys_addr_t addr, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +100051{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100052 return (addr + (size - 1)) & ~(size - 1);
Yinghai Lu95f72d12010-07-12 14:36:09 +100053}
54
Yinghai Lu10d06432010-07-28 15:43:02 +100055static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +100056 phys_addr_t base2, phys_addr_t size2)
Yinghai Lu95f72d12010-07-12 14:36:09 +100057{
58 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
59}
60
Yinghai Lu10d06432010-07-28 15:43:02 +100061static long __init_memblock memblock_addrs_adjacent(phys_addr_t base1, phys_addr_t size1,
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +100062 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
Yinghai Lu10d06432010-07-28 15:43:02 +100072static long __init_memblock 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
Yinghai Lu10d06432010-07-28 15:43:02 +100083long __init_memblock memblock_overlaps_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +100084{
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
Yinghai Luf1af98c2010-10-04 14:57:39 -0700108 /* In case, huge size is requested */
109 if (end < size)
110 return MEMBLOCK_ERROR;
111
112 base = memblock_align_down((end - size), align);
113
Benjamin Herrenschmidt25818f02010-07-28 15:25:10 +1000114 /* Prevent allocations returning 0 as it's also used to
115 * indicate an allocation failure
116 */
117 if (start == 0)
118 start = PAGE_SIZE;
119
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000120 while (start <= base) {
121 j = memblock_overlaps_region(&memblock.reserved, base, size);
122 if (j < 0)
123 return base;
124 res_base = memblock.reserved.regions[j].base;
125 if (res_base < size)
126 break;
127 base = memblock_align_down(res_base - size, align);
128 }
129
130 return MEMBLOCK_ERROR;
131}
132
Yinghai Lu3661ca62010-09-15 13:05:29 -0700133static phys_addr_t __init_memblock memblock_find_base(phys_addr_t size,
134 phys_addr_t align, phys_addr_t start, phys_addr_t end)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000135{
136 long i;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000137
138 BUG_ON(0 == size);
139
140 size = memblock_align_up(size, align);
141
142 /* Pump up max_addr */
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000143 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
144 end = memblock.current_limit;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000145
146 /* We do a top-down search, this tends to limit memory
147 * fragmentation by keeping early boot allocs near the
148 * top of memory
149 */
150 for (i = memblock.memory.cnt - 1; i >= 0; i--) {
151 phys_addr_t memblockbase = memblock.memory.regions[i].base;
152 phys_addr_t memblocksize = memblock.memory.regions[i].size;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000153 phys_addr_t bottom, top, found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000154
155 if (memblocksize < size)
156 continue;
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000157 if ((memblockbase + memblocksize) <= start)
158 break;
159 bottom = max(memblockbase, start);
160 top = min(memblockbase + memblocksize, end);
161 if (bottom >= top)
162 continue;
163 found = memblock_find_region(bottom, top, size, align);
164 if (found != MEMBLOCK_ERROR)
165 return found;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000166 }
167 return MEMBLOCK_ERROR;
168}
169
Yinghai Lu5303b682010-07-28 15:38:40 +1000170/*
171 * Find a free area with specified alignment in a specific range.
172 */
173u64 __init_memblock memblock_find_in_range(u64 start, u64 end, u64 size, u64 align)
174{
175 return memblock_find_base(size, align, start, end);
176}
177
Yinghai Lu7950c402010-08-25 13:39:14 -0700178/*
179 * Free memblock.reserved.regions
180 */
181int __init_memblock memblock_free_reserved_regions(void)
182{
183 if (memblock.reserved.regions == memblock_reserved_init_regions)
184 return 0;
185
186 return memblock_free(__pa(memblock.reserved.regions),
187 sizeof(struct memblock_region) * memblock.reserved.max);
188}
189
190/*
191 * Reserve memblock.reserved.regions
192 */
193int __init_memblock memblock_reserve_reserved_regions(void)
194{
195 if (memblock.reserved.regions == memblock_reserved_init_regions)
196 return 0;
197
198 return memblock_reserve(__pa(memblock.reserved.regions),
199 sizeof(struct memblock_region) * memblock.reserved.max);
200}
201
Yinghai Lu10d06432010-07-28 15:43:02 +1000202static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000203{
204 unsigned long i;
205
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000206 for (i = r; i < type->cnt - 1; i++) {
207 type->regions[i].base = type->regions[i + 1].base;
208 type->regions[i].size = type->regions[i + 1].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000209 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000210 type->cnt--;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000211}
212
213/* Assumption: base addr of region 1 < base addr of region 2 */
Yinghai Lu10d06432010-07-28 15:43:02 +1000214static void __init_memblock memblock_coalesce_regions(struct memblock_type *type,
Yinghai Lu95f72d12010-07-12 14:36:09 +1000215 unsigned long r1, unsigned long r2)
216{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000217 type->regions[r1].size += type->regions[r2].size;
218 memblock_remove_region(type, r2);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000219}
220
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700221/* Defined below but needed now */
222static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size);
223
Yinghai Lu10d06432010-07-28 15:43:02 +1000224static int __init_memblock memblock_double_array(struct memblock_type *type)
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700225{
226 struct memblock_region *new_array, *old_array;
227 phys_addr_t old_size, new_size, addr;
228 int use_slab = slab_is_available();
229
230 /* We don't allow resizing until we know about the reserved regions
231 * of memory that aren't suitable for allocation
232 */
233 if (!memblock_can_resize)
234 return -1;
235
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700236 /* Calculate new doubled size */
237 old_size = type->max * sizeof(struct memblock_region);
238 new_size = old_size << 1;
239
240 /* Try to find some space for it.
241 *
242 * WARNING: We assume that either slab_is_available() and we use it or
243 * we use MEMBLOCK for allocations. That means that this is unsafe to use
244 * when bootmem is currently active (unless bootmem itself is implemented
245 * on top of MEMBLOCK which isn't the case yet)
246 *
247 * This should however not be an issue for now, as we currently only
248 * call into MEMBLOCK while it's still active, or much later when slab is
249 * active for memory hotplug operations
250 */
251 if (use_slab) {
252 new_array = kmalloc(new_size, GFP_KERNEL);
253 addr = new_array == NULL ? MEMBLOCK_ERROR : __pa(new_array);
254 } else
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000255 addr = memblock_find_base(new_size, sizeof(phys_addr_t), 0, MEMBLOCK_ALLOC_ACCESSIBLE);
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700256 if (addr == MEMBLOCK_ERROR) {
257 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
258 memblock_type_name(type), type->max, type->max * 2);
259 return -1;
260 }
261 new_array = __va(addr);
262
Yinghai Luea9e4372010-07-28 15:13:22 +1000263 memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]",
264 memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1);
265
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700266 /* Found space, we now need to move the array over before
267 * we add the reserved region since it may be our reserved
268 * array itself that is full.
269 */
270 memcpy(new_array, type->regions, old_size);
271 memset(new_array + type->max, 0, old_size);
272 old_array = type->regions;
273 type->regions = new_array;
274 type->max <<= 1;
275
276 /* If we use SLAB that's it, we are done */
277 if (use_slab)
278 return 0;
279
280 /* Add the new reserved region now. Should not fail ! */
281 BUG_ON(memblock_add_region(&memblock.reserved, addr, new_size) < 0);
282
283 /* If the array wasn't our static init one, then free it. We only do
284 * that before SLAB is available as later on, we don't know whether
285 * to use kfree or free_bootmem_pages(). Shouldn't be a big deal
286 * anyways
287 */
288 if (old_array != memblock_memory_init_regions &&
289 old_array != memblock_reserved_init_regions)
290 memblock_free(__pa(old_array), old_size);
291
292 return 0;
293}
294
Yinghai Lu10d06432010-07-28 15:43:02 +1000295extern int __init_memblock __weak memblock_memory_can_coalesce(phys_addr_t addr1, phys_addr_t size1,
Benjamin Herrenschmidtd2cd5632010-07-06 15:39:14 -0700296 phys_addr_t addr2, phys_addr_t size2)
297{
298 return 1;
299}
300
Yinghai Lu10d06432010-07-28 15:43:02 +1000301static long __init_memblock memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000302{
303 unsigned long coalesced = 0;
304 long adjacent, i;
305
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000306 if ((type->cnt == 1) && (type->regions[0].size == 0)) {
307 type->regions[0].base = base;
308 type->regions[0].size = size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000309 return 0;
310 }
311
312 /* First try and coalesce this MEMBLOCK with another. */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000313 for (i = 0; i < type->cnt; i++) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000314 phys_addr_t rgnbase = type->regions[i].base;
315 phys_addr_t rgnsize = type->regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000316
317 if ((rgnbase == base) && (rgnsize == size))
318 /* Already have this region, so we're done */
319 return 0;
320
321 adjacent = memblock_addrs_adjacent(base, size, rgnbase, rgnsize);
Benjamin Herrenschmidtd2cd5632010-07-06 15:39:14 -0700322 /* Check if arch allows coalescing */
323 if (adjacent != 0 && type == &memblock.memory &&
324 !memblock_memory_can_coalesce(base, size, rgnbase, rgnsize))
325 break;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000326 if (adjacent > 0) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000327 type->regions[i].base -= size;
328 type->regions[i].size += size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000329 coalesced++;
330 break;
331 } else if (adjacent < 0) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000332 type->regions[i].size += size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000333 coalesced++;
334 break;
335 }
336 }
337
Benjamin Herrenschmidtd2cd5632010-07-06 15:39:14 -0700338 /* If we plugged a hole, we may want to also coalesce with the
339 * next region
340 */
341 if ((i < type->cnt - 1) && memblock_regions_adjacent(type, i, i+1) &&
342 ((type != &memblock.memory || memblock_memory_can_coalesce(type->regions[i].base,
343 type->regions[i].size,
344 type->regions[i+1].base,
345 type->regions[i+1].size)))) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000346 memblock_coalesce_regions(type, i, i+1);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000347 coalesced++;
348 }
349
350 if (coalesced)
351 return coalesced;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700352
353 /* If we are out of space, we fail. It's too late to resize the array
354 * but then this shouldn't have happened in the first place.
355 */
356 if (WARN_ON(type->cnt >= type->max))
Yinghai Lu95f72d12010-07-12 14:36:09 +1000357 return -1;
358
359 /* Couldn't coalesce the MEMBLOCK, so add it to the sorted table. */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000360 for (i = type->cnt - 1; i >= 0; i--) {
361 if (base < type->regions[i].base) {
362 type->regions[i+1].base = type->regions[i].base;
363 type->regions[i+1].size = type->regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000364 } else {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000365 type->regions[i+1].base = base;
366 type->regions[i+1].size = size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000367 break;
368 }
369 }
370
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000371 if (base < type->regions[0].base) {
372 type->regions[0].base = base;
373 type->regions[0].size = size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000374 }
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000375 type->cnt++;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000376
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700377 /* The array is full ? Try to resize it. If that fails, we undo
378 * our allocation and return an error
379 */
380 if (type->cnt == type->max && memblock_double_array(type)) {
381 type->cnt--;
382 return -1;
383 }
384
Yinghai Lu95f72d12010-07-12 14:36:09 +1000385 return 0;
386}
387
Yinghai Lu10d06432010-07-28 15:43:02 +1000388long __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000389{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000390 return memblock_add_region(&memblock.memory, base, size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000391
392}
393
Yinghai Lu10d06432010-07-28 15:43:02 +1000394static long __init_memblock __memblock_remove(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000395{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000396 phys_addr_t rgnbegin, rgnend;
397 phys_addr_t end = base + size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000398 int i;
399
400 rgnbegin = rgnend = 0; /* supress gcc warnings */
401
402 /* Find the region where (base, size) belongs to */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000403 for (i=0; i < type->cnt; i++) {
404 rgnbegin = type->regions[i].base;
405 rgnend = rgnbegin + type->regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000406
407 if ((rgnbegin <= base) && (end <= rgnend))
408 break;
409 }
410
411 /* Didn't find the region */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000412 if (i == type->cnt)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000413 return -1;
414
415 /* Check to see if we are removing entire region */
416 if ((rgnbegin == base) && (rgnend == end)) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000417 memblock_remove_region(type, i);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000418 return 0;
419 }
420
421 /* Check to see if region is matching at the front */
422 if (rgnbegin == base) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000423 type->regions[i].base = end;
424 type->regions[i].size -= size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000425 return 0;
426 }
427
428 /* Check to see if the region is matching at the end */
429 if (rgnend == end) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000430 type->regions[i].size -= size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000431 return 0;
432 }
433
434 /*
435 * We need to split the entry - adjust the current one to the
436 * beginging of the hole and add the region after hole.
437 */
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000438 type->regions[i].size = base - type->regions[i].base;
439 return memblock_add_region(type, end, rgnend - end);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000440}
441
Yinghai Lu10d06432010-07-28 15:43:02 +1000442long __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000443{
444 return __memblock_remove(&memblock.memory, base, size);
445}
446
Yinghai Lu3661ca62010-09-15 13:05:29 -0700447long __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000448{
449 return __memblock_remove(&memblock.reserved, base, size);
450}
451
Yinghai Lu3661ca62010-09-15 13:05:29 -0700452long __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000453{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000454 struct memblock_type *_rgn = &memblock.reserved;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000455
456 BUG_ON(0 == size);
457
458 return memblock_add_region(_rgn, base, size);
459}
460
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000461phys_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 +1000462{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000463 phys_addr_t found;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000464
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000465 /* We align the size to limit fragmentation. Without this, a lot of
466 * small allocs quickly eat up the whole reserve array on sparc
467 */
468 size = memblock_align_up(size, align);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000469
Benjamin Herrenschmidtfef501d2010-07-12 15:00:34 +1000470 found = memblock_find_base(size, align, 0, max_addr);
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000471 if (found != MEMBLOCK_ERROR &&
472 memblock_add_region(&memblock.reserved, found, size) >= 0)
473 return found;
474
475 return 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000476}
477
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000478phys_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 +1000479{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000480 phys_addr_t alloc;
481
482 alloc = __memblock_alloc_base(size, align, max_addr);
483
484 if (alloc == 0)
485 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
486 (unsigned long long) size, (unsigned long long) max_addr);
487
488 return alloc;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000489}
490
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000491phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000492{
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000493 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000494}
495
Yinghai Lu95f72d12010-07-12 14:36:09 +1000496
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000497/*
498 * Additional node-local allocators. Search for node memory is bottom up
499 * and walks memblock regions within that node bottom-up as well, but allocation
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700500 * within an memblock region is top-down. XXX I plan to fix that at some stage
501 *
502 * WARNING: Only available after early_node_map[] has been populated,
503 * on some architectures, that is after all the calls to add_active_range()
504 * have been done to populate it.
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000505 */
Yinghai Lu95f72d12010-07-12 14:36:09 +1000506
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000507phys_addr_t __weak __init memblock_nid_range(phys_addr_t start, phys_addr_t end, int *nid)
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700508{
Benjamin Herrenschmidtc196f762010-07-06 15:39:16 -0700509#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
510 /*
511 * This code originates from sparc which really wants use to walk by addresses
512 * and returns the nid. This is not very convenient for early_pfn_map[] users
513 * as the map isn't sorted yet, and it really wants to be walked by nid.
514 *
515 * For now, I implement the inefficient method below which walks the early
516 * map multiple times. Eventually we may want to use an ARCH config option
517 * to implement a completely different method for both case.
518 */
519 unsigned long start_pfn, end_pfn;
520 int i;
521
522 for (i = 0; i < MAX_NUMNODES; i++) {
523 get_pfn_range_for_nid(i, &start_pfn, &end_pfn);
524 if (start < PFN_PHYS(start_pfn) || start >= PFN_PHYS(end_pfn))
525 continue;
526 *nid = i;
527 return min(end, PFN_PHYS(end_pfn));
528 }
529#endif
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700530 *nid = 0;
531
532 return end;
533}
534
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000535static phys_addr_t __init memblock_alloc_nid_region(struct memblock_region *mp,
536 phys_addr_t size,
537 phys_addr_t align, int nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000538{
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000539 phys_addr_t start, end;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000540
541 start = mp->base;
542 end = start + mp->size;
543
544 start = memblock_align_up(start, align);
545 while (start < end) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000546 phys_addr_t this_end;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000547 int this_nid;
548
Benjamin Herrenschmidt35a1f0b2010-07-06 15:38:58 -0700549 this_end = memblock_nid_range(start, end, &this_nid);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000550 if (this_nid == nid) {
Benjamin Herrenschmidt3a9c2c82010-07-12 13:28:15 +1000551 phys_addr_t ret = memblock_find_region(start, this_end, size, align);
Benjamin Herrenschmidt4d629f92010-07-06 15:39:09 -0700552 if (ret != MEMBLOCK_ERROR &&
Benjamin Herrenschmidt3a9c2c82010-07-12 13:28:15 +1000553 memblock_add_region(&memblock.reserved, ret, size) >= 0)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000554 return ret;
555 }
556 start = this_end;
557 }
558
Benjamin Herrenschmidt4d629f92010-07-06 15:39:09 -0700559 return MEMBLOCK_ERROR;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000560}
561
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000562phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000563{
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000564 struct memblock_type *mem = &memblock.memory;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000565 int i;
566
567 BUG_ON(0 == size);
568
Benjamin Herrenschmidt7f219c72010-07-12 14:24:57 +1000569 /* We align the size to limit fragmentation. Without this, a lot of
570 * small allocs quickly eat up the whole reserve array on sparc
571 */
572 size = memblock_align_up(size, align);
573
Benjamin Herrenschmidtc3f72b52010-07-06 15:38:59 -0700574 /* We do a bottom-up search for a region with the right
575 * nid since that's easier considering how memblock_nid_range()
576 * works
577 */
Yinghai Lu95f72d12010-07-12 14:36:09 +1000578 for (i = 0; i < mem->cnt; i++) {
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000579 phys_addr_t ret = memblock_alloc_nid_region(&mem->regions[i],
Yinghai Lu95f72d12010-07-12 14:36:09 +1000580 size, align, nid);
Benjamin Herrenschmidt4d629f92010-07-06 15:39:09 -0700581 if (ret != MEMBLOCK_ERROR)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000582 return ret;
583 }
584
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700585 return 0;
586}
587
588phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
589{
590 phys_addr_t res = memblock_alloc_nid(size, align, nid);
591
592 if (res)
593 return res;
Benjamin Herrenschmidt918fe8d2010-07-06 15:39:18 -0700594 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000595}
596
Benjamin Herrenschmidt9d1e2492010-07-06 15:39:17 -0700597
598/*
599 * Remaining API functions
600 */
601
Yinghai Lu95f72d12010-07-12 14:36:09 +1000602/* You must call memblock_analyze() before this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000603phys_addr_t __init memblock_phys_mem_size(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000604{
Benjamin Herrenschmidt4734b592010-07-28 14:31:29 +1000605 return memblock.memory_size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000606}
607
Yinghai Lu10d06432010-07-28 15:43:02 +1000608phys_addr_t __init_memblock memblock_end_of_DRAM(void)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000609{
610 int idx = memblock.memory.cnt - 1;
611
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000612 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000613}
614
615/* You must call memblock_analyze() after this. */
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000616void __init memblock_enforce_memory_limit(phys_addr_t memory_limit)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000617{
618 unsigned long i;
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000619 phys_addr_t limit;
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000620 struct memblock_region *p;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000621
622 if (!memory_limit)
623 return;
624
625 /* Truncate the memblock regions to satisfy the memory limit. */
626 limit = memory_limit;
627 for (i = 0; i < memblock.memory.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000628 if (limit > memblock.memory.regions[i].size) {
629 limit -= memblock.memory.regions[i].size;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000630 continue;
631 }
632
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000633 memblock.memory.regions[i].size = limit;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000634 memblock.memory.cnt = i + 1;
635 break;
636 }
637
Yinghai Lu95f72d12010-07-12 14:36:09 +1000638 memory_limit = memblock_end_of_DRAM();
639
640 /* And truncate any reserves above the limit also. */
641 for (i = 0; i < memblock.reserved.cnt; i++) {
Benjamin Herrenschmidte3239ff2010-08-04 14:06:41 +1000642 p = &memblock.reserved.regions[i];
Yinghai Lu95f72d12010-07-12 14:36:09 +1000643
644 if (p->base > memory_limit)
645 p->size = 0;
646 else if ((p->base + p->size) > memory_limit)
647 p->size = memory_limit - p->base;
648
649 if (p->size == 0) {
650 memblock_remove_region(&memblock.reserved, i);
651 i--;
652 }
653 }
654}
655
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000656static int memblock_search(struct memblock_type *type, phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000657{
658 unsigned int left = 0, right = type->cnt;
659
660 do {
661 unsigned int mid = (right + left) / 2;
662
663 if (addr < type->regions[mid].base)
664 right = mid;
665 else if (addr >= (type->regions[mid].base +
666 type->regions[mid].size))
667 left = mid + 1;
668 else
669 return mid;
670 } while (left < right);
671 return -1;
672}
673
Benjamin Herrenschmidt2898cc42010-08-04 13:34:42 +1000674int __init memblock_is_reserved(phys_addr_t addr)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000675{
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000676 return memblock_search(&memblock.reserved, addr) != -1;
677}
Yinghai Lu95f72d12010-07-12 14:36:09 +1000678
Yinghai Lu3661ca62010-09-15 13:05:29 -0700679int __init_memblock memblock_is_memory(phys_addr_t addr)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000680{
681 return memblock_search(&memblock.memory, addr) != -1;
682}
683
Yinghai Lu3661ca62010-09-15 13:05:29 -0700684int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
Benjamin Herrenschmidt72d4b0b2010-08-04 14:38:47 +1000685{
686 int idx = memblock_search(&memblock.reserved, base);
687
688 if (idx == -1)
689 return 0;
690 return memblock.reserved.regions[idx].base <= base &&
691 (memblock.reserved.regions[idx].base +
692 memblock.reserved.regions[idx].size) >= (base + size);
Yinghai Lu95f72d12010-07-12 14:36:09 +1000693}
694
Yinghai Lu10d06432010-07-28 15:43:02 +1000695int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
Yinghai Lu95f72d12010-07-12 14:36:09 +1000696{
Benjamin Herrenschmidtf1c2c192010-08-04 14:17:17 +1000697 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
Yinghai Lu95f72d12010-07-12 14:36:09 +1000698}
699
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700700
Yinghai Lu3661ca62010-09-15 13:05:29 -0700701void __init_memblock memblock_set_current_limit(phys_addr_t limit)
Benjamin Herrenschmidte63075a2010-07-06 15:39:01 -0700702{
703 memblock.current_limit = limit;
704}
705
Yinghai Lu10d06432010-07-28 15:43:02 +1000706static void __init_memblock memblock_dump(struct memblock_type *region, char *name)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000707{
708 unsigned long long base, size;
709 int i;
710
711 pr_info(" %s.cnt = 0x%lx\n", name, region->cnt);
712
713 for (i = 0; i < region->cnt; i++) {
714 base = region->regions[i].base;
715 size = region->regions[i].size;
716
Yinghai Luea9e4372010-07-28 15:13:22 +1000717 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes\n",
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000718 name, i, base, base + size - 1, size);
719 }
720}
721
Yinghai Lu10d06432010-07-28 15:43:02 +1000722void __init_memblock memblock_dump_all(void)
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000723{
724 if (!memblock_debug)
725 return;
726
727 pr_info("MEMBLOCK configuration:\n");
728 pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size);
729
730 memblock_dump(&memblock.memory, "memory");
731 memblock_dump(&memblock.reserved, "reserved");
732}
733
734void __init memblock_analyze(void)
735{
736 int i;
737
738 /* Check marker in the unused last array entry */
739 WARN_ON(memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS].base
740 != (phys_addr_t)RED_INACTIVE);
741 WARN_ON(memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS].base
742 != (phys_addr_t)RED_INACTIVE);
743
744 memblock.memory_size = 0;
745
746 for (i = 0; i < memblock.memory.cnt; i++)
747 memblock.memory_size += memblock.memory.regions[i].size;
Benjamin Herrenschmidt142b45a2010-07-06 15:39:13 -0700748
749 /* We allow resizing from there */
750 memblock_can_resize = 1;
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000751}
752
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700753void __init memblock_init(void)
754{
Jeremy Fitzhardinge236260b2010-10-06 15:52:29 -0700755 static int init_done __initdata = 0;
756
757 if (init_done)
758 return;
759 init_done = 1;
760
Benjamin Herrenschmidt7590abe2010-07-06 15:39:10 -0700761 /* Hookup the initial arrays */
762 memblock.memory.regions = memblock_memory_init_regions;
763 memblock.memory.max = INIT_MEMBLOCK_REGIONS;
764 memblock.reserved.regions = memblock_reserved_init_regions;
765 memblock.reserved.max = INIT_MEMBLOCK_REGIONS;
766
767 /* Write a marker in the unused last array entry */
768 memblock.memory.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
769 memblock.reserved.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
770
771 /* Create a dummy zero size MEMBLOCK which will get coalesced away later.
772 * This simplifies the memblock_add() code below...
773 */
774 memblock.memory.regions[0].base = 0;
775 memblock.memory.regions[0].size = 0;
776 memblock.memory.cnt = 1;
777
778 /* Ditto. */
779 memblock.reserved.regions[0].base = 0;
780 memblock.reserved.regions[0].size = 0;
781 memblock.reserved.cnt = 1;
782
783 memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE;
784}
785
Benjamin Herrenschmidt6ed311b2010-07-12 14:36:48 +1000786static int __init early_memblock(char *p)
787{
788 if (p && strstr(p, "debug"))
789 memblock_debug = 1;
790 return 0;
791}
792early_param("memblock", early_memblock);
793
Yinghai Lu10d06432010-07-28 15:43:02 +1000794#if defined(CONFIG_DEBUG_FS) && !defined(ARCH_DISCARD_MEMBLOCK)
Benjamin Herrenschmidt6d03b882010-07-06 15:39:19 -0700795
796static int memblock_debug_show(struct seq_file *m, void *private)
797{
798 struct memblock_type *type = m->private;
799 struct memblock_region *reg;
800 int i;
801
802 for (i = 0; i < type->cnt; i++) {
803 reg = &type->regions[i];
804 seq_printf(m, "%4d: ", i);
805 if (sizeof(phys_addr_t) == 4)
806 seq_printf(m, "0x%08lx..0x%08lx\n",
807 (unsigned long)reg->base,
808 (unsigned long)(reg->base + reg->size - 1));
809 else
810 seq_printf(m, "0x%016llx..0x%016llx\n",
811 (unsigned long long)reg->base,
812 (unsigned long long)(reg->base + reg->size - 1));
813
814 }
815 return 0;
816}
817
818static int memblock_debug_open(struct inode *inode, struct file *file)
819{
820 return single_open(file, memblock_debug_show, inode->i_private);
821}
822
823static const struct file_operations memblock_debug_fops = {
824 .open = memblock_debug_open,
825 .read = seq_read,
826 .llseek = seq_lseek,
827 .release = single_release,
828};
829
830static int __init memblock_init_debugfs(void)
831{
832 struct dentry *root = debugfs_create_dir("memblock", NULL);
833 if (!root)
834 return -ENXIO;
835 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
836 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
837
838 return 0;
839}
840__initcall(memblock_init_debugfs);
841
842#endif /* CONFIG_DEBUG_FS */