memblock: Separate out memblock_isolate_range() from memblock_set_node()
[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
YL
38int memblock_debug __initdata_memblock;
39int 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
6ed311b2
BH
52/*
53 * Address comparison utilities
54 */
10d06439 55static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
2898cc4c 56 phys_addr_t base2, phys_addr_t size2)
95f72d1e
YL
57{
58 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
59}
60
2d7d3eb2
HS
61static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
62 phys_addr_t base, phys_addr_t size)
6ed311b2
BH
63{
64 unsigned long i;
65
66 for (i = 0; i < type->cnt; i++) {
67 phys_addr_t rgnbase = type->regions[i].base;
68 phys_addr_t rgnsize = type->regions[i].size;
69 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
70 break;
71 }
72
73 return (i < type->cnt) ? i : -1;
74}
75
76/*
77 * Find, allocate, deallocate or reserve unreserved regions. All allocations
78 * are top-down.
79 */
80
cd79481d 81static phys_addr_t __init_memblock memblock_find_region(phys_addr_t start, phys_addr_t end,
6ed311b2
BH
82 phys_addr_t size, phys_addr_t align)
83{
84 phys_addr_t base, res_base;
85 long j;
86
f1af98c7
YL
87 /* In case, huge size is requested */
88 if (end < size)
1f5026a7 89 return 0;
f1af98c7 90
348968eb 91 base = round_down(end - size, align);
f1af98c7 92
25818f0f
BH
93 /* Prevent allocations returning 0 as it's also used to
94 * indicate an allocation failure
95 */
96 if (start == 0)
97 start = PAGE_SIZE;
98
6ed311b2
BH
99 while (start <= base) {
100 j = memblock_overlaps_region(&memblock.reserved, base, size);
101 if (j < 0)
102 return base;
103 res_base = memblock.reserved.regions[j].base;
104 if (res_base < size)
105 break;
348968eb 106 base = round_down(res_base - size, align);
6ed311b2
BH
107 }
108
1f5026a7 109 return 0;
6ed311b2
BH
110}
111
fc769a8e
TH
112/*
113 * Find a free area with specified alignment in a specific range.
114 */
115phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, phys_addr_t end,
116 phys_addr_t size, phys_addr_t align)
6ed311b2
BH
117{
118 long i;
6ed311b2
BH
119
120 BUG_ON(0 == size);
121
6ed311b2 122 /* Pump up max_addr */
fef501d4
BH
123 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
124 end = memblock.current_limit;
6ed311b2
BH
125
126 /* We do a top-down search, this tends to limit memory
127 * fragmentation by keeping early boot allocs near the
128 * top of memory
129 */
130 for (i = memblock.memory.cnt - 1; i >= 0; i--) {
131 phys_addr_t memblockbase = memblock.memory.regions[i].base;
132 phys_addr_t memblocksize = memblock.memory.regions[i].size;
fef501d4 133 phys_addr_t bottom, top, found;
6ed311b2
BH
134
135 if (memblocksize < size)
136 continue;
fef501d4
BH
137 if ((memblockbase + memblocksize) <= start)
138 break;
139 bottom = max(memblockbase, start);
140 top = min(memblockbase + memblocksize, end);
141 if (bottom >= top)
142 continue;
143 found = memblock_find_region(bottom, top, size, align);
1f5026a7 144 if (found)
fef501d4 145 return found;
6ed311b2 146 }
1f5026a7 147 return 0;
6ed311b2
BH
148}
149
7950c407
YL
150/*
151 * Free memblock.reserved.regions
152 */
153int __init_memblock memblock_free_reserved_regions(void)
154{
155 if (memblock.reserved.regions == memblock_reserved_init_regions)
156 return 0;
157
158 return memblock_free(__pa(memblock.reserved.regions),
159 sizeof(struct memblock_region) * memblock.reserved.max);
160}
161
162/*
163 * Reserve memblock.reserved.regions
164 */
165int __init_memblock memblock_reserve_reserved_regions(void)
166{
167 if (memblock.reserved.regions == memblock_reserved_init_regions)
168 return 0;
169
170 return memblock_reserve(__pa(memblock.reserved.regions),
171 sizeof(struct memblock_region) * memblock.reserved.max);
172}
173
10d06439 174static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
95f72d1e 175{
7c0caeb8
TH
176 memmove(&type->regions[r], &type->regions[r + 1],
177 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
e3239ff9 178 type->cnt--;
95f72d1e 179
8f7a6605
BH
180 /* Special case for empty arrays */
181 if (type->cnt == 0) {
182 type->cnt = 1;
183 type->regions[0].base = 0;
184 type->regions[0].size = 0;
7c0caeb8 185 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
8f7a6605 186 }
95f72d1e
YL
187}
188
10d06439 189static int __init_memblock memblock_double_array(struct memblock_type *type)
142b45a7
BH
190{
191 struct memblock_region *new_array, *old_array;
192 phys_addr_t old_size, new_size, addr;
193 int use_slab = slab_is_available();
194
195 /* We don't allow resizing until we know about the reserved regions
196 * of memory that aren't suitable for allocation
197 */
198 if (!memblock_can_resize)
199 return -1;
200
142b45a7
BH
201 /* Calculate new doubled size */
202 old_size = type->max * sizeof(struct memblock_region);
203 new_size = old_size << 1;
204
205 /* Try to find some space for it.
206 *
207 * WARNING: We assume that either slab_is_available() and we use it or
208 * we use MEMBLOCK for allocations. That means that this is unsafe to use
209 * when bootmem is currently active (unless bootmem itself is implemented
210 * on top of MEMBLOCK which isn't the case yet)
211 *
212 * This should however not be an issue for now, as we currently only
213 * call into MEMBLOCK while it's still active, or much later when slab is
214 * active for memory hotplug operations
215 */
216 if (use_slab) {
217 new_array = kmalloc(new_size, GFP_KERNEL);
1f5026a7 218 addr = new_array ? __pa(new_array) : 0;
142b45a7 219 } else
fc769a8e 220 addr = memblock_find_in_range(0, MEMBLOCK_ALLOC_ACCESSIBLE, new_size, sizeof(phys_addr_t));
1f5026a7 221 if (!addr) {
142b45a7
BH
222 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
223 memblock_type_name(type), type->max, type->max * 2);
224 return -1;
225 }
226 new_array = __va(addr);
227
ea9e4376
YL
228 memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]",
229 memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1);
230
142b45a7
BH
231 /* Found space, we now need to move the array over before
232 * we add the reserved region since it may be our reserved
233 * array itself that is full.
234 */
235 memcpy(new_array, type->regions, old_size);
236 memset(new_array + type->max, 0, old_size);
237 old_array = type->regions;
238 type->regions = new_array;
239 type->max <<= 1;
240
241 /* If we use SLAB that's it, we are done */
242 if (use_slab)
243 return 0;
244
245 /* Add the new reserved region now. Should not fail ! */
9c8c27e2 246 BUG_ON(memblock_reserve(addr, new_size));
142b45a7
BH
247
248 /* If the array wasn't our static init one, then free it. We only do
249 * that before SLAB is available as later on, we don't know whether
250 * to use kfree or free_bootmem_pages(). Shouldn't be a big deal
251 * anyways
252 */
253 if (old_array != memblock_memory_init_regions &&
254 old_array != memblock_reserved_init_regions)
255 memblock_free(__pa(old_array), old_size);
256
257 return 0;
258}
259
784656f9
TH
260/**
261 * memblock_merge_regions - merge neighboring compatible regions
262 * @type: memblock type to scan
263 *
264 * Scan @type and merge neighboring compatible regions.
265 */
266static void __init_memblock memblock_merge_regions(struct memblock_type *type)
95f72d1e 267{
784656f9 268 int i = 0;
95f72d1e 269
784656f9
TH
270 /* cnt never goes below 1 */
271 while (i < type->cnt - 1) {
272 struct memblock_region *this = &type->regions[i];
273 struct memblock_region *next = &type->regions[i + 1];
95f72d1e 274
7c0caeb8
TH
275 if (this->base + this->size != next->base ||
276 memblock_get_region_node(this) !=
277 memblock_get_region_node(next)) {
784656f9
TH
278 BUG_ON(this->base + this->size > next->base);
279 i++;
280 continue;
8f7a6605
BH
281 }
282
784656f9
TH
283 this->size += next->size;
284 memmove(next, next + 1, (type->cnt - (i + 1)) * sizeof(*next));
285 type->cnt--;
95f72d1e 286 }
784656f9 287}
95f72d1e 288
784656f9
TH
289/**
290 * memblock_insert_region - insert new memblock region
291 * @type: memblock type to insert into
292 * @idx: index for the insertion point
293 * @base: base address of the new region
294 * @size: size of the new region
295 *
296 * Insert new memblock region [@base,@base+@size) into @type at @idx.
297 * @type must already have extra room to accomodate the new region.
298 */
299static void __init_memblock memblock_insert_region(struct memblock_type *type,
300 int idx, phys_addr_t base,
7c0caeb8 301 phys_addr_t size, int nid)
784656f9
TH
302{
303 struct memblock_region *rgn = &type->regions[idx];
304
305 BUG_ON(type->cnt >= type->max);
306 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
307 rgn->base = base;
308 rgn->size = size;
7c0caeb8 309 memblock_set_region_node(rgn, nid);
784656f9
TH
310 type->cnt++;
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
318 *
319 * Add new memblock region [@base,@base+@size) into @type. The new region
320 * is allowed to overlap with existing ones - overlaps don't affect already
321 * existing regions. @type is guaranteed to be minimal (all neighbouring
322 * compatible regions are merged) after the addition.
323 *
324 * RETURNS:
325 * 0 on success, -errno on failure.
326 */
581adcbe
TH
327static int __init_memblock memblock_add_region(struct memblock_type *type,
328 phys_addr_t base, phys_addr_t size)
784656f9
TH
329{
330 bool insert = false;
331 phys_addr_t obase = base, end = base + size;
332 int i, nr_new;
333
334 /* special case for empty array */
335 if (type->regions[0].size == 0) {
336 WARN_ON(type->cnt != 1);
8f7a6605
BH
337 type->regions[0].base = base;
338 type->regions[0].size = size;
7c0caeb8 339 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
8f7a6605 340 return 0;
95f72d1e 341 }
784656f9
TH
342repeat:
343 /*
344 * The following is executed twice. Once with %false @insert and
345 * then with %true. The first counts the number of regions needed
346 * to accomodate the new area. The second actually inserts them.
142b45a7 347 */
784656f9
TH
348 base = obase;
349 nr_new = 0;
95f72d1e 350
784656f9
TH
351 for (i = 0; i < type->cnt; i++) {
352 struct memblock_region *rgn = &type->regions[i];
353 phys_addr_t rbase = rgn->base;
354 phys_addr_t rend = rbase + rgn->size;
355
356 if (rbase >= end)
95f72d1e 357 break;
784656f9
TH
358 if (rend <= base)
359 continue;
360 /*
361 * @rgn overlaps. If it separates the lower part of new
362 * area, insert that portion.
363 */
364 if (rbase > base) {
365 nr_new++;
366 if (insert)
367 memblock_insert_region(type, i++, base,
7c0caeb8 368 rbase - base, MAX_NUMNODES);
95f72d1e 369 }
784656f9
TH
370 /* area below @rend is dealt with, forget about it */
371 base = min(rend, end);
95f72d1e 372 }
784656f9
TH
373
374 /* insert the remaining portion */
375 if (base < end) {
376 nr_new++;
377 if (insert)
7c0caeb8
TH
378 memblock_insert_region(type, i, base, end - base,
379 MAX_NUMNODES);
95f72d1e 380 }
95f72d1e 381
784656f9
TH
382 /*
383 * If this was the first round, resize array and repeat for actual
384 * insertions; otherwise, merge and return.
142b45a7 385 */
784656f9
TH
386 if (!insert) {
387 while (type->cnt + nr_new > type->max)
388 if (memblock_double_array(type) < 0)
389 return -ENOMEM;
390 insert = true;
391 goto repeat;
392 } else {
393 memblock_merge_regions(type);
394 return 0;
142b45a7 395 }
95f72d1e
YL
396}
397
581adcbe 398int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
95f72d1e 399{
e3239ff9 400 return memblock_add_region(&memblock.memory, base, size);
95f72d1e
YL
401}
402
6a9ceb31
TH
403#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
404/**
405 * memblock_isolate_range - isolate given range into disjoint memblocks
406 * @type: memblock type to isolate range for
407 * @base: base of range to isolate
408 * @size: size of range to isolate
409 * @start_rgn: out parameter for the start of isolated region
410 * @end_rgn: out parameter for the end of isolated region
411 *
412 * Walk @type and ensure that regions don't cross the boundaries defined by
413 * [@base,@base+@size). Crossing regions are split at the boundaries,
414 * which may create at most two more regions. The index of the first
415 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
416 *
417 * RETURNS:
418 * 0 on success, -errno on failure.
419 */
420static int __init_memblock memblock_isolate_range(struct memblock_type *type,
421 phys_addr_t base, phys_addr_t size,
422 int *start_rgn, int *end_rgn)
423{
424 phys_addr_t end = base + size;
425 int i;
426
427 *start_rgn = *end_rgn = 0;
428
429 /* we'll create at most two more regions */
430 while (type->cnt + 2 > type->max)
431 if (memblock_double_array(type) < 0)
432 return -ENOMEM;
433
434 for (i = 0; i < type->cnt; i++) {
435 struct memblock_region *rgn = &type->regions[i];
436 phys_addr_t rbase = rgn->base;
437 phys_addr_t rend = rbase + rgn->size;
438
439 if (rbase >= end)
440 break;
441 if (rend <= base)
442 continue;
443
444 if (rbase < base) {
445 /*
446 * @rgn intersects from below. Split and continue
447 * to process the next region - the new top half.
448 */
449 rgn->base = base;
450 rgn->size = rend - rgn->base;
451 memblock_insert_region(type, i, rbase, base - rbase,
452 rgn->nid);
453 } else if (rend > end) {
454 /*
455 * @rgn intersects from above. Split and redo the
456 * current region - the new bottom half.
457 */
458 rgn->base = end;
459 rgn->size = rend - rgn->base;
460 memblock_insert_region(type, i--, rbase, end - rbase,
461 rgn->nid);
462 } else {
463 /* @rgn is fully contained, record it */
464 if (!*end_rgn)
465 *start_rgn = i;
466 *end_rgn = i + 1;
467 }
468 }
469
470 return 0;
471}
472#endif
473
581adcbe
TH
474static int __init_memblock __memblock_remove(struct memblock_type *type,
475 phys_addr_t base, phys_addr_t size)
95f72d1e 476{
2898cc4c 477 phys_addr_t end = base + size;
95f72d1e
YL
478 int i;
479
8f7a6605
BH
480 /* Walk through the array for collisions */
481 for (i = 0; i < type->cnt; i++) {
482 struct memblock_region *rgn = &type->regions[i];
483 phys_addr_t rend = rgn->base + rgn->size;
95f72d1e 484
8f7a6605
BH
485 /* Nothing more to do, exit */
486 if (rgn->base > end || rgn->size == 0)
95f72d1e 487 break;
95f72d1e 488
8f7a6605
BH
489 /* If we fully enclose the block, drop it */
490 if (base <= rgn->base && end >= rend) {
491 memblock_remove_region(type, i--);
492 continue;
493 }
95f72d1e 494
8f7a6605
BH
495 /* If we are fully enclosed within a block
496 * then we need to split it and we are done
497 */
498 if (base > rgn->base && end < rend) {
499 rgn->size = base - rgn->base;
500 if (!memblock_add_region(type, end, rend - end))
501 return 0;
502 /* Failure to split is bad, we at least
503 * restore the block before erroring
504 */
505 rgn->size = rend - rgn->base;
506 WARN_ON(1);
507 return -1;
508 }
95f72d1e 509
8f7a6605
BH
510 /* Check if we need to trim the bottom of a block */
511 if (rgn->base < end && rend > end) {
512 rgn->size -= end - rgn->base;
513 rgn->base = end;
514 break;
515 }
95f72d1e 516
8f7a6605
BH
517 /* And check if we need to trim the top of a block */
518 if (base < rend)
519 rgn->size -= rend - base;
95f72d1e 520
8f7a6605
BH
521 }
522 return 0;
95f72d1e
YL
523}
524
581adcbe 525int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
95f72d1e
YL
526{
527 return __memblock_remove(&memblock.memory, base, size);
528}
529
581adcbe 530int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
95f72d1e 531{
24aa0788 532 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
a150439c
PA
533 (unsigned long long)base,
534 (unsigned long long)base + size,
535 (void *)_RET_IP_);
24aa0788 536
95f72d1e
YL
537 return __memblock_remove(&memblock.reserved, base, size);
538}
539
581adcbe 540int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
95f72d1e 541{
e3239ff9 542 struct memblock_type *_rgn = &memblock.reserved;
95f72d1e 543
24aa0788 544 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] %pF\n",
a150439c
PA
545 (unsigned long long)base,
546 (unsigned long long)base + size,
547 (void *)_RET_IP_);
95f72d1e
YL
548 BUG_ON(0 == size);
549
550 return memblock_add_region(_rgn, base, size);
551}
552
35fd0808
TH
553/**
554 * __next_free_mem_range - next function for for_each_free_mem_range()
555 * @idx: pointer to u64 loop variable
556 * @nid: nid: node selector, %MAX_NUMNODES for all nodes
557 * @p_start: ptr to phys_addr_t for start address of the range, can be %NULL
558 * @p_end: ptr to phys_addr_t for end address of the range, can be %NULL
559 * @p_nid: ptr to int for nid of the range, can be %NULL
560 *
561 * Find the first free area from *@idx which matches @nid, fill the out
562 * parameters, and update *@idx for the next iteration. The lower 32bit of
563 * *@idx contains index into memory region and the upper 32bit indexes the
564 * areas before each reserved region. For example, if reserved regions
565 * look like the following,
566 *
567 * 0:[0-16), 1:[32-48), 2:[128-130)
568 *
569 * The upper 32bit indexes the following regions.
570 *
571 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
572 *
573 * As both region arrays are sorted, the function advances the two indices
574 * in lockstep and returns each intersection.
575 */
576void __init_memblock __next_free_mem_range(u64 *idx, int nid,
577 phys_addr_t *out_start,
578 phys_addr_t *out_end, int *out_nid)
579{
580 struct memblock_type *mem = &memblock.memory;
581 struct memblock_type *rsv = &memblock.reserved;
582 int mi = *idx & 0xffffffff;
583 int ri = *idx >> 32;
584
585 for ( ; mi < mem->cnt; mi++) {
586 struct memblock_region *m = &mem->regions[mi];
587 phys_addr_t m_start = m->base;
588 phys_addr_t m_end = m->base + m->size;
589
590 /* only memory regions are associated with nodes, check it */
591 if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m))
592 continue;
593
594 /* scan areas before each reservation for intersection */
595 for ( ; ri < rsv->cnt + 1; ri++) {
596 struct memblock_region *r = &rsv->regions[ri];
597 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
598 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
599
600 /* if ri advanced past mi, break out to advance mi */
601 if (r_start >= m_end)
602 break;
603 /* if the two regions intersect, we're done */
604 if (m_start < r_end) {
605 if (out_start)
606 *out_start = max(m_start, r_start);
607 if (out_end)
608 *out_end = min(m_end, r_end);
609 if (out_nid)
610 *out_nid = memblock_get_region_node(m);
611 /*
612 * The region which ends first is advanced
613 * for the next iteration.
614 */
615 if (m_end <= r_end)
616 mi++;
617 else
618 ri++;
619 *idx = (u32)mi | (u64)ri << 32;
620 return;
621 }
622 }
623 }
624
625 /* signal end of iteration */
626 *idx = ULLONG_MAX;
627}
628
7c0caeb8
TH
629#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
630/*
631 * Common iterator interface used to define for_each_mem_range().
632 */
633void __init_memblock __next_mem_pfn_range(int *idx, int nid,
634 unsigned long *out_start_pfn,
635 unsigned long *out_end_pfn, int *out_nid)
636{
637 struct memblock_type *type = &memblock.memory;
638 struct memblock_region *r;
639
640 while (++*idx < type->cnt) {
641 r = &type->regions[*idx];
642
643 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
644 continue;
645 if (nid == MAX_NUMNODES || nid == r->nid)
646 break;
647 }
648 if (*idx >= type->cnt) {
649 *idx = -1;
650 return;
651 }
652
653 if (out_start_pfn)
654 *out_start_pfn = PFN_UP(r->base);
655 if (out_end_pfn)
656 *out_end_pfn = PFN_DOWN(r->base + r->size);
657 if (out_nid)
658 *out_nid = r->nid;
659}
660
661/**
662 * memblock_set_node - set node ID on memblock regions
663 * @base: base of area to set node ID for
664 * @size: size of area to set node ID for
665 * @nid: node ID to set
666 *
667 * Set the nid of memblock memory regions in [@base,@base+@size) to @nid.
668 * Regions which cross the area boundaries are split as necessary.
669 *
670 * RETURNS:
671 * 0 on success, -errno on failure.
672 */
673int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
674 int nid)
675{
676 struct memblock_type *type = &memblock.memory;
6a9ceb31
TH
677 int start_rgn, end_rgn;
678 int i, ret;
7c0caeb8 679
6a9ceb31
TH
680 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
681 if (ret)
682 return ret;
7c0caeb8 683
6a9ceb31
TH
684 for (i = start_rgn; i < end_rgn; i++)
685 type->regions[i].nid = nid;
7c0caeb8
TH
686
687 memblock_merge_regions(type);
688 return 0;
689}
690#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
691
6ed311b2 692phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
95f72d1e 693{
6ed311b2 694 phys_addr_t found;
95f72d1e 695
6ed311b2
BH
696 /* We align the size to limit fragmentation. Without this, a lot of
697 * small allocs quickly eat up the whole reserve array on sparc
698 */
348968eb 699 size = round_up(size, align);
95f72d1e 700
fc769a8e 701 found = memblock_find_in_range(0, max_addr, size, align);
9c8c27e2 702 if (found && !memblock_reserve(found, size))
6ed311b2 703 return found;
95f72d1e 704
6ed311b2 705 return 0;
95f72d1e
YL
706}
707
6ed311b2 708phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
95f72d1e 709{
6ed311b2
BH
710 phys_addr_t alloc;
711
712 alloc = __memblock_alloc_base(size, align, max_addr);
713
714 if (alloc == 0)
715 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
716 (unsigned long long) size, (unsigned long long) max_addr);
717
718 return alloc;
95f72d1e
YL
719}
720
6ed311b2 721phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
95f72d1e 722{
6ed311b2
BH
723 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
724}
95f72d1e 725
95f72d1e 726
6ed311b2 727/*
34e18455 728 * Additional node-local top-down allocators.
c196f76f
BH
729 *
730 * WARNING: Only available after early_node_map[] has been populated,
731 * on some architectures, that is after all the calls to add_active_range()
732 * have been done to populate it.
6ed311b2 733 */
95f72d1e 734
34e18455
TH
735static phys_addr_t __init memblock_nid_range_rev(phys_addr_t start,
736 phys_addr_t end, int *nid)
c3f72b57 737{
c196f76f 738#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
c196f76f
BH
739 unsigned long start_pfn, end_pfn;
740 int i;
741
b2fea988 742 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, nid)
34e18455
TH
743 if (end > PFN_PHYS(start_pfn) && end <= PFN_PHYS(end_pfn))
744 return max(start, PFN_PHYS(start_pfn));
c196f76f 745#endif
c3f72b57 746 *nid = 0;
34e18455 747 return start;
c3f72b57
BH
748}
749
e6498040
TH
750phys_addr_t __init memblock_find_in_range_node(phys_addr_t start,
751 phys_addr_t end,
2898cc4c
BH
752 phys_addr_t size,
753 phys_addr_t align, int nid)
95f72d1e 754{
e6498040
TH
755 struct memblock_type *mem = &memblock.memory;
756 int i;
95f72d1e 757
e6498040 758 BUG_ON(0 == size);
95f72d1e 759
e6498040
TH
760 /* Pump up max_addr */
761 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
762 end = memblock.current_limit;
95f72d1e 763
e6498040
TH
764 for (i = mem->cnt - 1; i >= 0; i--) {
765 struct memblock_region *r = &mem->regions[i];
766 phys_addr_t base = max(start, r->base);
767 phys_addr_t top = min(end, r->base + r->size);
768
769 while (base < top) {
770 phys_addr_t tbase, ret;
771 int tnid;
772
773 tbase = memblock_nid_range_rev(base, top, &tnid);
774 if (nid == MAX_NUMNODES || tnid == nid) {
775 ret = memblock_find_region(tbase, top, size, align);
776 if (ret)
777 return ret;
778 }
779 top = tbase;
95f72d1e 780 }
95f72d1e 781 }
e6498040 782
1f5026a7 783 return 0;
95f72d1e
YL
784}
785
2898cc4c 786phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
95f72d1e 787{
e6498040 788 phys_addr_t found;
95f72d1e 789
e6498040
TH
790 /*
791 * We align the size to limit fragmentation. Without this, a lot of
7f219c73
BH
792 * small allocs quickly eat up the whole reserve array on sparc
793 */
348968eb 794 size = round_up(size, align);
7f219c73 795
e6498040
TH
796 found = memblock_find_in_range_node(0, MEMBLOCK_ALLOC_ACCESSIBLE,
797 size, align, nid);
9c8c27e2 798 if (found && !memblock_reserve(found, size))
e6498040 799 return found;
95f72d1e 800
9d1e2492
BH
801 return 0;
802}
803
804phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
805{
806 phys_addr_t res = memblock_alloc_nid(size, align, nid);
807
808 if (res)
809 return res;
15fb0972 810 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
95f72d1e
YL
811}
812
9d1e2492
BH
813
814/*
815 * Remaining API functions
816 */
817
95f72d1e 818/* You must call memblock_analyze() before this. */
2898cc4c 819phys_addr_t __init memblock_phys_mem_size(void)
95f72d1e 820{
4734b594 821 return memblock.memory_size;
95f72d1e
YL
822}
823
0a93ebef
SR
824/* lowest address */
825phys_addr_t __init_memblock memblock_start_of_DRAM(void)
826{
827 return memblock.memory.regions[0].base;
828}
829
10d06439 830phys_addr_t __init_memblock memblock_end_of_DRAM(void)
95f72d1e
YL
831{
832 int idx = memblock.memory.cnt - 1;
833
e3239ff9 834 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
95f72d1e
YL
835}
836
837/* You must call memblock_analyze() after this. */
2898cc4c 838void __init memblock_enforce_memory_limit(phys_addr_t memory_limit)
95f72d1e
YL
839{
840 unsigned long i;
2898cc4c 841 phys_addr_t limit;
e3239ff9 842 struct memblock_region *p;
95f72d1e
YL
843
844 if (!memory_limit)
845 return;
846
847 /* Truncate the memblock regions to satisfy the memory limit. */
848 limit = memory_limit;
849 for (i = 0; i < memblock.memory.cnt; i++) {
e3239ff9
BH
850 if (limit > memblock.memory.regions[i].size) {
851 limit -= memblock.memory.regions[i].size;
95f72d1e
YL
852 continue;
853 }
854
e3239ff9 855 memblock.memory.regions[i].size = limit;
95f72d1e
YL
856 memblock.memory.cnt = i + 1;
857 break;
858 }
859
95f72d1e
YL
860 memory_limit = memblock_end_of_DRAM();
861
862 /* And truncate any reserves above the limit also. */
863 for (i = 0; i < memblock.reserved.cnt; i++) {
e3239ff9 864 p = &memblock.reserved.regions[i];
95f72d1e
YL
865
866 if (p->base > memory_limit)
867 p->size = 0;
868 else if ((p->base + p->size) > memory_limit)
869 p->size = memory_limit - p->base;
870
871 if (p->size == 0) {
872 memblock_remove_region(&memblock.reserved, i);
873 i--;
874 }
875 }
876}
877
cd79481d 878static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
72d4b0b4
BH
879{
880 unsigned int left = 0, right = type->cnt;
881
882 do {
883 unsigned int mid = (right + left) / 2;
884
885 if (addr < type->regions[mid].base)
886 right = mid;
887 else if (addr >= (type->regions[mid].base +
888 type->regions[mid].size))
889 left = mid + 1;
890 else
891 return mid;
892 } while (left < right);
893 return -1;
894}
895
2898cc4c 896int __init memblock_is_reserved(phys_addr_t addr)
95f72d1e 897{
72d4b0b4
BH
898 return memblock_search(&memblock.reserved, addr) != -1;
899}
95f72d1e 900
3661ca66 901int __init_memblock memblock_is_memory(phys_addr_t addr)
72d4b0b4
BH
902{
903 return memblock_search(&memblock.memory, addr) != -1;
904}
905
3661ca66 906int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
72d4b0b4 907{
abb65272 908 int idx = memblock_search(&memblock.memory, base);
72d4b0b4
BH
909
910 if (idx == -1)
911 return 0;
abb65272
TV
912 return memblock.memory.regions[idx].base <= base &&
913 (memblock.memory.regions[idx].base +
914 memblock.memory.regions[idx].size) >= (base + size);
95f72d1e
YL
915}
916
10d06439 917int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
95f72d1e 918{
f1c2c19c 919 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
95f72d1e
YL
920}
921
e63075a3 922
3661ca66 923void __init_memblock memblock_set_current_limit(phys_addr_t limit)
e63075a3
BH
924{
925 memblock.current_limit = limit;
926}
927
7c0caeb8 928static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
6ed311b2
BH
929{
930 unsigned long long base, size;
931 int i;
932
7c0caeb8 933 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
6ed311b2 934
7c0caeb8
TH
935 for (i = 0; i < type->cnt; i++) {
936 struct memblock_region *rgn = &type->regions[i];
937 char nid_buf[32] = "";
938
939 base = rgn->base;
940 size = rgn->size;
941#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
942 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
943 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
944 memblock_get_region_node(rgn));
945#endif
946 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s\n",
947 name, i, base, base + size - 1, size, nid_buf);
6ed311b2
BH
948 }
949}
950
4ff7b82f 951void __init_memblock __memblock_dump_all(void)
6ed311b2 952{
6ed311b2
BH
953 pr_info("MEMBLOCK configuration:\n");
954 pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size);
955
956 memblock_dump(&memblock.memory, "memory");
957 memblock_dump(&memblock.reserved, "reserved");
958}
959
960void __init memblock_analyze(void)
961{
962 int i;
963
6ed311b2
BH
964 memblock.memory_size = 0;
965
966 for (i = 0; i < memblock.memory.cnt; i++)
967 memblock.memory_size += memblock.memory.regions[i].size;
142b45a7
BH
968
969 /* We allow resizing from there */
970 memblock_can_resize = 1;
6ed311b2
BH
971}
972
6ed311b2
BH
973static int __init early_memblock(char *p)
974{
975 if (p && strstr(p, "debug"))
976 memblock_debug = 1;
977 return 0;
978}
979early_param("memblock", early_memblock);
980
c378ddd5 981#if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
6d03b885
BH
982
983static int memblock_debug_show(struct seq_file *m, void *private)
984{
985 struct memblock_type *type = m->private;
986 struct memblock_region *reg;
987 int i;
988
989 for (i = 0; i < type->cnt; i++) {
990 reg = &type->regions[i];
991 seq_printf(m, "%4d: ", i);
992 if (sizeof(phys_addr_t) == 4)
993 seq_printf(m, "0x%08lx..0x%08lx\n",
994 (unsigned long)reg->base,
995 (unsigned long)(reg->base + reg->size - 1));
996 else
997 seq_printf(m, "0x%016llx..0x%016llx\n",
998 (unsigned long long)reg->base,
999 (unsigned long long)(reg->base + reg->size - 1));
1000
1001 }
1002 return 0;
1003}
1004
1005static int memblock_debug_open(struct inode *inode, struct file *file)
1006{
1007 return single_open(file, memblock_debug_show, inode->i_private);
1008}
1009
1010static const struct file_operations memblock_debug_fops = {
1011 .open = memblock_debug_open,
1012 .read = seq_read,
1013 .llseek = seq_lseek,
1014 .release = single_release,
1015};
1016
1017static int __init memblock_init_debugfs(void)
1018{
1019 struct dentry *root = debugfs_create_dir("memblock", NULL);
1020 if (!root)
1021 return -ENXIO;
1022 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1023 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1024
1025 return 0;
1026}
1027__initcall(memblock_init_debugfs);
1028
1029#endif /* CONFIG_DEBUG_FS */