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// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Procedures for maintaining information about logical memory blocks.
4 *
5 * Peter Bergner, IBM Corp. June 2001.
6 * Copyright (C) 2001 Peter Bergner.
7 */
8
9#include <linux/kernel.h>
10#include <linux/slab.h>
11#include <linux/init.h>
12#include <linux/bitops.h>
13#include <linux/poison.h>
14#include <linux/pfn.h>
15#include <linux/debugfs.h>
16#include <linux/kmemleak.h>
17#include <linux/seq_file.h>
18#include <linux/memblock.h>
19
20#include <asm/sections.h>
21#include <linux/io.h>
22
23#include "internal.h"
24
25#define INIT_MEMBLOCK_REGIONS 128
26#define INIT_PHYSMEM_REGIONS 4
27
28#ifndef INIT_MEMBLOCK_RESERVED_REGIONS
29# define INIT_MEMBLOCK_RESERVED_REGIONS INIT_MEMBLOCK_REGIONS
30#endif
31
32/**
33 * DOC: memblock overview
34 *
35 * Memblock is a method of managing memory regions during the early
36 * boot period when the usual kernel memory allocators are not up and
37 * running.
38 *
39 * Memblock views the system memory as collections of contiguous
40 * regions. There are several types of these collections:
41 *
42 * * ``memory`` - describes the physical memory available to the
43 * kernel; this may differ from the actual physical memory installed
44 * in the system, for instance when the memory is restricted with
45 * ``mem=`` command line parameter
46 * * ``reserved`` - describes the regions that were allocated
47 * * ``physmem`` - describes the actual physical memory available during
48 * boot regardless of the possible restrictions and memory hot(un)plug;
49 * the ``physmem`` type is only available on some architectures.
50 *
51 * Each region is represented by :c:type:`struct memblock_region` that
52 * defines the region extents, its attributes and NUMA node id on NUMA
53 * systems. Every memory type is described by the :c:type:`struct
54 * memblock_type` which contains an array of memory regions along with
55 * the allocator metadata. The "memory" and "reserved" types are nicely
56 * wrapped with :c:type:`struct memblock`. This structure is statically
57 * initialized at build time. The region arrays are initially sized to
58 * %INIT_MEMBLOCK_REGIONS for "memory" and %INIT_MEMBLOCK_RESERVED_REGIONS
59 * for "reserved". The region array for "physmem" is initially sized to
60 * %INIT_PHYSMEM_REGIONS.
61 * The memblock_allow_resize() enables automatic resizing of the region
62 * arrays during addition of new regions. This feature should be used
63 * with care so that memory allocated for the region array will not
64 * overlap with areas that should be reserved, for example initrd.
65 *
66 * The early architecture setup should tell memblock what the physical
67 * memory layout is by using memblock_add() or memblock_add_node()
68 * functions. The first function does not assign the region to a NUMA
69 * node and it is appropriate for UMA systems. Yet, it is possible to
70 * use it on NUMA systems as well and assign the region to a NUMA node
71 * later in the setup process using memblock_set_node(). The
72 * memblock_add_node() performs such an assignment directly.
73 *
74 * Once memblock is setup the memory can be allocated using one of the
75 * API variants:
76 *
77 * * memblock_phys_alloc*() - these functions return the **physical**
78 * address of the allocated memory
79 * * memblock_alloc*() - these functions return the **virtual** address
80 * of the allocated memory.
81 *
82 * Note, that both API variants use implicit assumptions about allowed
83 * memory ranges and the fallback methods. Consult the documentation
84 * of memblock_alloc_internal() and memblock_alloc_range_nid()
85 * functions for more elaborate description.
86 *
87 * As the system boot progresses, the architecture specific mem_init()
88 * function frees all the memory to the buddy page allocator.
89 *
90 * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
91 * memblock data structures (except "physmem") will be discarded after the
92 * system initialization completes.
93 */
94
95#ifndef CONFIG_NEED_MULTIPLE_NODES
96struct pglist_data __refdata contig_page_data;
97EXPORT_SYMBOL(contig_page_data);
98#endif
99
100unsigned long max_low_pfn;
101unsigned long min_low_pfn;
102unsigned long max_pfn;
103unsigned long long max_possible_pfn;
104
105static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
106static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock;
107#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
108static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS];
109#endif
110
111struct memblock memblock __initdata_memblock = {
112 .memory.regions = memblock_memory_init_regions,
113 .memory.cnt = 1, /* empty dummy entry */
114 .memory.max = INIT_MEMBLOCK_REGIONS,
115 .memory.name = "memory",
116
117 .reserved.regions = memblock_reserved_init_regions,
118 .reserved.cnt = 1, /* empty dummy entry */
119 .reserved.max = INIT_MEMBLOCK_RESERVED_REGIONS,
120 .reserved.name = "reserved",
121
122 .bottom_up = false,
123 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
124};
125
126#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
127struct memblock_type physmem = {
128 .regions = memblock_physmem_init_regions,
129 .cnt = 1, /* empty dummy entry */
130 .max = INIT_PHYSMEM_REGIONS,
131 .name = "physmem",
132};
133#endif
134
135int memblock_debug __initdata_memblock;
136static bool system_has_some_mirror __initdata_memblock = false;
137static int memblock_can_resize __initdata_memblock;
138static int memblock_memory_in_slab __initdata_memblock = 0;
139static int memblock_reserved_in_slab __initdata_memblock = 0;
140
141static enum memblock_flags __init_memblock choose_memblock_flags(void)
142{
143 return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
144}
145
146/* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
147static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
148{
149 return *size = min(*size, PHYS_ADDR_MAX - base);
150}
151
152/*
153 * Address comparison utilities
154 */
155static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
156 phys_addr_t base2, phys_addr_t size2)
157{
158 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
159}
160
161bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
162 phys_addr_t base, phys_addr_t size)
163{
164 unsigned long i;
165
166 for (i = 0; i < type->cnt; i++)
167 if (memblock_addrs_overlap(base, size, type->regions[i].base,
168 type->regions[i].size))
169 break;
170 return i < type->cnt;
171}
172
173/**
174 * __memblock_find_range_bottom_up - find free area utility in bottom-up
175 * @start: start of candidate range
176 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
177 * %MEMBLOCK_ALLOC_ACCESSIBLE
178 * @size: size of free area to find
179 * @align: alignment of free area to find
180 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
181 * @flags: pick from blocks based on memory attributes
182 *
183 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
184 *
185 * Return:
186 * Found address on success, 0 on failure.
187 */
188static phys_addr_t __init_memblock
189__memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
190 phys_addr_t size, phys_addr_t align, int nid,
191 enum memblock_flags flags)
192{
193 phys_addr_t this_start, this_end, cand;
194 u64 i;
195
196 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
197 this_start = clamp(this_start, start, end);
198 this_end = clamp(this_end, start, end);
199
200 cand = round_up(this_start, align);
201 if (cand < this_end && this_end - cand >= size)
202 return cand;
203 }
204
205 return 0;
206}
207
208/**
209 * __memblock_find_range_top_down - find free area utility, in top-down
210 * @start: start of candidate range
211 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
212 * %MEMBLOCK_ALLOC_ACCESSIBLE
213 * @size: size of free area to find
214 * @align: alignment of free area to find
215 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
216 * @flags: pick from blocks based on memory attributes
217 *
218 * Utility called from memblock_find_in_range_node(), find free area top-down.
219 *
220 * Return:
221 * Found address on success, 0 on failure.
222 */
223static phys_addr_t __init_memblock
224__memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
225 phys_addr_t size, phys_addr_t align, int nid,
226 enum memblock_flags flags)
227{
228 phys_addr_t this_start, this_end, cand;
229 u64 i;
230
231 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
232 NULL) {
233 this_start = clamp(this_start, start, end);
234 this_end = clamp(this_end, start, end);
235
236 if (this_end < size)
237 continue;
238
239 cand = round_down(this_end - size, align);
240 if (cand >= this_start)
241 return cand;
242 }
243
244 return 0;
245}
246
247/**
248 * memblock_find_in_range_node - find free area in given range and node
249 * @size: size of free area to find
250 * @align: alignment of free area to find
251 * @start: start of candidate range
252 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
253 * %MEMBLOCK_ALLOC_ACCESSIBLE
254 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
255 * @flags: pick from blocks based on memory attributes
256 *
257 * Find @size free area aligned to @align in the specified range and node.
258 *
259 * When allocation direction is bottom-up, the @start should be greater
260 * than the end of the kernel image. Otherwise, it will be trimmed. The
261 * reason is that we want the bottom-up allocation just near the kernel
262 * image so it is highly likely that the allocated memory and the kernel
263 * will reside in the same node.
264 *
265 * If bottom-up allocation failed, will try to allocate memory top-down.
266 *
267 * Return:
268 * Found address on success, 0 on failure.
269 */
270static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
271 phys_addr_t align, phys_addr_t start,
272 phys_addr_t end, int nid,
273 enum memblock_flags flags)
274{
275 phys_addr_t kernel_end, ret;
276
277 /* pump up @end */
278 if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
279 end == MEMBLOCK_ALLOC_KASAN)
280 end = memblock.current_limit;
281
282 /* avoid allocating the first page */
283 start = max_t(phys_addr_t, start, PAGE_SIZE);
284 end = max(start, end);
285 kernel_end = __pa_symbol(_end);
286
287 /*
288 * try bottom-up allocation only when bottom-up mode
289 * is set and @end is above the kernel image.
290 */
291 if (memblock_bottom_up() && end > kernel_end) {
292 phys_addr_t bottom_up_start;
293
294 /* make sure we will allocate above the kernel */
295 bottom_up_start = max(start, kernel_end);
296
297 /* ok, try bottom-up allocation first */
298 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
299 size, align, nid, flags);
300 if (ret)
301 return ret;
302
303 /*
304 * we always limit bottom-up allocation above the kernel,
305 * but top-down allocation doesn't have the limit, so
306 * retrying top-down allocation may succeed when bottom-up
307 * allocation failed.
308 *
309 * bottom-up allocation is expected to be fail very rarely,
310 * so we use WARN_ONCE() here to see the stack trace if
311 * fail happens.
312 */
313 WARN_ONCE(IS_ENABLED(CONFIG_MEMORY_HOTREMOVE),
314 "memblock: bottom-up allocation failed, memory hotremove may be affected\n");
315 }
316
317 return __memblock_find_range_top_down(start, end, size, align, nid,
318 flags);
319}
320
321/**
322 * memblock_find_in_range - find free area in given range
323 * @start: start of candidate range
324 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
325 * %MEMBLOCK_ALLOC_ACCESSIBLE
326 * @size: size of free area to find
327 * @align: alignment of free area to find
328 *
329 * Find @size free area aligned to @align in the specified range.
330 *
331 * Return:
332 * Found address on success, 0 on failure.
333 */
334phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
335 phys_addr_t end, phys_addr_t size,
336 phys_addr_t align)
337{
338 phys_addr_t ret;
339 enum memblock_flags flags = choose_memblock_flags();
340
341again:
342 ret = memblock_find_in_range_node(size, align, start, end,
343 NUMA_NO_NODE, flags);
344
345 if (!ret && (flags & MEMBLOCK_MIRROR)) {
346 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
347 &size);
348 flags &= ~MEMBLOCK_MIRROR;
349 goto again;
350 }
351
352 return ret;
353}
354
355static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
356{
357 type->total_size -= type->regions[r].size;
358 memmove(&type->regions[r], &type->regions[r + 1],
359 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
360 type->cnt--;
361
362 /* Special case for empty arrays */
363 if (type->cnt == 0) {
364 WARN_ON(type->total_size != 0);
365 type->cnt = 1;
366 type->regions[0].base = 0;
367 type->regions[0].size = 0;
368 type->regions[0].flags = 0;
369 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
370 }
371}
372
373#ifndef CONFIG_ARCH_KEEP_MEMBLOCK
374/**
375 * memblock_discard - discard memory and reserved arrays if they were allocated
376 */
377void __init memblock_discard(void)
378{
379 phys_addr_t addr, size;
380
381 if (memblock.reserved.regions != memblock_reserved_init_regions) {
382 addr = __pa(memblock.reserved.regions);
383 size = PAGE_ALIGN(sizeof(struct memblock_region) *
384 memblock.reserved.max);
385 __memblock_free_late(addr, size);
386 }
387
388 if (memblock.memory.regions != memblock_memory_init_regions) {
389 addr = __pa(memblock.memory.regions);
390 size = PAGE_ALIGN(sizeof(struct memblock_region) *
391 memblock.memory.max);
392 __memblock_free_late(addr, size);
393 }
394}
395#endif
396
397/**
398 * memblock_double_array - double the size of the memblock regions array
399 * @type: memblock type of the regions array being doubled
400 * @new_area_start: starting address of memory range to avoid overlap with
401 * @new_area_size: size of memory range to avoid overlap with
402 *
403 * Double the size of the @type regions array. If memblock is being used to
404 * allocate memory for a new reserved regions array and there is a previously
405 * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
406 * waiting to be reserved, ensure the memory used by the new array does
407 * not overlap.
408 *
409 * Return:
410 * 0 on success, -1 on failure.
411 */
412static int __init_memblock memblock_double_array(struct memblock_type *type,
413 phys_addr_t new_area_start,
414 phys_addr_t new_area_size)
415{
416 struct memblock_region *new_array, *old_array;
417 phys_addr_t old_alloc_size, new_alloc_size;
418 phys_addr_t old_size, new_size, addr, new_end;
419 int use_slab = slab_is_available();
420 int *in_slab;
421
422 /* We don't allow resizing until we know about the reserved regions
423 * of memory that aren't suitable for allocation
424 */
425 if (!memblock_can_resize)
426 return -1;
427
428 /* Calculate new doubled size */
429 old_size = type->max * sizeof(struct memblock_region);
430 new_size = old_size << 1;
431 /*
432 * We need to allocated new one align to PAGE_SIZE,
433 * so we can free them completely later.
434 */
435 old_alloc_size = PAGE_ALIGN(old_size);
436 new_alloc_size = PAGE_ALIGN(new_size);
437
438 /* Retrieve the slab flag */
439 if (type == &memblock.memory)
440 in_slab = &memblock_memory_in_slab;
441 else
442 in_slab = &memblock_reserved_in_slab;
443
444 /* Try to find some space for it */
445 if (use_slab) {
446 new_array = kmalloc(new_size, GFP_KERNEL);
447 addr = new_array ? __pa(new_array) : 0;
448 } else {
449 /* only exclude range when trying to double reserved.regions */
450 if (type != &memblock.reserved)
451 new_area_start = new_area_size = 0;
452
453 addr = memblock_find_in_range(new_area_start + new_area_size,
454 memblock.current_limit,
455 new_alloc_size, PAGE_SIZE);
456 if (!addr && new_area_size)
457 addr = memblock_find_in_range(0,
458 min(new_area_start, memblock.current_limit),
459 new_alloc_size, PAGE_SIZE);
460
461 new_array = addr ? __va(addr) : NULL;
462 }
463 if (!addr) {
464 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
465 type->name, type->max, type->max * 2);
466 return -1;
467 }
468
469 new_end = addr + new_size - 1;
470 memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
471 type->name, type->max * 2, &addr, &new_end);
472
473 /*
474 * Found space, we now need to move the array over before we add the
475 * reserved region since it may be our reserved array itself that is
476 * full.
477 */
478 memcpy(new_array, type->regions, old_size);
479 memset(new_array + type->max, 0, old_size);
480 old_array = type->regions;
481 type->regions = new_array;
482 type->max <<= 1;
483
484 /* Free old array. We needn't free it if the array is the static one */
485 if (*in_slab)
486 kfree(old_array);
487 else if (old_array != memblock_memory_init_regions &&
488 old_array != memblock_reserved_init_regions)
489 memblock_free(__pa(old_array), old_alloc_size);
490
491 /*
492 * Reserve the new array if that comes from the memblock. Otherwise, we
493 * needn't do it
494 */
495 if (!use_slab)
496 BUG_ON(memblock_reserve(addr, new_alloc_size));
497
498 /* Update slab flag */
499 *in_slab = use_slab;
500
501 return 0;
502}
503
504/**
505 * memblock_merge_regions - merge neighboring compatible regions
506 * @type: memblock type to scan
507 *
508 * Scan @type and merge neighboring compatible regions.
509 */
510static void __init_memblock memblock_merge_regions(struct memblock_type *type)
511{
512 int i = 0;
513
514 /* cnt never goes below 1 */
515 while (i < type->cnt - 1) {
516 struct memblock_region *this = &type->regions[i];
517 struct memblock_region *next = &type->regions[i + 1];
518
519 if (this->base + this->size != next->base ||
520 memblock_get_region_node(this) !=
521 memblock_get_region_node(next) ||
522 this->flags != next->flags) {
523 BUG_ON(this->base + this->size > next->base);
524 i++;
525 continue;
526 }
527
528 this->size += next->size;
529 /* move forward from next + 1, index of which is i + 2 */
530 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
531 type->cnt--;
532 }
533}
534
535/**
536 * memblock_insert_region - insert new memblock region
537 * @type: memblock type to insert into
538 * @idx: index for the insertion point
539 * @base: base address of the new region
540 * @size: size of the new region
541 * @nid: node id of the new region
542 * @flags: flags of the new region
543 *
544 * Insert new memblock region [@base, @base + @size) into @type at @idx.
545 * @type must already have extra room to accommodate the new region.
546 */
547static void __init_memblock memblock_insert_region(struct memblock_type *type,
548 int idx, phys_addr_t base,
549 phys_addr_t size,
550 int nid,
551 enum memblock_flags flags)
552{
553 struct memblock_region *rgn = &type->regions[idx];
554
555 BUG_ON(type->cnt >= type->max);
556 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
557 rgn->base = base;
558 rgn->size = size;
559 rgn->flags = flags;
560 memblock_set_region_node(rgn, nid);
561 type->cnt++;
562 type->total_size += size;
563}
564
565/**
566 * memblock_add_range - add new memblock region
567 * @type: memblock type to add new region into
568 * @base: base address of the new region
569 * @size: size of the new region
570 * @nid: nid of the new region
571 * @flags: flags of the new region
572 *
573 * Add new memblock region [@base, @base + @size) into @type. The new region
574 * is allowed to overlap with existing ones - overlaps don't affect already
575 * existing regions. @type is guaranteed to be minimal (all neighbouring
576 * compatible regions are merged) after the addition.
577 *
578 * Return:
579 * 0 on success, -errno on failure.
580 */
581static int __init_memblock memblock_add_range(struct memblock_type *type,
582 phys_addr_t base, phys_addr_t size,
583 int nid, enum memblock_flags flags)
584{
585 bool insert = false;
586 phys_addr_t obase = base;
587 phys_addr_t end = base + memblock_cap_size(base, &size);
588 int idx, nr_new;
589 struct memblock_region *rgn;
590
591 if (!size)
592 return 0;
593
594 /* special case for empty array */
595 if (type->regions[0].size == 0) {
596 WARN_ON(type->cnt != 1 || type->total_size);
597 type->regions[0].base = base;
598 type->regions[0].size = size;
599 type->regions[0].flags = flags;
600 memblock_set_region_node(&type->regions[0], nid);
601 type->total_size = size;
602 return 0;
603 }
604repeat:
605 /*
606 * The following is executed twice. Once with %false @insert and
607 * then with %true. The first counts the number of regions needed
608 * to accommodate the new area. The second actually inserts them.
609 */
610 base = obase;
611 nr_new = 0;
612
613 for_each_memblock_type(idx, type, rgn) {
614 phys_addr_t rbase = rgn->base;
615 phys_addr_t rend = rbase + rgn->size;
616
617 if (rbase >= end)
618 break;
619 if (rend <= base)
620 continue;
621 /*
622 * @rgn overlaps. If it separates the lower part of new
623 * area, insert that portion.
624 */
625 if (rbase > base) {
626#ifdef CONFIG_NEED_MULTIPLE_NODES
627 WARN_ON(nid != memblock_get_region_node(rgn));
628#endif
629 WARN_ON(flags != rgn->flags);
630 nr_new++;
631 if (insert)
632 memblock_insert_region(type, idx++, base,
633 rbase - base, nid,
634 flags);
635 }
636 /* area below @rend is dealt with, forget about it */
637 base = min(rend, end);
638 }
639
640 /* insert the remaining portion */
641 if (base < end) {
642 nr_new++;
643 if (insert)
644 memblock_insert_region(type, idx, base, end - base,
645 nid, flags);
646 }
647
648 if (!nr_new)
649 return 0;
650
651 /*
652 * If this was the first round, resize array and repeat for actual
653 * insertions; otherwise, merge and return.
654 */
655 if (!insert) {
656 while (type->cnt + nr_new > type->max)
657 if (memblock_double_array(type, obase, size) < 0)
658 return -ENOMEM;
659 insert = true;
660 goto repeat;
661 } else {
662 memblock_merge_regions(type);
663 return 0;
664 }
665}
666
667/**
668 * memblock_add_node - add new memblock region within a NUMA node
669 * @base: base address of the new region
670 * @size: size of the new region
671 * @nid: nid of the new region
672 *
673 * Add new memblock region [@base, @base + @size) to the "memory"
674 * type. See memblock_add_range() description for mode details
675 *
676 * Return:
677 * 0 on success, -errno on failure.
678 */
679int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
680 int nid)
681{
682 return memblock_add_range(&memblock.memory, base, size, nid, 0);
683}
684
685/**
686 * memblock_add - add new memblock region
687 * @base: base address of the new region
688 * @size: size of the new region
689 *
690 * Add new memblock region [@base, @base + @size) to the "memory"
691 * type. See memblock_add_range() description for mode details
692 *
693 * Return:
694 * 0 on success, -errno on failure.
695 */
696int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
697{
698 phys_addr_t end = base + size - 1;
699
700 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
701 &base, &end, (void *)_RET_IP_);
702
703 return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
704}
705
706/**
707 * memblock_isolate_range - isolate given range into disjoint memblocks
708 * @type: memblock type to isolate range for
709 * @base: base of range to isolate
710 * @size: size of range to isolate
711 * @start_rgn: out parameter for the start of isolated region
712 * @end_rgn: out parameter for the end of isolated region
713 *
714 * Walk @type and ensure that regions don't cross the boundaries defined by
715 * [@base, @base + @size). Crossing regions are split at the boundaries,
716 * which may create at most two more regions. The index of the first
717 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
718 *
719 * Return:
720 * 0 on success, -errno on failure.
721 */
722static int __init_memblock memblock_isolate_range(struct memblock_type *type,
723 phys_addr_t base, phys_addr_t size,
724 int *start_rgn, int *end_rgn)
725{
726 phys_addr_t end = base + memblock_cap_size(base, &size);
727 int idx;
728 struct memblock_region *rgn;
729
730 *start_rgn = *end_rgn = 0;
731
732 if (!size)
733 return 0;
734
735 /* we'll create at most two more regions */
736 while (type->cnt + 2 > type->max)
737 if (memblock_double_array(type, base, size) < 0)
738 return -ENOMEM;
739
740 for_each_memblock_type(idx, type, rgn) {
741 phys_addr_t rbase = rgn->base;
742 phys_addr_t rend = rbase + rgn->size;
743
744 if (rbase >= end)
745 break;
746 if (rend <= base)
747 continue;
748
749 if (rbase < base) {
750 /*
751 * @rgn intersects from below. Split and continue
752 * to process the next region - the new top half.
753 */
754 rgn->base = base;
755 rgn->size -= base - rbase;
756 type->total_size -= base - rbase;
757 memblock_insert_region(type, idx, rbase, base - rbase,
758 memblock_get_region_node(rgn),
759 rgn->flags);
760 } else if (rend > end) {
761 /*
762 * @rgn intersects from above. Split and redo the
763 * current region - the new bottom half.
764 */
765 rgn->base = end;
766 rgn->size -= end - rbase;
767 type->total_size -= end - rbase;
768 memblock_insert_region(type, idx--, rbase, end - rbase,
769 memblock_get_region_node(rgn),
770 rgn->flags);
771 } else {
772 /* @rgn is fully contained, record it */
773 if (!*end_rgn)
774 *start_rgn = idx;
775 *end_rgn = idx + 1;
776 }
777 }
778
779 return 0;
780}
781
782static int __init_memblock memblock_remove_range(struct memblock_type *type,
783 phys_addr_t base, phys_addr_t size)
784{
785 int start_rgn, end_rgn;
786 int i, ret;
787
788 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
789 if (ret)
790 return ret;
791
792 for (i = end_rgn - 1; i >= start_rgn; i--)
793 memblock_remove_region(type, i);
794 return 0;
795}
796
797int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
798{
799 phys_addr_t end = base + size - 1;
800
801 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
802 &base, &end, (void *)_RET_IP_);
803
804 return memblock_remove_range(&memblock.memory, base, size);
805}
806
807/**
808 * memblock_free - free boot memory block
809 * @base: phys starting address of the boot memory block
810 * @size: size of the boot memory block in bytes
811 *
812 * Free boot memory block previously allocated by memblock_alloc_xx() API.
813 * The freeing memory will not be released to the buddy allocator.
814 */
815int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
816{
817 phys_addr_t end = base + size - 1;
818
819 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
820 &base, &end, (void *)_RET_IP_);
821
822 kmemleak_free_part_phys(base, size);
823 return memblock_remove_range(&memblock.reserved, base, size);
824}
825
826int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
827{
828 phys_addr_t end = base + size - 1;
829
830 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
831 &base, &end, (void *)_RET_IP_);
832
833 return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
834}
835
836#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
837int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
838{
839 phys_addr_t end = base + size - 1;
840
841 memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
842 &base, &end, (void *)_RET_IP_);
843
844 return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
845}
846#endif
847
848/**
849 * memblock_setclr_flag - set or clear flag for a memory region
850 * @base: base address of the region
851 * @size: size of the region
852 * @set: set or clear the flag
853 * @flag: the flag to udpate
854 *
855 * This function isolates region [@base, @base + @size), and sets/clears flag
856 *
857 * Return: 0 on success, -errno on failure.
858 */
859static int __init_memblock memblock_setclr_flag(phys_addr_t base,
860 phys_addr_t size, int set, int flag)
861{
862 struct memblock_type *type = &memblock.memory;
863 int i, ret, start_rgn, end_rgn;
864
865 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
866 if (ret)
867 return ret;
868
869 for (i = start_rgn; i < end_rgn; i++) {
870 struct memblock_region *r = &type->regions[i];
871
872 if (set)
873 r->flags |= flag;
874 else
875 r->flags &= ~flag;
876 }
877
878 memblock_merge_regions(type);
879 return 0;
880}
881
882/**
883 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
884 * @base: the base phys addr of the region
885 * @size: the size of the region
886 *
887 * Return: 0 on success, -errno on failure.
888 */
889int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
890{
891 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
892}
893
894/**
895 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
896 * @base: the base phys addr of the region
897 * @size: the size of the region
898 *
899 * Return: 0 on success, -errno on failure.
900 */
901int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
902{
903 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
904}
905
906/**
907 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
908 * @base: the base phys addr of the region
909 * @size: the size of the region
910 *
911 * Return: 0 on success, -errno on failure.
912 */
913int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
914{
915 system_has_some_mirror = true;
916
917 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
918}
919
920/**
921 * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
922 * @base: the base phys addr of the region
923 * @size: the size of the region
924 *
925 * Return: 0 on success, -errno on failure.
926 */
927int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
928{
929 return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
930}
931
932/**
933 * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
934 * @base: the base phys addr of the region
935 * @size: the size of the region
936 *
937 * Return: 0 on success, -errno on failure.
938 */
939int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
940{
941 return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
942}
943
944/**
945 * __next_reserved_mem_region - next function for for_each_reserved_region()
946 * @idx: pointer to u64 loop variable
947 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
948 * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
949 *
950 * Iterate over all reserved memory regions.
951 */
952void __init_memblock __next_reserved_mem_region(u64 *idx,
953 phys_addr_t *out_start,
954 phys_addr_t *out_end)
955{
956 struct memblock_type *type = &memblock.reserved;
957
958 if (*idx < type->cnt) {
959 struct memblock_region *r = &type->regions[*idx];
960 phys_addr_t base = r->base;
961 phys_addr_t size = r->size;
962
963 if (out_start)
964 *out_start = base;
965 if (out_end)
966 *out_end = base + size - 1;
967
968 *idx += 1;
969 return;
970 }
971
972 /* signal end of iteration */
973 *idx = ULLONG_MAX;
974}
975
976static bool should_skip_region(struct memblock_region *m, int nid, int flags)
977{
978 int m_nid = memblock_get_region_node(m);
979
980 /* only memory regions are associated with nodes, check it */
981 if (nid != NUMA_NO_NODE && nid != m_nid)
982 return true;
983
984 /* skip hotpluggable memory regions if needed */
985 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
986 return true;
987
988 /* if we want mirror memory skip non-mirror memory regions */
989 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
990 return true;
991
992 /* skip nomap memory unless we were asked for it explicitly */
993 if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
994 return true;
995
996 return false;
997}
998
999/**
1000 * __next_mem_range - next function for for_each_free_mem_range() etc.
1001 * @idx: pointer to u64 loop variable
1002 * @nid: node selector, %NUMA_NO_NODE for all nodes
1003 * @flags: pick from blocks based on memory attributes
1004 * @type_a: pointer to memblock_type from where the range is taken
1005 * @type_b: pointer to memblock_type which excludes memory from being taken
1006 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1007 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1008 * @out_nid: ptr to int for nid of the range, can be %NULL
1009 *
1010 * Find the first area from *@idx which matches @nid, fill the out
1011 * parameters, and update *@idx for the next iteration. The lower 32bit of
1012 * *@idx contains index into type_a and the upper 32bit indexes the
1013 * areas before each region in type_b. For example, if type_b regions
1014 * look like the following,
1015 *
1016 * 0:[0-16), 1:[32-48), 2:[128-130)
1017 *
1018 * The upper 32bit indexes the following regions.
1019 *
1020 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1021 *
1022 * As both region arrays are sorted, the function advances the two indices
1023 * in lockstep and returns each intersection.
1024 */
1025void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
1026 struct memblock_type *type_a,
1027 struct memblock_type *type_b, phys_addr_t *out_start,
1028 phys_addr_t *out_end, int *out_nid)
1029{
1030 int idx_a = *idx & 0xffffffff;
1031 int idx_b = *idx >> 32;
1032
1033 if (WARN_ONCE(nid == MAX_NUMNODES,
1034 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1035 nid = NUMA_NO_NODE;
1036
1037 for (; idx_a < type_a->cnt; idx_a++) {
1038 struct memblock_region *m = &type_a->regions[idx_a];
1039
1040 phys_addr_t m_start = m->base;
1041 phys_addr_t m_end = m->base + m->size;
1042 int m_nid = memblock_get_region_node(m);
1043
1044 if (should_skip_region(m, nid, flags))
1045 continue;
1046
1047 if (!type_b) {
1048 if (out_start)
1049 *out_start = m_start;
1050 if (out_end)
1051 *out_end = m_end;
1052 if (out_nid)
1053 *out_nid = m_nid;
1054 idx_a++;
1055 *idx = (u32)idx_a | (u64)idx_b << 32;
1056 return;
1057 }
1058
1059 /* scan areas before each reservation */
1060 for (; idx_b < type_b->cnt + 1; idx_b++) {
1061 struct memblock_region *r;
1062 phys_addr_t r_start;
1063 phys_addr_t r_end;
1064
1065 r = &type_b->regions[idx_b];
1066 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1067 r_end = idx_b < type_b->cnt ?
1068 r->base : PHYS_ADDR_MAX;
1069
1070 /*
1071 * if idx_b advanced past idx_a,
1072 * break out to advance idx_a
1073 */
1074 if (r_start >= m_end)
1075 break;
1076 /* if the two regions intersect, we're done */
1077 if (m_start < r_end) {
1078 if (out_start)
1079 *out_start =
1080 max(m_start, r_start);
1081 if (out_end)
1082 *out_end = min(m_end, r_end);
1083 if (out_nid)
1084 *out_nid = m_nid;
1085 /*
1086 * The region which ends first is
1087 * advanced for the next iteration.
1088 */
1089 if (m_end <= r_end)
1090 idx_a++;
1091 else
1092 idx_b++;
1093 *idx = (u32)idx_a | (u64)idx_b << 32;
1094 return;
1095 }
1096 }
1097 }
1098
1099 /* signal end of iteration */
1100 *idx = ULLONG_MAX;
1101}
1102
1103/**
1104 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1105 *
1106 * @idx: pointer to u64 loop variable
1107 * @nid: node selector, %NUMA_NO_NODE for all nodes
1108 * @flags: pick from blocks based on memory attributes
1109 * @type_a: pointer to memblock_type from where the range is taken
1110 * @type_b: pointer to memblock_type which excludes memory from being taken
1111 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1112 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1113 * @out_nid: ptr to int for nid of the range, can be %NULL
1114 *
1115 * Finds the next range from type_a which is not marked as unsuitable
1116 * in type_b.
1117 *
1118 * Reverse of __next_mem_range().
1119 */
1120void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1121 enum memblock_flags flags,
1122 struct memblock_type *type_a,
1123 struct memblock_type *type_b,
1124 phys_addr_t *out_start,
1125 phys_addr_t *out_end, int *out_nid)
1126{
1127 int idx_a = *idx & 0xffffffff;
1128 int idx_b = *idx >> 32;
1129
1130 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1131 nid = NUMA_NO_NODE;
1132
1133 if (*idx == (u64)ULLONG_MAX) {
1134 idx_a = type_a->cnt - 1;
1135 if (type_b != NULL)
1136 idx_b = type_b->cnt;
1137 else
1138 idx_b = 0;
1139 }
1140
1141 for (; idx_a >= 0; idx_a--) {
1142 struct memblock_region *m = &type_a->regions[idx_a];
1143
1144 phys_addr_t m_start = m->base;
1145 phys_addr_t m_end = m->base + m->size;
1146 int m_nid = memblock_get_region_node(m);
1147
1148 if (should_skip_region(m, nid, flags))
1149 continue;
1150
1151 if (!type_b) {
1152 if (out_start)
1153 *out_start = m_start;
1154 if (out_end)
1155 *out_end = m_end;
1156 if (out_nid)
1157 *out_nid = m_nid;
1158 idx_a--;
1159 *idx = (u32)idx_a | (u64)idx_b << 32;
1160 return;
1161 }
1162
1163 /* scan areas before each reservation */
1164 for (; idx_b >= 0; idx_b--) {
1165 struct memblock_region *r;
1166 phys_addr_t r_start;
1167 phys_addr_t r_end;
1168
1169 r = &type_b->regions[idx_b];
1170 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1171 r_end = idx_b < type_b->cnt ?
1172 r->base : PHYS_ADDR_MAX;
1173 /*
1174 * if idx_b advanced past idx_a,
1175 * break out to advance idx_a
1176 */
1177
1178 if (r_end <= m_start)
1179 break;
1180 /* if the two regions intersect, we're done */
1181 if (m_end > r_start) {
1182 if (out_start)
1183 *out_start = max(m_start, r_start);
1184 if (out_end)
1185 *out_end = min(m_end, r_end);
1186 if (out_nid)
1187 *out_nid = m_nid;
1188 if (m_start >= r_start)
1189 idx_a--;
1190 else
1191 idx_b--;
1192 *idx = (u32)idx_a | (u64)idx_b << 32;
1193 return;
1194 }
1195 }
1196 }
1197 /* signal end of iteration */
1198 *idx = ULLONG_MAX;
1199}
1200
1201/*
1202 * Common iterator interface used to define for_each_mem_pfn_range().
1203 */
1204void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1205 unsigned long *out_start_pfn,
1206 unsigned long *out_end_pfn, int *out_nid)
1207{
1208 struct memblock_type *type = &memblock.memory;
1209 struct memblock_region *r;
1210 int r_nid;
1211
1212 while (++*idx < type->cnt) {
1213 r = &type->regions[*idx];
1214 r_nid = memblock_get_region_node(r);
1215
1216 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1217 continue;
1218 if (nid == MAX_NUMNODES || nid == r_nid)
1219 break;
1220 }
1221 if (*idx >= type->cnt) {
1222 *idx = -1;
1223 return;
1224 }
1225
1226 if (out_start_pfn)
1227 *out_start_pfn = PFN_UP(r->base);
1228 if (out_end_pfn)
1229 *out_end_pfn = PFN_DOWN(r->base + r->size);
1230 if (out_nid)
1231 *out_nid = r_nid;
1232}
1233
1234/**
1235 * memblock_set_node - set node ID on memblock regions
1236 * @base: base of area to set node ID for
1237 * @size: size of area to set node ID for
1238 * @type: memblock type to set node ID for
1239 * @nid: node ID to set
1240 *
1241 * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1242 * Regions which cross the area boundaries are split as necessary.
1243 *
1244 * Return:
1245 * 0 on success, -errno on failure.
1246 */
1247int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1248 struct memblock_type *type, int nid)
1249{
1250#ifdef CONFIG_NEED_MULTIPLE_NODES
1251 int start_rgn, end_rgn;
1252 int i, ret;
1253
1254 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1255 if (ret)
1256 return ret;
1257
1258 for (i = start_rgn; i < end_rgn; i++)
1259 memblock_set_region_node(&type->regions[i], nid);
1260
1261 memblock_merge_regions(type);
1262#endif
1263 return 0;
1264}
1265
1266#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1267/**
1268 * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1269 *
1270 * @idx: pointer to u64 loop variable
1271 * @zone: zone in which all of the memory blocks reside
1272 * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1273 * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1274 *
1275 * This function is meant to be a zone/pfn specific wrapper for the
1276 * for_each_mem_range type iterators. Specifically they are used in the
1277 * deferred memory init routines and as such we were duplicating much of
1278 * this logic throughout the code. So instead of having it in multiple
1279 * locations it seemed like it would make more sense to centralize this to
1280 * one new iterator that does everything they need.
1281 */
1282void __init_memblock
1283__next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1284 unsigned long *out_spfn, unsigned long *out_epfn)
1285{
1286 int zone_nid = zone_to_nid(zone);
1287 phys_addr_t spa, epa;
1288 int nid;
1289
1290 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1291 &memblock.memory, &memblock.reserved,
1292 &spa, &epa, &nid);
1293
1294 while (*idx != U64_MAX) {
1295 unsigned long epfn = PFN_DOWN(epa);
1296 unsigned long spfn = PFN_UP(spa);
1297
1298 /*
1299 * Verify the end is at least past the start of the zone and
1300 * that we have at least one PFN to initialize.
1301 */
1302 if (zone->zone_start_pfn < epfn && spfn < epfn) {
1303 /* if we went too far just stop searching */
1304 if (zone_end_pfn(zone) <= spfn) {
1305 *idx = U64_MAX;
1306 break;
1307 }
1308
1309 if (out_spfn)
1310 *out_spfn = max(zone->zone_start_pfn, spfn);
1311 if (out_epfn)
1312 *out_epfn = min(zone_end_pfn(zone), epfn);
1313
1314 return;
1315 }
1316
1317 __next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1318 &memblock.memory, &memblock.reserved,
1319 &spa, &epa, &nid);
1320 }
1321
1322 /* signal end of iteration */
1323 if (out_spfn)
1324 *out_spfn = ULONG_MAX;
1325 if (out_epfn)
1326 *out_epfn = 0;
1327}
1328
1329#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1330
1331/**
1332 * memblock_alloc_range_nid - 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 * @start: the lower bound of the memory region to allocate (phys address)
1336 * @end: the upper bound of the memory region to allocate (phys address)
1337 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1338 * @exact_nid: control the allocation fall back to other nodes
1339 *
1340 * The allocation is performed from memory region limited by
1341 * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1342 *
1343 * If the specified node can not hold the requested memory and @exact_nid
1344 * is false, the allocation falls back to any node in the system.
1345 *
1346 * For systems with memory mirroring, the allocation is attempted first
1347 * from the regions with mirroring enabled and then retried from any
1348 * memory region.
1349 *
1350 * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for
1351 * allocated boot memory block, so that it is never reported as leaks.
1352 *
1353 * Return:
1354 * Physical address of allocated memory block on success, %0 on failure.
1355 */
1356phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1357 phys_addr_t align, phys_addr_t start,
1358 phys_addr_t end, int nid,
1359 bool exact_nid)
1360{
1361 enum memblock_flags flags = choose_memblock_flags();
1362 phys_addr_t found;
1363
1364 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1365 nid = NUMA_NO_NODE;
1366
1367 if (!align) {
1368 /* Can't use WARNs this early in boot on powerpc */
1369 dump_stack();
1370 align = SMP_CACHE_BYTES;
1371 }
1372
1373again:
1374 found = memblock_find_in_range_node(size, align, start, end, nid,
1375 flags);
1376 if (found && !memblock_reserve(found, size))
1377 goto done;
1378
1379 if (nid != NUMA_NO_NODE && !exact_nid) {
1380 found = memblock_find_in_range_node(size, align, start,
1381 end, NUMA_NO_NODE,
1382 flags);
1383 if (found && !memblock_reserve(found, size))
1384 goto done;
1385 }
1386
1387 if (flags & MEMBLOCK_MIRROR) {
1388 flags &= ~MEMBLOCK_MIRROR;
1389 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1390 &size);
1391 goto again;
1392 }
1393
1394 return 0;
1395
1396done:
1397 /* Skip kmemleak for kasan_init() due to high volume. */
1398 if (end != MEMBLOCK_ALLOC_KASAN)
1399 /*
1400 * The min_count is set to 0 so that memblock allocated
1401 * blocks are never reported as leaks. This is because many
1402 * of these blocks are only referred via the physical
1403 * address which is not looked up by kmemleak.
1404 */
1405 kmemleak_alloc_phys(found, size, 0, 0);
1406
1407 return found;
1408}
1409
1410/**
1411 * memblock_phys_alloc_range - allocate a memory block inside specified range
1412 * @size: size of memory block to be allocated in bytes
1413 * @align: alignment of the region and block's size
1414 * @start: the lower bound of the memory region to allocate (physical address)
1415 * @end: the upper bound of the memory region to allocate (physical address)
1416 *
1417 * Allocate @size bytes in the between @start and @end.
1418 *
1419 * Return: physical address of the allocated memory block on success,
1420 * %0 on failure.
1421 */
1422phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1423 phys_addr_t align,
1424 phys_addr_t start,
1425 phys_addr_t end)
1426{
1427 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1428 false);
1429}
1430
1431/**
1432 * memblock_phys_alloc_try_nid - allocate a memory block from specified MUMA node
1433 * @size: size of memory block to be allocated in bytes
1434 * @align: alignment of the region and block's size
1435 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1436 *
1437 * Allocates memory block from the specified NUMA node. If the node
1438 * has no available memory, attempts to allocated from any node in the
1439 * system.
1440 *
1441 * Return: physical address of the allocated memory block on success,
1442 * %0 on failure.
1443 */
1444phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1445{
1446 return memblock_alloc_range_nid(size, align, 0,
1447 MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
1448}
1449
1450/**
1451 * memblock_alloc_internal - allocate boot memory block
1452 * @size: size of memory block to be allocated in bytes
1453 * @align: alignment of the region and block's size
1454 * @min_addr: the lower bound of the memory region to allocate (phys address)
1455 * @max_addr: the upper bound of the memory region to allocate (phys address)
1456 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1457 * @exact_nid: control the allocation fall back to other nodes
1458 *
1459 * Allocates memory block using memblock_alloc_range_nid() and
1460 * converts the returned physical address to virtual.
1461 *
1462 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1463 * will fall back to memory below @min_addr. Other constraints, such
1464 * as node and mirrored memory will be handled again in
1465 * memblock_alloc_range_nid().
1466 *
1467 * Return:
1468 * Virtual address of allocated memory block on success, NULL on failure.
1469 */
1470static void * __init memblock_alloc_internal(
1471 phys_addr_t size, phys_addr_t align,
1472 phys_addr_t min_addr, phys_addr_t max_addr,
1473 int nid, bool exact_nid)
1474{
1475 phys_addr_t alloc;
1476
1477 /*
1478 * Detect any accidental use of these APIs after slab is ready, as at
1479 * this moment memblock may be deinitialized already and its
1480 * internal data may be destroyed (after execution of memblock_free_all)
1481 */
1482 if (WARN_ON_ONCE(slab_is_available()))
1483 return kzalloc_node(size, GFP_NOWAIT, nid);
1484
1485 if (max_addr > memblock.current_limit)
1486 max_addr = memblock.current_limit;
1487
1488 alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1489 exact_nid);
1490
1491 /* retry allocation without lower limit */
1492 if (!alloc && min_addr)
1493 alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1494 exact_nid);
1495
1496 if (!alloc)
1497 return NULL;
1498
1499 return phys_to_virt(alloc);
1500}
1501
1502/**
1503 * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1504 * without zeroing memory
1505 * @size: size of memory block to be allocated in bytes
1506 * @align: alignment of the region and block's size
1507 * @min_addr: the lower bound of the memory region from where the allocation
1508 * is preferred (phys address)
1509 * @max_addr: the upper bound of the memory region from where the allocation
1510 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1511 * allocate only from memory limited by memblock.current_limit value
1512 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1513 *
1514 * Public function, provides additional debug information (including caller
1515 * info), if enabled. Does not zero allocated memory.
1516 *
1517 * Return:
1518 * Virtual address of allocated memory block on success, NULL on failure.
1519 */
1520void * __init memblock_alloc_exact_nid_raw(
1521 phys_addr_t size, phys_addr_t align,
1522 phys_addr_t min_addr, phys_addr_t max_addr,
1523 int nid)
1524{
1525 void *ptr;
1526
1527 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1528 __func__, (u64)size, (u64)align, nid, &min_addr,
1529 &max_addr, (void *)_RET_IP_);
1530
1531 ptr = memblock_alloc_internal(size, align,
1532 min_addr, max_addr, nid, true);
1533 if (ptr && size > 0)
1534 page_init_poison(ptr, size);
1535
1536 return ptr;
1537}
1538
1539/**
1540 * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1541 * memory and without panicking
1542 * @size: size of memory block to be allocated in bytes
1543 * @align: alignment of the region and block's size
1544 * @min_addr: the lower bound of the memory region from where the allocation
1545 * is preferred (phys address)
1546 * @max_addr: the upper bound of the memory region from where the allocation
1547 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1548 * allocate only from memory limited by memblock.current_limit value
1549 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1550 *
1551 * Public function, provides additional debug information (including caller
1552 * info), if enabled. Does not zero allocated memory, does not panic if request
1553 * cannot be satisfied.
1554 *
1555 * Return:
1556 * Virtual address of allocated memory block on success, NULL on failure.
1557 */
1558void * __init memblock_alloc_try_nid_raw(
1559 phys_addr_t size, phys_addr_t align,
1560 phys_addr_t min_addr, phys_addr_t max_addr,
1561 int nid)
1562{
1563 void *ptr;
1564
1565 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1566 __func__, (u64)size, (u64)align, nid, &min_addr,
1567 &max_addr, (void *)_RET_IP_);
1568
1569 ptr = memblock_alloc_internal(size, align,
1570 min_addr, max_addr, nid, false);
1571 if (ptr && size > 0)
1572 page_init_poison(ptr, size);
1573
1574 return ptr;
1575}
1576
1577/**
1578 * memblock_alloc_try_nid - allocate boot memory block
1579 * @size: size of memory block to be allocated in bytes
1580 * @align: alignment of the region and block's size
1581 * @min_addr: the lower bound of the memory region from where the allocation
1582 * is preferred (phys address)
1583 * @max_addr: the upper bound of the memory region from where the allocation
1584 * is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1585 * allocate only from memory limited by memblock.current_limit value
1586 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1587 *
1588 * Public function, provides additional debug information (including caller
1589 * info), if enabled. This function zeroes the allocated memory.
1590 *
1591 * Return:
1592 * Virtual address of allocated memory block on success, NULL on failure.
1593 */
1594void * __init memblock_alloc_try_nid(
1595 phys_addr_t size, phys_addr_t align,
1596 phys_addr_t min_addr, phys_addr_t max_addr,
1597 int nid)
1598{
1599 void *ptr;
1600
1601 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1602 __func__, (u64)size, (u64)align, nid, &min_addr,
1603 &max_addr, (void *)_RET_IP_);
1604 ptr = memblock_alloc_internal(size, align,
1605 min_addr, max_addr, nid, false);
1606 if (ptr)
1607 memset(ptr, 0, size);
1608
1609 return ptr;
1610}
1611
1612/**
1613 * __memblock_free_late - free pages directly to buddy allocator
1614 * @base: phys starting address of the boot memory block
1615 * @size: size of the boot memory block in bytes
1616 *
1617 * This is only useful when the memblock allocator has already been torn
1618 * down, but we are still initializing the system. Pages are released directly
1619 * to the buddy allocator.
1620 */
1621void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1622{
1623 phys_addr_t cursor, end;
1624
1625 end = base + size - 1;
1626 memblock_dbg("%s: [%pa-%pa] %pS\n",
1627 __func__, &base, &end, (void *)_RET_IP_);
1628 kmemleak_free_part_phys(base, size);
1629 cursor = PFN_UP(base);
1630 end = PFN_DOWN(base + size);
1631
1632 for (; cursor < end; cursor++) {
1633 memblock_free_pages(pfn_to_page(cursor), cursor, 0);
1634 totalram_pages_inc();
1635 }
1636}
1637
1638/*
1639 * Remaining API functions
1640 */
1641
1642phys_addr_t __init_memblock memblock_phys_mem_size(void)
1643{
1644 return memblock.memory.total_size;
1645}
1646
1647phys_addr_t __init_memblock memblock_reserved_size(void)
1648{
1649 return memblock.reserved.total_size;
1650}
1651
1652phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1653{
1654 unsigned long pages = 0;
1655 struct memblock_region *r;
1656 unsigned long start_pfn, end_pfn;
1657
1658 for_each_memblock(memory, r) {
1659 start_pfn = memblock_region_memory_base_pfn(r);
1660 end_pfn = memblock_region_memory_end_pfn(r);
1661 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1662 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1663 pages += end_pfn - start_pfn;
1664 }
1665
1666 return PFN_PHYS(pages);
1667}
1668
1669/* lowest address */
1670phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1671{
1672 return memblock.memory.regions[0].base;
1673}
1674
1675phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1676{
1677 int idx = memblock.memory.cnt - 1;
1678
1679 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1680}
1681
1682static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1683{
1684 phys_addr_t max_addr = PHYS_ADDR_MAX;
1685 struct memblock_region *r;
1686
1687 /*
1688 * translate the memory @limit size into the max address within one of
1689 * the memory memblock regions, if the @limit exceeds the total size
1690 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1691 */
1692 for_each_memblock(memory, r) {
1693 if (limit <= r->size) {
1694 max_addr = r->base + limit;
1695 break;
1696 }
1697 limit -= r->size;
1698 }
1699
1700 return max_addr;
1701}
1702
1703void __init memblock_enforce_memory_limit(phys_addr_t limit)
1704{
1705 phys_addr_t max_addr;
1706
1707 if (!limit)
1708 return;
1709
1710 max_addr = __find_max_addr(limit);
1711
1712 /* @limit exceeds the total size of the memory, do nothing */
1713 if (max_addr == PHYS_ADDR_MAX)
1714 return;
1715
1716 /* truncate both memory and reserved regions */
1717 memblock_remove_range(&memblock.memory, max_addr,
1718 PHYS_ADDR_MAX);
1719 memblock_remove_range(&memblock.reserved, max_addr,
1720 PHYS_ADDR_MAX);
1721}
1722
1723void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1724{
1725 int start_rgn, end_rgn;
1726 int i, ret;
1727
1728 if (!size)
1729 return;
1730
1731 ret = memblock_isolate_range(&memblock.memory, base, size,
1732 &start_rgn, &end_rgn);
1733 if (ret)
1734 return;
1735
1736 /* remove all the MAP regions */
1737 for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1738 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1739 memblock_remove_region(&memblock.memory, i);
1740
1741 for (i = start_rgn - 1; i >= 0; i--)
1742 if (!memblock_is_nomap(&memblock.memory.regions[i]))
1743 memblock_remove_region(&memblock.memory, i);
1744
1745 /* truncate the reserved regions */
1746 memblock_remove_range(&memblock.reserved, 0, base);
1747 memblock_remove_range(&memblock.reserved,
1748 base + size, PHYS_ADDR_MAX);
1749}
1750
1751void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1752{
1753 phys_addr_t max_addr;
1754
1755 if (!limit)
1756 return;
1757
1758 max_addr = __find_max_addr(limit);
1759
1760 /* @limit exceeds the total size of the memory, do nothing */
1761 if (max_addr == PHYS_ADDR_MAX)
1762 return;
1763
1764 memblock_cap_memory_range(0, max_addr);
1765}
1766
1767static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1768{
1769 unsigned int left = 0, right = type->cnt;
1770
1771 do {
1772 unsigned int mid = (right + left) / 2;
1773
1774 if (addr < type->regions[mid].base)
1775 right = mid;
1776 else if (addr >= (type->regions[mid].base +
1777 type->regions[mid].size))
1778 left = mid + 1;
1779 else
1780 return mid;
1781 } while (left < right);
1782 return -1;
1783}
1784
1785bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1786{
1787 return memblock_search(&memblock.reserved, addr) != -1;
1788}
1789
1790bool __init_memblock memblock_is_memory(phys_addr_t addr)
1791{
1792 return memblock_search(&memblock.memory, addr) != -1;
1793}
1794
1795bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1796{
1797 int i = memblock_search(&memblock.memory, addr);
1798
1799 if (i == -1)
1800 return false;
1801 return !memblock_is_nomap(&memblock.memory.regions[i]);
1802}
1803
1804int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1805 unsigned long *start_pfn, unsigned long *end_pfn)
1806{
1807 struct memblock_type *type = &memblock.memory;
1808 int mid = memblock_search(type, PFN_PHYS(pfn));
1809
1810 if (mid == -1)
1811 return -1;
1812
1813 *start_pfn = PFN_DOWN(type->regions[mid].base);
1814 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1815
1816 return memblock_get_region_node(&type->regions[mid]);
1817}
1818
1819/**
1820 * memblock_is_region_memory - check if a region is a subset of memory
1821 * @base: base of region to check
1822 * @size: size of region to check
1823 *
1824 * Check if the region [@base, @base + @size) is a subset of a memory block.
1825 *
1826 * Return:
1827 * 0 if false, non-zero if true
1828 */
1829bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1830{
1831 int idx = memblock_search(&memblock.memory, base);
1832 phys_addr_t end = base + memblock_cap_size(base, &size);
1833
1834 if (idx == -1)
1835 return false;
1836 return (memblock.memory.regions[idx].base +
1837 memblock.memory.regions[idx].size) >= end;
1838}
1839
1840/**
1841 * memblock_is_region_reserved - check if a region intersects reserved memory
1842 * @base: base of region to check
1843 * @size: size of region to check
1844 *
1845 * Check if the region [@base, @base + @size) intersects a reserved
1846 * memory block.
1847 *
1848 * Return:
1849 * True if they intersect, false if not.
1850 */
1851bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1852{
1853 memblock_cap_size(base, &size);
1854 return memblock_overlaps_region(&memblock.reserved, base, size);
1855}
1856
1857void __init_memblock memblock_trim_memory(phys_addr_t align)
1858{
1859 phys_addr_t start, end, orig_start, orig_end;
1860 struct memblock_region *r;
1861
1862 for_each_memblock(memory, r) {
1863 orig_start = r->base;
1864 orig_end = r->base + r->size;
1865 start = round_up(orig_start, align);
1866 end = round_down(orig_end, align);
1867
1868 if (start == orig_start && end == orig_end)
1869 continue;
1870
1871 if (start < end) {
1872 r->base = start;
1873 r->size = end - start;
1874 } else {
1875 memblock_remove_region(&memblock.memory,
1876 r - memblock.memory.regions);
1877 r--;
1878 }
1879 }
1880}
1881
1882void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1883{
1884 memblock.current_limit = limit;
1885}
1886
1887phys_addr_t __init_memblock memblock_get_current_limit(void)
1888{
1889 return memblock.current_limit;
1890}
1891
1892static void __init_memblock memblock_dump(struct memblock_type *type)
1893{
1894 phys_addr_t base, end, size;
1895 enum memblock_flags flags;
1896 int idx;
1897 struct memblock_region *rgn;
1898
1899 pr_info(" %s.cnt = 0x%lx\n", type->name, type->cnt);
1900
1901 for_each_memblock_type(idx, type, rgn) {
1902 char nid_buf[32] = "";
1903
1904 base = rgn->base;
1905 size = rgn->size;
1906 end = base + size - 1;
1907 flags = rgn->flags;
1908#ifdef CONFIG_NEED_MULTIPLE_NODES
1909 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1910 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1911 memblock_get_region_node(rgn));
1912#endif
1913 pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1914 type->name, idx, &base, &end, &size, nid_buf, flags);
1915 }
1916}
1917
1918void __init_memblock __memblock_dump_all(void)
1919{
1920 pr_info("MEMBLOCK configuration:\n");
1921 pr_info(" memory size = %pa reserved size = %pa\n",
1922 &memblock.memory.total_size,
1923 &memblock.reserved.total_size);
1924
1925 memblock_dump(&memblock.memory);
1926 memblock_dump(&memblock.reserved);
1927#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1928 memblock_dump(&physmem);
1929#endif
1930}
1931
1932void __init memblock_allow_resize(void)
1933{
1934 memblock_can_resize = 1;
1935}
1936
1937static int __init early_memblock(char *p)
1938{
1939 if (p && strstr(p, "debug"))
1940 memblock_debug = 1;
1941 return 0;
1942}
1943early_param("memblock", early_memblock);
1944
1945static void __init __free_pages_memory(unsigned long start, unsigned long end)
1946{
1947 int order;
1948
1949 while (start < end) {
1950 order = min(MAX_ORDER - 1UL, __ffs(start));
1951
1952 while (start + (1UL << order) > end)
1953 order--;
1954
1955 memblock_free_pages(pfn_to_page(start), start, order);
1956
1957 start += (1UL << order);
1958 }
1959}
1960
1961static unsigned long __init __free_memory_core(phys_addr_t start,
1962 phys_addr_t end)
1963{
1964 unsigned long start_pfn = PFN_UP(start);
1965 unsigned long end_pfn = min_t(unsigned long,
1966 PFN_DOWN(end), max_low_pfn);
1967
1968 if (start_pfn >= end_pfn)
1969 return 0;
1970
1971 __free_pages_memory(start_pfn, end_pfn);
1972
1973 return end_pfn - start_pfn;
1974}
1975
1976static unsigned long __init free_low_memory_core_early(void)
1977{
1978 unsigned long count = 0;
1979 phys_addr_t start, end;
1980 u64 i;
1981
1982 memblock_clear_hotplug(0, -1);
1983
1984 for_each_reserved_mem_region(i, &start, &end)
1985 reserve_bootmem_region(start, end);
1986
1987 /*
1988 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
1989 * because in some case like Node0 doesn't have RAM installed
1990 * low ram will be on Node1
1991 */
1992 for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
1993 NULL)
1994 count += __free_memory_core(start, end);
1995
1996 return count;
1997}
1998
1999static int reset_managed_pages_done __initdata;
2000
2001void reset_node_managed_pages(pg_data_t *pgdat)
2002{
2003 struct zone *z;
2004
2005 for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
2006 atomic_long_set(&z->managed_pages, 0);
2007}
2008
2009void __init reset_all_zones_managed_pages(void)
2010{
2011 struct pglist_data *pgdat;
2012
2013 if (reset_managed_pages_done)
2014 return;
2015
2016 for_each_online_pgdat(pgdat)
2017 reset_node_managed_pages(pgdat);
2018
2019 reset_managed_pages_done = 1;
2020}
2021
2022/**
2023 * memblock_free_all - release free pages to the buddy allocator
2024 *
2025 * Return: the number of pages actually released.
2026 */
2027unsigned long __init memblock_free_all(void)
2028{
2029 unsigned long pages;
2030
2031 reset_all_zones_managed_pages();
2032
2033 pages = free_low_memory_core_early();
2034 totalram_pages_add(pages);
2035
2036 return pages;
2037}
2038
2039#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2040
2041static int memblock_debug_show(struct seq_file *m, void *private)
2042{
2043 struct memblock_type *type = m->private;
2044 struct memblock_region *reg;
2045 int i;
2046 phys_addr_t end;
2047
2048 for (i = 0; i < type->cnt; i++) {
2049 reg = &type->regions[i];
2050 end = reg->base + reg->size - 1;
2051
2052 seq_printf(m, "%4d: ", i);
2053 seq_printf(m, "%pa..%pa\n", ®->base, &end);
2054 }
2055 return 0;
2056}
2057DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2058
2059static int __init memblock_init_debugfs(void)
2060{
2061 struct dentry *root = debugfs_create_dir("memblock", NULL);
2062
2063 debugfs_create_file("memory", 0444, root,
2064 &memblock.memory, &memblock_debug_fops);
2065 debugfs_create_file("reserved", 0444, root,
2066 &memblock.reserved, &memblock_debug_fops);
2067#ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2068 debugfs_create_file("physmem", 0444, root, &physmem,
2069 &memblock_debug_fops);
2070#endif
2071
2072 return 0;
2073}
2074__initcall(memblock_init_debugfs);
2075
2076#endif /* CONFIG_DEBUG_FS */