Loading...
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>
14#include <linux/slab.h>
15#include <linux/init.h>
16#include <linux/bitops.h>
17#include <linux/poison.h>
18#include <linux/pfn.h>
19#include <linux/debugfs.h>
20#include <linux/seq_file.h>
21#include <linux/memblock.h>
22
23#include <asm-generic/sections.h>
24#include <linux/io.h>
25
26#include "internal.h"
27
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#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
31static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
32#endif
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
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
49 .bottom_up = false,
50 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
51};
52
53int memblock_debug __initdata_memblock;
54#ifdef CONFIG_MOVABLE_NODE
55bool movable_node_enabled __initdata_memblock = false;
56#endif
57static bool system_has_some_mirror __initdata_memblock = false;
58static int memblock_can_resize __initdata_memblock;
59static int memblock_memory_in_slab __initdata_memblock = 0;
60static int memblock_reserved_in_slab __initdata_memblock = 0;
61
62ulong __init_memblock choose_memblock_flags(void)
63{
64 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
65}
66
67/* inline so we don't get a warning when pr_debug is compiled out */
68static __init_memblock const char *
69memblock_type_name(struct memblock_type *type)
70{
71 if (type == &memblock.memory)
72 return "memory";
73 else if (type == &memblock.reserved)
74 return "reserved";
75 else
76 return "unknown";
77}
78
79/* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
80static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
81{
82 return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
83}
84
85/*
86 * Address comparison utilities
87 */
88static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
89 phys_addr_t base2, phys_addr_t size2)
90{
91 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
92}
93
94bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
95 phys_addr_t base, phys_addr_t size)
96{
97 unsigned long i;
98
99 for (i = 0; i < type->cnt; i++)
100 if (memblock_addrs_overlap(base, size, type->regions[i].base,
101 type->regions[i].size))
102 break;
103 return i < type->cnt;
104}
105
106/*
107 * __memblock_find_range_bottom_up - find free area utility in bottom-up
108 * @start: start of candidate range
109 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
110 * @size: size of free area to find
111 * @align: alignment of free area to find
112 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
113 * @flags: pick from blocks based on memory attributes
114 *
115 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
116 *
117 * RETURNS:
118 * Found address on success, 0 on failure.
119 */
120static phys_addr_t __init_memblock
121__memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
122 phys_addr_t size, phys_addr_t align, int nid,
123 ulong flags)
124{
125 phys_addr_t this_start, this_end, cand;
126 u64 i;
127
128 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
129 this_start = clamp(this_start, start, end);
130 this_end = clamp(this_end, start, end);
131
132 cand = round_up(this_start, align);
133 if (cand < this_end && this_end - cand >= size)
134 return cand;
135 }
136
137 return 0;
138}
139
140/**
141 * __memblock_find_range_top_down - find free area utility, in top-down
142 * @start: start of candidate range
143 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
144 * @size: size of free area to find
145 * @align: alignment of free area to find
146 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
147 * @flags: pick from blocks based on memory attributes
148 *
149 * Utility called from memblock_find_in_range_node(), find free area top-down.
150 *
151 * RETURNS:
152 * Found address on success, 0 on failure.
153 */
154static phys_addr_t __init_memblock
155__memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
156 phys_addr_t size, phys_addr_t align, int nid,
157 ulong flags)
158{
159 phys_addr_t this_start, this_end, cand;
160 u64 i;
161
162 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
163 NULL) {
164 this_start = clamp(this_start, start, end);
165 this_end = clamp(this_end, start, end);
166
167 if (this_end < size)
168 continue;
169
170 cand = round_down(this_end - size, align);
171 if (cand >= this_start)
172 return cand;
173 }
174
175 return 0;
176}
177
178/**
179 * memblock_find_in_range_node - find free area in given range and node
180 * @size: size of free area to find
181 * @align: alignment of free area to find
182 * @start: start of candidate range
183 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
184 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
185 * @flags: pick from blocks based on memory attributes
186 *
187 * Find @size free area aligned to @align in the specified range and node.
188 *
189 * When allocation direction is bottom-up, the @start should be greater
190 * than the end of the kernel image. Otherwise, it will be trimmed. The
191 * reason is that we want the bottom-up allocation just near the kernel
192 * image so it is highly likely that the allocated memory and the kernel
193 * will reside in the same node.
194 *
195 * If bottom-up allocation failed, will try to allocate memory top-down.
196 *
197 * RETURNS:
198 * Found address on success, 0 on failure.
199 */
200phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
201 phys_addr_t align, phys_addr_t start,
202 phys_addr_t end, int nid, ulong flags)
203{
204 phys_addr_t kernel_end, ret;
205
206 /* pump up @end */
207 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
208 end = memblock.current_limit;
209
210 /* avoid allocating the first page */
211 start = max_t(phys_addr_t, start, PAGE_SIZE);
212 end = max(start, end);
213 kernel_end = __pa_symbol(_end);
214
215 /*
216 * try bottom-up allocation only when bottom-up mode
217 * is set and @end is above the kernel image.
218 */
219 if (memblock_bottom_up() && end > kernel_end) {
220 phys_addr_t bottom_up_start;
221
222 /* make sure we will allocate above the kernel */
223 bottom_up_start = max(start, kernel_end);
224
225 /* ok, try bottom-up allocation first */
226 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
227 size, align, nid, flags);
228 if (ret)
229 return ret;
230
231 /*
232 * we always limit bottom-up allocation above the kernel,
233 * but top-down allocation doesn't have the limit, so
234 * retrying top-down allocation may succeed when bottom-up
235 * allocation failed.
236 *
237 * bottom-up allocation is expected to be fail very rarely,
238 * so we use WARN_ONCE() here to see the stack trace if
239 * fail happens.
240 */
241 WARN_ONCE(1, "memblock: bottom-up allocation failed, memory hotunplug may be affected\n");
242 }
243
244 return __memblock_find_range_top_down(start, end, size, align, nid,
245 flags);
246}
247
248/**
249 * memblock_find_in_range - find free area in given range
250 * @start: start of candidate range
251 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
252 * @size: size of free area to find
253 * @align: alignment of free area to find
254 *
255 * Find @size free area aligned to @align in the specified range.
256 *
257 * RETURNS:
258 * Found address on success, 0 on failure.
259 */
260phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
261 phys_addr_t end, phys_addr_t size,
262 phys_addr_t align)
263{
264 phys_addr_t ret;
265 ulong flags = choose_memblock_flags();
266
267again:
268 ret = memblock_find_in_range_node(size, align, start, end,
269 NUMA_NO_NODE, flags);
270
271 if (!ret && (flags & MEMBLOCK_MIRROR)) {
272 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
273 &size);
274 flags &= ~MEMBLOCK_MIRROR;
275 goto again;
276 }
277
278 return ret;
279}
280
281static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
282{
283 type->total_size -= type->regions[r].size;
284 memmove(&type->regions[r], &type->regions[r + 1],
285 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
286 type->cnt--;
287
288 /* Special case for empty arrays */
289 if (type->cnt == 0) {
290 WARN_ON(type->total_size != 0);
291 type->cnt = 1;
292 type->regions[0].base = 0;
293 type->regions[0].size = 0;
294 type->regions[0].flags = 0;
295 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
296 }
297}
298
299#ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
300
301phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
302 phys_addr_t *addr)
303{
304 if (memblock.reserved.regions == memblock_reserved_init_regions)
305 return 0;
306
307 *addr = __pa(memblock.reserved.regions);
308
309 return PAGE_ALIGN(sizeof(struct memblock_region) *
310 memblock.reserved.max);
311}
312
313phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
314 phys_addr_t *addr)
315{
316 if (memblock.memory.regions == memblock_memory_init_regions)
317 return 0;
318
319 *addr = __pa(memblock.memory.regions);
320
321 return PAGE_ALIGN(sizeof(struct memblock_region) *
322 memblock.memory.max);
323}
324
325#endif
326
327/**
328 * memblock_double_array - double the size of the memblock regions array
329 * @type: memblock type of the regions array being doubled
330 * @new_area_start: starting address of memory range to avoid overlap with
331 * @new_area_size: size of memory range to avoid overlap with
332 *
333 * Double the size of the @type regions array. If memblock is being used to
334 * allocate memory for a new reserved regions array and there is a previously
335 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
336 * waiting to be reserved, ensure the memory used by the new array does
337 * not overlap.
338 *
339 * RETURNS:
340 * 0 on success, -1 on failure.
341 */
342static int __init_memblock memblock_double_array(struct memblock_type *type,
343 phys_addr_t new_area_start,
344 phys_addr_t new_area_size)
345{
346 struct memblock_region *new_array, *old_array;
347 phys_addr_t old_alloc_size, new_alloc_size;
348 phys_addr_t old_size, new_size, addr;
349 int use_slab = slab_is_available();
350 int *in_slab;
351
352 /* We don't allow resizing until we know about the reserved regions
353 * of memory that aren't suitable for allocation
354 */
355 if (!memblock_can_resize)
356 return -1;
357
358 /* Calculate new doubled size */
359 old_size = type->max * sizeof(struct memblock_region);
360 new_size = old_size << 1;
361 /*
362 * We need to allocated new one align to PAGE_SIZE,
363 * so we can free them completely later.
364 */
365 old_alloc_size = PAGE_ALIGN(old_size);
366 new_alloc_size = PAGE_ALIGN(new_size);
367
368 /* Retrieve the slab flag */
369 if (type == &memblock.memory)
370 in_slab = &memblock_memory_in_slab;
371 else
372 in_slab = &memblock_reserved_in_slab;
373
374 /* Try to find some space for it.
375 *
376 * WARNING: We assume that either slab_is_available() and we use it or
377 * we use MEMBLOCK for allocations. That means that this is unsafe to
378 * use when bootmem is currently active (unless bootmem itself is
379 * implemented on top of MEMBLOCK which isn't the case yet)
380 *
381 * This should however not be an issue for now, as we currently only
382 * call into MEMBLOCK while it's still active, or much later when slab
383 * is active for memory hotplug operations
384 */
385 if (use_slab) {
386 new_array = kmalloc(new_size, GFP_KERNEL);
387 addr = new_array ? __pa(new_array) : 0;
388 } else {
389 /* only exclude range when trying to double reserved.regions */
390 if (type != &memblock.reserved)
391 new_area_start = new_area_size = 0;
392
393 addr = memblock_find_in_range(new_area_start + new_area_size,
394 memblock.current_limit,
395 new_alloc_size, PAGE_SIZE);
396 if (!addr && new_area_size)
397 addr = memblock_find_in_range(0,
398 min(new_area_start, memblock.current_limit),
399 new_alloc_size, PAGE_SIZE);
400
401 new_array = addr ? __va(addr) : NULL;
402 }
403 if (!addr) {
404 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
405 memblock_type_name(type), type->max, type->max * 2);
406 return -1;
407 }
408
409 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
410 memblock_type_name(type), type->max * 2, (u64)addr,
411 (u64)addr + new_size - 1);
412
413 /*
414 * Found space, we now need to move the array over before we add the
415 * reserved region since it may be our reserved array itself that is
416 * full.
417 */
418 memcpy(new_array, type->regions, old_size);
419 memset(new_array + type->max, 0, old_size);
420 old_array = type->regions;
421 type->regions = new_array;
422 type->max <<= 1;
423
424 /* Free old array. We needn't free it if the array is the static one */
425 if (*in_slab)
426 kfree(old_array);
427 else if (old_array != memblock_memory_init_regions &&
428 old_array != memblock_reserved_init_regions)
429 memblock_free(__pa(old_array), old_alloc_size);
430
431 /*
432 * Reserve the new array if that comes from the memblock. Otherwise, we
433 * needn't do it
434 */
435 if (!use_slab)
436 BUG_ON(memblock_reserve(addr, new_alloc_size));
437
438 /* Update slab flag */
439 *in_slab = use_slab;
440
441 return 0;
442}
443
444/**
445 * memblock_merge_regions - merge neighboring compatible regions
446 * @type: memblock type to scan
447 *
448 * Scan @type and merge neighboring compatible regions.
449 */
450static void __init_memblock memblock_merge_regions(struct memblock_type *type)
451{
452 int i = 0;
453
454 /* cnt never goes below 1 */
455 while (i < type->cnt - 1) {
456 struct memblock_region *this = &type->regions[i];
457 struct memblock_region *next = &type->regions[i + 1];
458
459 if (this->base + this->size != next->base ||
460 memblock_get_region_node(this) !=
461 memblock_get_region_node(next) ||
462 this->flags != next->flags) {
463 BUG_ON(this->base + this->size > next->base);
464 i++;
465 continue;
466 }
467
468 this->size += next->size;
469 /* move forward from next + 1, index of which is i + 2 */
470 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
471 type->cnt--;
472 }
473}
474
475/**
476 * memblock_insert_region - insert new memblock region
477 * @type: memblock type to insert into
478 * @idx: index for the insertion point
479 * @base: base address of the new region
480 * @size: size of the new region
481 * @nid: node id of the new region
482 * @flags: flags of the new region
483 *
484 * Insert new memblock region [@base,@base+@size) into @type at @idx.
485 * @type must already have extra room to accomodate the new region.
486 */
487static void __init_memblock memblock_insert_region(struct memblock_type *type,
488 int idx, phys_addr_t base,
489 phys_addr_t size,
490 int nid, unsigned long flags)
491{
492 struct memblock_region *rgn = &type->regions[idx];
493
494 BUG_ON(type->cnt >= type->max);
495 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
496 rgn->base = base;
497 rgn->size = size;
498 rgn->flags = flags;
499 memblock_set_region_node(rgn, nid);
500 type->cnt++;
501 type->total_size += size;
502}
503
504/**
505 * memblock_add_range - add new memblock region
506 * @type: memblock type to add new region into
507 * @base: base address of the new region
508 * @size: size of the new region
509 * @nid: nid of the new region
510 * @flags: flags of the new region
511 *
512 * Add new memblock region [@base,@base+@size) into @type. The new region
513 * is allowed to overlap with existing ones - overlaps don't affect already
514 * existing regions. @type is guaranteed to be minimal (all neighbouring
515 * compatible regions are merged) after the addition.
516 *
517 * RETURNS:
518 * 0 on success, -errno on failure.
519 */
520int __init_memblock memblock_add_range(struct memblock_type *type,
521 phys_addr_t base, phys_addr_t size,
522 int nid, unsigned long flags)
523{
524 bool insert = false;
525 phys_addr_t obase = base;
526 phys_addr_t end = base + memblock_cap_size(base, &size);
527 int idx, nr_new;
528 struct memblock_region *rgn;
529
530 if (!size)
531 return 0;
532
533 /* special case for empty array */
534 if (type->regions[0].size == 0) {
535 WARN_ON(type->cnt != 1 || type->total_size);
536 type->regions[0].base = base;
537 type->regions[0].size = size;
538 type->regions[0].flags = flags;
539 memblock_set_region_node(&type->regions[0], nid);
540 type->total_size = size;
541 return 0;
542 }
543repeat:
544 /*
545 * The following is executed twice. Once with %false @insert and
546 * then with %true. The first counts the number of regions needed
547 * to accomodate the new area. The second actually inserts them.
548 */
549 base = obase;
550 nr_new = 0;
551
552 for_each_memblock_type(type, rgn) {
553 phys_addr_t rbase = rgn->base;
554 phys_addr_t rend = rbase + rgn->size;
555
556 if (rbase >= end)
557 break;
558 if (rend <= base)
559 continue;
560 /*
561 * @rgn overlaps. If it separates the lower part of new
562 * area, insert that portion.
563 */
564 if (rbase > base) {
565#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
566 WARN_ON(nid != memblock_get_region_node(rgn));
567#endif
568 WARN_ON(flags != rgn->flags);
569 nr_new++;
570 if (insert)
571 memblock_insert_region(type, idx++, base,
572 rbase - base, nid,
573 flags);
574 }
575 /* area below @rend is dealt with, forget about it */
576 base = min(rend, end);
577 }
578
579 /* insert the remaining portion */
580 if (base < end) {
581 nr_new++;
582 if (insert)
583 memblock_insert_region(type, idx, base, end - base,
584 nid, flags);
585 }
586
587 /*
588 * If this was the first round, resize array and repeat for actual
589 * insertions; otherwise, merge and return.
590 */
591 if (!insert) {
592 while (type->cnt + nr_new > type->max)
593 if (memblock_double_array(type, obase, size) < 0)
594 return -ENOMEM;
595 insert = true;
596 goto repeat;
597 } else {
598 memblock_merge_regions(type);
599 return 0;
600 }
601}
602
603int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
604 int nid)
605{
606 return memblock_add_range(&memblock.memory, base, size, nid, 0);
607}
608
609static int __init_memblock memblock_add_region(phys_addr_t base,
610 phys_addr_t size,
611 int nid,
612 unsigned long flags)
613{
614 memblock_dbg("memblock_add: [%#016llx-%#016llx] flags %#02lx %pF\n",
615 (unsigned long long)base,
616 (unsigned long long)base + size - 1,
617 flags, (void *)_RET_IP_);
618
619 return memblock_add_range(&memblock.memory, base, size, nid, flags);
620}
621
622int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
623{
624 return memblock_add_region(base, size, MAX_NUMNODES, 0);
625}
626
627/**
628 * memblock_isolate_range - isolate given range into disjoint memblocks
629 * @type: memblock type to isolate range for
630 * @base: base of range to isolate
631 * @size: size of range to isolate
632 * @start_rgn: out parameter for the start of isolated region
633 * @end_rgn: out parameter for the end of isolated region
634 *
635 * Walk @type and ensure that regions don't cross the boundaries defined by
636 * [@base,@base+@size). Crossing regions are split at the boundaries,
637 * which may create at most two more regions. The index of the first
638 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
639 *
640 * RETURNS:
641 * 0 on success, -errno on failure.
642 */
643static int __init_memblock memblock_isolate_range(struct memblock_type *type,
644 phys_addr_t base, phys_addr_t size,
645 int *start_rgn, int *end_rgn)
646{
647 phys_addr_t end = base + memblock_cap_size(base, &size);
648 int idx;
649 struct memblock_region *rgn;
650
651 *start_rgn = *end_rgn = 0;
652
653 if (!size)
654 return 0;
655
656 /* we'll create at most two more regions */
657 while (type->cnt + 2 > type->max)
658 if (memblock_double_array(type, base, size) < 0)
659 return -ENOMEM;
660
661 for_each_memblock_type(type, rgn) {
662 phys_addr_t rbase = rgn->base;
663 phys_addr_t rend = rbase + rgn->size;
664
665 if (rbase >= end)
666 break;
667 if (rend <= base)
668 continue;
669
670 if (rbase < base) {
671 /*
672 * @rgn intersects from below. Split and continue
673 * to process the next region - the new top half.
674 */
675 rgn->base = base;
676 rgn->size -= base - rbase;
677 type->total_size -= base - rbase;
678 memblock_insert_region(type, idx, rbase, base - rbase,
679 memblock_get_region_node(rgn),
680 rgn->flags);
681 } else if (rend > end) {
682 /*
683 * @rgn intersects from above. Split and redo the
684 * current region - the new bottom half.
685 */
686 rgn->base = end;
687 rgn->size -= end - rbase;
688 type->total_size -= end - rbase;
689 memblock_insert_region(type, idx--, rbase, end - rbase,
690 memblock_get_region_node(rgn),
691 rgn->flags);
692 } else {
693 /* @rgn is fully contained, record it */
694 if (!*end_rgn)
695 *start_rgn = idx;
696 *end_rgn = idx + 1;
697 }
698 }
699
700 return 0;
701}
702
703static int __init_memblock memblock_remove_range(struct memblock_type *type,
704 phys_addr_t base, phys_addr_t size)
705{
706 int start_rgn, end_rgn;
707 int i, ret;
708
709 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
710 if (ret)
711 return ret;
712
713 for (i = end_rgn - 1; i >= start_rgn; i--)
714 memblock_remove_region(type, i);
715 return 0;
716}
717
718int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
719{
720 return memblock_remove_range(&memblock.memory, base, size);
721}
722
723
724int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
725{
726 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
727 (unsigned long long)base,
728 (unsigned long long)base + size - 1,
729 (void *)_RET_IP_);
730
731 kmemleak_free_part(__va(base), size);
732 return memblock_remove_range(&memblock.reserved, base, size);
733}
734
735static int __init_memblock memblock_reserve_region(phys_addr_t base,
736 phys_addr_t size,
737 int nid,
738 unsigned long flags)
739{
740 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
741 (unsigned long long)base,
742 (unsigned long long)base + size - 1,
743 flags, (void *)_RET_IP_);
744
745 return memblock_add_range(&memblock.reserved, base, size, nid, flags);
746}
747
748int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
749{
750 return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
751}
752
753/**
754 *
755 * This function isolates region [@base, @base + @size), and sets/clears flag
756 *
757 * Return 0 on success, -errno on failure.
758 */
759static int __init_memblock memblock_setclr_flag(phys_addr_t base,
760 phys_addr_t size, int set, int flag)
761{
762 struct memblock_type *type = &memblock.memory;
763 int i, ret, start_rgn, end_rgn;
764
765 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
766 if (ret)
767 return ret;
768
769 for (i = start_rgn; i < end_rgn; i++)
770 if (set)
771 memblock_set_region_flags(&type->regions[i], flag);
772 else
773 memblock_clear_region_flags(&type->regions[i], flag);
774
775 memblock_merge_regions(type);
776 return 0;
777}
778
779/**
780 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
781 * @base: the base phys addr of the region
782 * @size: the size of the region
783 *
784 * Return 0 on success, -errno on failure.
785 */
786int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
787{
788 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
789}
790
791/**
792 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
793 * @base: the base phys addr of the region
794 * @size: the size of the region
795 *
796 * Return 0 on success, -errno on failure.
797 */
798int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
799{
800 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
801}
802
803/**
804 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
805 * @base: the base phys addr of the region
806 * @size: the size of the region
807 *
808 * Return 0 on success, -errno on failure.
809 */
810int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
811{
812 system_has_some_mirror = true;
813
814 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
815}
816
817/**
818 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
819 * @base: the base phys addr of the region
820 * @size: the size of the region
821 *
822 * Return 0 on success, -errno on failure.
823 */
824int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
825{
826 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
827}
828
829/**
830 * __next_reserved_mem_region - next function for for_each_reserved_region()
831 * @idx: pointer to u64 loop variable
832 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
833 * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
834 *
835 * Iterate over all reserved memory regions.
836 */
837void __init_memblock __next_reserved_mem_region(u64 *idx,
838 phys_addr_t *out_start,
839 phys_addr_t *out_end)
840{
841 struct memblock_type *type = &memblock.reserved;
842
843 if (*idx >= 0 && *idx < type->cnt) {
844 struct memblock_region *r = &type->regions[*idx];
845 phys_addr_t base = r->base;
846 phys_addr_t size = r->size;
847
848 if (out_start)
849 *out_start = base;
850 if (out_end)
851 *out_end = base + size - 1;
852
853 *idx += 1;
854 return;
855 }
856
857 /* signal end of iteration */
858 *idx = ULLONG_MAX;
859}
860
861/**
862 * __next__mem_range - next function for for_each_free_mem_range() etc.
863 * @idx: pointer to u64 loop variable
864 * @nid: node selector, %NUMA_NO_NODE for all nodes
865 * @flags: pick from blocks based on memory attributes
866 * @type_a: pointer to memblock_type from where the range is taken
867 * @type_b: pointer to memblock_type which excludes memory from being taken
868 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
869 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
870 * @out_nid: ptr to int for nid of the range, can be %NULL
871 *
872 * Find the first area from *@idx which matches @nid, fill the out
873 * parameters, and update *@idx for the next iteration. The lower 32bit of
874 * *@idx contains index into type_a and the upper 32bit indexes the
875 * areas before each region in type_b. For example, if type_b regions
876 * look like the following,
877 *
878 * 0:[0-16), 1:[32-48), 2:[128-130)
879 *
880 * The upper 32bit indexes the following regions.
881 *
882 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
883 *
884 * As both region arrays are sorted, the function advances the two indices
885 * in lockstep and returns each intersection.
886 */
887void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
888 struct memblock_type *type_a,
889 struct memblock_type *type_b,
890 phys_addr_t *out_start,
891 phys_addr_t *out_end, int *out_nid)
892{
893 int idx_a = *idx & 0xffffffff;
894 int idx_b = *idx >> 32;
895
896 if (WARN_ONCE(nid == MAX_NUMNODES,
897 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
898 nid = NUMA_NO_NODE;
899
900 for (; idx_a < type_a->cnt; idx_a++) {
901 struct memblock_region *m = &type_a->regions[idx_a];
902
903 phys_addr_t m_start = m->base;
904 phys_addr_t m_end = m->base + m->size;
905 int m_nid = memblock_get_region_node(m);
906
907 /* only memory regions are associated with nodes, check it */
908 if (nid != NUMA_NO_NODE && nid != m_nid)
909 continue;
910
911 /* skip hotpluggable memory regions if needed */
912 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
913 continue;
914
915 /* if we want mirror memory skip non-mirror memory regions */
916 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
917 continue;
918
919 /* skip nomap memory unless we were asked for it explicitly */
920 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
921 continue;
922
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 < type_b->cnt + 1; 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 /*
947 * if idx_b advanced past idx_a,
948 * break out to advance idx_a
949 */
950 if (r_start >= m_end)
951 break;
952 /* if the two regions intersect, we're done */
953 if (m_start < r_end) {
954 if (out_start)
955 *out_start =
956 max(m_start, r_start);
957 if (out_end)
958 *out_end = min(m_end, r_end);
959 if (out_nid)
960 *out_nid = m_nid;
961 /*
962 * The region which ends first is
963 * advanced for the next iteration.
964 */
965 if (m_end <= r_end)
966 idx_a++;
967 else
968 idx_b++;
969 *idx = (u32)idx_a | (u64)idx_b << 32;
970 return;
971 }
972 }
973 }
974
975 /* signal end of iteration */
976 *idx = ULLONG_MAX;
977}
978
979/**
980 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
981 *
982 * Finds the next range from type_a which is not marked as unsuitable
983 * in type_b.
984 *
985 * @idx: pointer to u64 loop variable
986 * @nid: node selector, %NUMA_NO_NODE for all nodes
987 * @flags: pick from blocks based on memory attributes
988 * @type_a: pointer to memblock_type from where the range is taken
989 * @type_b: pointer to memblock_type which excludes memory from being taken
990 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
991 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
992 * @out_nid: ptr to int for nid of the range, can be %NULL
993 *
994 * Reverse of __next_mem_range().
995 */
996void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
997 struct memblock_type *type_a,
998 struct memblock_type *type_b,
999 phys_addr_t *out_start,
1000 phys_addr_t *out_end, int *out_nid)
1001{
1002 int idx_a = *idx & 0xffffffff;
1003 int idx_b = *idx >> 32;
1004
1005 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1006 nid = NUMA_NO_NODE;
1007
1008 if (*idx == (u64)ULLONG_MAX) {
1009 idx_a = type_a->cnt - 1;
1010 idx_b = type_b->cnt;
1011 }
1012
1013 for (; idx_a >= 0; idx_a--) {
1014 struct memblock_region *m = &type_a->regions[idx_a];
1015
1016 phys_addr_t m_start = m->base;
1017 phys_addr_t m_end = m->base + m->size;
1018 int m_nid = memblock_get_region_node(m);
1019
1020 /* only memory regions are associated with nodes, check it */
1021 if (nid != NUMA_NO_NODE && nid != m_nid)
1022 continue;
1023
1024 /* skip hotpluggable memory regions if needed */
1025 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1026 continue;
1027
1028 /* if we want mirror memory skip non-mirror memory regions */
1029 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1030 continue;
1031
1032 /* skip nomap memory unless we were asked for it explicitly */
1033 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1034 continue;
1035
1036 if (!type_b) {
1037 if (out_start)
1038 *out_start = m_start;
1039 if (out_end)
1040 *out_end = m_end;
1041 if (out_nid)
1042 *out_nid = m_nid;
1043 idx_a++;
1044 *idx = (u32)idx_a | (u64)idx_b << 32;
1045 return;
1046 }
1047
1048 /* scan areas before each reservation */
1049 for (; idx_b >= 0; idx_b--) {
1050 struct memblock_region *r;
1051 phys_addr_t r_start;
1052 phys_addr_t r_end;
1053
1054 r = &type_b->regions[idx_b];
1055 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1056 r_end = idx_b < type_b->cnt ?
1057 r->base : ULLONG_MAX;
1058 /*
1059 * if idx_b advanced past idx_a,
1060 * break out to advance idx_a
1061 */
1062
1063 if (r_end <= m_start)
1064 break;
1065 /* if the two regions intersect, we're done */
1066 if (m_end > r_start) {
1067 if (out_start)
1068 *out_start = max(m_start, r_start);
1069 if (out_end)
1070 *out_end = min(m_end, r_end);
1071 if (out_nid)
1072 *out_nid = m_nid;
1073 if (m_start >= r_start)
1074 idx_a--;
1075 else
1076 idx_b--;
1077 *idx = (u32)idx_a | (u64)idx_b << 32;
1078 return;
1079 }
1080 }
1081 }
1082 /* signal end of iteration */
1083 *idx = ULLONG_MAX;
1084}
1085
1086#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1087/*
1088 * Common iterator interface used to define for_each_mem_range().
1089 */
1090void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1091 unsigned long *out_start_pfn,
1092 unsigned long *out_end_pfn, int *out_nid)
1093{
1094 struct memblock_type *type = &memblock.memory;
1095 struct memblock_region *r;
1096
1097 while (++*idx < type->cnt) {
1098 r = &type->regions[*idx];
1099
1100 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1101 continue;
1102 if (nid == MAX_NUMNODES || nid == r->nid)
1103 break;
1104 }
1105 if (*idx >= type->cnt) {
1106 *idx = -1;
1107 return;
1108 }
1109
1110 if (out_start_pfn)
1111 *out_start_pfn = PFN_UP(r->base);
1112 if (out_end_pfn)
1113 *out_end_pfn = PFN_DOWN(r->base + r->size);
1114 if (out_nid)
1115 *out_nid = r->nid;
1116}
1117
1118/**
1119 * memblock_set_node - set node ID on memblock regions
1120 * @base: base of area to set node ID for
1121 * @size: size of area to set node ID for
1122 * @type: memblock type to set node ID for
1123 * @nid: node ID to set
1124 *
1125 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1126 * Regions which cross the area boundaries are split as necessary.
1127 *
1128 * RETURNS:
1129 * 0 on success, -errno on failure.
1130 */
1131int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1132 struct memblock_type *type, int nid)
1133{
1134 int start_rgn, end_rgn;
1135 int i, ret;
1136
1137 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1138 if (ret)
1139 return ret;
1140
1141 for (i = start_rgn; i < end_rgn; i++)
1142 memblock_set_region_node(&type->regions[i], nid);
1143
1144 memblock_merge_regions(type);
1145 return 0;
1146}
1147#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1148
1149static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1150 phys_addr_t align, phys_addr_t start,
1151 phys_addr_t end, int nid, ulong flags)
1152{
1153 phys_addr_t found;
1154
1155 if (!align)
1156 align = SMP_CACHE_BYTES;
1157
1158 found = memblock_find_in_range_node(size, align, start, end, nid,
1159 flags);
1160 if (found && !memblock_reserve(found, size)) {
1161 /*
1162 * The min_count is set to 0 so that memblock allocations are
1163 * never reported as leaks.
1164 */
1165 kmemleak_alloc(__va(found), size, 0, 0);
1166 return found;
1167 }
1168 return 0;
1169}
1170
1171phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1172 phys_addr_t start, phys_addr_t end,
1173 ulong flags)
1174{
1175 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1176 flags);
1177}
1178
1179static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1180 phys_addr_t align, phys_addr_t max_addr,
1181 int nid, ulong flags)
1182{
1183 return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1184}
1185
1186phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1187{
1188 ulong flags = choose_memblock_flags();
1189 phys_addr_t ret;
1190
1191again:
1192 ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1193 nid, flags);
1194
1195 if (!ret && (flags & MEMBLOCK_MIRROR)) {
1196 flags &= ~MEMBLOCK_MIRROR;
1197 goto again;
1198 }
1199 return ret;
1200}
1201
1202phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1203{
1204 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1205 MEMBLOCK_NONE);
1206}
1207
1208phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1209{
1210 phys_addr_t alloc;
1211
1212 alloc = __memblock_alloc_base(size, align, max_addr);
1213
1214 if (alloc == 0)
1215 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1216 (unsigned long long) size, (unsigned long long) max_addr);
1217
1218 return alloc;
1219}
1220
1221phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1222{
1223 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1224}
1225
1226phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1227{
1228 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1229
1230 if (res)
1231 return res;
1232 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1233}
1234
1235/**
1236 * memblock_virt_alloc_internal - allocate boot memory block
1237 * @size: size of memory block to be allocated in bytes
1238 * @align: alignment of the region and block's size
1239 * @min_addr: the lower bound of the memory region to allocate (phys address)
1240 * @max_addr: the upper bound of the memory region to allocate (phys address)
1241 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1242 *
1243 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1244 * will fall back to memory below @min_addr. Also, allocation may fall back
1245 * to any node in the system if the specified node can not
1246 * hold the requested memory.
1247 *
1248 * The allocation is performed from memory region limited by
1249 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1250 *
1251 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1252 *
1253 * The phys address of allocated boot memory block is converted to virtual and
1254 * allocated memory is reset to 0.
1255 *
1256 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1257 * allocated boot memory block, so that it is never reported as leaks.
1258 *
1259 * RETURNS:
1260 * Virtual address of allocated memory block on success, NULL on failure.
1261 */
1262static void * __init memblock_virt_alloc_internal(
1263 phys_addr_t size, phys_addr_t align,
1264 phys_addr_t min_addr, phys_addr_t max_addr,
1265 int nid)
1266{
1267 phys_addr_t alloc;
1268 void *ptr;
1269 ulong flags = choose_memblock_flags();
1270
1271 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1272 nid = NUMA_NO_NODE;
1273
1274 /*
1275 * Detect any accidental use of these APIs after slab is ready, as at
1276 * this moment memblock may be deinitialized already and its
1277 * internal data may be destroyed (after execution of free_all_bootmem)
1278 */
1279 if (WARN_ON_ONCE(slab_is_available()))
1280 return kzalloc_node(size, GFP_NOWAIT, nid);
1281
1282 if (!align)
1283 align = SMP_CACHE_BYTES;
1284
1285 if (max_addr > memblock.current_limit)
1286 max_addr = memblock.current_limit;
1287
1288again:
1289 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1290 nid, flags);
1291 if (alloc)
1292 goto done;
1293
1294 if (nid != NUMA_NO_NODE) {
1295 alloc = memblock_find_in_range_node(size, align, min_addr,
1296 max_addr, NUMA_NO_NODE,
1297 flags);
1298 if (alloc)
1299 goto done;
1300 }
1301
1302 if (min_addr) {
1303 min_addr = 0;
1304 goto again;
1305 }
1306
1307 if (flags & MEMBLOCK_MIRROR) {
1308 flags &= ~MEMBLOCK_MIRROR;
1309 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1310 &size);
1311 goto again;
1312 }
1313
1314 return NULL;
1315done:
1316 memblock_reserve(alloc, size);
1317 ptr = phys_to_virt(alloc);
1318 memset(ptr, 0, size);
1319
1320 /*
1321 * The min_count is set to 0 so that bootmem allocated blocks
1322 * are never reported as leaks. This is because many of these blocks
1323 * are only referred via the physical address which is not
1324 * looked up by kmemleak.
1325 */
1326 kmemleak_alloc(ptr, size, 0, 0);
1327
1328 return ptr;
1329}
1330
1331/**
1332 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1333 * @size: size of memory block to be allocated in bytes
1334 * @align: alignment of the region and block's size
1335 * @min_addr: the lower bound of the memory region from where the allocation
1336 * is preferred (phys address)
1337 * @max_addr: the upper bound of the memory region from where the allocation
1338 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1339 * allocate only from memory limited by memblock.current_limit value
1340 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1341 *
1342 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1343 * additional debug information (including caller info), if enabled.
1344 *
1345 * RETURNS:
1346 * Virtual address of allocated memory block on success, NULL on failure.
1347 */
1348void * __init memblock_virt_alloc_try_nid_nopanic(
1349 phys_addr_t size, phys_addr_t align,
1350 phys_addr_t min_addr, phys_addr_t max_addr,
1351 int nid)
1352{
1353 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1354 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1355 (u64)max_addr, (void *)_RET_IP_);
1356 return memblock_virt_alloc_internal(size, align, min_addr,
1357 max_addr, nid);
1358}
1359
1360/**
1361 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1362 * @size: size of memory block to be allocated in bytes
1363 * @align: alignment of the region and block's size
1364 * @min_addr: the lower bound of the memory region from where the allocation
1365 * is preferred (phys address)
1366 * @max_addr: the upper bound of the memory region from where the allocation
1367 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1368 * allocate only from memory limited by memblock.current_limit value
1369 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1370 *
1371 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1372 * which provides debug information (including caller info), if enabled,
1373 * and panics if the request can not be satisfied.
1374 *
1375 * RETURNS:
1376 * Virtual address of allocated memory block on success, NULL on failure.
1377 */
1378void * __init memblock_virt_alloc_try_nid(
1379 phys_addr_t size, phys_addr_t align,
1380 phys_addr_t min_addr, phys_addr_t max_addr,
1381 int nid)
1382{
1383 void *ptr;
1384
1385 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1386 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1387 (u64)max_addr, (void *)_RET_IP_);
1388 ptr = memblock_virt_alloc_internal(size, align,
1389 min_addr, max_addr, nid);
1390 if (ptr)
1391 return ptr;
1392
1393 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1394 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1395 (u64)max_addr);
1396 return NULL;
1397}
1398
1399/**
1400 * __memblock_free_early - free boot memory block
1401 * @base: phys starting address of the boot memory block
1402 * @size: size of the boot memory block in bytes
1403 *
1404 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1405 * The freeing memory will not be released to the buddy allocator.
1406 */
1407void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1408{
1409 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1410 __func__, (u64)base, (u64)base + size - 1,
1411 (void *)_RET_IP_);
1412 kmemleak_free_part(__va(base), size);
1413 memblock_remove_range(&memblock.reserved, base, size);
1414}
1415
1416/*
1417 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1418 * @addr: phys starting address of the boot memory block
1419 * @size: size of the boot memory block in bytes
1420 *
1421 * This is only useful when the bootmem allocator has already been torn
1422 * down, but we are still initializing the system. Pages are released directly
1423 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1424 */
1425void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1426{
1427 u64 cursor, end;
1428
1429 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1430 __func__, (u64)base, (u64)base + size - 1,
1431 (void *)_RET_IP_);
1432 kmemleak_free_part(__va(base), size);
1433 cursor = PFN_UP(base);
1434 end = PFN_DOWN(base + size);
1435
1436 for (; cursor < end; cursor++) {
1437 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1438 totalram_pages++;
1439 }
1440}
1441
1442/*
1443 * Remaining API functions
1444 */
1445
1446phys_addr_t __init_memblock memblock_phys_mem_size(void)
1447{
1448 return memblock.memory.total_size;
1449}
1450
1451phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1452{
1453 unsigned long pages = 0;
1454 struct memblock_region *r;
1455 unsigned long start_pfn, end_pfn;
1456
1457 for_each_memblock(memory, r) {
1458 start_pfn = memblock_region_memory_base_pfn(r);
1459 end_pfn = memblock_region_memory_end_pfn(r);
1460 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1461 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1462 pages += end_pfn - start_pfn;
1463 }
1464
1465 return PFN_PHYS(pages);
1466}
1467
1468/* lowest address */
1469phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1470{
1471 return memblock.memory.regions[0].base;
1472}
1473
1474phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1475{
1476 int idx = memblock.memory.cnt - 1;
1477
1478 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1479}
1480
1481void __init memblock_enforce_memory_limit(phys_addr_t limit)
1482{
1483 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1484 struct memblock_region *r;
1485
1486 if (!limit)
1487 return;
1488
1489 /* find out max address */
1490 for_each_memblock(memory, r) {
1491 if (limit <= r->size) {
1492 max_addr = r->base + limit;
1493 break;
1494 }
1495 limit -= r->size;
1496 }
1497
1498 /* truncate both memory and reserved regions */
1499 memblock_remove_range(&memblock.memory, max_addr,
1500 (phys_addr_t)ULLONG_MAX);
1501 memblock_remove_range(&memblock.reserved, max_addr,
1502 (phys_addr_t)ULLONG_MAX);
1503}
1504
1505static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1506{
1507 unsigned int left = 0, right = type->cnt;
1508
1509 do {
1510 unsigned int mid = (right + left) / 2;
1511
1512 if (addr < type->regions[mid].base)
1513 right = mid;
1514 else if (addr >= (type->regions[mid].base +
1515 type->regions[mid].size))
1516 left = mid + 1;
1517 else
1518 return mid;
1519 } while (left < right);
1520 return -1;
1521}
1522
1523bool __init memblock_is_reserved(phys_addr_t addr)
1524{
1525 return memblock_search(&memblock.reserved, addr) != -1;
1526}
1527
1528bool __init_memblock memblock_is_memory(phys_addr_t addr)
1529{
1530 return memblock_search(&memblock.memory, addr) != -1;
1531}
1532
1533int __init_memblock memblock_is_map_memory(phys_addr_t addr)
1534{
1535 int i = memblock_search(&memblock.memory, addr);
1536
1537 if (i == -1)
1538 return false;
1539 return !memblock_is_nomap(&memblock.memory.regions[i]);
1540}
1541
1542#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1543int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1544 unsigned long *start_pfn, unsigned long *end_pfn)
1545{
1546 struct memblock_type *type = &memblock.memory;
1547 int mid = memblock_search(type, PFN_PHYS(pfn));
1548
1549 if (mid == -1)
1550 return -1;
1551
1552 *start_pfn = PFN_DOWN(type->regions[mid].base);
1553 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1554
1555 return type->regions[mid].nid;
1556}
1557#endif
1558
1559/**
1560 * memblock_is_region_memory - check if a region is a subset of memory
1561 * @base: base of region to check
1562 * @size: size of region to check
1563 *
1564 * Check if the region [@base, @base+@size) is a subset of a memory block.
1565 *
1566 * RETURNS:
1567 * 0 if false, non-zero if true
1568 */
1569int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1570{
1571 int idx = memblock_search(&memblock.memory, base);
1572 phys_addr_t end = base + memblock_cap_size(base, &size);
1573
1574 if (idx == -1)
1575 return 0;
1576 return memblock.memory.regions[idx].base <= base &&
1577 (memblock.memory.regions[idx].base +
1578 memblock.memory.regions[idx].size) >= end;
1579}
1580
1581/**
1582 * memblock_is_region_reserved - check if a region intersects reserved memory
1583 * @base: base of region to check
1584 * @size: size of region to check
1585 *
1586 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1587 *
1588 * RETURNS:
1589 * True if they intersect, false if not.
1590 */
1591bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1592{
1593 memblock_cap_size(base, &size);
1594 return memblock_overlaps_region(&memblock.reserved, base, size);
1595}
1596
1597void __init_memblock memblock_trim_memory(phys_addr_t align)
1598{
1599 phys_addr_t start, end, orig_start, orig_end;
1600 struct memblock_region *r;
1601
1602 for_each_memblock(memory, r) {
1603 orig_start = r->base;
1604 orig_end = r->base + r->size;
1605 start = round_up(orig_start, align);
1606 end = round_down(orig_end, align);
1607
1608 if (start == orig_start && end == orig_end)
1609 continue;
1610
1611 if (start < end) {
1612 r->base = start;
1613 r->size = end - start;
1614 } else {
1615 memblock_remove_region(&memblock.memory,
1616 r - memblock.memory.regions);
1617 r--;
1618 }
1619 }
1620}
1621
1622void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1623{
1624 memblock.current_limit = limit;
1625}
1626
1627phys_addr_t __init_memblock memblock_get_current_limit(void)
1628{
1629 return memblock.current_limit;
1630}
1631
1632static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1633{
1634 unsigned long long base, size;
1635 unsigned long flags;
1636 int idx;
1637 struct memblock_region *rgn;
1638
1639 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
1640
1641 for_each_memblock_type(type, rgn) {
1642 char nid_buf[32] = "";
1643
1644 base = rgn->base;
1645 size = rgn->size;
1646 flags = rgn->flags;
1647#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1648 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1649 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1650 memblock_get_region_node(rgn));
1651#endif
1652 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1653 name, idx, base, base + size - 1, size, nid_buf, flags);
1654 }
1655}
1656
1657void __init_memblock __memblock_dump_all(void)
1658{
1659 pr_info("MEMBLOCK configuration:\n");
1660 pr_info(" memory size = %#llx reserved size = %#llx\n",
1661 (unsigned long long)memblock.memory.total_size,
1662 (unsigned long long)memblock.reserved.total_size);
1663
1664 memblock_dump(&memblock.memory, "memory");
1665 memblock_dump(&memblock.reserved, "reserved");
1666}
1667
1668void __init memblock_allow_resize(void)
1669{
1670 memblock_can_resize = 1;
1671}
1672
1673static int __init early_memblock(char *p)
1674{
1675 if (p && strstr(p, "debug"))
1676 memblock_debug = 1;
1677 return 0;
1678}
1679early_param("memblock", early_memblock);
1680
1681#if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1682
1683static int memblock_debug_show(struct seq_file *m, void *private)
1684{
1685 struct memblock_type *type = m->private;
1686 struct memblock_region *reg;
1687 int i;
1688
1689 for (i = 0; i < type->cnt; i++) {
1690 reg = &type->regions[i];
1691 seq_printf(m, "%4d: ", i);
1692 if (sizeof(phys_addr_t) == 4)
1693 seq_printf(m, "0x%08lx..0x%08lx\n",
1694 (unsigned long)reg->base,
1695 (unsigned long)(reg->base + reg->size - 1));
1696 else
1697 seq_printf(m, "0x%016llx..0x%016llx\n",
1698 (unsigned long long)reg->base,
1699 (unsigned long long)(reg->base + reg->size - 1));
1700
1701 }
1702 return 0;
1703}
1704
1705static int memblock_debug_open(struct inode *inode, struct file *file)
1706{
1707 return single_open(file, memblock_debug_show, inode->i_private);
1708}
1709
1710static const struct file_operations memblock_debug_fops = {
1711 .open = memblock_debug_open,
1712 .read = seq_read,
1713 .llseek = seq_lseek,
1714 .release = single_release,
1715};
1716
1717static int __init memblock_init_debugfs(void)
1718{
1719 struct dentry *root = debugfs_create_dir("memblock", NULL);
1720 if (!root)
1721 return -ENXIO;
1722 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1723 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1724#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1725 debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1726#endif
1727
1728 return 0;
1729}
1730__initcall(memblock_init_debugfs);
1731
1732#endif /* CONFIG_DEBUG_FS */
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>
14#include <linux/slab.h>
15#include <linux/init.h>
16#include <linux/bitops.h>
17#include <linux/poison.h>
18#include <linux/pfn.h>
19#include <linux/debugfs.h>
20#include <linux/seq_file.h>
21#include <linux/memblock.h>
22
23#include <asm-generic/sections.h>
24#include <linux/io.h>
25
26#include "internal.h"
27
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
40 .bottom_up = false,
41 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
42};
43
44int memblock_debug __initdata_memblock;
45#ifdef CONFIG_MOVABLE_NODE
46bool movable_node_enabled __initdata_memblock = false;
47#endif
48static int memblock_can_resize __initdata_memblock;
49static int memblock_memory_in_slab __initdata_memblock = 0;
50static int memblock_reserved_in_slab __initdata_memblock = 0;
51
52/* inline so we don't get a warning when pr_debug is compiled out */
53static __init_memblock const char *
54memblock_type_name(struct memblock_type *type)
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
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
70/*
71 * Address comparison utilities
72 */
73static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
74 phys_addr_t base2, phys_addr_t size2)
75{
76 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
77}
78
79static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
80 phys_addr_t base, phys_addr_t size)
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
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
100 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
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
126/**
127 * __memblock_find_range_top_down - find free area utility, in top-down
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
132 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
133 *
134 * Utility called from memblock_find_in_range_node(), find free area top-down.
135 *
136 * RETURNS:
137 * Found address on success, 0 on failure.
138 */
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)
142{
143 phys_addr_t this_start, this_end, cand;
144 u64 i;
145
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 }
157
158 return 0;
159}
160
161/**
162 * memblock_find_in_range_node - find free area in given range and node
163 * @size: size of free area to find
164 * @align: alignment of free area to find
165 * @start: start of candidate range
166 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
167 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
168 *
169 * Find @size free area aligned to @align in the specified range and node.
170 *
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 *
179 * RETURNS:
180 * Found address on success, 0 on failure.
181 */
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)
185{
186 int ret;
187 phys_addr_t kernel_end;
188
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);
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 }
227
228 return __memblock_find_range_top_down(start, end, size, align, nid);
229}
230
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:
241 * Found address on success, 0 on failure.
242 */
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)
246{
247 return memblock_find_in_range_node(size, align, start, end,
248 NUMA_NO_NODE);
249}
250
251static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
252{
253 type->total_size -= type->regions[r].size;
254 memmove(&type->regions[r], &type->regions[r + 1],
255 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
256 type->cnt--;
257
258 /* Special case for empty arrays */
259 if (type->cnt == 0) {
260 WARN_ON(type->total_size != 0);
261 type->cnt = 1;
262 type->regions[0].base = 0;
263 type->regions[0].size = 0;
264 type->regions[0].flags = 0;
265 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
266 }
267}
268
269#ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
270
271phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
272 phys_addr_t *addr)
273{
274 if (memblock.reserved.regions == memblock_reserved_init_regions)
275 return 0;
276
277 *addr = __pa(memblock.reserved.regions);
278
279 return PAGE_ALIGN(sizeof(struct memblock_region) *
280 memblock.reserved.max);
281}
282
283phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
284 phys_addr_t *addr)
285{
286 if (memblock.memory.regions == memblock_memory_init_regions)
287 return 0;
288
289 *addr = __pa(memblock.memory.regions);
290
291 return PAGE_ALIGN(sizeof(struct memblock_region) *
292 memblock.memory.max);
293}
294
295#endif
296
297/**
298 * memblock_double_array - double the size of the memblock regions array
299 * @type: memblock type of the regions array being doubled
300 * @new_area_start: starting address of memory range to avoid overlap with
301 * @new_area_size: size of memory range to avoid overlap with
302 *
303 * Double the size of the @type regions array. If memblock is being used to
304 * allocate memory for a new reserved regions array and there is a previously
305 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
306 * waiting to be reserved, ensure the memory used by the new array does
307 * not overlap.
308 *
309 * RETURNS:
310 * 0 on success, -1 on failure.
311 */
312static int __init_memblock memblock_double_array(struct memblock_type *type,
313 phys_addr_t new_area_start,
314 phys_addr_t new_area_size)
315{
316 struct memblock_region *new_array, *old_array;
317 phys_addr_t old_alloc_size, new_alloc_size;
318 phys_addr_t old_size, new_size, addr;
319 int use_slab = slab_is_available();
320 int *in_slab;
321
322 /* We don't allow resizing until we know about the reserved regions
323 * of memory that aren't suitable for allocation
324 */
325 if (!memblock_can_resize)
326 return -1;
327
328 /* Calculate new doubled size */
329 old_size = type->max * sizeof(struct memblock_region);
330 new_size = old_size << 1;
331 /*
332 * We need to allocated new one align to PAGE_SIZE,
333 * so we can free them completely later.
334 */
335 old_alloc_size = PAGE_ALIGN(old_size);
336 new_alloc_size = PAGE_ALIGN(new_size);
337
338 /* Retrieve the slab flag */
339 if (type == &memblock.memory)
340 in_slab = &memblock_memory_in_slab;
341 else
342 in_slab = &memblock_reserved_in_slab;
343
344 /* Try to find some space for it.
345 *
346 * WARNING: We assume that either slab_is_available() and we use it or
347 * we use MEMBLOCK for allocations. That means that this is unsafe to
348 * use when bootmem is currently active (unless bootmem itself is
349 * implemented on top of MEMBLOCK which isn't the case yet)
350 *
351 * This should however not be an issue for now, as we currently only
352 * call into MEMBLOCK while it's still active, or much later when slab
353 * is active for memory hotplug operations
354 */
355 if (use_slab) {
356 new_array = kmalloc(new_size, GFP_KERNEL);
357 addr = new_array ? __pa(new_array) : 0;
358 } else {
359 /* only exclude range when trying to double reserved.regions */
360 if (type != &memblock.reserved)
361 new_area_start = new_area_size = 0;
362
363 addr = memblock_find_in_range(new_area_start + new_area_size,
364 memblock.current_limit,
365 new_alloc_size, PAGE_SIZE);
366 if (!addr && new_area_size)
367 addr = memblock_find_in_range(0,
368 min(new_area_start, memblock.current_limit),
369 new_alloc_size, PAGE_SIZE);
370
371 new_array = addr ? __va(addr) : NULL;
372 }
373 if (!addr) {
374 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
375 memblock_type_name(type), type->max, type->max * 2);
376 return -1;
377 }
378
379 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
380 memblock_type_name(type), type->max * 2, (u64)addr,
381 (u64)addr + new_size - 1);
382
383 /*
384 * Found space, we now need to move the array over before we add the
385 * reserved region since it may be our reserved array itself that is
386 * full.
387 */
388 memcpy(new_array, type->regions, old_size);
389 memset(new_array + type->max, 0, old_size);
390 old_array = type->regions;
391 type->regions = new_array;
392 type->max <<= 1;
393
394 /* Free old array. We needn't free it if the array is the static one */
395 if (*in_slab)
396 kfree(old_array);
397 else if (old_array != memblock_memory_init_regions &&
398 old_array != memblock_reserved_init_regions)
399 memblock_free(__pa(old_array), old_alloc_size);
400
401 /*
402 * Reserve the new array if that comes from the memblock. Otherwise, we
403 * needn't do it
404 */
405 if (!use_slab)
406 BUG_ON(memblock_reserve(addr, new_alloc_size));
407
408 /* Update slab flag */
409 *in_slab = use_slab;
410
411 return 0;
412}
413
414/**
415 * memblock_merge_regions - merge neighboring compatible regions
416 * @type: memblock type to scan
417 *
418 * Scan @type and merge neighboring compatible regions.
419 */
420static void __init_memblock memblock_merge_regions(struct memblock_type *type)
421{
422 int i = 0;
423
424 /* cnt never goes below 1 */
425 while (i < type->cnt - 1) {
426 struct memblock_region *this = &type->regions[i];
427 struct memblock_region *next = &type->regions[i + 1];
428
429 if (this->base + this->size != next->base ||
430 memblock_get_region_node(this) !=
431 memblock_get_region_node(next) ||
432 this->flags != next->flags) {
433 BUG_ON(this->base + this->size > next->base);
434 i++;
435 continue;
436 }
437
438 this->size += next->size;
439 /* move forward from next + 1, index of which is i + 2 */
440 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
441 type->cnt--;
442 }
443}
444
445/**
446 * memblock_insert_region - insert new memblock region
447 * @type: memblock type to insert into
448 * @idx: index for the insertion point
449 * @base: base address of the new region
450 * @size: size of the new region
451 * @nid: node id of the new region
452 * @flags: flags of the new region
453 *
454 * Insert new memblock region [@base,@base+@size) into @type at @idx.
455 * @type must already have extra room to accomodate the new region.
456 */
457static void __init_memblock memblock_insert_region(struct memblock_type *type,
458 int idx, phys_addr_t base,
459 phys_addr_t size,
460 int nid, unsigned long flags)
461{
462 struct memblock_region *rgn = &type->regions[idx];
463
464 BUG_ON(type->cnt >= type->max);
465 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
466 rgn->base = base;
467 rgn->size = size;
468 rgn->flags = flags;
469 memblock_set_region_node(rgn, nid);
470 type->cnt++;
471 type->total_size += size;
472}
473
474/**
475 * memblock_add_region - add new memblock region
476 * @type: memblock type to add new region into
477 * @base: base address of the new region
478 * @size: size of the new region
479 * @nid: nid of the new region
480 * @flags: flags of the new region
481 *
482 * Add new memblock region [@base,@base+@size) into @type. The new region
483 * is allowed to overlap with existing ones - overlaps don't affect already
484 * existing regions. @type is guaranteed to be minimal (all neighbouring
485 * compatible regions are merged) after the addition.
486 *
487 * RETURNS:
488 * 0 on success, -errno on failure.
489 */
490static int __init_memblock memblock_add_region(struct memblock_type *type,
491 phys_addr_t base, phys_addr_t size,
492 int nid, unsigned long flags)
493{
494 bool insert = false;
495 phys_addr_t obase = base;
496 phys_addr_t end = base + memblock_cap_size(base, &size);
497 int i, nr_new;
498
499 if (!size)
500 return 0;
501
502 /* special case for empty array */
503 if (type->regions[0].size == 0) {
504 WARN_ON(type->cnt != 1 || type->total_size);
505 type->regions[0].base = base;
506 type->regions[0].size = size;
507 type->regions[0].flags = flags;
508 memblock_set_region_node(&type->regions[0], nid);
509 type->total_size = size;
510 return 0;
511 }
512repeat:
513 /*
514 * The following is executed twice. Once with %false @insert and
515 * then with %true. The first counts the number of regions needed
516 * to accomodate the new area. The second actually inserts them.
517 */
518 base = obase;
519 nr_new = 0;
520
521 for (i = 0; i < type->cnt; i++) {
522 struct memblock_region *rgn = &type->regions[i];
523 phys_addr_t rbase = rgn->base;
524 phys_addr_t rend = rbase + rgn->size;
525
526 if (rbase >= end)
527 break;
528 if (rend <= base)
529 continue;
530 /*
531 * @rgn overlaps. If it separates the lower part of new
532 * area, insert that portion.
533 */
534 if (rbase > base) {
535 nr_new++;
536 if (insert)
537 memblock_insert_region(type, i++, base,
538 rbase - base, nid,
539 flags);
540 }
541 /* area below @rend is dealt with, forget about it */
542 base = min(rend, end);
543 }
544
545 /* insert the remaining portion */
546 if (base < end) {
547 nr_new++;
548 if (insert)
549 memblock_insert_region(type, i, base, end - base,
550 nid, flags);
551 }
552
553 /*
554 * If this was the first round, resize array and repeat for actual
555 * insertions; otherwise, merge and return.
556 */
557 if (!insert) {
558 while (type->cnt + nr_new > type->max)
559 if (memblock_double_array(type, obase, size) < 0)
560 return -ENOMEM;
561 insert = true;
562 goto repeat;
563 } else {
564 memblock_merge_regions(type);
565 return 0;
566 }
567}
568
569int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
570 int nid)
571{
572 return memblock_add_region(&memblock.memory, base, size, nid, 0);
573}
574
575int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
576{
577 return memblock_add_region(&memblock.memory, base, size,
578 MAX_NUMNODES, 0);
579}
580
581/**
582 * memblock_isolate_range - isolate given range into disjoint memblocks
583 * @type: memblock type to isolate range for
584 * @base: base of range to isolate
585 * @size: size of range to isolate
586 * @start_rgn: out parameter for the start of isolated region
587 * @end_rgn: out parameter for the end of isolated region
588 *
589 * Walk @type and ensure that regions don't cross the boundaries defined by
590 * [@base,@base+@size). Crossing regions are split at the boundaries,
591 * which may create at most two more regions. The index of the first
592 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
593 *
594 * RETURNS:
595 * 0 on success, -errno on failure.
596 */
597static int __init_memblock memblock_isolate_range(struct memblock_type *type,
598 phys_addr_t base, phys_addr_t size,
599 int *start_rgn, int *end_rgn)
600{
601 phys_addr_t end = base + memblock_cap_size(base, &size);
602 int i;
603
604 *start_rgn = *end_rgn = 0;
605
606 if (!size)
607 return 0;
608
609 /* we'll create at most two more regions */
610 while (type->cnt + 2 > type->max)
611 if (memblock_double_array(type, base, size) < 0)
612 return -ENOMEM;
613
614 for (i = 0; i < type->cnt; i++) {
615 struct memblock_region *rgn = &type->regions[i];
616 phys_addr_t rbase = rgn->base;
617 phys_addr_t rend = rbase + rgn->size;
618
619 if (rbase >= end)
620 break;
621 if (rend <= base)
622 continue;
623
624 if (rbase < base) {
625 /*
626 * @rgn intersects from below. Split and continue
627 * to process the next region - the new top half.
628 */
629 rgn->base = base;
630 rgn->size -= base - rbase;
631 type->total_size -= base - rbase;
632 memblock_insert_region(type, i, rbase, base - rbase,
633 memblock_get_region_node(rgn),
634 rgn->flags);
635 } else if (rend > end) {
636 /*
637 * @rgn intersects from above. Split and redo the
638 * current region - the new bottom half.
639 */
640 rgn->base = end;
641 rgn->size -= end - rbase;
642 type->total_size -= end - rbase;
643 memblock_insert_region(type, i--, rbase, end - rbase,
644 memblock_get_region_node(rgn),
645 rgn->flags);
646 } else {
647 /* @rgn is fully contained, record it */
648 if (!*end_rgn)
649 *start_rgn = i;
650 *end_rgn = i + 1;
651 }
652 }
653
654 return 0;
655}
656
657static int __init_memblock __memblock_remove(struct memblock_type *type,
658 phys_addr_t base, phys_addr_t size)
659{
660 int start_rgn, end_rgn;
661 int i, ret;
662
663 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
664 if (ret)
665 return ret;
666
667 for (i = end_rgn - 1; i >= start_rgn; i--)
668 memblock_remove_region(type, i);
669 return 0;
670}
671
672int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
673{
674 return __memblock_remove(&memblock.memory, base, size);
675}
676
677int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
678{
679 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
680 (unsigned long long)base,
681 (unsigned long long)base + size - 1,
682 (void *)_RET_IP_);
683
684 return __memblock_remove(&memblock.reserved, base, size);
685}
686
687static int __init_memblock memblock_reserve_region(phys_addr_t base,
688 phys_addr_t size,
689 int nid,
690 unsigned long flags)
691{
692 struct memblock_type *_rgn = &memblock.reserved;
693
694 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
695 (unsigned long long)base,
696 (unsigned long long)base + size - 1,
697 flags, (void *)_RET_IP_);
698
699 return memblock_add_region(_rgn, base, size, nid, flags);
700}
701
702int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
703{
704 return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
705}
706
707/**
708 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
709 * @base: the base phys addr of the region
710 * @size: the size of the region
711 *
712 * This function isolates region [@base, @base + @size), and mark it with flag
713 * MEMBLOCK_HOTPLUG.
714 *
715 * Return 0 on succees, -errno on failure.
716 */
717int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
718{
719 struct memblock_type *type = &memblock.memory;
720 int i, ret, start_rgn, end_rgn;
721
722 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
723 if (ret)
724 return ret;
725
726 for (i = start_rgn; i < end_rgn; i++)
727 memblock_set_region_flags(&type->regions[i], MEMBLOCK_HOTPLUG);
728
729 memblock_merge_regions(type);
730 return 0;
731}
732
733/**
734 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
735 * @base: the base phys addr of the region
736 * @size: the size of the region
737 *
738 * This function isolates region [@base, @base + @size), and clear flag
739 * MEMBLOCK_HOTPLUG for the isolated regions.
740 *
741 * Return 0 on succees, -errno on failure.
742 */
743int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
744{
745 struct memblock_type *type = &memblock.memory;
746 int i, ret, start_rgn, end_rgn;
747
748 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
749 if (ret)
750 return ret;
751
752 for (i = start_rgn; i < end_rgn; i++)
753 memblock_clear_region_flags(&type->regions[i],
754 MEMBLOCK_HOTPLUG);
755
756 memblock_merge_regions(type);
757 return 0;
758}
759
760/**
761 * __next_free_mem_range - next function for for_each_free_mem_range()
762 * @idx: pointer to u64 loop variable
763 * @nid: node selector, %NUMA_NO_NODE for all nodes
764 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
765 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
766 * @out_nid: ptr to int for nid of the range, can be %NULL
767 *
768 * Find the first free area from *@idx which matches @nid, fill the out
769 * parameters, and update *@idx for the next iteration. The lower 32bit of
770 * *@idx contains index into memory region and the upper 32bit indexes the
771 * areas before each reserved region. For example, if reserved regions
772 * look like the following,
773 *
774 * 0:[0-16), 1:[32-48), 2:[128-130)
775 *
776 * The upper 32bit indexes the following regions.
777 *
778 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
779 *
780 * As both region arrays are sorted, the function advances the two indices
781 * in lockstep and returns each intersection.
782 */
783void __init_memblock __next_free_mem_range(u64 *idx, int nid,
784 phys_addr_t *out_start,
785 phys_addr_t *out_end, int *out_nid)
786{
787 struct memblock_type *mem = &memblock.memory;
788 struct memblock_type *rsv = &memblock.reserved;
789 int mi = *idx & 0xffffffff;
790 int ri = *idx >> 32;
791
792 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
793 nid = NUMA_NO_NODE;
794
795 for ( ; mi < mem->cnt; mi++) {
796 struct memblock_region *m = &mem->regions[mi];
797 phys_addr_t m_start = m->base;
798 phys_addr_t m_end = m->base + m->size;
799
800 /* only memory regions are associated with nodes, check it */
801 if (nid != NUMA_NO_NODE && nid != memblock_get_region_node(m))
802 continue;
803
804 /* scan areas before each reservation for intersection */
805 for ( ; ri < rsv->cnt + 1; ri++) {
806 struct memblock_region *r = &rsv->regions[ri];
807 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
808 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
809
810 /* if ri advanced past mi, break out to advance mi */
811 if (r_start >= m_end)
812 break;
813 /* if the two regions intersect, we're done */
814 if (m_start < r_end) {
815 if (out_start)
816 *out_start = max(m_start, r_start);
817 if (out_end)
818 *out_end = min(m_end, r_end);
819 if (out_nid)
820 *out_nid = memblock_get_region_node(m);
821 /*
822 * The region which ends first is advanced
823 * for the next iteration.
824 */
825 if (m_end <= r_end)
826 mi++;
827 else
828 ri++;
829 *idx = (u32)mi | (u64)ri << 32;
830 return;
831 }
832 }
833 }
834
835 /* signal end of iteration */
836 *idx = ULLONG_MAX;
837}
838
839/**
840 * __next_free_mem_range_rev - next function for for_each_free_mem_range_reverse()
841 * @idx: pointer to u64 loop variable
842 * @nid: nid: node selector, %NUMA_NO_NODE for all nodes
843 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
844 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
845 * @out_nid: ptr to int for nid of the range, can be %NULL
846 *
847 * Reverse of __next_free_mem_range().
848 *
849 * Linux kernel cannot migrate pages used by itself. Memory hotplug users won't
850 * be able to hot-remove hotpluggable memory used by the kernel. So this
851 * function skip hotpluggable regions if needed when allocating memory for the
852 * kernel.
853 */
854void __init_memblock __next_free_mem_range_rev(u64 *idx, int nid,
855 phys_addr_t *out_start,
856 phys_addr_t *out_end, int *out_nid)
857{
858 struct memblock_type *mem = &memblock.memory;
859 struct memblock_type *rsv = &memblock.reserved;
860 int mi = *idx & 0xffffffff;
861 int ri = *idx >> 32;
862
863 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
864 nid = NUMA_NO_NODE;
865
866 if (*idx == (u64)ULLONG_MAX) {
867 mi = mem->cnt - 1;
868 ri = rsv->cnt;
869 }
870
871 for ( ; mi >= 0; mi--) {
872 struct memblock_region *m = &mem->regions[mi];
873 phys_addr_t m_start = m->base;
874 phys_addr_t m_end = m->base + m->size;
875
876 /* only memory regions are associated with nodes, check it */
877 if (nid != NUMA_NO_NODE && nid != memblock_get_region_node(m))
878 continue;
879
880 /* skip hotpluggable memory regions if needed */
881 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
882 continue;
883
884 /* scan areas before each reservation for intersection */
885 for ( ; ri >= 0; ri--) {
886 struct memblock_region *r = &rsv->regions[ri];
887 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
888 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
889
890 /* if ri advanced past mi, break out to advance mi */
891 if (r_end <= m_start)
892 break;
893 /* if the two regions intersect, we're done */
894 if (m_end > r_start) {
895 if (out_start)
896 *out_start = max(m_start, r_start);
897 if (out_end)
898 *out_end = min(m_end, r_end);
899 if (out_nid)
900 *out_nid = memblock_get_region_node(m);
901
902 if (m_start >= r_start)
903 mi--;
904 else
905 ri--;
906 *idx = (u32)mi | (u64)ri << 32;
907 return;
908 }
909 }
910 }
911
912 *idx = ULLONG_MAX;
913}
914
915#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
916/*
917 * Common iterator interface used to define for_each_mem_range().
918 */
919void __init_memblock __next_mem_pfn_range(int *idx, int nid,
920 unsigned long *out_start_pfn,
921 unsigned long *out_end_pfn, int *out_nid)
922{
923 struct memblock_type *type = &memblock.memory;
924 struct memblock_region *r;
925
926 while (++*idx < type->cnt) {
927 r = &type->regions[*idx];
928
929 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
930 continue;
931 if (nid == MAX_NUMNODES || nid == r->nid)
932 break;
933 }
934 if (*idx >= type->cnt) {
935 *idx = -1;
936 return;
937 }
938
939 if (out_start_pfn)
940 *out_start_pfn = PFN_UP(r->base);
941 if (out_end_pfn)
942 *out_end_pfn = PFN_DOWN(r->base + r->size);
943 if (out_nid)
944 *out_nid = r->nid;
945}
946
947/**
948 * memblock_set_node - set node ID on memblock regions
949 * @base: base of area to set node ID for
950 * @size: size of area to set node ID for
951 * @type: memblock type to set node ID for
952 * @nid: node ID to set
953 *
954 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
955 * Regions which cross the area boundaries are split as necessary.
956 *
957 * RETURNS:
958 * 0 on success, -errno on failure.
959 */
960int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
961 struct memblock_type *type, int nid)
962{
963 int start_rgn, end_rgn;
964 int i, ret;
965
966 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
967 if (ret)
968 return ret;
969
970 for (i = start_rgn; i < end_rgn; i++)
971 memblock_set_region_node(&type->regions[i], nid);
972
973 memblock_merge_regions(type);
974 return 0;
975}
976#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
977
978static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
979 phys_addr_t align, phys_addr_t max_addr,
980 int nid)
981{
982 phys_addr_t found;
983
984 if (!align)
985 align = SMP_CACHE_BYTES;
986
987 found = memblock_find_in_range_node(size, align, 0, max_addr, nid);
988 if (found && !memblock_reserve(found, size))
989 return found;
990
991 return 0;
992}
993
994phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
995{
996 return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
997}
998
999phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1000{
1001 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE);
1002}
1003
1004phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1005{
1006 phys_addr_t alloc;
1007
1008 alloc = __memblock_alloc_base(size, align, max_addr);
1009
1010 if (alloc == 0)
1011 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1012 (unsigned long long) size, (unsigned long long) max_addr);
1013
1014 return alloc;
1015}
1016
1017phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1018{
1019 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1020}
1021
1022phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1023{
1024 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1025
1026 if (res)
1027 return res;
1028 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1029}
1030
1031/**
1032 * memblock_virt_alloc_internal - allocate boot memory block
1033 * @size: size of memory block to be allocated in bytes
1034 * @align: alignment of the region and block's size
1035 * @min_addr: the lower bound of the memory region to allocate (phys address)
1036 * @max_addr: the upper bound of the memory region to allocate (phys address)
1037 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1038 *
1039 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1040 * will fall back to memory below @min_addr. Also, allocation may fall back
1041 * to any node in the system if the specified node can not
1042 * hold the requested memory.
1043 *
1044 * The allocation is performed from memory region limited by
1045 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1046 *
1047 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1048 *
1049 * The phys address of allocated boot memory block is converted to virtual and
1050 * allocated memory is reset to 0.
1051 *
1052 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1053 * allocated boot memory block, so that it is never reported as leaks.
1054 *
1055 * RETURNS:
1056 * Virtual address of allocated memory block on success, NULL on failure.
1057 */
1058static void * __init memblock_virt_alloc_internal(
1059 phys_addr_t size, phys_addr_t align,
1060 phys_addr_t min_addr, phys_addr_t max_addr,
1061 int nid)
1062{
1063 phys_addr_t alloc;
1064 void *ptr;
1065
1066 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1067 nid = NUMA_NO_NODE;
1068
1069 /*
1070 * Detect any accidental use of these APIs after slab is ready, as at
1071 * this moment memblock may be deinitialized already and its
1072 * internal data may be destroyed (after execution of free_all_bootmem)
1073 */
1074 if (WARN_ON_ONCE(slab_is_available()))
1075 return kzalloc_node(size, GFP_NOWAIT, nid);
1076
1077 if (!align)
1078 align = SMP_CACHE_BYTES;
1079
1080 if (max_addr > memblock.current_limit)
1081 max_addr = memblock.current_limit;
1082
1083again:
1084 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1085 nid);
1086 if (alloc)
1087 goto done;
1088
1089 if (nid != NUMA_NO_NODE) {
1090 alloc = memblock_find_in_range_node(size, align, min_addr,
1091 max_addr, NUMA_NO_NODE);
1092 if (alloc)
1093 goto done;
1094 }
1095
1096 if (min_addr) {
1097 min_addr = 0;
1098 goto again;
1099 } else {
1100 goto error;
1101 }
1102
1103done:
1104 memblock_reserve(alloc, size);
1105 ptr = phys_to_virt(alloc);
1106 memset(ptr, 0, size);
1107
1108 /*
1109 * The min_count is set to 0 so that bootmem allocated blocks
1110 * are never reported as leaks. This is because many of these blocks
1111 * are only referred via the physical address which is not
1112 * looked up by kmemleak.
1113 */
1114 kmemleak_alloc(ptr, size, 0, 0);
1115
1116 return ptr;
1117
1118error:
1119 return NULL;
1120}
1121
1122/**
1123 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1124 * @size: size of memory block to be allocated in bytes
1125 * @align: alignment of the region and block's size
1126 * @min_addr: the lower bound of the memory region from where the allocation
1127 * is preferred (phys address)
1128 * @max_addr: the upper bound of the memory region from where the allocation
1129 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1130 * allocate only from memory limited by memblock.current_limit value
1131 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1132 *
1133 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1134 * additional debug information (including caller info), if enabled.
1135 *
1136 * RETURNS:
1137 * Virtual address of allocated memory block on success, NULL on failure.
1138 */
1139void * __init memblock_virt_alloc_try_nid_nopanic(
1140 phys_addr_t size, phys_addr_t align,
1141 phys_addr_t min_addr, phys_addr_t max_addr,
1142 int nid)
1143{
1144 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1145 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1146 (u64)max_addr, (void *)_RET_IP_);
1147 return memblock_virt_alloc_internal(size, align, min_addr,
1148 max_addr, nid);
1149}
1150
1151/**
1152 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1153 * @size: size of memory block to be allocated in bytes
1154 * @align: alignment of the region and block's size
1155 * @min_addr: the lower bound of the memory region from where the allocation
1156 * is preferred (phys address)
1157 * @max_addr: the upper bound of the memory region from where the allocation
1158 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1159 * allocate only from memory limited by memblock.current_limit value
1160 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1161 *
1162 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1163 * which provides debug information (including caller info), if enabled,
1164 * and panics if the request can not be satisfied.
1165 *
1166 * RETURNS:
1167 * Virtual address of allocated memory block on success, NULL on failure.
1168 */
1169void * __init memblock_virt_alloc_try_nid(
1170 phys_addr_t size, phys_addr_t align,
1171 phys_addr_t min_addr, phys_addr_t max_addr,
1172 int nid)
1173{
1174 void *ptr;
1175
1176 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1177 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1178 (u64)max_addr, (void *)_RET_IP_);
1179 ptr = memblock_virt_alloc_internal(size, align,
1180 min_addr, max_addr, nid);
1181 if (ptr)
1182 return ptr;
1183
1184 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1185 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1186 (u64)max_addr);
1187 return NULL;
1188}
1189
1190/**
1191 * __memblock_free_early - free boot memory block
1192 * @base: phys starting address of the boot memory block
1193 * @size: size of the boot memory block in bytes
1194 *
1195 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1196 * The freeing memory will not be released to the buddy allocator.
1197 */
1198void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1199{
1200 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1201 __func__, (u64)base, (u64)base + size - 1,
1202 (void *)_RET_IP_);
1203 kmemleak_free_part(__va(base), size);
1204 __memblock_remove(&memblock.reserved, base, size);
1205}
1206
1207/*
1208 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1209 * @addr: phys starting address of the boot memory block
1210 * @size: size of the boot memory block in bytes
1211 *
1212 * This is only useful when the bootmem allocator has already been torn
1213 * down, but we are still initializing the system. Pages are released directly
1214 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1215 */
1216void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1217{
1218 u64 cursor, end;
1219
1220 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1221 __func__, (u64)base, (u64)base + size - 1,
1222 (void *)_RET_IP_);
1223 kmemleak_free_part(__va(base), size);
1224 cursor = PFN_UP(base);
1225 end = PFN_DOWN(base + size);
1226
1227 for (; cursor < end; cursor++) {
1228 __free_pages_bootmem(pfn_to_page(cursor), 0);
1229 totalram_pages++;
1230 }
1231}
1232
1233/*
1234 * Remaining API functions
1235 */
1236
1237phys_addr_t __init memblock_phys_mem_size(void)
1238{
1239 return memblock.memory.total_size;
1240}
1241
1242phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1243{
1244 unsigned long pages = 0;
1245 struct memblock_region *r;
1246 unsigned long start_pfn, end_pfn;
1247
1248 for_each_memblock(memory, r) {
1249 start_pfn = memblock_region_memory_base_pfn(r);
1250 end_pfn = memblock_region_memory_end_pfn(r);
1251 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1252 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1253 pages += end_pfn - start_pfn;
1254 }
1255
1256 return PFN_PHYS(pages);
1257}
1258
1259/* lowest address */
1260phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1261{
1262 return memblock.memory.regions[0].base;
1263}
1264
1265phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1266{
1267 int idx = memblock.memory.cnt - 1;
1268
1269 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1270}
1271
1272void __init memblock_enforce_memory_limit(phys_addr_t limit)
1273{
1274 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1275 struct memblock_region *r;
1276
1277 if (!limit)
1278 return;
1279
1280 /* find out max address */
1281 for_each_memblock(memory, r) {
1282 if (limit <= r->size) {
1283 max_addr = r->base + limit;
1284 break;
1285 }
1286 limit -= r->size;
1287 }
1288
1289 /* truncate both memory and reserved regions */
1290 __memblock_remove(&memblock.memory, max_addr, (phys_addr_t)ULLONG_MAX);
1291 __memblock_remove(&memblock.reserved, max_addr, (phys_addr_t)ULLONG_MAX);
1292}
1293
1294static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1295{
1296 unsigned int left = 0, right = type->cnt;
1297
1298 do {
1299 unsigned int mid = (right + left) / 2;
1300
1301 if (addr < type->regions[mid].base)
1302 right = mid;
1303 else if (addr >= (type->regions[mid].base +
1304 type->regions[mid].size))
1305 left = mid + 1;
1306 else
1307 return mid;
1308 } while (left < right);
1309 return -1;
1310}
1311
1312int __init memblock_is_reserved(phys_addr_t addr)
1313{
1314 return memblock_search(&memblock.reserved, addr) != -1;
1315}
1316
1317int __init_memblock memblock_is_memory(phys_addr_t addr)
1318{
1319 return memblock_search(&memblock.memory, addr) != -1;
1320}
1321
1322#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1323int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1324 unsigned long *start_pfn, unsigned long *end_pfn)
1325{
1326 struct memblock_type *type = &memblock.memory;
1327 int mid = memblock_search(type, PFN_PHYS(pfn));
1328
1329 if (mid == -1)
1330 return -1;
1331
1332 *start_pfn = type->regions[mid].base >> PAGE_SHIFT;
1333 *end_pfn = (type->regions[mid].base + type->regions[mid].size)
1334 >> PAGE_SHIFT;
1335
1336 return type->regions[mid].nid;
1337}
1338#endif
1339
1340/**
1341 * memblock_is_region_memory - check if a region is a subset of memory
1342 * @base: base of region to check
1343 * @size: size of region to check
1344 *
1345 * Check if the region [@base, @base+@size) is a subset of a memory block.
1346 *
1347 * RETURNS:
1348 * 0 if false, non-zero if true
1349 */
1350int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1351{
1352 int idx = memblock_search(&memblock.memory, base);
1353 phys_addr_t end = base + memblock_cap_size(base, &size);
1354
1355 if (idx == -1)
1356 return 0;
1357 return memblock.memory.regions[idx].base <= base &&
1358 (memblock.memory.regions[idx].base +
1359 memblock.memory.regions[idx].size) >= end;
1360}
1361
1362/**
1363 * memblock_is_region_reserved - check if a region intersects reserved memory
1364 * @base: base of region to check
1365 * @size: size of region to check
1366 *
1367 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1368 *
1369 * RETURNS:
1370 * 0 if false, non-zero if true
1371 */
1372int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1373{
1374 memblock_cap_size(base, &size);
1375 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
1376}
1377
1378void __init_memblock memblock_trim_memory(phys_addr_t align)
1379{
1380 phys_addr_t start, end, orig_start, orig_end;
1381 struct memblock_region *r;
1382
1383 for_each_memblock(memory, r) {
1384 orig_start = r->base;
1385 orig_end = r->base + r->size;
1386 start = round_up(orig_start, align);
1387 end = round_down(orig_end, align);
1388
1389 if (start == orig_start && end == orig_end)
1390 continue;
1391
1392 if (start < end) {
1393 r->base = start;
1394 r->size = end - start;
1395 } else {
1396 memblock_remove_region(&memblock.memory,
1397 r - memblock.memory.regions);
1398 r--;
1399 }
1400 }
1401}
1402
1403void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1404{
1405 memblock.current_limit = limit;
1406}
1407
1408phys_addr_t __init_memblock memblock_get_current_limit(void)
1409{
1410 return memblock.current_limit;
1411}
1412
1413static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1414{
1415 unsigned long long base, size;
1416 unsigned long flags;
1417 int i;
1418
1419 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
1420
1421 for (i = 0; i < type->cnt; i++) {
1422 struct memblock_region *rgn = &type->regions[i];
1423 char nid_buf[32] = "";
1424
1425 base = rgn->base;
1426 size = rgn->size;
1427 flags = rgn->flags;
1428#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1429 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1430 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1431 memblock_get_region_node(rgn));
1432#endif
1433 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1434 name, i, base, base + size - 1, size, nid_buf, flags);
1435 }
1436}
1437
1438void __init_memblock __memblock_dump_all(void)
1439{
1440 pr_info("MEMBLOCK configuration:\n");
1441 pr_info(" memory size = %#llx reserved size = %#llx\n",
1442 (unsigned long long)memblock.memory.total_size,
1443 (unsigned long long)memblock.reserved.total_size);
1444
1445 memblock_dump(&memblock.memory, "memory");
1446 memblock_dump(&memblock.reserved, "reserved");
1447}
1448
1449void __init memblock_allow_resize(void)
1450{
1451 memblock_can_resize = 1;
1452}
1453
1454static int __init early_memblock(char *p)
1455{
1456 if (p && strstr(p, "debug"))
1457 memblock_debug = 1;
1458 return 0;
1459}
1460early_param("memblock", early_memblock);
1461
1462#if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1463
1464static int memblock_debug_show(struct seq_file *m, void *private)
1465{
1466 struct memblock_type *type = m->private;
1467 struct memblock_region *reg;
1468 int i;
1469
1470 for (i = 0; i < type->cnt; i++) {
1471 reg = &type->regions[i];
1472 seq_printf(m, "%4d: ", i);
1473 if (sizeof(phys_addr_t) == 4)
1474 seq_printf(m, "0x%08lx..0x%08lx\n",
1475 (unsigned long)reg->base,
1476 (unsigned long)(reg->base + reg->size - 1));
1477 else
1478 seq_printf(m, "0x%016llx..0x%016llx\n",
1479 (unsigned long long)reg->base,
1480 (unsigned long long)(reg->base + reg->size - 1));
1481
1482 }
1483 return 0;
1484}
1485
1486static int memblock_debug_open(struct inode *inode, struct file *file)
1487{
1488 return single_open(file, memblock_debug_show, inode->i_private);
1489}
1490
1491static const struct file_operations memblock_debug_fops = {
1492 .open = memblock_debug_open,
1493 .read = seq_read,
1494 .llseek = seq_lseek,
1495 .release = single_release,
1496};
1497
1498static int __init memblock_init_debugfs(void)
1499{
1500 struct dentry *root = debugfs_create_dir("memblock", NULL);
1501 if (!root)
1502 return -ENXIO;
1503 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1504 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1505
1506 return 0;
1507}
1508__initcall(memblock_init_debugfs);
1509
1510#endif /* CONFIG_DEBUG_FS */