mm: vmscan: call NUMA-unaware shrinkers irrespective of nodemask
[deliverable/linux.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;
30
31struct memblock memblock __initdata_memblock = {
32 .memory.regions = memblock_memory_init_regions,
33 .memory.cnt = 1, /* empty dummy entry */
34 .memory.max = INIT_MEMBLOCK_REGIONS,
35
36 .reserved.regions = memblock_reserved_init_regions,
37 .reserved.cnt = 1, /* empty dummy entry */
38 .reserved.max = INIT_MEMBLOCK_REGIONS,
39
79442ed1 40 .bottom_up = false,
fe091c20
TH
41 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
42};
95f72d1e 43
10d06439 44int memblock_debug __initdata_memblock;
55ac590c
TC
45#ifdef CONFIG_MOVABLE_NODE
46bool movable_node_enabled __initdata_memblock = false;
47#endif
1aadc056 48static int memblock_can_resize __initdata_memblock;
181eb394
GS
49static int memblock_memory_in_slab __initdata_memblock = 0;
50static int memblock_reserved_in_slab __initdata_memblock = 0;
95f72d1e 51
142b45a7 52/* inline so we don't get a warning when pr_debug is compiled out */
c2233116
RP
53static __init_memblock const char *
54memblock_type_name(struct memblock_type *type)
142b45a7
BH
55{
56 if (type == &memblock.memory)
57 return "memory";
58 else if (type == &memblock.reserved)
59 return "reserved";
60 else
61 return "unknown";
62}
63
eb18f1b5
TH
64/* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
65static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
66{
67 return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
68}
69
6ed311b2
BH
70/*
71 * Address comparison utilities
72 */
10d06439 73static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
2898cc4c 74 phys_addr_t base2, phys_addr_t size2)
95f72d1e
YL
75{
76 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
77}
78
2d7d3eb2
HS
79static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
80 phys_addr_t base, phys_addr_t size)
6ed311b2
BH
81{
82 unsigned long i;
83
84 for (i = 0; i < type->cnt; i++) {
85 phys_addr_t rgnbase = type->regions[i].base;
86 phys_addr_t rgnsize = type->regions[i].size;
87 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
88 break;
89 }
90
91 return (i < type->cnt) ? i : -1;
92}
93
79442ed1
TC
94/*
95 * __memblock_find_range_bottom_up - find free area utility in bottom-up
96 * @start: start of candidate range
97 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
98 * @size: size of free area to find
99 * @align: alignment of free area to find
b1154233 100 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
79442ed1
TC
101 *
102 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
103 *
104 * RETURNS:
105 * Found address on success, 0 on failure.
106 */
107static phys_addr_t __init_memblock
108__memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
109 phys_addr_t size, phys_addr_t align, int nid)
110{
111 phys_addr_t this_start, this_end, cand;
112 u64 i;
113
114 for_each_free_mem_range(i, nid, &this_start, &this_end, NULL) {
115 this_start = clamp(this_start, start, end);
116 this_end = clamp(this_end, start, end);
117
118 cand = round_up(this_start, align);
119 if (cand < this_end && this_end - cand >= size)
120 return cand;
121 }
122
123 return 0;
124}
125
7bd0b0f0 126/**
1402899e 127 * __memblock_find_range_top_down - find free area utility, in top-down
7bd0b0f0
TH
128 * @start: start of candidate range
129 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
130 * @size: size of free area to find
131 * @align: alignment of free area to find
b1154233 132 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
7bd0b0f0 133 *
1402899e 134 * Utility called from memblock_find_in_range_node(), find free area top-down.
7bd0b0f0
TH
135 *
136 * RETURNS:
79442ed1 137 * Found address on success, 0 on failure.
6ed311b2 138 */
1402899e
TC
139static phys_addr_t __init_memblock
140__memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
141 phys_addr_t size, phys_addr_t align, int nid)
f7210e6c
TC
142{
143 phys_addr_t this_start, this_end, cand;
144 u64 i;
145
f7210e6c
TC
146 for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) {
147 this_start = clamp(this_start, start, end);
148 this_end = clamp(this_end, start, end);
149
150 if (this_end < size)
151 continue;
152
153 cand = round_down(this_end - size, align);
154 if (cand >= this_start)
155 return cand;
156 }
1402899e 157
f7210e6c
TC
158 return 0;
159}
6ed311b2 160
1402899e
TC
161/**
162 * memblock_find_in_range_node - find free area in given range and node
1402899e
TC
163 * @size: size of free area to find
164 * @align: alignment of free area to find
87029ee9
GS
165 * @start: start of candidate range
166 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
b1154233 167 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1402899e
TC
168 *
169 * Find @size free area aligned to @align in the specified range and node.
170 *
79442ed1
TC
171 * When allocation direction is bottom-up, the @start should be greater
172 * than the end of the kernel image. Otherwise, it will be trimmed. The
173 * reason is that we want the bottom-up allocation just near the kernel
174 * image so it is highly likely that the allocated memory and the kernel
175 * will reside in the same node.
176 *
177 * If bottom-up allocation failed, will try to allocate memory top-down.
178 *
1402899e 179 * RETURNS:
79442ed1 180 * Found address on success, 0 on failure.
1402899e 181 */
87029ee9
GS
182phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
183 phys_addr_t align, phys_addr_t start,
184 phys_addr_t end, int nid)
1402899e 185{
79442ed1
TC
186 int ret;
187 phys_addr_t kernel_end;
188
1402899e
TC
189 /* pump up @end */
190 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
191 end = memblock.current_limit;
192
193 /* avoid allocating the first page */
194 start = max_t(phys_addr_t, start, PAGE_SIZE);
195 end = max(start, end);
79442ed1
TC
196 kernel_end = __pa_symbol(_end);
197
198 /*
199 * try bottom-up allocation only when bottom-up mode
200 * is set and @end is above the kernel image.
201 */
202 if (memblock_bottom_up() && end > kernel_end) {
203 phys_addr_t bottom_up_start;
204
205 /* make sure we will allocate above the kernel */
206 bottom_up_start = max(start, kernel_end);
207
208 /* ok, try bottom-up allocation first */
209 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
210 size, align, nid);
211 if (ret)
212 return ret;
213
214 /*
215 * we always limit bottom-up allocation above the kernel,
216 * but top-down allocation doesn't have the limit, so
217 * retrying top-down allocation may succeed when bottom-up
218 * allocation failed.
219 *
220 * bottom-up allocation is expected to be fail very rarely,
221 * so we use WARN_ONCE() here to see the stack trace if
222 * fail happens.
223 */
224 WARN_ONCE(1, "memblock: bottom-up allocation failed, "
225 "memory hotunplug may be affected\n");
226 }
1402899e
TC
227
228 return __memblock_find_range_top_down(start, end, size, align, nid);
229}
230
7bd0b0f0
TH
231/**
232 * memblock_find_in_range - find free area in given range
233 * @start: start of candidate range
234 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
235 * @size: size of free area to find
236 * @align: alignment of free area to find
237 *
238 * Find @size free area aligned to @align in the specified range.
239 *
240 * RETURNS:
79442ed1 241 * Found address on success, 0 on failure.
fc769a8e 242 */
7bd0b0f0
TH
243phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
244 phys_addr_t end, phys_addr_t size,
245 phys_addr_t align)
6ed311b2 246{
87029ee9 247 return memblock_find_in_range_node(size, align, start, end,
b1154233 248 NUMA_NO_NODE);
6ed311b2
BH
249}
250
10d06439 251static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
95f72d1e 252{
1440c4e2 253 type->total_size -= type->regions[r].size;
7c0caeb8
TH
254 memmove(&type->regions[r], &type->regions[r + 1],
255 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
e3239ff9 256 type->cnt--;
95f72d1e 257
8f7a6605
BH
258 /* Special case for empty arrays */
259 if (type->cnt == 0) {
1440c4e2 260 WARN_ON(type->total_size != 0);
8f7a6605
BH
261 type->cnt = 1;
262 type->regions[0].base = 0;
263 type->regions[0].size = 0;
66a20757 264 type->regions[0].flags = 0;
7c0caeb8 265 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
8f7a6605 266 }
95f72d1e
YL
267}
268
29f67386
YL
269phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
270 phys_addr_t *addr)
271{
272 if (memblock.reserved.regions == memblock_reserved_init_regions)
273 return 0;
274
fd615c4e
GS
275 /*
276 * Don't allow nobootmem allocator to free reserved memory regions
277 * array if
278 * - CONFIG_DEBUG_FS is enabled;
279 * - CONFIG_ARCH_DISCARD_MEMBLOCK is not enabled;
280 * - reserved memory regions array have been resized during boot.
281 * Otherwise debug_fs entry "sys/kernel/debug/memblock/reserved"
282 * will show garbage instead of state of memory reservations.
283 */
284 if (IS_ENABLED(CONFIG_DEBUG_FS) &&
285 !IS_ENABLED(CONFIG_ARCH_DISCARD_MEMBLOCK))
286 return 0;
287
29f67386
YL
288 *addr = __pa(memblock.reserved.regions);
289
290 return PAGE_ALIGN(sizeof(struct memblock_region) *
291 memblock.reserved.max);
292}
293
5e270e25
PH
294#ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
295
296phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
297 phys_addr_t *addr)
298{
299 if (memblock.memory.regions == memblock_memory_init_regions)
300 return 0;
301
302 *addr = __pa(memblock.memory.regions);
303
304 return PAGE_ALIGN(sizeof(struct memblock_region) *
305 memblock.memory.max);
306}
307
308#endif
309
48c3b583
GP
310/**
311 * memblock_double_array - double the size of the memblock regions array
312 * @type: memblock type of the regions array being doubled
313 * @new_area_start: starting address of memory range to avoid overlap with
314 * @new_area_size: size of memory range to avoid overlap with
315 *
316 * Double the size of the @type regions array. If memblock is being used to
317 * allocate memory for a new reserved regions array and there is a previously
318 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
319 * waiting to be reserved, ensure the memory used by the new array does
320 * not overlap.
321 *
322 * RETURNS:
323 * 0 on success, -1 on failure.
324 */
325static int __init_memblock memblock_double_array(struct memblock_type *type,
326 phys_addr_t new_area_start,
327 phys_addr_t new_area_size)
142b45a7
BH
328{
329 struct memblock_region *new_array, *old_array;
29f67386 330 phys_addr_t old_alloc_size, new_alloc_size;
142b45a7
BH
331 phys_addr_t old_size, new_size, addr;
332 int use_slab = slab_is_available();
181eb394 333 int *in_slab;
142b45a7
BH
334
335 /* We don't allow resizing until we know about the reserved regions
336 * of memory that aren't suitable for allocation
337 */
338 if (!memblock_can_resize)
339 return -1;
340
142b45a7
BH
341 /* Calculate new doubled size */
342 old_size = type->max * sizeof(struct memblock_region);
343 new_size = old_size << 1;
29f67386
YL
344 /*
345 * We need to allocated new one align to PAGE_SIZE,
346 * so we can free them completely later.
347 */
348 old_alloc_size = PAGE_ALIGN(old_size);
349 new_alloc_size = PAGE_ALIGN(new_size);
142b45a7 350
181eb394
GS
351 /* Retrieve the slab flag */
352 if (type == &memblock.memory)
353 in_slab = &memblock_memory_in_slab;
354 else
355 in_slab = &memblock_reserved_in_slab;
356
142b45a7
BH
357 /* Try to find some space for it.
358 *
359 * WARNING: We assume that either slab_is_available() and we use it or
fd07383b
AM
360 * we use MEMBLOCK for allocations. That means that this is unsafe to
361 * use when bootmem is currently active (unless bootmem itself is
362 * implemented on top of MEMBLOCK which isn't the case yet)
142b45a7
BH
363 *
364 * This should however not be an issue for now, as we currently only
fd07383b
AM
365 * call into MEMBLOCK while it's still active, or much later when slab
366 * is active for memory hotplug operations
142b45a7
BH
367 */
368 if (use_slab) {
369 new_array = kmalloc(new_size, GFP_KERNEL);
1f5026a7 370 addr = new_array ? __pa(new_array) : 0;
4e2f0775 371 } else {
48c3b583
GP
372 /* only exclude range when trying to double reserved.regions */
373 if (type != &memblock.reserved)
374 new_area_start = new_area_size = 0;
375
376 addr = memblock_find_in_range(new_area_start + new_area_size,
377 memblock.current_limit,
29f67386 378 new_alloc_size, PAGE_SIZE);
48c3b583
GP
379 if (!addr && new_area_size)
380 addr = memblock_find_in_range(0,
fd07383b
AM
381 min(new_area_start, memblock.current_limit),
382 new_alloc_size, PAGE_SIZE);
48c3b583 383
15674868 384 new_array = addr ? __va(addr) : NULL;
4e2f0775 385 }
1f5026a7 386 if (!addr) {
142b45a7
BH
387 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
388 memblock_type_name(type), type->max, type->max * 2);
389 return -1;
390 }
142b45a7 391
fd07383b
AM
392 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
393 memblock_type_name(type), type->max * 2, (u64)addr,
394 (u64)addr + new_size - 1);
ea9e4376 395
fd07383b
AM
396 /*
397 * Found space, we now need to move the array over before we add the
398 * reserved region since it may be our reserved array itself that is
399 * full.
142b45a7
BH
400 */
401 memcpy(new_array, type->regions, old_size);
402 memset(new_array + type->max, 0, old_size);
403 old_array = type->regions;
404 type->regions = new_array;
405 type->max <<= 1;
406
fd07383b 407 /* Free old array. We needn't free it if the array is the static one */
181eb394
GS
408 if (*in_slab)
409 kfree(old_array);
410 else if (old_array != memblock_memory_init_regions &&
411 old_array != memblock_reserved_init_regions)
29f67386 412 memblock_free(__pa(old_array), old_alloc_size);
142b45a7 413
fd07383b
AM
414 /*
415 * Reserve the new array if that comes from the memblock. Otherwise, we
416 * needn't do it
181eb394
GS
417 */
418 if (!use_slab)
29f67386 419 BUG_ON(memblock_reserve(addr, new_alloc_size));
181eb394
GS
420
421 /* Update slab flag */
422 *in_slab = use_slab;
423
142b45a7
BH
424 return 0;
425}
426
784656f9
TH
427/**
428 * memblock_merge_regions - merge neighboring compatible regions
429 * @type: memblock type to scan
430 *
431 * Scan @type and merge neighboring compatible regions.
432 */
433static void __init_memblock memblock_merge_regions(struct memblock_type *type)
95f72d1e 434{
784656f9 435 int i = 0;
95f72d1e 436
784656f9
TH
437 /* cnt never goes below 1 */
438 while (i < type->cnt - 1) {
439 struct memblock_region *this = &type->regions[i];
440 struct memblock_region *next = &type->regions[i + 1];
95f72d1e 441
7c0caeb8
TH
442 if (this->base + this->size != next->base ||
443 memblock_get_region_node(this) !=
66a20757
TC
444 memblock_get_region_node(next) ||
445 this->flags != next->flags) {
784656f9
TH
446 BUG_ON(this->base + this->size > next->base);
447 i++;
448 continue;
8f7a6605
BH
449 }
450
784656f9 451 this->size += next->size;
c0232ae8
LF
452 /* move forward from next + 1, index of which is i + 2 */
453 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
784656f9 454 type->cnt--;
95f72d1e 455 }
784656f9 456}
95f72d1e 457
784656f9
TH
458/**
459 * memblock_insert_region - insert new memblock region
209ff86d
TC
460 * @type: memblock type to insert into
461 * @idx: index for the insertion point
462 * @base: base address of the new region
463 * @size: size of the new region
464 * @nid: node id of the new region
66a20757 465 * @flags: flags of the new region
784656f9
TH
466 *
467 * Insert new memblock region [@base,@base+@size) into @type at @idx.
468 * @type must already have extra room to accomodate the new region.
469 */
470static void __init_memblock memblock_insert_region(struct memblock_type *type,
471 int idx, phys_addr_t base,
66a20757
TC
472 phys_addr_t size,
473 int nid, unsigned long flags)
784656f9
TH
474{
475 struct memblock_region *rgn = &type->regions[idx];
476
477 BUG_ON(type->cnt >= type->max);
478 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
479 rgn->base = base;
480 rgn->size = size;
66a20757 481 rgn->flags = flags;
7c0caeb8 482 memblock_set_region_node(rgn, nid);
784656f9 483 type->cnt++;
1440c4e2 484 type->total_size += size;
784656f9
TH
485}
486
487/**
488 * memblock_add_region - add new memblock region
489 * @type: memblock type to add new region into
490 * @base: base address of the new region
491 * @size: size of the new region
7fb0bc3f 492 * @nid: nid of the new region
66a20757 493 * @flags: flags of the new region
784656f9
TH
494 *
495 * Add new memblock region [@base,@base+@size) into @type. The new region
496 * is allowed to overlap with existing ones - overlaps don't affect already
497 * existing regions. @type is guaranteed to be minimal (all neighbouring
498 * compatible regions are merged) after the addition.
499 *
500 * RETURNS:
501 * 0 on success, -errno on failure.
502 */
581adcbe 503static int __init_memblock memblock_add_region(struct memblock_type *type,
66a20757
TC
504 phys_addr_t base, phys_addr_t size,
505 int nid, unsigned long flags)
784656f9
TH
506{
507 bool insert = false;
eb18f1b5
TH
508 phys_addr_t obase = base;
509 phys_addr_t end = base + memblock_cap_size(base, &size);
784656f9
TH
510 int i, nr_new;
511
b3dc627c
TH
512 if (!size)
513 return 0;
514
784656f9
TH
515 /* special case for empty array */
516 if (type->regions[0].size == 0) {
1440c4e2 517 WARN_ON(type->cnt != 1 || type->total_size);
8f7a6605
BH
518 type->regions[0].base = base;
519 type->regions[0].size = size;
66a20757 520 type->regions[0].flags = flags;
7fb0bc3f 521 memblock_set_region_node(&type->regions[0], nid);
1440c4e2 522 type->total_size = size;
8f7a6605 523 return 0;
95f72d1e 524 }
784656f9
TH
525repeat:
526 /*
527 * The following is executed twice. Once with %false @insert and
528 * then with %true. The first counts the number of regions needed
529 * to accomodate the new area. The second actually inserts them.
142b45a7 530 */
784656f9
TH
531 base = obase;
532 nr_new = 0;
95f72d1e 533
784656f9
TH
534 for (i = 0; i < type->cnt; i++) {
535 struct memblock_region *rgn = &type->regions[i];
536 phys_addr_t rbase = rgn->base;
537 phys_addr_t rend = rbase + rgn->size;
538
539 if (rbase >= end)
95f72d1e 540 break;
784656f9
TH
541 if (rend <= base)
542 continue;
543 /*
544 * @rgn overlaps. If it separates the lower part of new
545 * area, insert that portion.
546 */
547 if (rbase > base) {
548 nr_new++;
549 if (insert)
550 memblock_insert_region(type, i++, base,
66a20757
TC
551 rbase - base, nid,
552 flags);
95f72d1e 553 }
784656f9
TH
554 /* area below @rend is dealt with, forget about it */
555 base = min(rend, end);
95f72d1e 556 }
784656f9
TH
557
558 /* insert the remaining portion */
559 if (base < end) {
560 nr_new++;
561 if (insert)
66a20757
TC
562 memblock_insert_region(type, i, base, end - base,
563 nid, flags);
95f72d1e 564 }
95f72d1e 565
784656f9
TH
566 /*
567 * If this was the first round, resize array and repeat for actual
568 * insertions; otherwise, merge and return.
142b45a7 569 */
784656f9
TH
570 if (!insert) {
571 while (type->cnt + nr_new > type->max)
48c3b583 572 if (memblock_double_array(type, obase, size) < 0)
784656f9
TH
573 return -ENOMEM;
574 insert = true;
575 goto repeat;
576 } else {
577 memblock_merge_regions(type);
578 return 0;
142b45a7 579 }
95f72d1e
YL
580}
581
7fb0bc3f
TH
582int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
583 int nid)
584{
66a20757 585 return memblock_add_region(&memblock.memory, base, size, nid, 0);
7fb0bc3f
TH
586}
587
581adcbe 588int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
95f72d1e 589{
66a20757
TC
590 return memblock_add_region(&memblock.memory, base, size,
591 MAX_NUMNODES, 0);
95f72d1e
YL
592}
593
6a9ceb31
TH
594/**
595 * memblock_isolate_range - isolate given range into disjoint memblocks
596 * @type: memblock type to isolate range for
597 * @base: base of range to isolate
598 * @size: size of range to isolate
599 * @start_rgn: out parameter for the start of isolated region
600 * @end_rgn: out parameter for the end of isolated region
601 *
602 * Walk @type and ensure that regions don't cross the boundaries defined by
603 * [@base,@base+@size). Crossing regions are split at the boundaries,
604 * which may create at most two more regions. The index of the first
605 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
606 *
607 * RETURNS:
608 * 0 on success, -errno on failure.
609 */
610static int __init_memblock memblock_isolate_range(struct memblock_type *type,
611 phys_addr_t base, phys_addr_t size,
612 int *start_rgn, int *end_rgn)
613{
eb18f1b5 614 phys_addr_t end = base + memblock_cap_size(base, &size);
6a9ceb31
TH
615 int i;
616
617 *start_rgn = *end_rgn = 0;
618
b3dc627c
TH
619 if (!size)
620 return 0;
621
6a9ceb31
TH
622 /* we'll create at most two more regions */
623 while (type->cnt + 2 > type->max)
48c3b583 624 if (memblock_double_array(type, base, size) < 0)
6a9ceb31
TH
625 return -ENOMEM;
626
627 for (i = 0; i < type->cnt; i++) {
628 struct memblock_region *rgn = &type->regions[i];
629 phys_addr_t rbase = rgn->base;
630 phys_addr_t rend = rbase + rgn->size;
631
632 if (rbase >= end)
633 break;
634 if (rend <= base)
635 continue;
636
637 if (rbase < base) {
638 /*
639 * @rgn intersects from below. Split and continue
640 * to process the next region - the new top half.
641 */
642 rgn->base = base;
1440c4e2
TH
643 rgn->size -= base - rbase;
644 type->total_size -= base - rbase;
6a9ceb31 645 memblock_insert_region(type, i, rbase, base - rbase,
66a20757
TC
646 memblock_get_region_node(rgn),
647 rgn->flags);
6a9ceb31
TH
648 } else if (rend > end) {
649 /*
650 * @rgn intersects from above. Split and redo the
651 * current region - the new bottom half.
652 */
653 rgn->base = end;
1440c4e2
TH
654 rgn->size -= end - rbase;
655 type->total_size -= end - rbase;
6a9ceb31 656 memblock_insert_region(type, i--, rbase, end - rbase,
66a20757
TC
657 memblock_get_region_node(rgn),
658 rgn->flags);
6a9ceb31
TH
659 } else {
660 /* @rgn is fully contained, record it */
661 if (!*end_rgn)
662 *start_rgn = i;
663 *end_rgn = i + 1;
664 }
665 }
666
667 return 0;
668}
6a9ceb31 669
581adcbe
TH
670static int __init_memblock __memblock_remove(struct memblock_type *type,
671 phys_addr_t base, phys_addr_t size)
95f72d1e 672{
71936180
TH
673 int start_rgn, end_rgn;
674 int i, ret;
95f72d1e 675
71936180
TH
676 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
677 if (ret)
678 return ret;
95f72d1e 679
71936180
TH
680 for (i = end_rgn - 1; i >= start_rgn; i--)
681 memblock_remove_region(type, i);
8f7a6605 682 return 0;
95f72d1e
YL
683}
684
581adcbe 685int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
95f72d1e
YL
686{
687 return __memblock_remove(&memblock.memory, base, size);
688}
689
581adcbe 690int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
95f72d1e 691{
24aa0788 692 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
a150439c 693 (unsigned long long)base,
931d13f5 694 (unsigned long long)base + size - 1,
a150439c 695 (void *)_RET_IP_);
24aa0788 696
95f72d1e
YL
697 return __memblock_remove(&memblock.reserved, base, size);
698}
699
66a20757
TC
700static int __init_memblock memblock_reserve_region(phys_addr_t base,
701 phys_addr_t size,
702 int nid,
703 unsigned long flags)
95f72d1e 704{
e3239ff9 705 struct memblock_type *_rgn = &memblock.reserved;
95f72d1e 706
66a20757 707 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
a150439c 708 (unsigned long long)base,
931d13f5 709 (unsigned long long)base + size - 1,
66a20757
TC
710 flags, (void *)_RET_IP_);
711
712 return memblock_add_region(_rgn, base, size, nid, flags);
713}
95f72d1e 714
66a20757
TC
715int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
716{
717 return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
95f72d1e
YL
718}
719
66b16edf
TC
720/**
721 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
722 * @base: the base phys addr of the region
723 * @size: the size of the region
724 *
725 * This function isolates region [@base, @base + @size), and mark it with flag
726 * MEMBLOCK_HOTPLUG.
727 *
728 * Return 0 on succees, -errno on failure.
729 */
730int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
731{
732 struct memblock_type *type = &memblock.memory;
733 int i, ret, start_rgn, end_rgn;
734
735 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
736 if (ret)
737 return ret;
738
739 for (i = start_rgn; i < end_rgn; i++)
740 memblock_set_region_flags(&type->regions[i], MEMBLOCK_HOTPLUG);
741
742 memblock_merge_regions(type);
743 return 0;
744}
745
746/**
747 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
748 * @base: the base phys addr of the region
749 * @size: the size of the region
750 *
751 * This function isolates region [@base, @base + @size), and clear flag
752 * MEMBLOCK_HOTPLUG for the isolated regions.
753 *
754 * Return 0 on succees, -errno on failure.
755 */
756int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
757{
758 struct memblock_type *type = &memblock.memory;
759 int i, ret, start_rgn, end_rgn;
760
761 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
762 if (ret)
763 return ret;
764
765 for (i = start_rgn; i < end_rgn; i++)
766 memblock_clear_region_flags(&type->regions[i],
767 MEMBLOCK_HOTPLUG);
768
769 memblock_merge_regions(type);
770 return 0;
771}
772
35fd0808
TH
773/**
774 * __next_free_mem_range - next function for for_each_free_mem_range()
775 * @idx: pointer to u64 loop variable
b1154233 776 * @nid: node selector, %NUMA_NO_NODE for all nodes
dad7557e
WL
777 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
778 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
779 * @out_nid: ptr to int for nid of the range, can be %NULL
35fd0808
TH
780 *
781 * Find the first free area from *@idx which matches @nid, fill the out
782 * parameters, and update *@idx for the next iteration. The lower 32bit of
783 * *@idx contains index into memory region and the upper 32bit indexes the
784 * areas before each reserved region. For example, if reserved regions
785 * look like the following,
786 *
787 * 0:[0-16), 1:[32-48), 2:[128-130)
788 *
789 * The upper 32bit indexes the following regions.
790 *
791 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
792 *
793 * As both region arrays are sorted, the function advances the two indices
794 * in lockstep and returns each intersection.
795 */
796void __init_memblock __next_free_mem_range(u64 *idx, int nid,
797 phys_addr_t *out_start,
798 phys_addr_t *out_end, int *out_nid)
799{
800 struct memblock_type *mem = &memblock.memory;
801 struct memblock_type *rsv = &memblock.reserved;
802 int mi = *idx & 0xffffffff;
803 int ri = *idx >> 32;
b1154233 804
560dca27
GS
805 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
806 nid = NUMA_NO_NODE;
35fd0808
TH
807
808 for ( ; mi < mem->cnt; mi++) {
809 struct memblock_region *m = &mem->regions[mi];
810 phys_addr_t m_start = m->base;
811 phys_addr_t m_end = m->base + m->size;
812
813 /* only memory regions are associated with nodes, check it */
560dca27 814 if (nid != NUMA_NO_NODE && nid != memblock_get_region_node(m))
35fd0808
TH
815 continue;
816
817 /* scan areas before each reservation for intersection */
818 for ( ; ri < rsv->cnt + 1; ri++) {
819 struct memblock_region *r = &rsv->regions[ri];
820 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
821 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
822
823 /* if ri advanced past mi, break out to advance mi */
824 if (r_start >= m_end)
825 break;
826 /* if the two regions intersect, we're done */
827 if (m_start < r_end) {
828 if (out_start)
829 *out_start = max(m_start, r_start);
830 if (out_end)
831 *out_end = min(m_end, r_end);
832 if (out_nid)
833 *out_nid = memblock_get_region_node(m);
834 /*
835 * The region which ends first is advanced
836 * for the next iteration.
837 */
838 if (m_end <= r_end)
839 mi++;
840 else
841 ri++;
842 *idx = (u32)mi | (u64)ri << 32;
843 return;
844 }
845 }
846 }
847
848 /* signal end of iteration */
849 *idx = ULLONG_MAX;
850}
851
7bd0b0f0
TH
852/**
853 * __next_free_mem_range_rev - next function for for_each_free_mem_range_reverse()
854 * @idx: pointer to u64 loop variable
b1154233 855 * @nid: nid: node selector, %NUMA_NO_NODE for all nodes
dad7557e
WL
856 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
857 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
858 * @out_nid: ptr to int for nid of the range, can be %NULL
7bd0b0f0
TH
859 *
860 * Reverse of __next_free_mem_range().
55ac590c
TC
861 *
862 * Linux kernel cannot migrate pages used by itself. Memory hotplug users won't
863 * be able to hot-remove hotpluggable memory used by the kernel. So this
864 * function skip hotpluggable regions if needed when allocating memory for the
865 * kernel.
7bd0b0f0
TH
866 */
867void __init_memblock __next_free_mem_range_rev(u64 *idx, int nid,
868 phys_addr_t *out_start,
869 phys_addr_t *out_end, int *out_nid)
870{
871 struct memblock_type *mem = &memblock.memory;
872 struct memblock_type *rsv = &memblock.reserved;
873 int mi = *idx & 0xffffffff;
874 int ri = *idx >> 32;
b1154233 875
560dca27
GS
876 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
877 nid = NUMA_NO_NODE;
7bd0b0f0
TH
878
879 if (*idx == (u64)ULLONG_MAX) {
880 mi = mem->cnt - 1;
881 ri = rsv->cnt;
882 }
883
884 for ( ; mi >= 0; mi--) {
885 struct memblock_region *m = &mem->regions[mi];
886 phys_addr_t m_start = m->base;
887 phys_addr_t m_end = m->base + m->size;
888
889 /* only memory regions are associated with nodes, check it */
560dca27 890 if (nid != NUMA_NO_NODE && nid != memblock_get_region_node(m))
7bd0b0f0
TH
891 continue;
892
55ac590c
TC
893 /* skip hotpluggable memory regions if needed */
894 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
895 continue;
896
7bd0b0f0
TH
897 /* scan areas before each reservation for intersection */
898 for ( ; ri >= 0; ri--) {
899 struct memblock_region *r = &rsv->regions[ri];
900 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
901 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
902
903 /* if ri advanced past mi, break out to advance mi */
904 if (r_end <= m_start)
905 break;
906 /* if the two regions intersect, we're done */
907 if (m_end > r_start) {
908 if (out_start)
909 *out_start = max(m_start, r_start);
910 if (out_end)
911 *out_end = min(m_end, r_end);
912 if (out_nid)
913 *out_nid = memblock_get_region_node(m);
914
915 if (m_start >= r_start)
916 mi--;
917 else
918 ri--;
919 *idx = (u32)mi | (u64)ri << 32;
920 return;
921 }
922 }
923 }
924
925 *idx = ULLONG_MAX;
926}
927
7c0caeb8
TH
928#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
929/*
930 * Common iterator interface used to define for_each_mem_range().
931 */
932void __init_memblock __next_mem_pfn_range(int *idx, int nid,
933 unsigned long *out_start_pfn,
934 unsigned long *out_end_pfn, int *out_nid)
935{
936 struct memblock_type *type = &memblock.memory;
937 struct memblock_region *r;
938
939 while (++*idx < type->cnt) {
940 r = &type->regions[*idx];
941
942 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
943 continue;
944 if (nid == MAX_NUMNODES || nid == r->nid)
945 break;
946 }
947 if (*idx >= type->cnt) {
948 *idx = -1;
949 return;
950 }
951
952 if (out_start_pfn)
953 *out_start_pfn = PFN_UP(r->base);
954 if (out_end_pfn)
955 *out_end_pfn = PFN_DOWN(r->base + r->size);
956 if (out_nid)
957 *out_nid = r->nid;
958}
959
960/**
961 * memblock_set_node - set node ID on memblock regions
962 * @base: base of area to set node ID for
963 * @size: size of area to set node ID for
e7e8de59 964 * @type: memblock type to set node ID for
7c0caeb8
TH
965 * @nid: node ID to set
966 *
e7e8de59 967 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
7c0caeb8
TH
968 * Regions which cross the area boundaries are split as necessary.
969 *
970 * RETURNS:
971 * 0 on success, -errno on failure.
972 */
973int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
e7e8de59 974 struct memblock_type *type, int nid)
7c0caeb8 975{
6a9ceb31
TH
976 int start_rgn, end_rgn;
977 int i, ret;
7c0caeb8 978
6a9ceb31
TH
979 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
980 if (ret)
981 return ret;
7c0caeb8 982
6a9ceb31 983 for (i = start_rgn; i < end_rgn; i++)
e9d24ad3 984 memblock_set_region_node(&type->regions[i], nid);
7c0caeb8
TH
985
986 memblock_merge_regions(type);
987 return 0;
988}
989#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
990
7bd0b0f0
TH
991static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
992 phys_addr_t align, phys_addr_t max_addr,
993 int nid)
95f72d1e 994{
6ed311b2 995 phys_addr_t found;
95f72d1e 996
79f40fab
GS
997 if (!align)
998 align = SMP_CACHE_BYTES;
94f3d3af 999
847854f5
TH
1000 /* align @size to avoid excessive fragmentation on reserved array */
1001 size = round_up(size, align);
1002
87029ee9 1003 found = memblock_find_in_range_node(size, align, 0, max_addr, nid);
9c8c27e2 1004 if (found && !memblock_reserve(found, size))
6ed311b2 1005 return found;
95f72d1e 1006
6ed311b2 1007 return 0;
95f72d1e
YL
1008}
1009
7bd0b0f0
TH
1010phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1011{
1012 return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
1013}
1014
1015phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1016{
b1154233 1017 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE);
7bd0b0f0
TH
1018}
1019
6ed311b2 1020phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
95f72d1e 1021{
6ed311b2
BH
1022 phys_addr_t alloc;
1023
1024 alloc = __memblock_alloc_base(size, align, max_addr);
1025
1026 if (alloc == 0)
1027 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1028 (unsigned long long) size, (unsigned long long) max_addr);
1029
1030 return alloc;
95f72d1e
YL
1031}
1032
6ed311b2 1033phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
95f72d1e 1034{
6ed311b2
BH
1035 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1036}
95f72d1e 1037
9d1e2492
BH
1038phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1039{
1040 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1041
1042 if (res)
1043 return res;
15fb0972 1044 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
95f72d1e
YL
1045}
1046
26f09e9b
SS
1047/**
1048 * memblock_virt_alloc_internal - allocate boot memory block
1049 * @size: size of memory block to be allocated in bytes
1050 * @align: alignment of the region and block's size
1051 * @min_addr: the lower bound of the memory region to allocate (phys address)
1052 * @max_addr: the upper bound of the memory region to allocate (phys address)
1053 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1054 *
1055 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1056 * will fall back to memory below @min_addr. Also, allocation may fall back
1057 * to any node in the system if the specified node can not
1058 * hold the requested memory.
1059 *
1060 * The allocation is performed from memory region limited by
1061 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1062 *
1063 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1064 *
1065 * The phys address of allocated boot memory block is converted to virtual and
1066 * allocated memory is reset to 0.
1067 *
1068 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1069 * allocated boot memory block, so that it is never reported as leaks.
1070 *
1071 * RETURNS:
1072 * Virtual address of allocated memory block on success, NULL on failure.
1073 */
1074static void * __init memblock_virt_alloc_internal(
1075 phys_addr_t size, phys_addr_t align,
1076 phys_addr_t min_addr, phys_addr_t max_addr,
1077 int nid)
1078{
1079 phys_addr_t alloc;
1080 void *ptr;
1081
560dca27
GS
1082 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1083 nid = NUMA_NO_NODE;
26f09e9b
SS
1084
1085 /*
1086 * Detect any accidental use of these APIs after slab is ready, as at
1087 * this moment memblock may be deinitialized already and its
1088 * internal data may be destroyed (after execution of free_all_bootmem)
1089 */
1090 if (WARN_ON_ONCE(slab_is_available()))
1091 return kzalloc_node(size, GFP_NOWAIT, nid);
1092
1093 if (!align)
1094 align = SMP_CACHE_BYTES;
1095
1096 /* align @size to avoid excessive fragmentation on reserved array */
1097 size = round_up(size, align);
1098
1099again:
1100 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1101 nid);
1102 if (alloc)
1103 goto done;
1104
1105 if (nid != NUMA_NO_NODE) {
1106 alloc = memblock_find_in_range_node(size, align, min_addr,
1107 max_addr, NUMA_NO_NODE);
1108 if (alloc)
1109 goto done;
1110 }
1111
1112 if (min_addr) {
1113 min_addr = 0;
1114 goto again;
1115 } else {
1116 goto error;
1117 }
1118
1119done:
1120 memblock_reserve(alloc, size);
1121 ptr = phys_to_virt(alloc);
1122 memset(ptr, 0, size);
1123
1124 /*
1125 * The min_count is set to 0 so that bootmem allocated blocks
1126 * are never reported as leaks. This is because many of these blocks
1127 * are only referred via the physical address which is not
1128 * looked up by kmemleak.
1129 */
1130 kmemleak_alloc(ptr, size, 0, 0);
1131
1132 return ptr;
1133
1134error:
1135 return NULL;
1136}
1137
1138/**
1139 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1140 * @size: size of memory block to be allocated in bytes
1141 * @align: alignment of the region and block's size
1142 * @min_addr: the lower bound of the memory region from where the allocation
1143 * is preferred (phys address)
1144 * @max_addr: the upper bound of the memory region from where the allocation
1145 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1146 * allocate only from memory limited by memblock.current_limit value
1147 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1148 *
1149 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1150 * additional debug information (including caller info), if enabled.
1151 *
1152 * RETURNS:
1153 * Virtual address of allocated memory block on success, NULL on failure.
1154 */
1155void * __init memblock_virt_alloc_try_nid_nopanic(
1156 phys_addr_t size, phys_addr_t align,
1157 phys_addr_t min_addr, phys_addr_t max_addr,
1158 int nid)
1159{
1160 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1161 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1162 (u64)max_addr, (void *)_RET_IP_);
1163 return memblock_virt_alloc_internal(size, align, min_addr,
1164 max_addr, nid);
1165}
1166
1167/**
1168 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1169 * @size: size of memory block to be allocated in bytes
1170 * @align: alignment of the region and block's size
1171 * @min_addr: the lower bound of the memory region from where the allocation
1172 * is preferred (phys address)
1173 * @max_addr: the upper bound of the memory region from where the allocation
1174 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1175 * allocate only from memory limited by memblock.current_limit value
1176 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1177 *
1178 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1179 * which provides debug information (including caller info), if enabled,
1180 * and panics if the request can not be satisfied.
1181 *
1182 * RETURNS:
1183 * Virtual address of allocated memory block on success, NULL on failure.
1184 */
1185void * __init memblock_virt_alloc_try_nid(
1186 phys_addr_t size, phys_addr_t align,
1187 phys_addr_t min_addr, phys_addr_t max_addr,
1188 int nid)
1189{
1190 void *ptr;
1191
1192 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1193 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1194 (u64)max_addr, (void *)_RET_IP_);
1195 ptr = memblock_virt_alloc_internal(size, align,
1196 min_addr, max_addr, nid);
1197 if (ptr)
1198 return ptr;
1199
1200 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1201 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1202 (u64)max_addr);
1203 return NULL;
1204}
1205
1206/**
1207 * __memblock_free_early - free boot memory block
1208 * @base: phys starting address of the boot memory block
1209 * @size: size of the boot memory block in bytes
1210 *
1211 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1212 * The freeing memory will not be released to the buddy allocator.
1213 */
1214void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1215{
1216 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1217 __func__, (u64)base, (u64)base + size - 1,
1218 (void *)_RET_IP_);
1219 kmemleak_free_part(__va(base), size);
1220 __memblock_remove(&memblock.reserved, base, size);
1221}
1222
1223/*
1224 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1225 * @addr: phys starting address of the boot memory block
1226 * @size: size of the boot memory block in bytes
1227 *
1228 * This is only useful when the bootmem allocator has already been torn
1229 * down, but we are still initializing the system. Pages are released directly
1230 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1231 */
1232void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1233{
1234 u64 cursor, end;
1235
1236 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1237 __func__, (u64)base, (u64)base + size - 1,
1238 (void *)_RET_IP_);
1239 kmemleak_free_part(__va(base), size);
1240 cursor = PFN_UP(base);
1241 end = PFN_DOWN(base + size);
1242
1243 for (; cursor < end; cursor++) {
1244 __free_pages_bootmem(pfn_to_page(cursor), 0);
1245 totalram_pages++;
1246 }
1247}
9d1e2492
BH
1248
1249/*
1250 * Remaining API functions
1251 */
1252
2898cc4c 1253phys_addr_t __init memblock_phys_mem_size(void)
95f72d1e 1254{
1440c4e2 1255 return memblock.memory.total_size;
95f72d1e
YL
1256}
1257
595ad9af
YL
1258phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1259{
1260 unsigned long pages = 0;
1261 struct memblock_region *r;
1262 unsigned long start_pfn, end_pfn;
1263
1264 for_each_memblock(memory, r) {
1265 start_pfn = memblock_region_memory_base_pfn(r);
1266 end_pfn = memblock_region_memory_end_pfn(r);
1267 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1268 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1269 pages += end_pfn - start_pfn;
1270 }
1271
1272 return (phys_addr_t)pages << PAGE_SHIFT;
1273}
1274
0a93ebef
SR
1275/* lowest address */
1276phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1277{
1278 return memblock.memory.regions[0].base;
1279}
1280
10d06439 1281phys_addr_t __init_memblock memblock_end_of_DRAM(void)
95f72d1e
YL
1282{
1283 int idx = memblock.memory.cnt - 1;
1284
e3239ff9 1285 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
95f72d1e
YL
1286}
1287
c0ce8fef 1288void __init memblock_enforce_memory_limit(phys_addr_t limit)
95f72d1e
YL
1289{
1290 unsigned long i;
c0ce8fef 1291 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
95f72d1e 1292
c0ce8fef 1293 if (!limit)
95f72d1e
YL
1294 return;
1295
c0ce8fef 1296 /* find out max address */
95f72d1e 1297 for (i = 0; i < memblock.memory.cnt; i++) {
c0ce8fef 1298 struct memblock_region *r = &memblock.memory.regions[i];
95f72d1e 1299
c0ce8fef
TH
1300 if (limit <= r->size) {
1301 max_addr = r->base + limit;
1302 break;
95f72d1e 1303 }
c0ce8fef 1304 limit -= r->size;
95f72d1e 1305 }
c0ce8fef
TH
1306
1307 /* truncate both memory and reserved regions */
1308 __memblock_remove(&memblock.memory, max_addr, (phys_addr_t)ULLONG_MAX);
1309 __memblock_remove(&memblock.reserved, max_addr, (phys_addr_t)ULLONG_MAX);
95f72d1e
YL
1310}
1311
cd79481d 1312static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
72d4b0b4
BH
1313{
1314 unsigned int left = 0, right = type->cnt;
1315
1316 do {
1317 unsigned int mid = (right + left) / 2;
1318
1319 if (addr < type->regions[mid].base)
1320 right = mid;
1321 else if (addr >= (type->regions[mid].base +
1322 type->regions[mid].size))
1323 left = mid + 1;
1324 else
1325 return mid;
1326 } while (left < right);
1327 return -1;
1328}
1329
2898cc4c 1330int __init memblock_is_reserved(phys_addr_t addr)
95f72d1e 1331{
72d4b0b4
BH
1332 return memblock_search(&memblock.reserved, addr) != -1;
1333}
95f72d1e 1334
3661ca66 1335int __init_memblock memblock_is_memory(phys_addr_t addr)
72d4b0b4
BH
1336{
1337 return memblock_search(&memblock.memory, addr) != -1;
1338}
1339
e76b63f8
YL
1340#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1341int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1342 unsigned long *start_pfn, unsigned long *end_pfn)
1343{
1344 struct memblock_type *type = &memblock.memory;
1345 int mid = memblock_search(type, (phys_addr_t)pfn << PAGE_SHIFT);
1346
1347 if (mid == -1)
1348 return -1;
1349
1350 *start_pfn = type->regions[mid].base >> PAGE_SHIFT;
1351 *end_pfn = (type->regions[mid].base + type->regions[mid].size)
1352 >> PAGE_SHIFT;
1353
1354 return type->regions[mid].nid;
1355}
1356#endif
1357
eab30949
SB
1358/**
1359 * memblock_is_region_memory - check if a region is a subset of memory
1360 * @base: base of region to check
1361 * @size: size of region to check
1362 *
1363 * Check if the region [@base, @base+@size) is a subset of a memory block.
1364 *
1365 * RETURNS:
1366 * 0 if false, non-zero if true
1367 */
3661ca66 1368int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
72d4b0b4 1369{
abb65272 1370 int idx = memblock_search(&memblock.memory, base);
eb18f1b5 1371 phys_addr_t end = base + memblock_cap_size(base, &size);
72d4b0b4
BH
1372
1373 if (idx == -1)
1374 return 0;
abb65272
TV
1375 return memblock.memory.regions[idx].base <= base &&
1376 (memblock.memory.regions[idx].base +
eb18f1b5 1377 memblock.memory.regions[idx].size) >= end;
95f72d1e
YL
1378}
1379
eab30949
SB
1380/**
1381 * memblock_is_region_reserved - check if a region intersects reserved memory
1382 * @base: base of region to check
1383 * @size: size of region to check
1384 *
1385 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1386 *
1387 * RETURNS:
1388 * 0 if false, non-zero if true
1389 */
10d06439 1390int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
95f72d1e 1391{
eb18f1b5 1392 memblock_cap_size(base, &size);
f1c2c19c 1393 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
95f72d1e
YL
1394}
1395
6ede1fd3
YL
1396void __init_memblock memblock_trim_memory(phys_addr_t align)
1397{
1398 int i;
1399 phys_addr_t start, end, orig_start, orig_end;
1400 struct memblock_type *mem = &memblock.memory;
1401
1402 for (i = 0; i < mem->cnt; i++) {
1403 orig_start = mem->regions[i].base;
1404 orig_end = mem->regions[i].base + mem->regions[i].size;
1405 start = round_up(orig_start, align);
1406 end = round_down(orig_end, align);
1407
1408 if (start == orig_start && end == orig_end)
1409 continue;
1410
1411 if (start < end) {
1412 mem->regions[i].base = start;
1413 mem->regions[i].size = end - start;
1414 } else {
1415 memblock_remove_region(mem, i);
1416 i--;
1417 }
1418 }
1419}
e63075a3 1420
3661ca66 1421void __init_memblock memblock_set_current_limit(phys_addr_t limit)
e63075a3
BH
1422{
1423 memblock.current_limit = limit;
1424}
1425
7c0caeb8 1426static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
6ed311b2
BH
1427{
1428 unsigned long long base, size;
66a20757 1429 unsigned long flags;
6ed311b2
BH
1430 int i;
1431
7c0caeb8 1432 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
6ed311b2 1433
7c0caeb8
TH
1434 for (i = 0; i < type->cnt; i++) {
1435 struct memblock_region *rgn = &type->regions[i];
1436 char nid_buf[32] = "";
1437
1438 base = rgn->base;
1439 size = rgn->size;
66a20757 1440 flags = rgn->flags;
7c0caeb8
TH
1441#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1442 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1443 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1444 memblock_get_region_node(rgn));
1445#endif
66a20757
TC
1446 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1447 name, i, base, base + size - 1, size, nid_buf, flags);
6ed311b2
BH
1448 }
1449}
1450
4ff7b82f 1451void __init_memblock __memblock_dump_all(void)
6ed311b2 1452{
6ed311b2 1453 pr_info("MEMBLOCK configuration:\n");
1440c4e2
TH
1454 pr_info(" memory size = %#llx reserved size = %#llx\n",
1455 (unsigned long long)memblock.memory.total_size,
1456 (unsigned long long)memblock.reserved.total_size);
6ed311b2
BH
1457
1458 memblock_dump(&memblock.memory, "memory");
1459 memblock_dump(&memblock.reserved, "reserved");
1460}
1461
1aadc056 1462void __init memblock_allow_resize(void)
6ed311b2 1463{
142b45a7 1464 memblock_can_resize = 1;
6ed311b2
BH
1465}
1466
6ed311b2
BH
1467static int __init early_memblock(char *p)
1468{
1469 if (p && strstr(p, "debug"))
1470 memblock_debug = 1;
1471 return 0;
1472}
1473early_param("memblock", early_memblock);
1474
c378ddd5 1475#if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
6d03b885
BH
1476
1477static int memblock_debug_show(struct seq_file *m, void *private)
1478{
1479 struct memblock_type *type = m->private;
1480 struct memblock_region *reg;
1481 int i;
1482
1483 for (i = 0; i < type->cnt; i++) {
1484 reg = &type->regions[i];
1485 seq_printf(m, "%4d: ", i);
1486 if (sizeof(phys_addr_t) == 4)
1487 seq_printf(m, "0x%08lx..0x%08lx\n",
1488 (unsigned long)reg->base,
1489 (unsigned long)(reg->base + reg->size - 1));
1490 else
1491 seq_printf(m, "0x%016llx..0x%016llx\n",
1492 (unsigned long long)reg->base,
1493 (unsigned long long)(reg->base + reg->size - 1));
1494
1495 }
1496 return 0;
1497}
1498
1499static int memblock_debug_open(struct inode *inode, struct file *file)
1500{
1501 return single_open(file, memblock_debug_show, inode->i_private);
1502}
1503
1504static const struct file_operations memblock_debug_fops = {
1505 .open = memblock_debug_open,
1506 .read = seq_read,
1507 .llseek = seq_lseek,
1508 .release = single_release,
1509};
1510
1511static int __init memblock_init_debugfs(void)
1512{
1513 struct dentry *root = debugfs_create_dir("memblock", NULL);
1514 if (!root)
1515 return -ENXIO;
1516 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1517 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1518
1519 return 0;
1520}
1521__initcall(memblock_init_debugfs);
1522
1523#endif /* CONFIG_DEBUG_FS */
This page took 0.277384 seconds and 5 git commands to generate.