]> git.ipfire.org Git - people/ms/linux.git/blob - mm/memblock.c
Merge branch 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[people/ms/linux.git] / mm / memblock.c
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
28 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
29 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
30 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
31 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
32 #endif
33
34 struct 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
53 int memblock_debug __initdata_memblock;
54 #ifdef CONFIG_MOVABLE_NODE
55 bool movable_node_enabled __initdata_memblock = false;
56 #endif
57 static bool system_has_some_mirror __initdata_memblock = false;
58 static int memblock_can_resize __initdata_memblock;
59 static int memblock_memory_in_slab __initdata_memblock = 0;
60 static int memblock_reserved_in_slab __initdata_memblock = 0;
61
62 ulong __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 */
68 static __init_memblock const char *
69 memblock_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 */
80 static 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 */
88 static 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
94 bool __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 phys_addr_t rgnbase = type->regions[i].base;
101 phys_addr_t rgnsize = type->regions[i].size;
102 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
103 break;
104 }
105
106 return i < type->cnt;
107 }
108
109 /*
110 * __memblock_find_range_bottom_up - find free area utility in bottom-up
111 * @start: start of candidate range
112 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
113 * @size: size of free area to find
114 * @align: alignment of free area to find
115 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
116 * @flags: pick from blocks based on memory attributes
117 *
118 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
119 *
120 * RETURNS:
121 * Found address on success, 0 on failure.
122 */
123 static phys_addr_t __init_memblock
124 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
125 phys_addr_t size, phys_addr_t align, int nid,
126 ulong flags)
127 {
128 phys_addr_t this_start, this_end, cand;
129 u64 i;
130
131 for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
132 this_start = clamp(this_start, start, end);
133 this_end = clamp(this_end, start, end);
134
135 cand = round_up(this_start, align);
136 if (cand < this_end && this_end - cand >= size)
137 return cand;
138 }
139
140 return 0;
141 }
142
143 /**
144 * __memblock_find_range_top_down - find free area utility, in top-down
145 * @start: start of candidate range
146 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
147 * @size: size of free area to find
148 * @align: alignment of free area to find
149 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
150 * @flags: pick from blocks based on memory attributes
151 *
152 * Utility called from memblock_find_in_range_node(), find free area top-down.
153 *
154 * RETURNS:
155 * Found address on success, 0 on failure.
156 */
157 static phys_addr_t __init_memblock
158 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
159 phys_addr_t size, phys_addr_t align, int nid,
160 ulong flags)
161 {
162 phys_addr_t this_start, this_end, cand;
163 u64 i;
164
165 for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
166 NULL) {
167 this_start = clamp(this_start, start, end);
168 this_end = clamp(this_end, start, end);
169
170 if (this_end < size)
171 continue;
172
173 cand = round_down(this_end - size, align);
174 if (cand >= this_start)
175 return cand;
176 }
177
178 return 0;
179 }
180
181 /**
182 * memblock_find_in_range_node - find free area in given range and node
183 * @size: size of free area to find
184 * @align: alignment of free area to find
185 * @start: start of candidate range
186 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
187 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
188 * @flags: pick from blocks based on memory attributes
189 *
190 * Find @size free area aligned to @align in the specified range and node.
191 *
192 * When allocation direction is bottom-up, the @start should be greater
193 * than the end of the kernel image. Otherwise, it will be trimmed. The
194 * reason is that we want the bottom-up allocation just near the kernel
195 * image so it is highly likely that the allocated memory and the kernel
196 * will reside in the same node.
197 *
198 * If bottom-up allocation failed, will try to allocate memory top-down.
199 *
200 * RETURNS:
201 * Found address on success, 0 on failure.
202 */
203 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
204 phys_addr_t align, phys_addr_t start,
205 phys_addr_t end, int nid, ulong flags)
206 {
207 phys_addr_t kernel_end, ret;
208
209 /* pump up @end */
210 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
211 end = memblock.current_limit;
212
213 /* avoid allocating the first page */
214 start = max_t(phys_addr_t, start, PAGE_SIZE);
215 end = max(start, end);
216 kernel_end = __pa_symbol(_end);
217
218 /*
219 * try bottom-up allocation only when bottom-up mode
220 * is set and @end is above the kernel image.
221 */
222 if (memblock_bottom_up() && end > kernel_end) {
223 phys_addr_t bottom_up_start;
224
225 /* make sure we will allocate above the kernel */
226 bottom_up_start = max(start, kernel_end);
227
228 /* ok, try bottom-up allocation first */
229 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
230 size, align, nid, flags);
231 if (ret)
232 return ret;
233
234 /*
235 * we always limit bottom-up allocation above the kernel,
236 * but top-down allocation doesn't have the limit, so
237 * retrying top-down allocation may succeed when bottom-up
238 * allocation failed.
239 *
240 * bottom-up allocation is expected to be fail very rarely,
241 * so we use WARN_ONCE() here to see the stack trace if
242 * fail happens.
243 */
244 WARN_ONCE(1, "memblock: bottom-up allocation failed, "
245 "memory hotunplug may be affected\n");
246 }
247
248 return __memblock_find_range_top_down(start, end, size, align, nid,
249 flags);
250 }
251
252 /**
253 * memblock_find_in_range - find free area in given range
254 * @start: start of candidate range
255 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
256 * @size: size of free area to find
257 * @align: alignment of free area to find
258 *
259 * Find @size free area aligned to @align in the specified range.
260 *
261 * RETURNS:
262 * Found address on success, 0 on failure.
263 */
264 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
265 phys_addr_t end, phys_addr_t size,
266 phys_addr_t align)
267 {
268 phys_addr_t ret;
269 ulong flags = choose_memblock_flags();
270
271 again:
272 ret = memblock_find_in_range_node(size, align, start, end,
273 NUMA_NO_NODE, flags);
274
275 if (!ret && (flags & MEMBLOCK_MIRROR)) {
276 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
277 &size);
278 flags &= ~MEMBLOCK_MIRROR;
279 goto again;
280 }
281
282 return ret;
283 }
284
285 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
286 {
287 type->total_size -= type->regions[r].size;
288 memmove(&type->regions[r], &type->regions[r + 1],
289 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
290 type->cnt--;
291
292 /* Special case for empty arrays */
293 if (type->cnt == 0) {
294 WARN_ON(type->total_size != 0);
295 type->cnt = 1;
296 type->regions[0].base = 0;
297 type->regions[0].size = 0;
298 type->regions[0].flags = 0;
299 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
300 }
301 }
302
303 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
304
305 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
306 phys_addr_t *addr)
307 {
308 if (memblock.reserved.regions == memblock_reserved_init_regions)
309 return 0;
310
311 *addr = __pa(memblock.reserved.regions);
312
313 return PAGE_ALIGN(sizeof(struct memblock_region) *
314 memblock.reserved.max);
315 }
316
317 phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
318 phys_addr_t *addr)
319 {
320 if (memblock.memory.regions == memblock_memory_init_regions)
321 return 0;
322
323 *addr = __pa(memblock.memory.regions);
324
325 return PAGE_ALIGN(sizeof(struct memblock_region) *
326 memblock.memory.max);
327 }
328
329 #endif
330
331 /**
332 * memblock_double_array - double the size of the memblock regions array
333 * @type: memblock type of the regions array being doubled
334 * @new_area_start: starting address of memory range to avoid overlap with
335 * @new_area_size: size of memory range to avoid overlap with
336 *
337 * Double the size of the @type regions array. If memblock is being used to
338 * allocate memory for a new reserved regions array and there is a previously
339 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
340 * waiting to be reserved, ensure the memory used by the new array does
341 * not overlap.
342 *
343 * RETURNS:
344 * 0 on success, -1 on failure.
345 */
346 static int __init_memblock memblock_double_array(struct memblock_type *type,
347 phys_addr_t new_area_start,
348 phys_addr_t new_area_size)
349 {
350 struct memblock_region *new_array, *old_array;
351 phys_addr_t old_alloc_size, new_alloc_size;
352 phys_addr_t old_size, new_size, addr;
353 int use_slab = slab_is_available();
354 int *in_slab;
355
356 /* We don't allow resizing until we know about the reserved regions
357 * of memory that aren't suitable for allocation
358 */
359 if (!memblock_can_resize)
360 return -1;
361
362 /* Calculate new doubled size */
363 old_size = type->max * sizeof(struct memblock_region);
364 new_size = old_size << 1;
365 /*
366 * We need to allocated new one align to PAGE_SIZE,
367 * so we can free them completely later.
368 */
369 old_alloc_size = PAGE_ALIGN(old_size);
370 new_alloc_size = PAGE_ALIGN(new_size);
371
372 /* Retrieve the slab flag */
373 if (type == &memblock.memory)
374 in_slab = &memblock_memory_in_slab;
375 else
376 in_slab = &memblock_reserved_in_slab;
377
378 /* Try to find some space for it.
379 *
380 * WARNING: We assume that either slab_is_available() and we use it or
381 * we use MEMBLOCK for allocations. That means that this is unsafe to
382 * use when bootmem is currently active (unless bootmem itself is
383 * implemented on top of MEMBLOCK which isn't the case yet)
384 *
385 * This should however not be an issue for now, as we currently only
386 * call into MEMBLOCK while it's still active, or much later when slab
387 * is active for memory hotplug operations
388 */
389 if (use_slab) {
390 new_array = kmalloc(new_size, GFP_KERNEL);
391 addr = new_array ? __pa(new_array) : 0;
392 } else {
393 /* only exclude range when trying to double reserved.regions */
394 if (type != &memblock.reserved)
395 new_area_start = new_area_size = 0;
396
397 addr = memblock_find_in_range(new_area_start + new_area_size,
398 memblock.current_limit,
399 new_alloc_size, PAGE_SIZE);
400 if (!addr && new_area_size)
401 addr = memblock_find_in_range(0,
402 min(new_area_start, memblock.current_limit),
403 new_alloc_size, PAGE_SIZE);
404
405 new_array = addr ? __va(addr) : NULL;
406 }
407 if (!addr) {
408 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
409 memblock_type_name(type), type->max, type->max * 2);
410 return -1;
411 }
412
413 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
414 memblock_type_name(type), type->max * 2, (u64)addr,
415 (u64)addr + new_size - 1);
416
417 /*
418 * Found space, we now need to move the array over before we add the
419 * reserved region since it may be our reserved array itself that is
420 * full.
421 */
422 memcpy(new_array, type->regions, old_size);
423 memset(new_array + type->max, 0, old_size);
424 old_array = type->regions;
425 type->regions = new_array;
426 type->max <<= 1;
427
428 /* Free old array. We needn't free it if the array is the static one */
429 if (*in_slab)
430 kfree(old_array);
431 else if (old_array != memblock_memory_init_regions &&
432 old_array != memblock_reserved_init_regions)
433 memblock_free(__pa(old_array), old_alloc_size);
434
435 /*
436 * Reserve the new array if that comes from the memblock. Otherwise, we
437 * needn't do it
438 */
439 if (!use_slab)
440 BUG_ON(memblock_reserve(addr, new_alloc_size));
441
442 /* Update slab flag */
443 *in_slab = use_slab;
444
445 return 0;
446 }
447
448 /**
449 * memblock_merge_regions - merge neighboring compatible regions
450 * @type: memblock type to scan
451 *
452 * Scan @type and merge neighboring compatible regions.
453 */
454 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
455 {
456 int i = 0;
457
458 /* cnt never goes below 1 */
459 while (i < type->cnt - 1) {
460 struct memblock_region *this = &type->regions[i];
461 struct memblock_region *next = &type->regions[i + 1];
462
463 if (this->base + this->size != next->base ||
464 memblock_get_region_node(this) !=
465 memblock_get_region_node(next) ||
466 this->flags != next->flags) {
467 BUG_ON(this->base + this->size > next->base);
468 i++;
469 continue;
470 }
471
472 this->size += next->size;
473 /* move forward from next + 1, index of which is i + 2 */
474 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
475 type->cnt--;
476 }
477 }
478
479 /**
480 * memblock_insert_region - insert new memblock region
481 * @type: memblock type to insert into
482 * @idx: index for the insertion point
483 * @base: base address of the new region
484 * @size: size of the new region
485 * @nid: node id of the new region
486 * @flags: flags of the new region
487 *
488 * Insert new memblock region [@base,@base+@size) into @type at @idx.
489 * @type must already have extra room to accomodate the new region.
490 */
491 static void __init_memblock memblock_insert_region(struct memblock_type *type,
492 int idx, phys_addr_t base,
493 phys_addr_t size,
494 int nid, unsigned long flags)
495 {
496 struct memblock_region *rgn = &type->regions[idx];
497
498 BUG_ON(type->cnt >= type->max);
499 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
500 rgn->base = base;
501 rgn->size = size;
502 rgn->flags = flags;
503 memblock_set_region_node(rgn, nid);
504 type->cnt++;
505 type->total_size += size;
506 }
507
508 /**
509 * memblock_add_range - add new memblock region
510 * @type: memblock type to add new region into
511 * @base: base address of the new region
512 * @size: size of the new region
513 * @nid: nid of the new region
514 * @flags: flags of the new region
515 *
516 * Add new memblock region [@base,@base+@size) into @type. The new region
517 * is allowed to overlap with existing ones - overlaps don't affect already
518 * existing regions. @type is guaranteed to be minimal (all neighbouring
519 * compatible regions are merged) after the addition.
520 *
521 * RETURNS:
522 * 0 on success, -errno on failure.
523 */
524 int __init_memblock memblock_add_range(struct memblock_type *type,
525 phys_addr_t base, phys_addr_t size,
526 int nid, unsigned long flags)
527 {
528 bool insert = false;
529 phys_addr_t obase = base;
530 phys_addr_t end = base + memblock_cap_size(base, &size);
531 int i, nr_new;
532
533 if (!size)
534 return 0;
535
536 /* special case for empty array */
537 if (type->regions[0].size == 0) {
538 WARN_ON(type->cnt != 1 || type->total_size);
539 type->regions[0].base = base;
540 type->regions[0].size = size;
541 type->regions[0].flags = flags;
542 memblock_set_region_node(&type->regions[0], nid);
543 type->total_size = size;
544 return 0;
545 }
546 repeat:
547 /*
548 * The following is executed twice. Once with %false @insert and
549 * then with %true. The first counts the number of regions needed
550 * to accomodate the new area. The second actually inserts them.
551 */
552 base = obase;
553 nr_new = 0;
554
555 for (i = 0; i < type->cnt; i++) {
556 struct memblock_region *rgn = &type->regions[i];
557 phys_addr_t rbase = rgn->base;
558 phys_addr_t rend = rbase + rgn->size;
559
560 if (rbase >= end)
561 break;
562 if (rend <= base)
563 continue;
564 /*
565 * @rgn overlaps. If it separates the lower part of new
566 * area, insert that portion.
567 */
568 if (rbase > base) {
569 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
570 WARN_ON(nid != memblock_get_region_node(rgn));
571 #endif
572 WARN_ON(flags != rgn->flags);
573 nr_new++;
574 if (insert)
575 memblock_insert_region(type, i++, base,
576 rbase - base, nid,
577 flags);
578 }
579 /* area below @rend is dealt with, forget about it */
580 base = min(rend, end);
581 }
582
583 /* insert the remaining portion */
584 if (base < end) {
585 nr_new++;
586 if (insert)
587 memblock_insert_region(type, i, base, end - base,
588 nid, flags);
589 }
590
591 /*
592 * If this was the first round, resize array and repeat for actual
593 * insertions; otherwise, merge and return.
594 */
595 if (!insert) {
596 while (type->cnt + nr_new > type->max)
597 if (memblock_double_array(type, obase, size) < 0)
598 return -ENOMEM;
599 insert = true;
600 goto repeat;
601 } else {
602 memblock_merge_regions(type);
603 return 0;
604 }
605 }
606
607 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
608 int nid)
609 {
610 return memblock_add_range(&memblock.memory, base, size, nid, 0);
611 }
612
613 static int __init_memblock memblock_add_region(phys_addr_t base,
614 phys_addr_t size,
615 int nid,
616 unsigned long flags)
617 {
618 struct memblock_type *type = &memblock.memory;
619
620 memblock_dbg("memblock_add: [%#016llx-%#016llx] flags %#02lx %pF\n",
621 (unsigned long long)base,
622 (unsigned long long)base + size - 1,
623 flags, (void *)_RET_IP_);
624
625 return memblock_add_range(type, base, size, nid, flags);
626 }
627
628 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
629 {
630 return memblock_add_region(base, size, MAX_NUMNODES, 0);
631 }
632
633 /**
634 * memblock_isolate_range - isolate given range into disjoint memblocks
635 * @type: memblock type to isolate range for
636 * @base: base of range to isolate
637 * @size: size of range to isolate
638 * @start_rgn: out parameter for the start of isolated region
639 * @end_rgn: out parameter for the end of isolated region
640 *
641 * Walk @type and ensure that regions don't cross the boundaries defined by
642 * [@base,@base+@size). Crossing regions are split at the boundaries,
643 * which may create at most two more regions. The index of the first
644 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
645 *
646 * RETURNS:
647 * 0 on success, -errno on failure.
648 */
649 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
650 phys_addr_t base, phys_addr_t size,
651 int *start_rgn, int *end_rgn)
652 {
653 phys_addr_t end = base + memblock_cap_size(base, &size);
654 int i;
655
656 *start_rgn = *end_rgn = 0;
657
658 if (!size)
659 return 0;
660
661 /* we'll create at most two more regions */
662 while (type->cnt + 2 > type->max)
663 if (memblock_double_array(type, base, size) < 0)
664 return -ENOMEM;
665
666 for (i = 0; i < type->cnt; i++) {
667 struct memblock_region *rgn = &type->regions[i];
668 phys_addr_t rbase = rgn->base;
669 phys_addr_t rend = rbase + rgn->size;
670
671 if (rbase >= end)
672 break;
673 if (rend <= base)
674 continue;
675
676 if (rbase < base) {
677 /*
678 * @rgn intersects from below. Split and continue
679 * to process the next region - the new top half.
680 */
681 rgn->base = base;
682 rgn->size -= base - rbase;
683 type->total_size -= base - rbase;
684 memblock_insert_region(type, i, rbase, base - rbase,
685 memblock_get_region_node(rgn),
686 rgn->flags);
687 } else if (rend > end) {
688 /*
689 * @rgn intersects from above. Split and redo the
690 * current region - the new bottom half.
691 */
692 rgn->base = end;
693 rgn->size -= end - rbase;
694 type->total_size -= end - rbase;
695 memblock_insert_region(type, i--, rbase, end - rbase,
696 memblock_get_region_node(rgn),
697 rgn->flags);
698 } else {
699 /* @rgn is fully contained, record it */
700 if (!*end_rgn)
701 *start_rgn = i;
702 *end_rgn = i + 1;
703 }
704 }
705
706 return 0;
707 }
708
709 static int __init_memblock memblock_remove_range(struct memblock_type *type,
710 phys_addr_t base, phys_addr_t size)
711 {
712 int start_rgn, end_rgn;
713 int i, ret;
714
715 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
716 if (ret)
717 return ret;
718
719 for (i = end_rgn - 1; i >= start_rgn; i--)
720 memblock_remove_region(type, i);
721 return 0;
722 }
723
724 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
725 {
726 return memblock_remove_range(&memblock.memory, base, size);
727 }
728
729
730 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
731 {
732 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
733 (unsigned long long)base,
734 (unsigned long long)base + size - 1,
735 (void *)_RET_IP_);
736
737 kmemleak_free_part(__va(base), size);
738 return memblock_remove_range(&memblock.reserved, base, size);
739 }
740
741 static int __init_memblock memblock_reserve_region(phys_addr_t base,
742 phys_addr_t size,
743 int nid,
744 unsigned long flags)
745 {
746 struct memblock_type *type = &memblock.reserved;
747
748 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
749 (unsigned long long)base,
750 (unsigned long long)base + size - 1,
751 flags, (void *)_RET_IP_);
752
753 return memblock_add_range(type, base, size, nid, flags);
754 }
755
756 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
757 {
758 return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
759 }
760
761 /**
762 *
763 * This function isolates region [@base, @base + @size), and sets/clears flag
764 *
765 * Return 0 on success, -errno on failure.
766 */
767 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
768 phys_addr_t size, int set, int flag)
769 {
770 struct memblock_type *type = &memblock.memory;
771 int i, ret, start_rgn, end_rgn;
772
773 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
774 if (ret)
775 return ret;
776
777 for (i = start_rgn; i < end_rgn; i++)
778 if (set)
779 memblock_set_region_flags(&type->regions[i], flag);
780 else
781 memblock_clear_region_flags(&type->regions[i], flag);
782
783 memblock_merge_regions(type);
784 return 0;
785 }
786
787 /**
788 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
789 * @base: the base phys addr of the region
790 * @size: the size of the region
791 *
792 * Return 0 on success, -errno on failure.
793 */
794 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
795 {
796 return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
797 }
798
799 /**
800 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
801 * @base: the base phys addr of the region
802 * @size: the size of the region
803 *
804 * Return 0 on success, -errno on failure.
805 */
806 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
807 {
808 return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
809 }
810
811 /**
812 * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
813 * @base: the base phys addr of the region
814 * @size: the size of the region
815 *
816 * Return 0 on success, -errno on failure.
817 */
818 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
819 {
820 system_has_some_mirror = true;
821
822 return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
823 }
824
825
826 /**
827 * __next_reserved_mem_region - next function for for_each_reserved_region()
828 * @idx: pointer to u64 loop variable
829 * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
830 * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
831 *
832 * Iterate over all reserved memory regions.
833 */
834 void __init_memblock __next_reserved_mem_region(u64 *idx,
835 phys_addr_t *out_start,
836 phys_addr_t *out_end)
837 {
838 struct memblock_type *type = &memblock.reserved;
839
840 if (*idx >= 0 && *idx < type->cnt) {
841 struct memblock_region *r = &type->regions[*idx];
842 phys_addr_t base = r->base;
843 phys_addr_t size = r->size;
844
845 if (out_start)
846 *out_start = base;
847 if (out_end)
848 *out_end = base + size - 1;
849
850 *idx += 1;
851 return;
852 }
853
854 /* signal end of iteration */
855 *idx = ULLONG_MAX;
856 }
857
858 /**
859 * __next__mem_range - next function for for_each_free_mem_range() etc.
860 * @idx: pointer to u64 loop variable
861 * @nid: node selector, %NUMA_NO_NODE for all nodes
862 * @flags: pick from blocks based on memory attributes
863 * @type_a: pointer to memblock_type from where the range is taken
864 * @type_b: pointer to memblock_type which excludes memory from being taken
865 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
866 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
867 * @out_nid: ptr to int for nid of the range, can be %NULL
868 *
869 * Find the first area from *@idx which matches @nid, fill the out
870 * parameters, and update *@idx for the next iteration. The lower 32bit of
871 * *@idx contains index into type_a and the upper 32bit indexes the
872 * areas before each region in type_b. For example, if type_b regions
873 * look like the following,
874 *
875 * 0:[0-16), 1:[32-48), 2:[128-130)
876 *
877 * The upper 32bit indexes the following regions.
878 *
879 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
880 *
881 * As both region arrays are sorted, the function advances the two indices
882 * in lockstep and returns each intersection.
883 */
884 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
885 struct memblock_type *type_a,
886 struct memblock_type *type_b,
887 phys_addr_t *out_start,
888 phys_addr_t *out_end, int *out_nid)
889 {
890 int idx_a = *idx & 0xffffffff;
891 int idx_b = *idx >> 32;
892
893 if (WARN_ONCE(nid == MAX_NUMNODES,
894 "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
895 nid = NUMA_NO_NODE;
896
897 for (; idx_a < type_a->cnt; idx_a++) {
898 struct memblock_region *m = &type_a->regions[idx_a];
899
900 phys_addr_t m_start = m->base;
901 phys_addr_t m_end = m->base + m->size;
902 int m_nid = memblock_get_region_node(m);
903
904 /* only memory regions are associated with nodes, check it */
905 if (nid != NUMA_NO_NODE && nid != m_nid)
906 continue;
907
908 /* skip hotpluggable memory regions if needed */
909 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
910 continue;
911
912 /* if we want mirror memory skip non-mirror memory regions */
913 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
914 continue;
915
916 if (!type_b) {
917 if (out_start)
918 *out_start = m_start;
919 if (out_end)
920 *out_end = m_end;
921 if (out_nid)
922 *out_nid = m_nid;
923 idx_a++;
924 *idx = (u32)idx_a | (u64)idx_b << 32;
925 return;
926 }
927
928 /* scan areas before each reservation */
929 for (; idx_b < type_b->cnt + 1; idx_b++) {
930 struct memblock_region *r;
931 phys_addr_t r_start;
932 phys_addr_t r_end;
933
934 r = &type_b->regions[idx_b];
935 r_start = idx_b ? r[-1].base + r[-1].size : 0;
936 r_end = idx_b < type_b->cnt ?
937 r->base : ULLONG_MAX;
938
939 /*
940 * if idx_b advanced past idx_a,
941 * break out to advance idx_a
942 */
943 if (r_start >= m_end)
944 break;
945 /* if the two regions intersect, we're done */
946 if (m_start < r_end) {
947 if (out_start)
948 *out_start =
949 max(m_start, r_start);
950 if (out_end)
951 *out_end = min(m_end, r_end);
952 if (out_nid)
953 *out_nid = m_nid;
954 /*
955 * The region which ends first is
956 * advanced for the next iteration.
957 */
958 if (m_end <= r_end)
959 idx_a++;
960 else
961 idx_b++;
962 *idx = (u32)idx_a | (u64)idx_b << 32;
963 return;
964 }
965 }
966 }
967
968 /* signal end of iteration */
969 *idx = ULLONG_MAX;
970 }
971
972 /**
973 * __next_mem_range_rev - generic next function for for_each_*_range_rev()
974 *
975 * Finds the next range from type_a which is not marked as unsuitable
976 * in type_b.
977 *
978 * @idx: pointer to u64 loop variable
979 * @nid: node selector, %NUMA_NO_NODE for all nodes
980 * @flags: pick from blocks based on memory attributes
981 * @type_a: pointer to memblock_type from where the range is taken
982 * @type_b: pointer to memblock_type which excludes memory from being taken
983 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
984 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
985 * @out_nid: ptr to int for nid of the range, can be %NULL
986 *
987 * Reverse of __next_mem_range().
988 */
989 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
990 struct memblock_type *type_a,
991 struct memblock_type *type_b,
992 phys_addr_t *out_start,
993 phys_addr_t *out_end, int *out_nid)
994 {
995 int idx_a = *idx & 0xffffffff;
996 int idx_b = *idx >> 32;
997
998 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
999 nid = NUMA_NO_NODE;
1000
1001 if (*idx == (u64)ULLONG_MAX) {
1002 idx_a = type_a->cnt - 1;
1003 idx_b = type_b->cnt;
1004 }
1005
1006 for (; idx_a >= 0; idx_a--) {
1007 struct memblock_region *m = &type_a->regions[idx_a];
1008
1009 phys_addr_t m_start = m->base;
1010 phys_addr_t m_end = m->base + m->size;
1011 int m_nid = memblock_get_region_node(m);
1012
1013 /* only memory regions are associated with nodes, check it */
1014 if (nid != NUMA_NO_NODE && nid != m_nid)
1015 continue;
1016
1017 /* skip hotpluggable memory regions if needed */
1018 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1019 continue;
1020
1021 /* if we want mirror memory skip non-mirror memory regions */
1022 if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1023 continue;
1024
1025 if (!type_b) {
1026 if (out_start)
1027 *out_start = m_start;
1028 if (out_end)
1029 *out_end = m_end;
1030 if (out_nid)
1031 *out_nid = m_nid;
1032 idx_a++;
1033 *idx = (u32)idx_a | (u64)idx_b << 32;
1034 return;
1035 }
1036
1037 /* scan areas before each reservation */
1038 for (; idx_b >= 0; idx_b--) {
1039 struct memblock_region *r;
1040 phys_addr_t r_start;
1041 phys_addr_t r_end;
1042
1043 r = &type_b->regions[idx_b];
1044 r_start = idx_b ? r[-1].base + r[-1].size : 0;
1045 r_end = idx_b < type_b->cnt ?
1046 r->base : ULLONG_MAX;
1047 /*
1048 * if idx_b advanced past idx_a,
1049 * break out to advance idx_a
1050 */
1051
1052 if (r_end <= m_start)
1053 break;
1054 /* if the two regions intersect, we're done */
1055 if (m_end > r_start) {
1056 if (out_start)
1057 *out_start = max(m_start, r_start);
1058 if (out_end)
1059 *out_end = min(m_end, r_end);
1060 if (out_nid)
1061 *out_nid = m_nid;
1062 if (m_start >= r_start)
1063 idx_a--;
1064 else
1065 idx_b--;
1066 *idx = (u32)idx_a | (u64)idx_b << 32;
1067 return;
1068 }
1069 }
1070 }
1071 /* signal end of iteration */
1072 *idx = ULLONG_MAX;
1073 }
1074
1075 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1076 /*
1077 * Common iterator interface used to define for_each_mem_range().
1078 */
1079 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1080 unsigned long *out_start_pfn,
1081 unsigned long *out_end_pfn, int *out_nid)
1082 {
1083 struct memblock_type *type = &memblock.memory;
1084 struct memblock_region *r;
1085
1086 while (++*idx < type->cnt) {
1087 r = &type->regions[*idx];
1088
1089 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1090 continue;
1091 if (nid == MAX_NUMNODES || nid == r->nid)
1092 break;
1093 }
1094 if (*idx >= type->cnt) {
1095 *idx = -1;
1096 return;
1097 }
1098
1099 if (out_start_pfn)
1100 *out_start_pfn = PFN_UP(r->base);
1101 if (out_end_pfn)
1102 *out_end_pfn = PFN_DOWN(r->base + r->size);
1103 if (out_nid)
1104 *out_nid = r->nid;
1105 }
1106
1107 /**
1108 * memblock_set_node - set node ID on memblock regions
1109 * @base: base of area to set node ID for
1110 * @size: size of area to set node ID for
1111 * @type: memblock type to set node ID for
1112 * @nid: node ID to set
1113 *
1114 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1115 * Regions which cross the area boundaries are split as necessary.
1116 *
1117 * RETURNS:
1118 * 0 on success, -errno on failure.
1119 */
1120 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1121 struct memblock_type *type, int nid)
1122 {
1123 int start_rgn, end_rgn;
1124 int i, ret;
1125
1126 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1127 if (ret)
1128 return ret;
1129
1130 for (i = start_rgn; i < end_rgn; i++)
1131 memblock_set_region_node(&type->regions[i], nid);
1132
1133 memblock_merge_regions(type);
1134 return 0;
1135 }
1136 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1137
1138 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1139 phys_addr_t align, phys_addr_t start,
1140 phys_addr_t end, int nid, ulong flags)
1141 {
1142 phys_addr_t found;
1143
1144 if (!align)
1145 align = SMP_CACHE_BYTES;
1146
1147 found = memblock_find_in_range_node(size, align, start, end, nid,
1148 flags);
1149 if (found && !memblock_reserve(found, size)) {
1150 /*
1151 * The min_count is set to 0 so that memblock allocations are
1152 * never reported as leaks.
1153 */
1154 kmemleak_alloc(__va(found), size, 0, 0);
1155 return found;
1156 }
1157 return 0;
1158 }
1159
1160 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1161 phys_addr_t start, phys_addr_t end,
1162 ulong flags)
1163 {
1164 return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1165 flags);
1166 }
1167
1168 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1169 phys_addr_t align, phys_addr_t max_addr,
1170 int nid, ulong flags)
1171 {
1172 return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1173 }
1174
1175 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1176 {
1177 ulong flags = choose_memblock_flags();
1178 phys_addr_t ret;
1179
1180 again:
1181 ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1182 nid, flags);
1183
1184 if (!ret && (flags & MEMBLOCK_MIRROR)) {
1185 flags &= ~MEMBLOCK_MIRROR;
1186 goto again;
1187 }
1188 return ret;
1189 }
1190
1191 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1192 {
1193 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1194 MEMBLOCK_NONE);
1195 }
1196
1197 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1198 {
1199 phys_addr_t alloc;
1200
1201 alloc = __memblock_alloc_base(size, align, max_addr);
1202
1203 if (alloc == 0)
1204 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1205 (unsigned long long) size, (unsigned long long) max_addr);
1206
1207 return alloc;
1208 }
1209
1210 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1211 {
1212 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1213 }
1214
1215 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1216 {
1217 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1218
1219 if (res)
1220 return res;
1221 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1222 }
1223
1224 /**
1225 * memblock_virt_alloc_internal - allocate boot memory block
1226 * @size: size of memory block to be allocated in bytes
1227 * @align: alignment of the region and block's size
1228 * @min_addr: the lower bound of the memory region to allocate (phys address)
1229 * @max_addr: the upper bound of the memory region to allocate (phys address)
1230 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1231 *
1232 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1233 * will fall back to memory below @min_addr. Also, allocation may fall back
1234 * to any node in the system if the specified node can not
1235 * hold the requested memory.
1236 *
1237 * The allocation is performed from memory region limited by
1238 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1239 *
1240 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1241 *
1242 * The phys address of allocated boot memory block is converted to virtual and
1243 * allocated memory is reset to 0.
1244 *
1245 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1246 * allocated boot memory block, so that it is never reported as leaks.
1247 *
1248 * RETURNS:
1249 * Virtual address of allocated memory block on success, NULL on failure.
1250 */
1251 static void * __init memblock_virt_alloc_internal(
1252 phys_addr_t size, phys_addr_t align,
1253 phys_addr_t min_addr, phys_addr_t max_addr,
1254 int nid)
1255 {
1256 phys_addr_t alloc;
1257 void *ptr;
1258 ulong flags = choose_memblock_flags();
1259
1260 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1261 nid = NUMA_NO_NODE;
1262
1263 /*
1264 * Detect any accidental use of these APIs after slab is ready, as at
1265 * this moment memblock may be deinitialized already and its
1266 * internal data may be destroyed (after execution of free_all_bootmem)
1267 */
1268 if (WARN_ON_ONCE(slab_is_available()))
1269 return kzalloc_node(size, GFP_NOWAIT, nid);
1270
1271 if (!align)
1272 align = SMP_CACHE_BYTES;
1273
1274 if (max_addr > memblock.current_limit)
1275 max_addr = memblock.current_limit;
1276
1277 again:
1278 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1279 nid, flags);
1280 if (alloc)
1281 goto done;
1282
1283 if (nid != NUMA_NO_NODE) {
1284 alloc = memblock_find_in_range_node(size, align, min_addr,
1285 max_addr, NUMA_NO_NODE,
1286 flags);
1287 if (alloc)
1288 goto done;
1289 }
1290
1291 if (min_addr) {
1292 min_addr = 0;
1293 goto again;
1294 }
1295
1296 if (flags & MEMBLOCK_MIRROR) {
1297 flags &= ~MEMBLOCK_MIRROR;
1298 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1299 &size);
1300 goto again;
1301 }
1302
1303 return NULL;
1304 done:
1305 memblock_reserve(alloc, size);
1306 ptr = phys_to_virt(alloc);
1307 memset(ptr, 0, size);
1308
1309 /*
1310 * The min_count is set to 0 so that bootmem allocated blocks
1311 * are never reported as leaks. This is because many of these blocks
1312 * are only referred via the physical address which is not
1313 * looked up by kmemleak.
1314 */
1315 kmemleak_alloc(ptr, size, 0, 0);
1316
1317 return ptr;
1318 }
1319
1320 /**
1321 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1322 * @size: size of memory block to be allocated in bytes
1323 * @align: alignment of the region and block's size
1324 * @min_addr: the lower bound of the memory region from where the allocation
1325 * is preferred (phys address)
1326 * @max_addr: the upper bound of the memory region from where the allocation
1327 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1328 * allocate only from memory limited by memblock.current_limit value
1329 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1330 *
1331 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1332 * additional debug information (including caller info), if enabled.
1333 *
1334 * RETURNS:
1335 * Virtual address of allocated memory block on success, NULL on failure.
1336 */
1337 void * __init memblock_virt_alloc_try_nid_nopanic(
1338 phys_addr_t size, phys_addr_t align,
1339 phys_addr_t min_addr, phys_addr_t max_addr,
1340 int nid)
1341 {
1342 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1343 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1344 (u64)max_addr, (void *)_RET_IP_);
1345 return memblock_virt_alloc_internal(size, align, min_addr,
1346 max_addr, nid);
1347 }
1348
1349 /**
1350 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1351 * @size: size of memory block to be allocated in bytes
1352 * @align: alignment of the region and block's size
1353 * @min_addr: the lower bound of the memory region from where the allocation
1354 * is preferred (phys address)
1355 * @max_addr: the upper bound of the memory region from where the allocation
1356 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1357 * allocate only from memory limited by memblock.current_limit value
1358 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1359 *
1360 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1361 * which provides debug information (including caller info), if enabled,
1362 * and panics if the request can not be satisfied.
1363 *
1364 * RETURNS:
1365 * Virtual address of allocated memory block on success, NULL on failure.
1366 */
1367 void * __init memblock_virt_alloc_try_nid(
1368 phys_addr_t size, phys_addr_t align,
1369 phys_addr_t min_addr, phys_addr_t max_addr,
1370 int nid)
1371 {
1372 void *ptr;
1373
1374 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1375 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1376 (u64)max_addr, (void *)_RET_IP_);
1377 ptr = memblock_virt_alloc_internal(size, align,
1378 min_addr, max_addr, nid);
1379 if (ptr)
1380 return ptr;
1381
1382 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1383 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1384 (u64)max_addr);
1385 return NULL;
1386 }
1387
1388 /**
1389 * __memblock_free_early - free boot memory block
1390 * @base: phys starting address of the boot memory block
1391 * @size: size of the boot memory block in bytes
1392 *
1393 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1394 * The freeing memory will not be released to the buddy allocator.
1395 */
1396 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1397 {
1398 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1399 __func__, (u64)base, (u64)base + size - 1,
1400 (void *)_RET_IP_);
1401 kmemleak_free_part(__va(base), size);
1402 memblock_remove_range(&memblock.reserved, base, size);
1403 }
1404
1405 /*
1406 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1407 * @addr: phys starting address of the boot memory block
1408 * @size: size of the boot memory block in bytes
1409 *
1410 * This is only useful when the bootmem allocator has already been torn
1411 * down, but we are still initializing the system. Pages are released directly
1412 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1413 */
1414 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1415 {
1416 u64 cursor, end;
1417
1418 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1419 __func__, (u64)base, (u64)base + size - 1,
1420 (void *)_RET_IP_);
1421 kmemleak_free_part(__va(base), size);
1422 cursor = PFN_UP(base);
1423 end = PFN_DOWN(base + size);
1424
1425 for (; cursor < end; cursor++) {
1426 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1427 totalram_pages++;
1428 }
1429 }
1430
1431 /*
1432 * Remaining API functions
1433 */
1434
1435 phys_addr_t __init memblock_phys_mem_size(void)
1436 {
1437 return memblock.memory.total_size;
1438 }
1439
1440 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1441 {
1442 unsigned long pages = 0;
1443 struct memblock_region *r;
1444 unsigned long start_pfn, end_pfn;
1445
1446 for_each_memblock(memory, r) {
1447 start_pfn = memblock_region_memory_base_pfn(r);
1448 end_pfn = memblock_region_memory_end_pfn(r);
1449 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1450 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1451 pages += end_pfn - start_pfn;
1452 }
1453
1454 return PFN_PHYS(pages);
1455 }
1456
1457 /* lowest address */
1458 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1459 {
1460 return memblock.memory.regions[0].base;
1461 }
1462
1463 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1464 {
1465 int idx = memblock.memory.cnt - 1;
1466
1467 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1468 }
1469
1470 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1471 {
1472 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1473 struct memblock_region *r;
1474
1475 if (!limit)
1476 return;
1477
1478 /* find out max address */
1479 for_each_memblock(memory, r) {
1480 if (limit <= r->size) {
1481 max_addr = r->base + limit;
1482 break;
1483 }
1484 limit -= r->size;
1485 }
1486
1487 /* truncate both memory and reserved regions */
1488 memblock_remove_range(&memblock.memory, max_addr,
1489 (phys_addr_t)ULLONG_MAX);
1490 memblock_remove_range(&memblock.reserved, max_addr,
1491 (phys_addr_t)ULLONG_MAX);
1492 }
1493
1494 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1495 {
1496 unsigned int left = 0, right = type->cnt;
1497
1498 do {
1499 unsigned int mid = (right + left) / 2;
1500
1501 if (addr < type->regions[mid].base)
1502 right = mid;
1503 else if (addr >= (type->regions[mid].base +
1504 type->regions[mid].size))
1505 left = mid + 1;
1506 else
1507 return mid;
1508 } while (left < right);
1509 return -1;
1510 }
1511
1512 int __init memblock_is_reserved(phys_addr_t addr)
1513 {
1514 return memblock_search(&memblock.reserved, addr) != -1;
1515 }
1516
1517 int __init_memblock memblock_is_memory(phys_addr_t addr)
1518 {
1519 return memblock_search(&memblock.memory, addr) != -1;
1520 }
1521
1522 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1523 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1524 unsigned long *start_pfn, unsigned long *end_pfn)
1525 {
1526 struct memblock_type *type = &memblock.memory;
1527 int mid = memblock_search(type, PFN_PHYS(pfn));
1528
1529 if (mid == -1)
1530 return -1;
1531
1532 *start_pfn = PFN_DOWN(type->regions[mid].base);
1533 *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1534
1535 return type->regions[mid].nid;
1536 }
1537 #endif
1538
1539 /**
1540 * memblock_is_region_memory - check if a region is a subset of memory
1541 * @base: base of region to check
1542 * @size: size of region to check
1543 *
1544 * Check if the region [@base, @base+@size) is a subset of a memory block.
1545 *
1546 * RETURNS:
1547 * 0 if false, non-zero if true
1548 */
1549 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1550 {
1551 int idx = memblock_search(&memblock.memory, base);
1552 phys_addr_t end = base + memblock_cap_size(base, &size);
1553
1554 if (idx == -1)
1555 return 0;
1556 return memblock.memory.regions[idx].base <= base &&
1557 (memblock.memory.regions[idx].base +
1558 memblock.memory.regions[idx].size) >= end;
1559 }
1560
1561 /**
1562 * memblock_is_region_reserved - check if a region intersects reserved memory
1563 * @base: base of region to check
1564 * @size: size of region to check
1565 *
1566 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1567 *
1568 * RETURNS:
1569 * True if they intersect, false if not.
1570 */
1571 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1572 {
1573 memblock_cap_size(base, &size);
1574 return memblock_overlaps_region(&memblock.reserved, base, size);
1575 }
1576
1577 void __init_memblock memblock_trim_memory(phys_addr_t align)
1578 {
1579 phys_addr_t start, end, orig_start, orig_end;
1580 struct memblock_region *r;
1581
1582 for_each_memblock(memory, r) {
1583 orig_start = r->base;
1584 orig_end = r->base + r->size;
1585 start = round_up(orig_start, align);
1586 end = round_down(orig_end, align);
1587
1588 if (start == orig_start && end == orig_end)
1589 continue;
1590
1591 if (start < end) {
1592 r->base = start;
1593 r->size = end - start;
1594 } else {
1595 memblock_remove_region(&memblock.memory,
1596 r - memblock.memory.regions);
1597 r--;
1598 }
1599 }
1600 }
1601
1602 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1603 {
1604 memblock.current_limit = limit;
1605 }
1606
1607 phys_addr_t __init_memblock memblock_get_current_limit(void)
1608 {
1609 return memblock.current_limit;
1610 }
1611
1612 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1613 {
1614 unsigned long long base, size;
1615 unsigned long flags;
1616 int i;
1617
1618 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
1619
1620 for (i = 0; i < type->cnt; i++) {
1621 struct memblock_region *rgn = &type->regions[i];
1622 char nid_buf[32] = "";
1623
1624 base = rgn->base;
1625 size = rgn->size;
1626 flags = rgn->flags;
1627 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1628 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1629 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1630 memblock_get_region_node(rgn));
1631 #endif
1632 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1633 name, i, base, base + size - 1, size, nid_buf, flags);
1634 }
1635 }
1636
1637 void __init_memblock __memblock_dump_all(void)
1638 {
1639 pr_info("MEMBLOCK configuration:\n");
1640 pr_info(" memory size = %#llx reserved size = %#llx\n",
1641 (unsigned long long)memblock.memory.total_size,
1642 (unsigned long long)memblock.reserved.total_size);
1643
1644 memblock_dump(&memblock.memory, "memory");
1645 memblock_dump(&memblock.reserved, "reserved");
1646 }
1647
1648 void __init memblock_allow_resize(void)
1649 {
1650 memblock_can_resize = 1;
1651 }
1652
1653 static int __init early_memblock(char *p)
1654 {
1655 if (p && strstr(p, "debug"))
1656 memblock_debug = 1;
1657 return 0;
1658 }
1659 early_param("memblock", early_memblock);
1660
1661 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1662
1663 static int memblock_debug_show(struct seq_file *m, void *private)
1664 {
1665 struct memblock_type *type = m->private;
1666 struct memblock_region *reg;
1667 int i;
1668
1669 for (i = 0; i < type->cnt; i++) {
1670 reg = &type->regions[i];
1671 seq_printf(m, "%4d: ", i);
1672 if (sizeof(phys_addr_t) == 4)
1673 seq_printf(m, "0x%08lx..0x%08lx\n",
1674 (unsigned long)reg->base,
1675 (unsigned long)(reg->base + reg->size - 1));
1676 else
1677 seq_printf(m, "0x%016llx..0x%016llx\n",
1678 (unsigned long long)reg->base,
1679 (unsigned long long)(reg->base + reg->size - 1));
1680
1681 }
1682 return 0;
1683 }
1684
1685 static int memblock_debug_open(struct inode *inode, struct file *file)
1686 {
1687 return single_open(file, memblock_debug_show, inode->i_private);
1688 }
1689
1690 static const struct file_operations memblock_debug_fops = {
1691 .open = memblock_debug_open,
1692 .read = seq_read,
1693 .llseek = seq_lseek,
1694 .release = single_release,
1695 };
1696
1697 static int __init memblock_init_debugfs(void)
1698 {
1699 struct dentry *root = debugfs_create_dir("memblock", NULL);
1700 if (!root)
1701 return -ENXIO;
1702 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1703 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1704 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1705 debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1706 #endif
1707
1708 return 0;
1709 }
1710 __initcall(memblock_init_debugfs);
1711
1712 #endif /* CONFIG_DEBUG_FS */