ext4: fix race between sync and completed io work
[linux-2.6-block.git] / mm / memblock.c
CommitLineData
95f72d1e
YL
1/*
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>
142b45a7 14#include <linux/slab.h>
95f72d1e
YL
15#include <linux/init.h>
16#include <linux/bitops.h>
449e8df3 17#include <linux/poison.h>
c196f76f 18#include <linux/pfn.h>
6d03b885
BH
19#include <linux/debugfs.h>
20#include <linux/seq_file.h>
95f72d1e
YL
21#include <linux/memblock.h>
22
fe091c20
TH
23static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
24static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
25
26struct memblock memblock __initdata_memblock = {
27 .memory.regions = memblock_memory_init_regions,
28 .memory.cnt = 1, /* empty dummy entry */
29 .memory.max = INIT_MEMBLOCK_REGIONS,
30
31 .reserved.regions = memblock_reserved_init_regions,
32 .reserved.cnt = 1, /* empty dummy entry */
33 .reserved.max = INIT_MEMBLOCK_REGIONS,
34
35 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
36};
95f72d1e 37
10d06439 38int memblock_debug __initdata_memblock;
1aadc056 39static int memblock_can_resize __initdata_memblock;
95f72d1e 40
142b45a7
BH
41/* inline so we don't get a warning when pr_debug is compiled out */
42static inline const char *memblock_type_name(struct memblock_type *type)
43{
44 if (type == &memblock.memory)
45 return "memory";
46 else if (type == &memblock.reserved)
47 return "reserved";
48 else
49 return "unknown";
50}
51
eb18f1b5
TH
52/* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
53static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
54{
55 return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
56}
57
6ed311b2
BH
58/*
59 * Address comparison utilities
60 */
10d06439 61static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
2898cc4c 62 phys_addr_t base2, phys_addr_t size2)
95f72d1e
YL
63{
64 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
65}
66
2d7d3eb2
HS
67static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
68 phys_addr_t base, phys_addr_t size)
6ed311b2
BH
69{
70 unsigned long i;
71
72 for (i = 0; i < type->cnt; i++) {
73 phys_addr_t rgnbase = type->regions[i].base;
74 phys_addr_t rgnsize = type->regions[i].size;
75 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
76 break;
77 }
78
79 return (i < type->cnt) ? i : -1;
80}
81
7bd0b0f0
TH
82/**
83 * memblock_find_in_range_node - find free area in given range and node
84 * @start: start of candidate range
85 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
86 * @size: size of free area to find
87 * @align: alignment of free area to find
88 * @nid: nid of the free area to find, %MAX_NUMNODES for any node
89 *
90 * Find @size free area aligned to @align in the specified range and node.
91 *
92 * RETURNS:
93 * Found address on success, %0 on failure.
6ed311b2 94 */
7bd0b0f0
TH
95phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t start,
96 phys_addr_t end, phys_addr_t size,
97 phys_addr_t align, int nid)
6ed311b2 98{
7bd0b0f0
TH
99 phys_addr_t this_start, this_end, cand;
100 u64 i;
6ed311b2 101
7bd0b0f0
TH
102 /* align @size to avoid excessive fragmentation on reserved array */
103 size = round_up(size, align);
104
105 /* pump up @end */
106 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
107 end = memblock.current_limit;
f1af98c7 108
5d53cb27
TH
109 /* avoid allocating the first page */
110 start = max_t(phys_addr_t, start, PAGE_SIZE);
7bd0b0f0 111 end = max(start, end);
f1af98c7 112
7bd0b0f0
TH
113 for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) {
114 this_start = clamp(this_start, start, end);
115 this_end = clamp(this_end, start, end);
6ed311b2 116
5d53cb27
TH
117 if (this_end < size)
118 continue;
119
7bd0b0f0
TH
120 cand = round_down(this_end - size, align);
121 if (cand >= this_start)
122 return cand;
123 }
1f5026a7 124 return 0;
6ed311b2
BH
125}
126
7bd0b0f0
TH
127/**
128 * memblock_find_in_range - find free area in given range
129 * @start: start of candidate range
130 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
131 * @size: size of free area to find
132 * @align: alignment of free area to find
133 *
134 * Find @size free area aligned to @align in the specified range.
135 *
136 * RETURNS:
137 * Found address on success, %0 on failure.
fc769a8e 138 */
7bd0b0f0
TH
139phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
140 phys_addr_t end, phys_addr_t size,
141 phys_addr_t align)
6ed311b2 142{
7bd0b0f0
TH
143 return memblock_find_in_range_node(start, end, size, align,
144 MAX_NUMNODES);
6ed311b2
BH
145}
146
7950c407
YL
147/*
148 * Free memblock.reserved.regions
149 */
150int __init_memblock memblock_free_reserved_regions(void)
151{
152 if (memblock.reserved.regions == memblock_reserved_init_regions)
153 return 0;
154
155 return memblock_free(__pa(memblock.reserved.regions),
156 sizeof(struct memblock_region) * memblock.reserved.max);
157}
158
159/*
160 * Reserve memblock.reserved.regions
161 */
162int __init_memblock memblock_reserve_reserved_regions(void)
163{
164 if (memblock.reserved.regions == memblock_reserved_init_regions)
165 return 0;
166
167 return memblock_reserve(__pa(memblock.reserved.regions),
168 sizeof(struct memblock_region) * memblock.reserved.max);
169}
170
10d06439 171static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
95f72d1e 172{
1440c4e2 173 type->total_size -= type->regions[r].size;
7c0caeb8
TH
174 memmove(&type->regions[r], &type->regions[r + 1],
175 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
e3239ff9 176 type->cnt--;
95f72d1e 177
8f7a6605
BH
178 /* Special case for empty arrays */
179 if (type->cnt == 0) {
1440c4e2 180 WARN_ON(type->total_size != 0);
8f7a6605
BH
181 type->cnt = 1;
182 type->regions[0].base = 0;
183 type->regions[0].size = 0;
7c0caeb8 184 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
8f7a6605 185 }
95f72d1e
YL
186}
187
10d06439 188static int __init_memblock memblock_double_array(struct memblock_type *type)
142b45a7
BH
189{
190 struct memblock_region *new_array, *old_array;
191 phys_addr_t old_size, new_size, addr;
192 int use_slab = slab_is_available();
193
194 /* We don't allow resizing until we know about the reserved regions
195 * of memory that aren't suitable for allocation
196 */
197 if (!memblock_can_resize)
198 return -1;
199
142b45a7
BH
200 /* Calculate new doubled size */
201 old_size = type->max * sizeof(struct memblock_region);
202 new_size = old_size << 1;
203
204 /* Try to find some space for it.
205 *
206 * WARNING: We assume that either slab_is_available() and we use it or
207 * we use MEMBLOCK for allocations. That means that this is unsafe to use
208 * when bootmem is currently active (unless bootmem itself is implemented
209 * on top of MEMBLOCK which isn't the case yet)
210 *
211 * This should however not be an issue for now, as we currently only
212 * call into MEMBLOCK while it's still active, or much later when slab is
213 * active for memory hotplug operations
214 */
215 if (use_slab) {
216 new_array = kmalloc(new_size, GFP_KERNEL);
1f5026a7 217 addr = new_array ? __pa(new_array) : 0;
142b45a7 218 } else
fc769a8e 219 addr = memblock_find_in_range(0, MEMBLOCK_ALLOC_ACCESSIBLE, new_size, sizeof(phys_addr_t));
1f5026a7 220 if (!addr) {
142b45a7
BH
221 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
222 memblock_type_name(type), type->max, type->max * 2);
223 return -1;
224 }
225 new_array = __va(addr);
226
ea9e4376
YL
227 memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]",
228 memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1);
229
142b45a7
BH
230 /* Found space, we now need to move the array over before
231 * we add the reserved region since it may be our reserved
232 * array itself that is full.
233 */
234 memcpy(new_array, type->regions, old_size);
235 memset(new_array + type->max, 0, old_size);
236 old_array = type->regions;
237 type->regions = new_array;
238 type->max <<= 1;
239
240 /* If we use SLAB that's it, we are done */
241 if (use_slab)
242 return 0;
243
244 /* Add the new reserved region now. Should not fail ! */
9c8c27e2 245 BUG_ON(memblock_reserve(addr, new_size));
142b45a7
BH
246
247 /* If the array wasn't our static init one, then free it. We only do
248 * that before SLAB is available as later on, we don't know whether
249 * to use kfree or free_bootmem_pages(). Shouldn't be a big deal
250 * anyways
251 */
252 if (old_array != memblock_memory_init_regions &&
253 old_array != memblock_reserved_init_regions)
254 memblock_free(__pa(old_array), old_size);
255
256 return 0;
257}
258
784656f9
TH
259/**
260 * memblock_merge_regions - merge neighboring compatible regions
261 * @type: memblock type to scan
262 *
263 * Scan @type and merge neighboring compatible regions.
264 */
265static void __init_memblock memblock_merge_regions(struct memblock_type *type)
95f72d1e 266{
784656f9 267 int i = 0;
95f72d1e 268
784656f9
TH
269 /* cnt never goes below 1 */
270 while (i < type->cnt - 1) {
271 struct memblock_region *this = &type->regions[i];
272 struct memblock_region *next = &type->regions[i + 1];
95f72d1e 273
7c0caeb8
TH
274 if (this->base + this->size != next->base ||
275 memblock_get_region_node(this) !=
276 memblock_get_region_node(next)) {
784656f9
TH
277 BUG_ON(this->base + this->size > next->base);
278 i++;
279 continue;
8f7a6605
BH
280 }
281
784656f9
TH
282 this->size += next->size;
283 memmove(next, next + 1, (type->cnt - (i + 1)) * sizeof(*next));
284 type->cnt--;
95f72d1e 285 }
784656f9 286}
95f72d1e 287
784656f9
TH
288/**
289 * memblock_insert_region - insert new memblock region
290 * @type: memblock type to insert into
291 * @idx: index for the insertion point
292 * @base: base address of the new region
293 * @size: size of the new region
294 *
295 * Insert new memblock region [@base,@base+@size) into @type at @idx.
296 * @type must already have extra room to accomodate the new region.
297 */
298static void __init_memblock memblock_insert_region(struct memblock_type *type,
299 int idx, phys_addr_t base,
7c0caeb8 300 phys_addr_t size, int nid)
784656f9
TH
301{
302 struct memblock_region *rgn = &type->regions[idx];
303
304 BUG_ON(type->cnt >= type->max);
305 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
306 rgn->base = base;
307 rgn->size = size;
7c0caeb8 308 memblock_set_region_node(rgn, nid);
784656f9 309 type->cnt++;
1440c4e2 310 type->total_size += size;
784656f9
TH
311}
312
313/**
314 * memblock_add_region - add new memblock region
315 * @type: memblock type to add new region into
316 * @base: base address of the new region
317 * @size: size of the new region
7fb0bc3f 318 * @nid: nid of the new region
784656f9
TH
319 *
320 * Add new memblock region [@base,@base+@size) into @type. The new region
321 * is allowed to overlap with existing ones - overlaps don't affect already
322 * existing regions. @type is guaranteed to be minimal (all neighbouring
323 * compatible regions are merged) after the addition.
324 *
325 * RETURNS:
326 * 0 on success, -errno on failure.
327 */
581adcbe 328static int __init_memblock memblock_add_region(struct memblock_type *type,
7fb0bc3f 329 phys_addr_t base, phys_addr_t size, int nid)
784656f9
TH
330{
331 bool insert = false;
eb18f1b5
TH
332 phys_addr_t obase = base;
333 phys_addr_t end = base + memblock_cap_size(base, &size);
784656f9
TH
334 int i, nr_new;
335
336 /* special case for empty array */
337 if (type->regions[0].size == 0) {
1440c4e2 338 WARN_ON(type->cnt != 1 || type->total_size);
8f7a6605
BH
339 type->regions[0].base = base;
340 type->regions[0].size = size;
7fb0bc3f 341 memblock_set_region_node(&type->regions[0], nid);
1440c4e2 342 type->total_size = size;
8f7a6605 343 return 0;
95f72d1e 344 }
784656f9
TH
345repeat:
346 /*
347 * The following is executed twice. Once with %false @insert and
348 * then with %true. The first counts the number of regions needed
349 * to accomodate the new area. The second actually inserts them.
142b45a7 350 */
784656f9
TH
351 base = obase;
352 nr_new = 0;
95f72d1e 353
784656f9
TH
354 for (i = 0; i < type->cnt; i++) {
355 struct memblock_region *rgn = &type->regions[i];
356 phys_addr_t rbase = rgn->base;
357 phys_addr_t rend = rbase + rgn->size;
358
359 if (rbase >= end)
95f72d1e 360 break;
784656f9
TH
361 if (rend <= base)
362 continue;
363 /*
364 * @rgn overlaps. If it separates the lower part of new
365 * area, insert that portion.
366 */
367 if (rbase > base) {
368 nr_new++;
369 if (insert)
370 memblock_insert_region(type, i++, base,
7fb0bc3f 371 rbase - base, nid);
95f72d1e 372 }
784656f9
TH
373 /* area below @rend is dealt with, forget about it */
374 base = min(rend, end);
95f72d1e 375 }
784656f9
TH
376
377 /* insert the remaining portion */
378 if (base < end) {
379 nr_new++;
380 if (insert)
7fb0bc3f 381 memblock_insert_region(type, i, base, end - base, nid);
95f72d1e 382 }
95f72d1e 383
784656f9
TH
384 /*
385 * If this was the first round, resize array and repeat for actual
386 * insertions; otherwise, merge and return.
142b45a7 387 */
784656f9
TH
388 if (!insert) {
389 while (type->cnt + nr_new > type->max)
390 if (memblock_double_array(type) < 0)
391 return -ENOMEM;
392 insert = true;
393 goto repeat;
394 } else {
395 memblock_merge_regions(type);
396 return 0;
142b45a7 397 }
95f72d1e
YL
398}
399
7fb0bc3f
TH
400int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
401 int nid)
402{
403 return memblock_add_region(&memblock.memory, base, size, nid);
404}
405
581adcbe 406int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
95f72d1e 407{
7fb0bc3f 408 return memblock_add_region(&memblock.memory, base, size, MAX_NUMNODES);
95f72d1e
YL
409}
410
6a9ceb31
TH
411/**
412 * memblock_isolate_range - isolate given range into disjoint memblocks
413 * @type: memblock type to isolate range for
414 * @base: base of range to isolate
415 * @size: size of range to isolate
416 * @start_rgn: out parameter for the start of isolated region
417 * @end_rgn: out parameter for the end of isolated region
418 *
419 * Walk @type and ensure that regions don't cross the boundaries defined by
420 * [@base,@base+@size). Crossing regions are split at the boundaries,
421 * which may create at most two more regions. The index of the first
422 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
423 *
424 * RETURNS:
425 * 0 on success, -errno on failure.
426 */
427static int __init_memblock memblock_isolate_range(struct memblock_type *type,
428 phys_addr_t base, phys_addr_t size,
429 int *start_rgn, int *end_rgn)
430{
eb18f1b5 431 phys_addr_t end = base + memblock_cap_size(base, &size);
6a9ceb31
TH
432 int i;
433
434 *start_rgn = *end_rgn = 0;
435
436 /* we'll create at most two more regions */
437 while (type->cnt + 2 > type->max)
438 if (memblock_double_array(type) < 0)
439 return -ENOMEM;
440
441 for (i = 0; i < type->cnt; i++) {
442 struct memblock_region *rgn = &type->regions[i];
443 phys_addr_t rbase = rgn->base;
444 phys_addr_t rend = rbase + rgn->size;
445
446 if (rbase >= end)
447 break;
448 if (rend <= base)
449 continue;
450
451 if (rbase < base) {
452 /*
453 * @rgn intersects from below. Split and continue
454 * to process the next region - the new top half.
455 */
456 rgn->base = base;
1440c4e2
TH
457 rgn->size -= base - rbase;
458 type->total_size -= base - rbase;
6a9ceb31 459 memblock_insert_region(type, i, rbase, base - rbase,
71936180 460 memblock_get_region_node(rgn));
6a9ceb31
TH
461 } else if (rend > end) {
462 /*
463 * @rgn intersects from above. Split and redo the
464 * current region - the new bottom half.
465 */
466 rgn->base = end;
1440c4e2
TH
467 rgn->size -= end - rbase;
468 type->total_size -= end - rbase;
6a9ceb31 469 memblock_insert_region(type, i--, rbase, end - rbase,
71936180 470 memblock_get_region_node(rgn));
6a9ceb31
TH
471 } else {
472 /* @rgn is fully contained, record it */
473 if (!*end_rgn)
474 *start_rgn = i;
475 *end_rgn = i + 1;
476 }
477 }
478
479 return 0;
480}
6a9ceb31 481
581adcbe
TH
482static int __init_memblock __memblock_remove(struct memblock_type *type,
483 phys_addr_t base, phys_addr_t size)
95f72d1e 484{
71936180
TH
485 int start_rgn, end_rgn;
486 int i, ret;
95f72d1e 487
71936180
TH
488 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
489 if (ret)
490 return ret;
95f72d1e 491
71936180
TH
492 for (i = end_rgn - 1; i >= start_rgn; i--)
493 memblock_remove_region(type, i);
8f7a6605 494 return 0;
95f72d1e
YL
495}
496
581adcbe 497int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
95f72d1e
YL
498{
499 return __memblock_remove(&memblock.memory, base, size);
500}
501
581adcbe 502int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
95f72d1e 503{
24aa0788 504 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
a150439c
PA
505 (unsigned long long)base,
506 (unsigned long long)base + size,
507 (void *)_RET_IP_);
24aa0788 508
95f72d1e
YL
509 return __memblock_remove(&memblock.reserved, base, size);
510}
511
581adcbe 512int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
95f72d1e 513{
e3239ff9 514 struct memblock_type *_rgn = &memblock.reserved;
95f72d1e 515
24aa0788 516 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] %pF\n",
a150439c
PA
517 (unsigned long long)base,
518 (unsigned long long)base + size,
519 (void *)_RET_IP_);
95f72d1e
YL
520 BUG_ON(0 == size);
521
7fb0bc3f 522 return memblock_add_region(_rgn, base, size, MAX_NUMNODES);
95f72d1e
YL
523}
524
35fd0808
TH
525/**
526 * __next_free_mem_range - next function for for_each_free_mem_range()
527 * @idx: pointer to u64 loop variable
528 * @nid: nid: node selector, %MAX_NUMNODES for all nodes
529 * @p_start: ptr to phys_addr_t for start address of the range, can be %NULL
530 * @p_end: ptr to phys_addr_t for end address of the range, can be %NULL
531 * @p_nid: ptr to int for nid of the range, can be %NULL
532 *
533 * Find the first free area from *@idx which matches @nid, fill the out
534 * parameters, and update *@idx for the next iteration. The lower 32bit of
535 * *@idx contains index into memory region and the upper 32bit indexes the
536 * areas before each reserved region. For example, if reserved regions
537 * look like the following,
538 *
539 * 0:[0-16), 1:[32-48), 2:[128-130)
540 *
541 * The upper 32bit indexes the following regions.
542 *
543 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
544 *
545 * As both region arrays are sorted, the function advances the two indices
546 * in lockstep and returns each intersection.
547 */
548void __init_memblock __next_free_mem_range(u64 *idx, int nid,
549 phys_addr_t *out_start,
550 phys_addr_t *out_end, int *out_nid)
551{
552 struct memblock_type *mem = &memblock.memory;
553 struct memblock_type *rsv = &memblock.reserved;
554 int mi = *idx & 0xffffffff;
555 int ri = *idx >> 32;
556
557 for ( ; mi < mem->cnt; mi++) {
558 struct memblock_region *m = &mem->regions[mi];
559 phys_addr_t m_start = m->base;
560 phys_addr_t m_end = m->base + m->size;
561
562 /* only memory regions are associated with nodes, check it */
563 if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m))
564 continue;
565
566 /* scan areas before each reservation for intersection */
567 for ( ; ri < rsv->cnt + 1; ri++) {
568 struct memblock_region *r = &rsv->regions[ri];
569 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
570 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
571
572 /* if ri advanced past mi, break out to advance mi */
573 if (r_start >= m_end)
574 break;
575 /* if the two regions intersect, we're done */
576 if (m_start < r_end) {
577 if (out_start)
578 *out_start = max(m_start, r_start);
579 if (out_end)
580 *out_end = min(m_end, r_end);
581 if (out_nid)
582 *out_nid = memblock_get_region_node(m);
583 /*
584 * The region which ends first is advanced
585 * for the next iteration.
586 */
587 if (m_end <= r_end)
588 mi++;
589 else
590 ri++;
591 *idx = (u32)mi | (u64)ri << 32;
592 return;
593 }
594 }
595 }
596
597 /* signal end of iteration */
598 *idx = ULLONG_MAX;
599}
600
7bd0b0f0
TH
601/**
602 * __next_free_mem_range_rev - next function for for_each_free_mem_range_reverse()
603 * @idx: pointer to u64 loop variable
604 * @nid: nid: node selector, %MAX_NUMNODES for all nodes
605 * @p_start: ptr to phys_addr_t for start address of the range, can be %NULL
606 * @p_end: ptr to phys_addr_t for end address of the range, can be %NULL
607 * @p_nid: ptr to int for nid of the range, can be %NULL
608 *
609 * Reverse of __next_free_mem_range().
610 */
611void __init_memblock __next_free_mem_range_rev(u64 *idx, int nid,
612 phys_addr_t *out_start,
613 phys_addr_t *out_end, int *out_nid)
614{
615 struct memblock_type *mem = &memblock.memory;
616 struct memblock_type *rsv = &memblock.reserved;
617 int mi = *idx & 0xffffffff;
618 int ri = *idx >> 32;
619
620 if (*idx == (u64)ULLONG_MAX) {
621 mi = mem->cnt - 1;
622 ri = rsv->cnt;
623 }
624
625 for ( ; mi >= 0; mi--) {
626 struct memblock_region *m = &mem->regions[mi];
627 phys_addr_t m_start = m->base;
628 phys_addr_t m_end = m->base + m->size;
629
630 /* only memory regions are associated with nodes, check it */
631 if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m))
632 continue;
633
634 /* scan areas before each reservation for intersection */
635 for ( ; ri >= 0; ri--) {
636 struct memblock_region *r = &rsv->regions[ri];
637 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
638 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
639
640 /* if ri advanced past mi, break out to advance mi */
641 if (r_end <= m_start)
642 break;
643 /* if the two regions intersect, we're done */
644 if (m_end > r_start) {
645 if (out_start)
646 *out_start = max(m_start, r_start);
647 if (out_end)
648 *out_end = min(m_end, r_end);
649 if (out_nid)
650 *out_nid = memblock_get_region_node(m);
651
652 if (m_start >= r_start)
653 mi--;
654 else
655 ri--;
656 *idx = (u32)mi | (u64)ri << 32;
657 return;
658 }
659 }
660 }
661
662 *idx = ULLONG_MAX;
663}
664
7c0caeb8
TH
665#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
666/*
667 * Common iterator interface used to define for_each_mem_range().
668 */
669void __init_memblock __next_mem_pfn_range(int *idx, int nid,
670 unsigned long *out_start_pfn,
671 unsigned long *out_end_pfn, int *out_nid)
672{
673 struct memblock_type *type = &memblock.memory;
674 struct memblock_region *r;
675
676 while (++*idx < type->cnt) {
677 r = &type->regions[*idx];
678
679 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
680 continue;
681 if (nid == MAX_NUMNODES || nid == r->nid)
682 break;
683 }
684 if (*idx >= type->cnt) {
685 *idx = -1;
686 return;
687 }
688
689 if (out_start_pfn)
690 *out_start_pfn = PFN_UP(r->base);
691 if (out_end_pfn)
692 *out_end_pfn = PFN_DOWN(r->base + r->size);
693 if (out_nid)
694 *out_nid = r->nid;
695}
696
697/**
698 * memblock_set_node - set node ID on memblock regions
699 * @base: base of area to set node ID for
700 * @size: size of area to set node ID for
701 * @nid: node ID to set
702 *
703 * Set the nid of memblock memory regions in [@base,@base+@size) to @nid.
704 * Regions which cross the area boundaries are split as necessary.
705 *
706 * RETURNS:
707 * 0 on success, -errno on failure.
708 */
709int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
710 int nid)
711{
712 struct memblock_type *type = &memblock.memory;
6a9ceb31
TH
713 int start_rgn, end_rgn;
714 int i, ret;
7c0caeb8 715
6a9ceb31
TH
716 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
717 if (ret)
718 return ret;
7c0caeb8 719
6a9ceb31
TH
720 for (i = start_rgn; i < end_rgn; i++)
721 type->regions[i].nid = nid;
7c0caeb8
TH
722
723 memblock_merge_regions(type);
724 return 0;
725}
726#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
727
7bd0b0f0
TH
728static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
729 phys_addr_t align, phys_addr_t max_addr,
730 int nid)
95f72d1e 731{
6ed311b2 732 phys_addr_t found;
95f72d1e 733
7bd0b0f0 734 found = memblock_find_in_range_node(0, max_addr, size, align, nid);
9c8c27e2 735 if (found && !memblock_reserve(found, size))
6ed311b2 736 return found;
95f72d1e 737
6ed311b2 738 return 0;
95f72d1e
YL
739}
740
7bd0b0f0
TH
741phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
742{
743 return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
744}
745
746phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
747{
748 return memblock_alloc_base_nid(size, align, max_addr, MAX_NUMNODES);
749}
750
6ed311b2 751phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
95f72d1e 752{
6ed311b2
BH
753 phys_addr_t alloc;
754
755 alloc = __memblock_alloc_base(size, align, max_addr);
756
757 if (alloc == 0)
758 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
759 (unsigned long long) size, (unsigned long long) max_addr);
760
761 return alloc;
95f72d1e
YL
762}
763
6ed311b2 764phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
95f72d1e 765{
6ed311b2
BH
766 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
767}
95f72d1e 768
9d1e2492
BH
769phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
770{
771 phys_addr_t res = memblock_alloc_nid(size, align, nid);
772
773 if (res)
774 return res;
15fb0972 775 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
95f72d1e
YL
776}
777
9d1e2492
BH
778
779/*
780 * Remaining API functions
781 */
782
2898cc4c 783phys_addr_t __init memblock_phys_mem_size(void)
95f72d1e 784{
1440c4e2 785 return memblock.memory.total_size;
95f72d1e
YL
786}
787
0a93ebef
SR
788/* lowest address */
789phys_addr_t __init_memblock memblock_start_of_DRAM(void)
790{
791 return memblock.memory.regions[0].base;
792}
793
10d06439 794phys_addr_t __init_memblock memblock_end_of_DRAM(void)
95f72d1e
YL
795{
796 int idx = memblock.memory.cnt - 1;
797
e3239ff9 798 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
95f72d1e
YL
799}
800
c0ce8fef 801void __init memblock_enforce_memory_limit(phys_addr_t limit)
95f72d1e
YL
802{
803 unsigned long i;
c0ce8fef 804 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
95f72d1e 805
c0ce8fef 806 if (!limit)
95f72d1e
YL
807 return;
808
c0ce8fef 809 /* find out max address */
95f72d1e 810 for (i = 0; i < memblock.memory.cnt; i++) {
c0ce8fef 811 struct memblock_region *r = &memblock.memory.regions[i];
95f72d1e 812
c0ce8fef
TH
813 if (limit <= r->size) {
814 max_addr = r->base + limit;
815 break;
95f72d1e 816 }
c0ce8fef 817 limit -= r->size;
95f72d1e 818 }
c0ce8fef
TH
819
820 /* truncate both memory and reserved regions */
821 __memblock_remove(&memblock.memory, max_addr, (phys_addr_t)ULLONG_MAX);
822 __memblock_remove(&memblock.reserved, max_addr, (phys_addr_t)ULLONG_MAX);
95f72d1e
YL
823}
824
cd79481d 825static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
72d4b0b4
BH
826{
827 unsigned int left = 0, right = type->cnt;
828
829 do {
830 unsigned int mid = (right + left) / 2;
831
832 if (addr < type->regions[mid].base)
833 right = mid;
834 else if (addr >= (type->regions[mid].base +
835 type->regions[mid].size))
836 left = mid + 1;
837 else
838 return mid;
839 } while (left < right);
840 return -1;
841}
842
2898cc4c 843int __init memblock_is_reserved(phys_addr_t addr)
95f72d1e 844{
72d4b0b4
BH
845 return memblock_search(&memblock.reserved, addr) != -1;
846}
95f72d1e 847
3661ca66 848int __init_memblock memblock_is_memory(phys_addr_t addr)
72d4b0b4
BH
849{
850 return memblock_search(&memblock.memory, addr) != -1;
851}
852
3661ca66 853int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
72d4b0b4 854{
abb65272 855 int idx = memblock_search(&memblock.memory, base);
eb18f1b5 856 phys_addr_t end = base + memblock_cap_size(base, &size);
72d4b0b4
BH
857
858 if (idx == -1)
859 return 0;
abb65272
TV
860 return memblock.memory.regions[idx].base <= base &&
861 (memblock.memory.regions[idx].base +
eb18f1b5 862 memblock.memory.regions[idx].size) >= end;
95f72d1e
YL
863}
864
10d06439 865int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
95f72d1e 866{
eb18f1b5 867 memblock_cap_size(base, &size);
f1c2c19c 868 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
95f72d1e
YL
869}
870
e63075a3 871
3661ca66 872void __init_memblock memblock_set_current_limit(phys_addr_t limit)
e63075a3
BH
873{
874 memblock.current_limit = limit;
875}
876
7c0caeb8 877static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
6ed311b2
BH
878{
879 unsigned long long base, size;
880 int i;
881
7c0caeb8 882 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
6ed311b2 883
7c0caeb8
TH
884 for (i = 0; i < type->cnt; i++) {
885 struct memblock_region *rgn = &type->regions[i];
886 char nid_buf[32] = "";
887
888 base = rgn->base;
889 size = rgn->size;
890#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
891 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
892 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
893 memblock_get_region_node(rgn));
894#endif
895 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s\n",
896 name, i, base, base + size - 1, size, nid_buf);
6ed311b2
BH
897 }
898}
899
4ff7b82f 900void __init_memblock __memblock_dump_all(void)
6ed311b2 901{
6ed311b2 902 pr_info("MEMBLOCK configuration:\n");
1440c4e2
TH
903 pr_info(" memory size = %#llx reserved size = %#llx\n",
904 (unsigned long long)memblock.memory.total_size,
905 (unsigned long long)memblock.reserved.total_size);
6ed311b2
BH
906
907 memblock_dump(&memblock.memory, "memory");
908 memblock_dump(&memblock.reserved, "reserved");
909}
910
1aadc056 911void __init memblock_allow_resize(void)
6ed311b2 912{
142b45a7 913 memblock_can_resize = 1;
6ed311b2
BH
914}
915
6ed311b2
BH
916static int __init early_memblock(char *p)
917{
918 if (p && strstr(p, "debug"))
919 memblock_debug = 1;
920 return 0;
921}
922early_param("memblock", early_memblock);
923
c378ddd5 924#if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
6d03b885
BH
925
926static int memblock_debug_show(struct seq_file *m, void *private)
927{
928 struct memblock_type *type = m->private;
929 struct memblock_region *reg;
930 int i;
931
932 for (i = 0; i < type->cnt; i++) {
933 reg = &type->regions[i];
934 seq_printf(m, "%4d: ", i);
935 if (sizeof(phys_addr_t) == 4)
936 seq_printf(m, "0x%08lx..0x%08lx\n",
937 (unsigned long)reg->base,
938 (unsigned long)(reg->base + reg->size - 1));
939 else
940 seq_printf(m, "0x%016llx..0x%016llx\n",
941 (unsigned long long)reg->base,
942 (unsigned long long)(reg->base + reg->size - 1));
943
944 }
945 return 0;
946}
947
948static int memblock_debug_open(struct inode *inode, struct file *file)
949{
950 return single_open(file, memblock_debug_show, inode->i_private);
951}
952
953static const struct file_operations memblock_debug_fops = {
954 .open = memblock_debug_open,
955 .read = seq_read,
956 .llseek = seq_lseek,
957 .release = single_release,
958};
959
960static int __init memblock_init_debugfs(void)
961{
962 struct dentry *root = debugfs_create_dir("memblock", NULL);
963 if (!root)
964 return -ENXIO;
965 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
966 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
967
968 return 0;
969}
970__initcall(memblock_init_debugfs);
971
972#endif /* CONFIG_DEBUG_FS */