mm/memory_hotplug.c: use PFN_DOWN()
[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
79442ed1 23#include <asm-generic/sections.h>
26f09e9b
SS
24#include <linux/io.h>
25
26#include "internal.h"
79442ed1 27
fe091c20
TH
28static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
29static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
70210ed9
PH
30#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
31static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
32#endif
fe091c20
TH
33
34struct memblock memblock __initdata_memblock = {
35 .memory.regions = memblock_memory_init_regions,
36 .memory.cnt = 1, /* empty dummy entry */
37 .memory.max = INIT_MEMBLOCK_REGIONS,
38
39 .reserved.regions = memblock_reserved_init_regions,
40 .reserved.cnt = 1, /* empty dummy entry */
41 .reserved.max = INIT_MEMBLOCK_REGIONS,
42
70210ed9
PH
43#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
44 .physmem.regions = memblock_physmem_init_regions,
45 .physmem.cnt = 1, /* empty dummy entry */
46 .physmem.max = INIT_PHYSMEM_REGIONS,
47#endif
48
79442ed1 49 .bottom_up = false,
fe091c20
TH
50 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
51};
95f72d1e 52
10d06439 53int memblock_debug __initdata_memblock;
55ac590c
TC
54#ifdef CONFIG_MOVABLE_NODE
55bool movable_node_enabled __initdata_memblock = false;
56#endif
1aadc056 57static int memblock_can_resize __initdata_memblock;
181eb394
GS
58static int memblock_memory_in_slab __initdata_memblock = 0;
59static int memblock_reserved_in_slab __initdata_memblock = 0;
95f72d1e 60
142b45a7 61/* inline so we don't get a warning when pr_debug is compiled out */
c2233116
RP
62static __init_memblock const char *
63memblock_type_name(struct memblock_type *type)
142b45a7
BH
64{
65 if (type == &memblock.memory)
66 return "memory";
67 else if (type == &memblock.reserved)
68 return "reserved";
69 else
70 return "unknown";
71}
72
eb18f1b5
TH
73/* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
74static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
75{
76 return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
77}
78
6ed311b2
BH
79/*
80 * Address comparison utilities
81 */
10d06439 82static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
2898cc4c 83 phys_addr_t base2, phys_addr_t size2)
95f72d1e
YL
84{
85 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
86}
87
2d7d3eb2
HS
88static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
89 phys_addr_t base, phys_addr_t size)
6ed311b2
BH
90{
91 unsigned long i;
92
93 for (i = 0; i < type->cnt; i++) {
94 phys_addr_t rgnbase = type->regions[i].base;
95 phys_addr_t rgnsize = type->regions[i].size;
96 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
97 break;
98 }
99
100 return (i < type->cnt) ? i : -1;
101}
102
79442ed1
TC
103/*
104 * __memblock_find_range_bottom_up - find free area utility in bottom-up
105 * @start: start of candidate range
106 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
107 * @size: size of free area to find
108 * @align: alignment of free area to find
b1154233 109 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
79442ed1
TC
110 *
111 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
112 *
113 * RETURNS:
114 * Found address on success, 0 on failure.
115 */
116static phys_addr_t __init_memblock
117__memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
118 phys_addr_t size, phys_addr_t align, int nid)
119{
120 phys_addr_t this_start, this_end, cand;
121 u64 i;
122
123 for_each_free_mem_range(i, nid, &this_start, &this_end, NULL) {
124 this_start = clamp(this_start, start, end);
125 this_end = clamp(this_end, start, end);
126
127 cand = round_up(this_start, align);
128 if (cand < this_end && this_end - cand >= size)
129 return cand;
130 }
131
132 return 0;
133}
134
7bd0b0f0 135/**
1402899e 136 * __memblock_find_range_top_down - find free area utility, in top-down
7bd0b0f0
TH
137 * @start: start of candidate range
138 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
139 * @size: size of free area to find
140 * @align: alignment of free area to find
b1154233 141 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
7bd0b0f0 142 *
1402899e 143 * Utility called from memblock_find_in_range_node(), find free area top-down.
7bd0b0f0
TH
144 *
145 * RETURNS:
79442ed1 146 * Found address on success, 0 on failure.
6ed311b2 147 */
1402899e
TC
148static phys_addr_t __init_memblock
149__memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
150 phys_addr_t size, phys_addr_t align, int nid)
f7210e6c
TC
151{
152 phys_addr_t this_start, this_end, cand;
153 u64 i;
154
f7210e6c
TC
155 for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) {
156 this_start = clamp(this_start, start, end);
157 this_end = clamp(this_end, start, end);
158
159 if (this_end < size)
160 continue;
161
162 cand = round_down(this_end - size, align);
163 if (cand >= this_start)
164 return cand;
165 }
1402899e 166
f7210e6c
TC
167 return 0;
168}
6ed311b2 169
1402899e
TC
170/**
171 * memblock_find_in_range_node - find free area in given range and node
1402899e
TC
172 * @size: size of free area to find
173 * @align: alignment of free area to find
87029ee9
GS
174 * @start: start of candidate range
175 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
b1154233 176 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1402899e
TC
177 *
178 * Find @size free area aligned to @align in the specified range and node.
179 *
79442ed1
TC
180 * When allocation direction is bottom-up, the @start should be greater
181 * than the end of the kernel image. Otherwise, it will be trimmed. The
182 * reason is that we want the bottom-up allocation just near the kernel
183 * image so it is highly likely that the allocated memory and the kernel
184 * will reside in the same node.
185 *
186 * If bottom-up allocation failed, will try to allocate memory top-down.
187 *
1402899e 188 * RETURNS:
79442ed1 189 * Found address on success, 0 on failure.
1402899e 190 */
87029ee9
GS
191phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
192 phys_addr_t align, phys_addr_t start,
193 phys_addr_t end, int nid)
1402899e 194{
79442ed1
TC
195 int ret;
196 phys_addr_t kernel_end;
197
1402899e
TC
198 /* pump up @end */
199 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
200 end = memblock.current_limit;
201
202 /* avoid allocating the first page */
203 start = max_t(phys_addr_t, start, PAGE_SIZE);
204 end = max(start, end);
79442ed1
TC
205 kernel_end = __pa_symbol(_end);
206
207 /*
208 * try bottom-up allocation only when bottom-up mode
209 * is set and @end is above the kernel image.
210 */
211 if (memblock_bottom_up() && end > kernel_end) {
212 phys_addr_t bottom_up_start;
213
214 /* make sure we will allocate above the kernel */
215 bottom_up_start = max(start, kernel_end);
216
217 /* ok, try bottom-up allocation first */
218 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
219 size, align, nid);
220 if (ret)
221 return ret;
222
223 /*
224 * we always limit bottom-up allocation above the kernel,
225 * but top-down allocation doesn't have the limit, so
226 * retrying top-down allocation may succeed when bottom-up
227 * allocation failed.
228 *
229 * bottom-up allocation is expected to be fail very rarely,
230 * so we use WARN_ONCE() here to see the stack trace if
231 * fail happens.
232 */
233 WARN_ONCE(1, "memblock: bottom-up allocation failed, "
234 "memory hotunplug may be affected\n");
235 }
1402899e
TC
236
237 return __memblock_find_range_top_down(start, end, size, align, nid);
238}
239
7bd0b0f0
TH
240/**
241 * memblock_find_in_range - find free area in given range
242 * @start: start of candidate range
243 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
244 * @size: size of free area to find
245 * @align: alignment of free area to find
246 *
247 * Find @size free area aligned to @align in the specified range.
248 *
249 * RETURNS:
79442ed1 250 * Found address on success, 0 on failure.
fc769a8e 251 */
7bd0b0f0
TH
252phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
253 phys_addr_t end, phys_addr_t size,
254 phys_addr_t align)
6ed311b2 255{
87029ee9 256 return memblock_find_in_range_node(size, align, start, end,
b1154233 257 NUMA_NO_NODE);
6ed311b2
BH
258}
259
10d06439 260static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
95f72d1e 261{
1440c4e2 262 type->total_size -= type->regions[r].size;
7c0caeb8
TH
263 memmove(&type->regions[r], &type->regions[r + 1],
264 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
e3239ff9 265 type->cnt--;
95f72d1e 266
8f7a6605
BH
267 /* Special case for empty arrays */
268 if (type->cnt == 0) {
1440c4e2 269 WARN_ON(type->total_size != 0);
8f7a6605
BH
270 type->cnt = 1;
271 type->regions[0].base = 0;
272 type->regions[0].size = 0;
66a20757 273 type->regions[0].flags = 0;
7c0caeb8 274 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
8f7a6605 275 }
95f72d1e
YL
276}
277
354f17e1
PH
278#ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
279
29f67386
YL
280phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
281 phys_addr_t *addr)
282{
283 if (memblock.reserved.regions == memblock_reserved_init_regions)
284 return 0;
285
286 *addr = __pa(memblock.reserved.regions);
287
288 return PAGE_ALIGN(sizeof(struct memblock_region) *
289 memblock.reserved.max);
290}
291
5e270e25
PH
292phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
293 phys_addr_t *addr)
294{
295 if (memblock.memory.regions == memblock_memory_init_regions)
296 return 0;
297
298 *addr = __pa(memblock.memory.regions);
299
300 return PAGE_ALIGN(sizeof(struct memblock_region) *
301 memblock.memory.max);
302}
303
304#endif
305
48c3b583
GP
306/**
307 * memblock_double_array - double the size of the memblock regions array
308 * @type: memblock type of the regions array being doubled
309 * @new_area_start: starting address of memory range to avoid overlap with
310 * @new_area_size: size of memory range to avoid overlap with
311 *
312 * Double the size of the @type regions array. If memblock is being used to
313 * allocate memory for a new reserved regions array and there is a previously
314 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
315 * waiting to be reserved, ensure the memory used by the new array does
316 * not overlap.
317 *
318 * RETURNS:
319 * 0 on success, -1 on failure.
320 */
321static int __init_memblock memblock_double_array(struct memblock_type *type,
322 phys_addr_t new_area_start,
323 phys_addr_t new_area_size)
142b45a7
BH
324{
325 struct memblock_region *new_array, *old_array;
29f67386 326 phys_addr_t old_alloc_size, new_alloc_size;
142b45a7
BH
327 phys_addr_t old_size, new_size, addr;
328 int use_slab = slab_is_available();
181eb394 329 int *in_slab;
142b45a7
BH
330
331 /* We don't allow resizing until we know about the reserved regions
332 * of memory that aren't suitable for allocation
333 */
334 if (!memblock_can_resize)
335 return -1;
336
142b45a7
BH
337 /* Calculate new doubled size */
338 old_size = type->max * sizeof(struct memblock_region);
339 new_size = old_size << 1;
29f67386
YL
340 /*
341 * We need to allocated new one align to PAGE_SIZE,
342 * so we can free them completely later.
343 */
344 old_alloc_size = PAGE_ALIGN(old_size);
345 new_alloc_size = PAGE_ALIGN(new_size);
142b45a7 346
181eb394
GS
347 /* Retrieve the slab flag */
348 if (type == &memblock.memory)
349 in_slab = &memblock_memory_in_slab;
350 else
351 in_slab = &memblock_reserved_in_slab;
352
142b45a7
BH
353 /* Try to find some space for it.
354 *
355 * WARNING: We assume that either slab_is_available() and we use it or
fd07383b
AM
356 * we use MEMBLOCK for allocations. That means that this is unsafe to
357 * use when bootmem is currently active (unless bootmem itself is
358 * implemented on top of MEMBLOCK which isn't the case yet)
142b45a7
BH
359 *
360 * This should however not be an issue for now, as we currently only
fd07383b
AM
361 * call into MEMBLOCK while it's still active, or much later when slab
362 * is active for memory hotplug operations
142b45a7
BH
363 */
364 if (use_slab) {
365 new_array = kmalloc(new_size, GFP_KERNEL);
1f5026a7 366 addr = new_array ? __pa(new_array) : 0;
4e2f0775 367 } else {
48c3b583
GP
368 /* only exclude range when trying to double reserved.regions */
369 if (type != &memblock.reserved)
370 new_area_start = new_area_size = 0;
371
372 addr = memblock_find_in_range(new_area_start + new_area_size,
373 memblock.current_limit,
29f67386 374 new_alloc_size, PAGE_SIZE);
48c3b583
GP
375 if (!addr && new_area_size)
376 addr = memblock_find_in_range(0,
fd07383b
AM
377 min(new_area_start, memblock.current_limit),
378 new_alloc_size, PAGE_SIZE);
48c3b583 379
15674868 380 new_array = addr ? __va(addr) : NULL;
4e2f0775 381 }
1f5026a7 382 if (!addr) {
142b45a7
BH
383 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
384 memblock_type_name(type), type->max, type->max * 2);
385 return -1;
386 }
142b45a7 387
fd07383b
AM
388 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
389 memblock_type_name(type), type->max * 2, (u64)addr,
390 (u64)addr + new_size - 1);
ea9e4376 391
fd07383b
AM
392 /*
393 * Found space, we now need to move the array over before we add the
394 * reserved region since it may be our reserved array itself that is
395 * full.
142b45a7
BH
396 */
397 memcpy(new_array, type->regions, old_size);
398 memset(new_array + type->max, 0, old_size);
399 old_array = type->regions;
400 type->regions = new_array;
401 type->max <<= 1;
402
fd07383b 403 /* Free old array. We needn't free it if the array is the static one */
181eb394
GS
404 if (*in_slab)
405 kfree(old_array);
406 else if (old_array != memblock_memory_init_regions &&
407 old_array != memblock_reserved_init_regions)
29f67386 408 memblock_free(__pa(old_array), old_alloc_size);
142b45a7 409
fd07383b
AM
410 /*
411 * Reserve the new array if that comes from the memblock. Otherwise, we
412 * needn't do it
181eb394
GS
413 */
414 if (!use_slab)
29f67386 415 BUG_ON(memblock_reserve(addr, new_alloc_size));
181eb394
GS
416
417 /* Update slab flag */
418 *in_slab = use_slab;
419
142b45a7
BH
420 return 0;
421}
422
784656f9
TH
423/**
424 * memblock_merge_regions - merge neighboring compatible regions
425 * @type: memblock type to scan
426 *
427 * Scan @type and merge neighboring compatible regions.
428 */
429static void __init_memblock memblock_merge_regions(struct memblock_type *type)
95f72d1e 430{
784656f9 431 int i = 0;
95f72d1e 432
784656f9
TH
433 /* cnt never goes below 1 */
434 while (i < type->cnt - 1) {
435 struct memblock_region *this = &type->regions[i];
436 struct memblock_region *next = &type->regions[i + 1];
95f72d1e 437
7c0caeb8
TH
438 if (this->base + this->size != next->base ||
439 memblock_get_region_node(this) !=
66a20757
TC
440 memblock_get_region_node(next) ||
441 this->flags != next->flags) {
784656f9
TH
442 BUG_ON(this->base + this->size > next->base);
443 i++;
444 continue;
8f7a6605
BH
445 }
446
784656f9 447 this->size += next->size;
c0232ae8
LF
448 /* move forward from next + 1, index of which is i + 2 */
449 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
784656f9 450 type->cnt--;
95f72d1e 451 }
784656f9 452}
95f72d1e 453
784656f9
TH
454/**
455 * memblock_insert_region - insert new memblock region
209ff86d
TC
456 * @type: memblock type to insert into
457 * @idx: index for the insertion point
458 * @base: base address of the new region
459 * @size: size of the new region
460 * @nid: node id of the new region
66a20757 461 * @flags: flags of the new region
784656f9
TH
462 *
463 * Insert new memblock region [@base,@base+@size) into @type at @idx.
464 * @type must already have extra room to accomodate the new region.
465 */
466static void __init_memblock memblock_insert_region(struct memblock_type *type,
467 int idx, phys_addr_t base,
66a20757
TC
468 phys_addr_t size,
469 int nid, unsigned long flags)
784656f9
TH
470{
471 struct memblock_region *rgn = &type->regions[idx];
472
473 BUG_ON(type->cnt >= type->max);
474 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
475 rgn->base = base;
476 rgn->size = size;
66a20757 477 rgn->flags = flags;
7c0caeb8 478 memblock_set_region_node(rgn, nid);
784656f9 479 type->cnt++;
1440c4e2 480 type->total_size += size;
784656f9
TH
481}
482
483/**
f1af9d3a 484 * memblock_add_range - add new memblock region
784656f9
TH
485 * @type: memblock type to add new region into
486 * @base: base address of the new region
487 * @size: size of the new region
7fb0bc3f 488 * @nid: nid of the new region
66a20757 489 * @flags: flags of the new region
784656f9
TH
490 *
491 * Add new memblock region [@base,@base+@size) into @type. The new region
492 * is allowed to overlap with existing ones - overlaps don't affect already
493 * existing regions. @type is guaranteed to be minimal (all neighbouring
494 * compatible regions are merged) after the addition.
495 *
496 * RETURNS:
497 * 0 on success, -errno on failure.
498 */
f1af9d3a 499int __init_memblock memblock_add_range(struct memblock_type *type,
66a20757
TC
500 phys_addr_t base, phys_addr_t size,
501 int nid, unsigned long flags)
784656f9
TH
502{
503 bool insert = false;
eb18f1b5
TH
504 phys_addr_t obase = base;
505 phys_addr_t end = base + memblock_cap_size(base, &size);
784656f9
TH
506 int i, nr_new;
507
b3dc627c
TH
508 if (!size)
509 return 0;
510
784656f9
TH
511 /* special case for empty array */
512 if (type->regions[0].size == 0) {
1440c4e2 513 WARN_ON(type->cnt != 1 || type->total_size);
8f7a6605
BH
514 type->regions[0].base = base;
515 type->regions[0].size = size;
66a20757 516 type->regions[0].flags = flags;
7fb0bc3f 517 memblock_set_region_node(&type->regions[0], nid);
1440c4e2 518 type->total_size = size;
8f7a6605 519 return 0;
95f72d1e 520 }
784656f9
TH
521repeat:
522 /*
523 * The following is executed twice. Once with %false @insert and
524 * then with %true. The first counts the number of regions needed
525 * to accomodate the new area. The second actually inserts them.
142b45a7 526 */
784656f9
TH
527 base = obase;
528 nr_new = 0;
95f72d1e 529
784656f9
TH
530 for (i = 0; i < type->cnt; i++) {
531 struct memblock_region *rgn = &type->regions[i];
532 phys_addr_t rbase = rgn->base;
533 phys_addr_t rend = rbase + rgn->size;
534
535 if (rbase >= end)
95f72d1e 536 break;
784656f9
TH
537 if (rend <= base)
538 continue;
539 /*
540 * @rgn overlaps. If it separates the lower part of new
541 * area, insert that portion.
542 */
543 if (rbase > base) {
544 nr_new++;
545 if (insert)
546 memblock_insert_region(type, i++, base,
66a20757
TC
547 rbase - base, nid,
548 flags);
95f72d1e 549 }
784656f9
TH
550 /* area below @rend is dealt with, forget about it */
551 base = min(rend, end);
95f72d1e 552 }
784656f9
TH
553
554 /* insert the remaining portion */
555 if (base < end) {
556 nr_new++;
557 if (insert)
66a20757
TC
558 memblock_insert_region(type, i, base, end - base,
559 nid, flags);
95f72d1e 560 }
95f72d1e 561
784656f9
TH
562 /*
563 * If this was the first round, resize array and repeat for actual
564 * insertions; otherwise, merge and return.
142b45a7 565 */
784656f9
TH
566 if (!insert) {
567 while (type->cnt + nr_new > type->max)
48c3b583 568 if (memblock_double_array(type, obase, size) < 0)
784656f9
TH
569 return -ENOMEM;
570 insert = true;
571 goto repeat;
572 } else {
573 memblock_merge_regions(type);
574 return 0;
142b45a7 575 }
95f72d1e
YL
576}
577
7fb0bc3f
TH
578int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
579 int nid)
580{
f1af9d3a 581 return memblock_add_range(&memblock.memory, base, size, nid, 0);
7fb0bc3f
TH
582}
583
581adcbe 584int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
95f72d1e 585{
f1af9d3a 586 return memblock_add_range(&memblock.memory, base, size,
66a20757 587 MAX_NUMNODES, 0);
95f72d1e
YL
588}
589
6a9ceb31
TH
590/**
591 * memblock_isolate_range - isolate given range into disjoint memblocks
592 * @type: memblock type to isolate range for
593 * @base: base of range to isolate
594 * @size: size of range to isolate
595 * @start_rgn: out parameter for the start of isolated region
596 * @end_rgn: out parameter for the end of isolated region
597 *
598 * Walk @type and ensure that regions don't cross the boundaries defined by
599 * [@base,@base+@size). Crossing regions are split at the boundaries,
600 * which may create at most two more regions. The index of the first
601 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
602 *
603 * RETURNS:
604 * 0 on success, -errno on failure.
605 */
606static int __init_memblock memblock_isolate_range(struct memblock_type *type,
607 phys_addr_t base, phys_addr_t size,
608 int *start_rgn, int *end_rgn)
609{
eb18f1b5 610 phys_addr_t end = base + memblock_cap_size(base, &size);
6a9ceb31
TH
611 int i;
612
613 *start_rgn = *end_rgn = 0;
614
b3dc627c
TH
615 if (!size)
616 return 0;
617
6a9ceb31
TH
618 /* we'll create at most two more regions */
619 while (type->cnt + 2 > type->max)
48c3b583 620 if (memblock_double_array(type, base, size) < 0)
6a9ceb31
TH
621 return -ENOMEM;
622
623 for (i = 0; i < type->cnt; i++) {
624 struct memblock_region *rgn = &type->regions[i];
625 phys_addr_t rbase = rgn->base;
626 phys_addr_t rend = rbase + rgn->size;
627
628 if (rbase >= end)
629 break;
630 if (rend <= base)
631 continue;
632
633 if (rbase < base) {
634 /*
635 * @rgn intersects from below. Split and continue
636 * to process the next region - the new top half.
637 */
638 rgn->base = base;
1440c4e2
TH
639 rgn->size -= base - rbase;
640 type->total_size -= base - rbase;
6a9ceb31 641 memblock_insert_region(type, i, rbase, base - rbase,
66a20757
TC
642 memblock_get_region_node(rgn),
643 rgn->flags);
6a9ceb31
TH
644 } else if (rend > end) {
645 /*
646 * @rgn intersects from above. Split and redo the
647 * current region - the new bottom half.
648 */
649 rgn->base = end;
1440c4e2
TH
650 rgn->size -= end - rbase;
651 type->total_size -= end - rbase;
6a9ceb31 652 memblock_insert_region(type, i--, rbase, end - rbase,
66a20757
TC
653 memblock_get_region_node(rgn),
654 rgn->flags);
6a9ceb31
TH
655 } else {
656 /* @rgn is fully contained, record it */
657 if (!*end_rgn)
658 *start_rgn = i;
659 *end_rgn = i + 1;
660 }
661 }
662
663 return 0;
664}
6a9ceb31 665
f1af9d3a
PH
666int __init_memblock memblock_remove_range(struct memblock_type *type,
667 phys_addr_t base, phys_addr_t size)
95f72d1e 668{
71936180
TH
669 int start_rgn, end_rgn;
670 int i, ret;
95f72d1e 671
71936180
TH
672 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
673 if (ret)
674 return ret;
95f72d1e 675
71936180
TH
676 for (i = end_rgn - 1; i >= start_rgn; i--)
677 memblock_remove_region(type, i);
8f7a6605 678 return 0;
95f72d1e
YL
679}
680
581adcbe 681int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
95f72d1e 682{
f1af9d3a 683 return memblock_remove_range(&memblock.memory, base, size);
95f72d1e
YL
684}
685
f1af9d3a 686
581adcbe 687int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
95f72d1e 688{
24aa0788 689 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
a150439c 690 (unsigned long long)base,
931d13f5 691 (unsigned long long)base + size - 1,
a150439c 692 (void *)_RET_IP_);
24aa0788 693
f1af9d3a 694 return memblock_remove_range(&memblock.reserved, base, size);
95f72d1e
YL
695}
696
66a20757
TC
697static int __init_memblock memblock_reserve_region(phys_addr_t base,
698 phys_addr_t size,
699 int nid,
700 unsigned long flags)
95f72d1e 701{
e3239ff9 702 struct memblock_type *_rgn = &memblock.reserved;
95f72d1e 703
66a20757 704 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
a150439c 705 (unsigned long long)base,
931d13f5 706 (unsigned long long)base + size - 1,
66a20757
TC
707 flags, (void *)_RET_IP_);
708
f1af9d3a 709 return memblock_add_range(_rgn, base, size, nid, flags);
66a20757 710}
95f72d1e 711
66a20757
TC
712int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
713{
714 return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
95f72d1e
YL
715}
716
66b16edf
TC
717/**
718 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
719 * @base: the base phys addr of the region
720 * @size: the size of the region
721 *
722 * This function isolates region [@base, @base + @size), and mark it with flag
723 * MEMBLOCK_HOTPLUG.
724 *
725 * Return 0 on succees, -errno on failure.
726 */
727int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
728{
729 struct memblock_type *type = &memblock.memory;
730 int i, ret, start_rgn, end_rgn;
731
732 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
733 if (ret)
734 return ret;
735
736 for (i = start_rgn; i < end_rgn; i++)
737 memblock_set_region_flags(&type->regions[i], MEMBLOCK_HOTPLUG);
738
739 memblock_merge_regions(type);
740 return 0;
741}
742
743/**
744 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
745 * @base: the base phys addr of the region
746 * @size: the size of the region
747 *
748 * This function isolates region [@base, @base + @size), and clear flag
749 * MEMBLOCK_HOTPLUG for the isolated regions.
750 *
751 * Return 0 on succees, -errno on failure.
752 */
753int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
754{
755 struct memblock_type *type = &memblock.memory;
756 int i, ret, start_rgn, end_rgn;
757
758 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
759 if (ret)
760 return ret;
761
762 for (i = start_rgn; i < end_rgn; i++)
763 memblock_clear_region_flags(&type->regions[i],
764 MEMBLOCK_HOTPLUG);
765
766 memblock_merge_regions(type);
767 return 0;
768}
769
35fd0808 770/**
f1af9d3a 771 * __next__mem_range - next function for for_each_free_mem_range() etc.
35fd0808 772 * @idx: pointer to u64 loop variable
b1154233 773 * @nid: node selector, %NUMA_NO_NODE for all nodes
f1af9d3a
PH
774 * @type_a: pointer to memblock_type from where the range is taken
775 * @type_b: pointer to memblock_type which excludes memory from being taken
dad7557e
WL
776 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
777 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
778 * @out_nid: ptr to int for nid of the range, can be %NULL
35fd0808 779 *
f1af9d3a 780 * Find the first area from *@idx which matches @nid, fill the out
35fd0808 781 * parameters, and update *@idx for the next iteration. The lower 32bit of
f1af9d3a
PH
782 * *@idx contains index into type_a and the upper 32bit indexes the
783 * areas before each region in type_b. For example, if type_b regions
35fd0808
TH
784 * look like the following,
785 *
786 * 0:[0-16), 1:[32-48), 2:[128-130)
787 *
788 * The upper 32bit indexes the following regions.
789 *
790 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
791 *
792 * As both region arrays are sorted, the function advances the two indices
793 * in lockstep and returns each intersection.
794 */
f1af9d3a
PH
795void __init_memblock __next_mem_range(u64 *idx, int nid,
796 struct memblock_type *type_a,
797 struct memblock_type *type_b,
798 phys_addr_t *out_start,
799 phys_addr_t *out_end, int *out_nid)
35fd0808 800{
f1af9d3a
PH
801 int idx_a = *idx & 0xffffffff;
802 int idx_b = *idx >> 32;
b1154233 803
f1af9d3a
PH
804 if (WARN_ONCE(nid == MAX_NUMNODES,
805 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
560dca27 806 nid = NUMA_NO_NODE;
35fd0808 807
f1af9d3a
PH
808 for (; idx_a < type_a->cnt; idx_a++) {
809 struct memblock_region *m = &type_a->regions[idx_a];
810
35fd0808
TH
811 phys_addr_t m_start = m->base;
812 phys_addr_t m_end = m->base + m->size;
f1af9d3a 813 int m_nid = memblock_get_region_node(m);
35fd0808
TH
814
815 /* only memory regions are associated with nodes, check it */
f1af9d3a 816 if (nid != NUMA_NO_NODE && nid != m_nid)
35fd0808
TH
817 continue;
818
f1af9d3a
PH
819 if (!type_b) {
820 if (out_start)
821 *out_start = m_start;
822 if (out_end)
823 *out_end = m_end;
824 if (out_nid)
825 *out_nid = m_nid;
826 idx_a++;
827 *idx = (u32)idx_a | (u64)idx_b << 32;
828 return;
829 }
830
831 /* scan areas before each reservation */
832 for (; idx_b < type_b->cnt + 1; idx_b++) {
833 struct memblock_region *r;
834 phys_addr_t r_start;
835 phys_addr_t r_end;
836
837 r = &type_b->regions[idx_b];
838 r_start = idx_b ? r[-1].base + r[-1].size : 0;
839 r_end = idx_b < type_b->cnt ?
840 r->base : ULLONG_MAX;
35fd0808 841
f1af9d3a
PH
842 /*
843 * if idx_b advanced past idx_a,
844 * break out to advance idx_a
845 */
35fd0808
TH
846 if (r_start >= m_end)
847 break;
848 /* if the two regions intersect, we're done */
849 if (m_start < r_end) {
850 if (out_start)
f1af9d3a
PH
851 *out_start =
852 max(m_start, r_start);
35fd0808
TH
853 if (out_end)
854 *out_end = min(m_end, r_end);
855 if (out_nid)
f1af9d3a 856 *out_nid = m_nid;
35fd0808 857 /*
f1af9d3a
PH
858 * The region which ends first is
859 * advanced for the next iteration.
35fd0808
TH
860 */
861 if (m_end <= r_end)
f1af9d3a 862 idx_a++;
35fd0808 863 else
f1af9d3a
PH
864 idx_b++;
865 *idx = (u32)idx_a | (u64)idx_b << 32;
35fd0808
TH
866 return;
867 }
868 }
869 }
870
871 /* signal end of iteration */
872 *idx = ULLONG_MAX;
873}
874
7bd0b0f0 875/**
f1af9d3a
PH
876 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
877 *
878 * Finds the next range from type_a which is not marked as unsuitable
879 * in type_b.
880 *
7bd0b0f0 881 * @idx: pointer to u64 loop variable
b1154233 882 * @nid: nid: node selector, %NUMA_NO_NODE for all nodes
f1af9d3a
PH
883 * @type_a: pointer to memblock_type from where the range is taken
884 * @type_b: pointer to memblock_type which excludes memory from being taken
dad7557e
WL
885 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
886 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
887 * @out_nid: ptr to int for nid of the range, can be %NULL
7bd0b0f0 888 *
f1af9d3a 889 * Reverse of __next_mem_range().
7bd0b0f0 890 */
f1af9d3a
PH
891void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
892 struct memblock_type *type_a,
893 struct memblock_type *type_b,
894 phys_addr_t *out_start,
895 phys_addr_t *out_end, int *out_nid)
7bd0b0f0 896{
f1af9d3a
PH
897 int idx_a = *idx & 0xffffffff;
898 int idx_b = *idx >> 32;
b1154233 899
560dca27
GS
900 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
901 nid = NUMA_NO_NODE;
7bd0b0f0
TH
902
903 if (*idx == (u64)ULLONG_MAX) {
f1af9d3a
PH
904 idx_a = type_a->cnt - 1;
905 idx_b = type_b->cnt;
7bd0b0f0
TH
906 }
907
f1af9d3a
PH
908 for (; idx_a >= 0; idx_a--) {
909 struct memblock_region *m = &type_a->regions[idx_a];
910
7bd0b0f0
TH
911 phys_addr_t m_start = m->base;
912 phys_addr_t m_end = m->base + m->size;
f1af9d3a 913 int m_nid = memblock_get_region_node(m);
7bd0b0f0
TH
914
915 /* only memory regions are associated with nodes, check it */
f1af9d3a 916 if (nid != NUMA_NO_NODE && nid != m_nid)
7bd0b0f0
TH
917 continue;
918
55ac590c
TC
919 /* skip hotpluggable memory regions if needed */
920 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
921 continue;
922
f1af9d3a
PH
923 if (!type_b) {
924 if (out_start)
925 *out_start = m_start;
926 if (out_end)
927 *out_end = m_end;
928 if (out_nid)
929 *out_nid = m_nid;
930 idx_a++;
931 *idx = (u32)idx_a | (u64)idx_b << 32;
932 return;
933 }
934
935 /* scan areas before each reservation */
936 for (; idx_b >= 0; idx_b--) {
937 struct memblock_region *r;
938 phys_addr_t r_start;
939 phys_addr_t r_end;
940
941 r = &type_b->regions[idx_b];
942 r_start = idx_b ? r[-1].base + r[-1].size : 0;
943 r_end = idx_b < type_b->cnt ?
944 r->base : ULLONG_MAX;
945 /*
946 * if idx_b advanced past idx_a,
947 * break out to advance idx_a
948 */
7bd0b0f0 949
7bd0b0f0
TH
950 if (r_end <= m_start)
951 break;
952 /* if the two regions intersect, we're done */
953 if (m_end > r_start) {
954 if (out_start)
955 *out_start = max(m_start, r_start);
956 if (out_end)
957 *out_end = min(m_end, r_end);
958 if (out_nid)
f1af9d3a 959 *out_nid = m_nid;
7bd0b0f0 960 if (m_start >= r_start)
f1af9d3a 961 idx_a--;
7bd0b0f0 962 else
f1af9d3a
PH
963 idx_b--;
964 *idx = (u32)idx_a | (u64)idx_b << 32;
7bd0b0f0
TH
965 return;
966 }
967 }
968 }
f1af9d3a 969 /* signal end of iteration */
7bd0b0f0
TH
970 *idx = ULLONG_MAX;
971}
972
7c0caeb8
TH
973#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
974/*
975 * Common iterator interface used to define for_each_mem_range().
976 */
977void __init_memblock __next_mem_pfn_range(int *idx, int nid,
978 unsigned long *out_start_pfn,
979 unsigned long *out_end_pfn, int *out_nid)
980{
981 struct memblock_type *type = &memblock.memory;
982 struct memblock_region *r;
983
984 while (++*idx < type->cnt) {
985 r = &type->regions[*idx];
986
987 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
988 continue;
989 if (nid == MAX_NUMNODES || nid == r->nid)
990 break;
991 }
992 if (*idx >= type->cnt) {
993 *idx = -1;
994 return;
995 }
996
997 if (out_start_pfn)
998 *out_start_pfn = PFN_UP(r->base);
999 if (out_end_pfn)
1000 *out_end_pfn = PFN_DOWN(r->base + r->size);
1001 if (out_nid)
1002 *out_nid = r->nid;
1003}
1004
1005/**
1006 * memblock_set_node - set node ID on memblock regions
1007 * @base: base of area to set node ID for
1008 * @size: size of area to set node ID for
e7e8de59 1009 * @type: memblock type to set node ID for
7c0caeb8
TH
1010 * @nid: node ID to set
1011 *
e7e8de59 1012 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
7c0caeb8
TH
1013 * Regions which cross the area boundaries are split as necessary.
1014 *
1015 * RETURNS:
1016 * 0 on success, -errno on failure.
1017 */
1018int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
e7e8de59 1019 struct memblock_type *type, int nid)
7c0caeb8 1020{
6a9ceb31
TH
1021 int start_rgn, end_rgn;
1022 int i, ret;
7c0caeb8 1023
6a9ceb31
TH
1024 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1025 if (ret)
1026 return ret;
7c0caeb8 1027
6a9ceb31 1028 for (i = start_rgn; i < end_rgn; i++)
e9d24ad3 1029 memblock_set_region_node(&type->regions[i], nid);
7c0caeb8
TH
1030
1031 memblock_merge_regions(type);
1032 return 0;
1033}
1034#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1035
2bfc2862
AM
1036static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1037 phys_addr_t align, phys_addr_t start,
1038 phys_addr_t end, int nid)
95f72d1e 1039{
6ed311b2 1040 phys_addr_t found;
95f72d1e 1041
79f40fab
GS
1042 if (!align)
1043 align = SMP_CACHE_BYTES;
94f3d3af 1044
2bfc2862 1045 found = memblock_find_in_range_node(size, align, start, end, nid);
9c8c27e2 1046 if (found && !memblock_reserve(found, size))
6ed311b2 1047 return found;
95f72d1e 1048
6ed311b2 1049 return 0;
95f72d1e
YL
1050}
1051
2bfc2862
AM
1052phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1053 phys_addr_t start, phys_addr_t end)
1054{
1055 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE);
1056}
1057
1058static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1059 phys_addr_t align, phys_addr_t max_addr,
1060 int nid)
1061{
1062 return memblock_alloc_range_nid(size, align, 0, max_addr, nid);
1063}
1064
7bd0b0f0
TH
1065phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1066{
1067 return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
1068}
1069
1070phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1071{
b1154233 1072 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE);
7bd0b0f0
TH
1073}
1074
6ed311b2 1075phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
95f72d1e 1076{
6ed311b2
BH
1077 phys_addr_t alloc;
1078
1079 alloc = __memblock_alloc_base(size, align, max_addr);
1080
1081 if (alloc == 0)
1082 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1083 (unsigned long long) size, (unsigned long long) max_addr);
1084
1085 return alloc;
95f72d1e
YL
1086}
1087
6ed311b2 1088phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
95f72d1e 1089{
6ed311b2
BH
1090 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1091}
95f72d1e 1092
9d1e2492
BH
1093phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1094{
1095 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1096
1097 if (res)
1098 return res;
15fb0972 1099 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
95f72d1e
YL
1100}
1101
26f09e9b
SS
1102/**
1103 * memblock_virt_alloc_internal - allocate boot memory block
1104 * @size: size of memory block to be allocated in bytes
1105 * @align: alignment of the region and block's size
1106 * @min_addr: the lower bound of the memory region to allocate (phys address)
1107 * @max_addr: the upper bound of the memory region to allocate (phys address)
1108 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1109 *
1110 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1111 * will fall back to memory below @min_addr. Also, allocation may fall back
1112 * to any node in the system if the specified node can not
1113 * hold the requested memory.
1114 *
1115 * The allocation is performed from memory region limited by
1116 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1117 *
1118 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1119 *
1120 * The phys address of allocated boot memory block is converted to virtual and
1121 * allocated memory is reset to 0.
1122 *
1123 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1124 * allocated boot memory block, so that it is never reported as leaks.
1125 *
1126 * RETURNS:
1127 * Virtual address of allocated memory block on success, NULL on failure.
1128 */
1129static void * __init memblock_virt_alloc_internal(
1130 phys_addr_t size, phys_addr_t align,
1131 phys_addr_t min_addr, phys_addr_t max_addr,
1132 int nid)
1133{
1134 phys_addr_t alloc;
1135 void *ptr;
1136
560dca27
GS
1137 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1138 nid = NUMA_NO_NODE;
26f09e9b
SS
1139
1140 /*
1141 * Detect any accidental use of these APIs after slab is ready, as at
1142 * this moment memblock may be deinitialized already and its
1143 * internal data may be destroyed (after execution of free_all_bootmem)
1144 */
1145 if (WARN_ON_ONCE(slab_is_available()))
1146 return kzalloc_node(size, GFP_NOWAIT, nid);
1147
1148 if (!align)
1149 align = SMP_CACHE_BYTES;
1150
f544e14f
YL
1151 if (max_addr > memblock.current_limit)
1152 max_addr = memblock.current_limit;
1153
26f09e9b
SS
1154again:
1155 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1156 nid);
1157 if (alloc)
1158 goto done;
1159
1160 if (nid != NUMA_NO_NODE) {
1161 alloc = memblock_find_in_range_node(size, align, min_addr,
1162 max_addr, NUMA_NO_NODE);
1163 if (alloc)
1164 goto done;
1165 }
1166
1167 if (min_addr) {
1168 min_addr = 0;
1169 goto again;
1170 } else {
1171 goto error;
1172 }
1173
1174done:
1175 memblock_reserve(alloc, size);
1176 ptr = phys_to_virt(alloc);
1177 memset(ptr, 0, size);
1178
1179 /*
1180 * The min_count is set to 0 so that bootmem allocated blocks
1181 * are never reported as leaks. This is because many of these blocks
1182 * are only referred via the physical address which is not
1183 * looked up by kmemleak.
1184 */
1185 kmemleak_alloc(ptr, size, 0, 0);
1186
1187 return ptr;
1188
1189error:
1190 return NULL;
1191}
1192
1193/**
1194 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1195 * @size: size of memory block to be allocated in bytes
1196 * @align: alignment of the region and block's size
1197 * @min_addr: the lower bound of the memory region from where the allocation
1198 * is preferred (phys address)
1199 * @max_addr: the upper bound of the memory region from where the allocation
1200 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1201 * allocate only from memory limited by memblock.current_limit value
1202 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1203 *
1204 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1205 * additional debug information (including caller info), if enabled.
1206 *
1207 * RETURNS:
1208 * Virtual address of allocated memory block on success, NULL on failure.
1209 */
1210void * __init memblock_virt_alloc_try_nid_nopanic(
1211 phys_addr_t size, phys_addr_t align,
1212 phys_addr_t min_addr, phys_addr_t max_addr,
1213 int nid)
1214{
1215 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1216 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1217 (u64)max_addr, (void *)_RET_IP_);
1218 return memblock_virt_alloc_internal(size, align, min_addr,
1219 max_addr, nid);
1220}
1221
1222/**
1223 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1224 * @size: size of memory block to be allocated in bytes
1225 * @align: alignment of the region and block's size
1226 * @min_addr: the lower bound of the memory region from where the allocation
1227 * is preferred (phys address)
1228 * @max_addr: the upper bound of the memory region from where the allocation
1229 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1230 * allocate only from memory limited by memblock.current_limit value
1231 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1232 *
1233 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1234 * which provides debug information (including caller info), if enabled,
1235 * and panics if the request can not be satisfied.
1236 *
1237 * RETURNS:
1238 * Virtual address of allocated memory block on success, NULL on failure.
1239 */
1240void * __init memblock_virt_alloc_try_nid(
1241 phys_addr_t size, phys_addr_t align,
1242 phys_addr_t min_addr, phys_addr_t max_addr,
1243 int nid)
1244{
1245 void *ptr;
1246
1247 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1248 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1249 (u64)max_addr, (void *)_RET_IP_);
1250 ptr = memblock_virt_alloc_internal(size, align,
1251 min_addr, max_addr, nid);
1252 if (ptr)
1253 return ptr;
1254
1255 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1256 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1257 (u64)max_addr);
1258 return NULL;
1259}
1260
1261/**
1262 * __memblock_free_early - free boot memory block
1263 * @base: phys starting address of the boot memory block
1264 * @size: size of the boot memory block in bytes
1265 *
1266 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1267 * The freeing memory will not be released to the buddy allocator.
1268 */
1269void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1270{
1271 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1272 __func__, (u64)base, (u64)base + size - 1,
1273 (void *)_RET_IP_);
1274 kmemleak_free_part(__va(base), size);
f1af9d3a 1275 memblock_remove_range(&memblock.reserved, base, size);
26f09e9b
SS
1276}
1277
1278/*
1279 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1280 * @addr: phys starting address of the boot memory block
1281 * @size: size of the boot memory block in bytes
1282 *
1283 * This is only useful when the bootmem allocator has already been torn
1284 * down, but we are still initializing the system. Pages are released directly
1285 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1286 */
1287void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1288{
1289 u64 cursor, end;
1290
1291 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1292 __func__, (u64)base, (u64)base + size - 1,
1293 (void *)_RET_IP_);
1294 kmemleak_free_part(__va(base), size);
1295 cursor = PFN_UP(base);
1296 end = PFN_DOWN(base + size);
1297
1298 for (; cursor < end; cursor++) {
1299 __free_pages_bootmem(pfn_to_page(cursor), 0);
1300 totalram_pages++;
1301 }
1302}
9d1e2492
BH
1303
1304/*
1305 * Remaining API functions
1306 */
1307
2898cc4c 1308phys_addr_t __init memblock_phys_mem_size(void)
95f72d1e 1309{
1440c4e2 1310 return memblock.memory.total_size;
95f72d1e
YL
1311}
1312
595ad9af
YL
1313phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1314{
1315 unsigned long pages = 0;
1316 struct memblock_region *r;
1317 unsigned long start_pfn, end_pfn;
1318
1319 for_each_memblock(memory, r) {
1320 start_pfn = memblock_region_memory_base_pfn(r);
1321 end_pfn = memblock_region_memory_end_pfn(r);
1322 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1323 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1324 pages += end_pfn - start_pfn;
1325 }
1326
16763230 1327 return PFN_PHYS(pages);
595ad9af
YL
1328}
1329
0a93ebef
SR
1330/* lowest address */
1331phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1332{
1333 return memblock.memory.regions[0].base;
1334}
1335
10d06439 1336phys_addr_t __init_memblock memblock_end_of_DRAM(void)
95f72d1e
YL
1337{
1338 int idx = memblock.memory.cnt - 1;
1339
e3239ff9 1340 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
95f72d1e
YL
1341}
1342
c0ce8fef 1343void __init memblock_enforce_memory_limit(phys_addr_t limit)
95f72d1e 1344{
c0ce8fef 1345 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
136199f0 1346 struct memblock_region *r;
95f72d1e 1347
c0ce8fef 1348 if (!limit)
95f72d1e
YL
1349 return;
1350
c0ce8fef 1351 /* find out max address */
136199f0 1352 for_each_memblock(memory, r) {
c0ce8fef
TH
1353 if (limit <= r->size) {
1354 max_addr = r->base + limit;
1355 break;
95f72d1e 1356 }
c0ce8fef 1357 limit -= r->size;
95f72d1e 1358 }
c0ce8fef
TH
1359
1360 /* truncate both memory and reserved regions */
f1af9d3a
PH
1361 memblock_remove_range(&memblock.memory, max_addr,
1362 (phys_addr_t)ULLONG_MAX);
1363 memblock_remove_range(&memblock.reserved, max_addr,
1364 (phys_addr_t)ULLONG_MAX);
95f72d1e
YL
1365}
1366
cd79481d 1367static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
72d4b0b4
BH
1368{
1369 unsigned int left = 0, right = type->cnt;
1370
1371 do {
1372 unsigned int mid = (right + left) / 2;
1373
1374 if (addr < type->regions[mid].base)
1375 right = mid;
1376 else if (addr >= (type->regions[mid].base +
1377 type->regions[mid].size))
1378 left = mid + 1;
1379 else
1380 return mid;
1381 } while (left < right);
1382 return -1;
1383}
1384
2898cc4c 1385int __init memblock_is_reserved(phys_addr_t addr)
95f72d1e 1386{
72d4b0b4
BH
1387 return memblock_search(&memblock.reserved, addr) != -1;
1388}
95f72d1e 1389
3661ca66 1390int __init_memblock memblock_is_memory(phys_addr_t addr)
72d4b0b4
BH
1391{
1392 return memblock_search(&memblock.memory, addr) != -1;
1393}
1394
e76b63f8
YL
1395#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1396int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1397 unsigned long *start_pfn, unsigned long *end_pfn)
1398{
1399 struct memblock_type *type = &memblock.memory;
16763230 1400 int mid = memblock_search(type, PFN_PHYS(pfn));
e76b63f8
YL
1401
1402 if (mid == -1)
1403 return -1;
1404
1405 *start_pfn = type->regions[mid].base >> PAGE_SHIFT;
1406 *end_pfn = (type->regions[mid].base + type->regions[mid].size)
1407 >> PAGE_SHIFT;
1408
1409 return type->regions[mid].nid;
1410}
1411#endif
1412
eab30949
SB
1413/**
1414 * memblock_is_region_memory - check if a region is a subset of memory
1415 * @base: base of region to check
1416 * @size: size of region to check
1417 *
1418 * Check if the region [@base, @base+@size) is a subset of a memory block.
1419 *
1420 * RETURNS:
1421 * 0 if false, non-zero if true
1422 */
3661ca66 1423int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
72d4b0b4 1424{
abb65272 1425 int idx = memblock_search(&memblock.memory, base);
eb18f1b5 1426 phys_addr_t end = base + memblock_cap_size(base, &size);
72d4b0b4
BH
1427
1428 if (idx == -1)
1429 return 0;
abb65272
TV
1430 return memblock.memory.regions[idx].base <= base &&
1431 (memblock.memory.regions[idx].base +
eb18f1b5 1432 memblock.memory.regions[idx].size) >= end;
95f72d1e
YL
1433}
1434
eab30949
SB
1435/**
1436 * memblock_is_region_reserved - check if a region intersects reserved memory
1437 * @base: base of region to check
1438 * @size: size of region to check
1439 *
1440 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1441 *
1442 * RETURNS:
1443 * 0 if false, non-zero if true
1444 */
10d06439 1445int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
95f72d1e 1446{
eb18f1b5 1447 memblock_cap_size(base, &size);
f1c2c19c 1448 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
95f72d1e
YL
1449}
1450
6ede1fd3
YL
1451void __init_memblock memblock_trim_memory(phys_addr_t align)
1452{
6ede1fd3 1453 phys_addr_t start, end, orig_start, orig_end;
136199f0 1454 struct memblock_region *r;
6ede1fd3 1455
136199f0
EM
1456 for_each_memblock(memory, r) {
1457 orig_start = r->base;
1458 orig_end = r->base + r->size;
6ede1fd3
YL
1459 start = round_up(orig_start, align);
1460 end = round_down(orig_end, align);
1461
1462 if (start == orig_start && end == orig_end)
1463 continue;
1464
1465 if (start < end) {
136199f0
EM
1466 r->base = start;
1467 r->size = end - start;
6ede1fd3 1468 } else {
136199f0
EM
1469 memblock_remove_region(&memblock.memory,
1470 r - memblock.memory.regions);
1471 r--;
6ede1fd3
YL
1472 }
1473 }
1474}
e63075a3 1475
3661ca66 1476void __init_memblock memblock_set_current_limit(phys_addr_t limit)
e63075a3
BH
1477{
1478 memblock.current_limit = limit;
1479}
1480
fec51014
LA
1481phys_addr_t __init_memblock memblock_get_current_limit(void)
1482{
1483 return memblock.current_limit;
1484}
1485
7c0caeb8 1486static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
6ed311b2
BH
1487{
1488 unsigned long long base, size;
66a20757 1489 unsigned long flags;
6ed311b2
BH
1490 int i;
1491
7c0caeb8 1492 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
6ed311b2 1493
7c0caeb8
TH
1494 for (i = 0; i < type->cnt; i++) {
1495 struct memblock_region *rgn = &type->regions[i];
1496 char nid_buf[32] = "";
1497
1498 base = rgn->base;
1499 size = rgn->size;
66a20757 1500 flags = rgn->flags;
7c0caeb8
TH
1501#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1502 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1503 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1504 memblock_get_region_node(rgn));
1505#endif
66a20757
TC
1506 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1507 name, i, base, base + size - 1, size, nid_buf, flags);
6ed311b2
BH
1508 }
1509}
1510
4ff7b82f 1511void __init_memblock __memblock_dump_all(void)
6ed311b2 1512{
6ed311b2 1513 pr_info("MEMBLOCK configuration:\n");
1440c4e2
TH
1514 pr_info(" memory size = %#llx reserved size = %#llx\n",
1515 (unsigned long long)memblock.memory.total_size,
1516 (unsigned long long)memblock.reserved.total_size);
6ed311b2
BH
1517
1518 memblock_dump(&memblock.memory, "memory");
1519 memblock_dump(&memblock.reserved, "reserved");
1520}
1521
1aadc056 1522void __init memblock_allow_resize(void)
6ed311b2 1523{
142b45a7 1524 memblock_can_resize = 1;
6ed311b2
BH
1525}
1526
6ed311b2
BH
1527static int __init early_memblock(char *p)
1528{
1529 if (p && strstr(p, "debug"))
1530 memblock_debug = 1;
1531 return 0;
1532}
1533early_param("memblock", early_memblock);
1534
c378ddd5 1535#if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
6d03b885
BH
1536
1537static int memblock_debug_show(struct seq_file *m, void *private)
1538{
1539 struct memblock_type *type = m->private;
1540 struct memblock_region *reg;
1541 int i;
1542
1543 for (i = 0; i < type->cnt; i++) {
1544 reg = &type->regions[i];
1545 seq_printf(m, "%4d: ", i);
1546 if (sizeof(phys_addr_t) == 4)
1547 seq_printf(m, "0x%08lx..0x%08lx\n",
1548 (unsigned long)reg->base,
1549 (unsigned long)(reg->base + reg->size - 1));
1550 else
1551 seq_printf(m, "0x%016llx..0x%016llx\n",
1552 (unsigned long long)reg->base,
1553 (unsigned long long)(reg->base + reg->size - 1));
1554
1555 }
1556 return 0;
1557}
1558
1559static int memblock_debug_open(struct inode *inode, struct file *file)
1560{
1561 return single_open(file, memblock_debug_show, inode->i_private);
1562}
1563
1564static const struct file_operations memblock_debug_fops = {
1565 .open = memblock_debug_open,
1566 .read = seq_read,
1567 .llseek = seq_lseek,
1568 .release = single_release,
1569};
1570
1571static int __init memblock_init_debugfs(void)
1572{
1573 struct dentry *root = debugfs_create_dir("memblock", NULL);
1574 if (!root)
1575 return -ENXIO;
1576 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1577 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
70210ed9
PH
1578#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1579 debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1580#endif
6d03b885
BH
1581
1582 return 0;
1583}
1584__initcall(memblock_init_debugfs);
1585
1586#endif /* CONFIG_DEBUG_FS */