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
23struct memblock memblock __initdata_memblock;
24
25int memblock_debug __initdata_memblock;
26int memblock_can_resize __initdata_memblock;
27static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS + 1] __initdata_memblock;
28static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS + 1] __initdata_memblock;
29
30/* inline so we don't get a warning when pr_debug is compiled out */
31static inline const char *memblock_type_name(struct memblock_type *type)
32{
33 if (type == &memblock.memory)
34 return "memory";
35 else if (type == &memblock.reserved)
36 return "reserved";
37 else
38 return "unknown";
39}
40
41/*
42 * Address comparison utilities
43 */
44
45static phys_addr_t __init_memblock memblock_align_down(phys_addr_t addr, phys_addr_t size)
46{
47 return addr & ~(size - 1);
48}
49
50static phys_addr_t __init_memblock memblock_align_up(phys_addr_t addr, phys_addr_t size)
51{
52 return (addr + (size - 1)) & ~(size - 1);
53}
54
55static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
56 phys_addr_t base2, phys_addr_t size2)
57{
58 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
59}
60
61long __init_memblock memblock_overlaps_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
62{
63 unsigned long i;
64
65 for (i = 0; i < type->cnt; i++) {
66 phys_addr_t rgnbase = type->regions[i].base;
67 phys_addr_t rgnsize = type->regions[i].size;
68 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
69 break;
70 }
71
72 return (i < type->cnt) ? i : -1;
73}
74
75/*
76 * Find, allocate, deallocate or reserve unreserved regions. All allocations
77 * are top-down.
78 */
79
80static phys_addr_t __init_memblock memblock_find_region(phys_addr_t start, phys_addr_t end,
81 phys_addr_t size, phys_addr_t align)
82{
83 phys_addr_t base, res_base;
84 long j;
85
86 /* In case, huge size is requested */
87 if (end < size)
88 return MEMBLOCK_ERROR;
89
90 base = memblock_align_down((end - size), align);
91
92 /* Prevent allocations returning 0 as it's also used to
93 * indicate an allocation failure
94 */
95 if (start == 0)
96 start = PAGE_SIZE;
97
98 while (start <= base) {
99 j = memblock_overlaps_region(&memblock.reserved, base, size);
100 if (j < 0)
101 return base;
102 res_base = memblock.reserved.regions[j].base;
103 if (res_base < size)
104 break;
105 base = memblock_align_down(res_base - size, align);
106 }
107
108 return MEMBLOCK_ERROR;
109}
110
111static phys_addr_t __init_memblock memblock_find_base(phys_addr_t size,
112 phys_addr_t align, phys_addr_t start, phys_addr_t end)
113{
114 long i;
115
116 BUG_ON(0 == size);
117
118 /* Pump up max_addr */
119 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
120 end = memblock.current_limit;
121
122 /* We do a top-down search, this tends to limit memory
123 * fragmentation by keeping early boot allocs near the
124 * top of memory
125 */
126 for (i = memblock.memory.cnt - 1; i >= 0; i--) {
127 phys_addr_t memblockbase = memblock.memory.regions[i].base;
128 phys_addr_t memblocksize = memblock.memory.regions[i].size;
129 phys_addr_t bottom, top, found;
130
131 if (memblocksize < size)
132 continue;
133 if ((memblockbase + memblocksize) <= start)
134 break;
135 bottom = max(memblockbase, start);
136 top = min(memblockbase + memblocksize, end);
137 if (bottom >= top)
138 continue;
139 found = memblock_find_region(bottom, top, size, align);
140 if (found != MEMBLOCK_ERROR)
141 return found;
142 }
143 return MEMBLOCK_ERROR;
144}
145
146/*
147 * Find a free area with specified alignment in a specific range.
148 */
149u64 __init_memblock memblock_find_in_range(u64 start, u64 end, u64 size, u64 align)
150{
151 return memblock_find_base(size, align, start, end);
152}
153
154/*
155 * Free memblock.reserved.regions
156 */
157int __init_memblock memblock_free_reserved_regions(void)
158{
159 if (memblock.reserved.regions == memblock_reserved_init_regions)
160 return 0;
161
162 return memblock_free(__pa(memblock.reserved.regions),
163 sizeof(struct memblock_region) * memblock.reserved.max);
164}
165
166/*
167 * Reserve memblock.reserved.regions
168 */
169int __init_memblock memblock_reserve_reserved_regions(void)
170{
171 if (memblock.reserved.regions == memblock_reserved_init_regions)
172 return 0;
173
174 return memblock_reserve(__pa(memblock.reserved.regions),
175 sizeof(struct memblock_region) * memblock.reserved.max);
176}
177
178static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
179{
180 unsigned long i;
181
182 for (i = r; i < type->cnt - 1; i++) {
183 type->regions[i].base = type->regions[i + 1].base;
184 type->regions[i].size = type->regions[i + 1].size;
185 }
186 type->cnt--;
187
188 /* Special case for empty arrays */
189 if (type->cnt == 0) {
190 type->cnt = 1;
191 type->regions[0].base = 0;
192 type->regions[0].size = 0;
193 }
194}
195
196/* Defined below but needed now */
197static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size);
198
199static int __init_memblock memblock_double_array(struct memblock_type *type)
200{
201 struct memblock_region *new_array, *old_array;
202 phys_addr_t old_size, new_size, addr;
203 int use_slab = slab_is_available();
204
205 /* We don't allow resizing until we know about the reserved regions
206 * of memory that aren't suitable for allocation
207 */
208 if (!memblock_can_resize)
209 return -1;
210
211 /* Calculate new doubled size */
212 old_size = type->max * sizeof(struct memblock_region);
213 new_size = old_size << 1;
214
215 /* Try to find some space for it.
216 *
217 * WARNING: We assume that either slab_is_available() and we use it or
218 * we use MEMBLOCK for allocations. That means that this is unsafe to use
219 * when bootmem is currently active (unless bootmem itself is implemented
220 * on top of MEMBLOCK which isn't the case yet)
221 *
222 * This should however not be an issue for now, as we currently only
223 * call into MEMBLOCK while it's still active, or much later when slab is
224 * active for memory hotplug operations
225 */
226 if (use_slab) {
227 new_array = kmalloc(new_size, GFP_KERNEL);
228 addr = new_array == NULL ? MEMBLOCK_ERROR : __pa(new_array);
229 } else
230 addr = memblock_find_base(new_size, sizeof(phys_addr_t), 0, MEMBLOCK_ALLOC_ACCESSIBLE);
231 if (addr == MEMBLOCK_ERROR) {
232 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
233 memblock_type_name(type), type->max, type->max * 2);
234 return -1;
235 }
236 new_array = __va(addr);
237
238 memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]",
239 memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1);
240
241 /* Found space, we now need to move the array over before
242 * we add the reserved region since it may be our reserved
243 * array itself that is full.
244 */
245 memcpy(new_array, type->regions, old_size);
246 memset(new_array + type->max, 0, old_size);
247 old_array = type->regions;
248 type->regions = new_array;
249 type->max <<= 1;
250
251 /* If we use SLAB that's it, we are done */
252 if (use_slab)
253 return 0;
254
255 /* Add the new reserved region now. Should not fail ! */
256 BUG_ON(memblock_add_region(&memblock.reserved, addr, new_size));
257
258 /* If the array wasn't our static init one, then free it. We only do
259 * that before SLAB is available as later on, we don't know whether
260 * to use kfree or free_bootmem_pages(). Shouldn't be a big deal
261 * anyways
262 */
263 if (old_array != memblock_memory_init_regions &&
264 old_array != memblock_reserved_init_regions)
265 memblock_free(__pa(old_array), old_size);
266
267 return 0;
268}
269
270extern int __init_memblock __weak memblock_memory_can_coalesce(phys_addr_t addr1, phys_addr_t size1,
271 phys_addr_t addr2, phys_addr_t size2)
272{
273 return 1;
274}
275
276static long __init_memblock memblock_add_region(struct memblock_type *type,
277 phys_addr_t base, phys_addr_t size)
278{
279 phys_addr_t end = base + size;
280 int i, slot = -1;
281
282 /* First try and coalesce this MEMBLOCK with others */
283 for (i = 0; i < type->cnt; i++) {
284 struct memblock_region *rgn = &type->regions[i];
285 phys_addr_t rend = rgn->base + rgn->size;
286
287 /* Exit if there's no possible hits */
288 if (rgn->base > end || rgn->size == 0)
289 break;
290
291 /* Check if we are fully enclosed within an existing
292 * block
293 */
294 if (rgn->base <= base && rend >= end)
295 return 0;
296
297 /* Check if we overlap or are adjacent with the bottom
298 * of a block.
299 */
300 if (base < rgn->base && end >= rgn->base) {
301 /* If we can't coalesce, create a new block */
302 if (!memblock_memory_can_coalesce(base, size,
303 rgn->base,
304 rgn->size)) {
305 /* Overlap & can't coalesce are mutually
306 * exclusive, if you do that, be prepared
307 * for trouble
308 */
309 WARN_ON(end != rgn->base);
310 goto new_block;
311 }
312 /* We extend the bottom of the block down to our
313 * base
314 */
315 rgn->base = base;
316 rgn->size = rend - base;
317
318 /* Return if we have nothing else to allocate
319 * (fully coalesced)
320 */
321 if (rend >= end)
322 return 0;
323
324 /* We continue processing from the end of the
325 * coalesced block.
326 */
327 base = rend;
328 size = end - base;
329 }
330
331 /* Now check if we overlap or are adjacent with the
332 * top of a block
333 */
334 if (base <= rend && end >= rend) {
335 /* If we can't coalesce, create a new block */
336 if (!memblock_memory_can_coalesce(rgn->base,
337 rgn->size,
338 base, size)) {
339 /* Overlap & can't coalesce are mutually
340 * exclusive, if you do that, be prepared
341 * for trouble
342 */
343 WARN_ON(rend != base);
344 goto new_block;
345 }
346 /* We adjust our base down to enclose the
347 * original block and destroy it. It will be
348 * part of our new allocation. Since we've
349 * freed an entry, we know we won't fail
350 * to allocate one later, so we won't risk
351 * losing the original block allocation.
352 */
353 size += (base - rgn->base);
354 base = rgn->base;
355 memblock_remove_region(type, i--);
356 }
357 }
358
359 /* If the array is empty, special case, replace the fake
360 * filler region and return
361 */
362 if ((type->cnt == 1) && (type->regions[0].size == 0)) {
363 type->regions[0].base = base;
364 type->regions[0].size = size;
365 return 0;
366 }
367
368 new_block:
369 /* If we are out of space, we fail. It's too late to resize the array
370 * but then this shouldn't have happened in the first place.
371 */
372 if (WARN_ON(type->cnt >= type->max))
373 return -1;
374
375 /* Couldn't coalesce the MEMBLOCK, so add it to the sorted table. */
376 for (i = type->cnt - 1; i >= 0; i--) {
377 if (base < type->regions[i].base) {
378 type->regions[i+1].base = type->regions[i].base;
379 type->regions[i+1].size = type->regions[i].size;
380 } else {
381 type->regions[i+1].base = base;
382 type->regions[i+1].size = size;
383 slot = i + 1;
384 break;
385 }
386 }
387 if (base < type->regions[0].base) {
388 type->regions[0].base = base;
389 type->regions[0].size = size;
390 slot = 0;
391 }
392 type->cnt++;
393
394 /* The array is full ? Try to resize it. If that fails, we undo
395 * our allocation and return an error
396 */
397 if (type->cnt == type->max && memblock_double_array(type)) {
398 BUG_ON(slot < 0);
399 memblock_remove_region(type, slot);
400 return -1;
401 }
402
403 return 0;
404}
405
406long __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
407{
408 return memblock_add_region(&memblock.memory, base, size);
409
410}
411
412static long __init_memblock __memblock_remove(struct memblock_type *type,
413 phys_addr_t base, phys_addr_t size)
414{
415 phys_addr_t end = base + size;
416 int i;
417
418 /* Walk through the array for collisions */
419 for (i = 0; i < type->cnt; i++) {
420 struct memblock_region *rgn = &type->regions[i];
421 phys_addr_t rend = rgn->base + rgn->size;
422
423 /* Nothing more to do, exit */
424 if (rgn->base > end || rgn->size == 0)
425 break;
426
427 /* If we fully enclose the block, drop it */
428 if (base <= rgn->base && end >= rend) {
429 memblock_remove_region(type, i--);
430 continue;
431 }
432
433 /* If we are fully enclosed within a block
434 * then we need to split it and we are done
435 */
436 if (base > rgn->base && end < rend) {
437 rgn->size = base - rgn->base;
438 if (!memblock_add_region(type, end, rend - end))
439 return 0;
440 /* Failure to split is bad, we at least
441 * restore the block before erroring
442 */
443 rgn->size = rend - rgn->base;
444 WARN_ON(1);
445 return -1;
446 }
447
448 /* Check if we need to trim the bottom of a block */
449 if (rgn->base < end && rend > end) {
450 rgn->size -= end - rgn->base;
451 rgn->base = end;
452 break;
453 }
454
455 /* And check if we need to trim the top of a block */
456 if (base < rend)
457 rgn->size -= rend - base;
458
459 }
460 return 0;
461}
462
463long __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
464{
465 return __memblock_remove(&memblock.memory, base, size);
466}
467
468long __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
469{
470 return __memblock_remove(&memblock.reserved, base, size);
471}
472
473long __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
474{
475 struct memblock_type *_rgn = &memblock.reserved;
476
477 BUG_ON(0 == size);
478
479 return memblock_add_region(_rgn, base, size);
480}
481
482phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
483{
484 phys_addr_t found;
485
486 /* We align the size to limit fragmentation. Without this, a lot of
487 * small allocs quickly eat up the whole reserve array on sparc
488 */
489 size = memblock_align_up(size, align);
490
491 found = memblock_find_base(size, align, 0, max_addr);
492 if (found != MEMBLOCK_ERROR &&
493 !memblock_add_region(&memblock.reserved, found, size))
494 return found;
495
496 return 0;
497}
498
499phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
500{
501 phys_addr_t alloc;
502
503 alloc = __memblock_alloc_base(size, align, max_addr);
504
505 if (alloc == 0)
506 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
507 (unsigned long long) size, (unsigned long long) max_addr);
508
509 return alloc;
510}
511
512phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
513{
514 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
515}
516
517
518/*
519 * Additional node-local allocators. Search for node memory is bottom up
520 * and walks memblock regions within that node bottom-up as well, but allocation
521 * within an memblock region is top-down. XXX I plan to fix that at some stage
522 *
523 * WARNING: Only available after early_node_map[] has been populated,
524 * on some architectures, that is after all the calls to add_active_range()
525 * have been done to populate it.
526 */
527
528phys_addr_t __weak __init memblock_nid_range(phys_addr_t start, phys_addr_t end, int *nid)
529{
530#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
531 /*
532 * This code originates from sparc which really wants use to walk by addresses
533 * and returns the nid. This is not very convenient for early_pfn_map[] users
534 * as the map isn't sorted yet, and it really wants to be walked by nid.
535 *
536 * For now, I implement the inefficient method below which walks the early
537 * map multiple times. Eventually we may want to use an ARCH config option
538 * to implement a completely different method for both case.
539 */
540 unsigned long start_pfn, end_pfn;
541 int i;
542
543 for (i = 0; i < MAX_NUMNODES; i++) {
544 get_pfn_range_for_nid(i, &start_pfn, &end_pfn);
545 if (start < PFN_PHYS(start_pfn) || start >= PFN_PHYS(end_pfn))
546 continue;
547 *nid = i;
548 return min(end, PFN_PHYS(end_pfn));
549 }
550#endif
551 *nid = 0;
552
553 return end;
554}
555
556static phys_addr_t __init memblock_alloc_nid_region(struct memblock_region *mp,
557 phys_addr_t size,
558 phys_addr_t align, int nid)
559{
560 phys_addr_t start, end;
561
562 start = mp->base;
563 end = start + mp->size;
564
565 start = memblock_align_up(start, align);
566 while (start < end) {
567 phys_addr_t this_end;
568 int this_nid;
569
570 this_end = memblock_nid_range(start, end, &this_nid);
571 if (this_nid == nid) {
572 phys_addr_t ret = memblock_find_region(start, this_end, size, align);
573 if (ret != MEMBLOCK_ERROR &&
574 !memblock_add_region(&memblock.reserved, ret, size))
575 return ret;
576 }
577 start = this_end;
578 }
579
580 return MEMBLOCK_ERROR;
581}
582
583phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
584{
585 struct memblock_type *mem = &memblock.memory;
586 int i;
587
588 BUG_ON(0 == size);
589
590 /* We align the size to limit fragmentation. Without this, a lot of
591 * small allocs quickly eat up the whole reserve array on sparc
592 */
593 size = memblock_align_up(size, align);
594
595 /* We do a bottom-up search for a region with the right
596 * nid since that's easier considering how memblock_nid_range()
597 * works
598 */
599 for (i = 0; i < mem->cnt; i++) {
600 phys_addr_t ret = memblock_alloc_nid_region(&mem->regions[i],
601 size, align, nid);
602 if (ret != MEMBLOCK_ERROR)
603 return ret;
604 }
605
606 return 0;
607}
608
609phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
610{
611 phys_addr_t res = memblock_alloc_nid(size, align, nid);
612
613 if (res)
614 return res;
615 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ANYWHERE);
616}
617
618
619/*
620 * Remaining API functions
621 */
622
623/* You must call memblock_analyze() before this. */
624phys_addr_t __init memblock_phys_mem_size(void)
625{
626 return memblock.memory_size;
627}
628
629phys_addr_t __init_memblock memblock_end_of_DRAM(void)
630{
631 int idx = memblock.memory.cnt - 1;
632
633 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
634}
635
636/* You must call memblock_analyze() after this. */
637void __init memblock_enforce_memory_limit(phys_addr_t memory_limit)
638{
639 unsigned long i;
640 phys_addr_t limit;
641 struct memblock_region *p;
642
643 if (!memory_limit)
644 return;
645
646 /* Truncate the memblock regions to satisfy the memory limit. */
647 limit = memory_limit;
648 for (i = 0; i < memblock.memory.cnt; i++) {
649 if (limit > memblock.memory.regions[i].size) {
650 limit -= memblock.memory.regions[i].size;
651 continue;
652 }
653
654 memblock.memory.regions[i].size = limit;
655 memblock.memory.cnt = i + 1;
656 break;
657 }
658
659 memory_limit = memblock_end_of_DRAM();
660
661 /* And truncate any reserves above the limit also. */
662 for (i = 0; i < memblock.reserved.cnt; i++) {
663 p = &memblock.reserved.regions[i];
664
665 if (p->base > memory_limit)
666 p->size = 0;
667 else if ((p->base + p->size) > memory_limit)
668 p->size = memory_limit - p->base;
669
670 if (p->size == 0) {
671 memblock_remove_region(&memblock.reserved, i);
672 i--;
673 }
674 }
675}
676
677static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
678{
679 unsigned int left = 0, right = type->cnt;
680
681 do {
682 unsigned int mid = (right + left) / 2;
683
684 if (addr < type->regions[mid].base)
685 right = mid;
686 else if (addr >= (type->regions[mid].base +
687 type->regions[mid].size))
688 left = mid + 1;
689 else
690 return mid;
691 } while (left < right);
692 return -1;
693}
694
695int __init memblock_is_reserved(phys_addr_t addr)
696{
697 return memblock_search(&memblock.reserved, addr) != -1;
698}
699
700int __init_memblock memblock_is_memory(phys_addr_t addr)
701{
702 return memblock_search(&memblock.memory, addr) != -1;
703}
704
705int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
706{
707 int idx = memblock_search(&memblock.memory, base);
708
709 if (idx == -1)
710 return 0;
711 return memblock.memory.regions[idx].base <= base &&
712 (memblock.memory.regions[idx].base +
713 memblock.memory.regions[idx].size) >= (base + size);
714}
715
716int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
717{
718 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
719}
720
721
722void __init_memblock memblock_set_current_limit(phys_addr_t limit)
723{
724 memblock.current_limit = limit;
725}
726
727static void __init_memblock memblock_dump(struct memblock_type *region, char *name)
728{
729 unsigned long long base, size;
730 int i;
731
732 pr_info(" %s.cnt = 0x%lx\n", name, region->cnt);
733
734 for (i = 0; i < region->cnt; i++) {
735 base = region->regions[i].base;
736 size = region->regions[i].size;
737
738 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes\n",
739 name, i, base, base + size - 1, size);
740 }
741}
742
743void __init_memblock memblock_dump_all(void)
744{
745 if (!memblock_debug)
746 return;
747
748 pr_info("MEMBLOCK configuration:\n");
749 pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size);
750
751 memblock_dump(&memblock.memory, "memory");
752 memblock_dump(&memblock.reserved, "reserved");
753}
754
755void __init memblock_analyze(void)
756{
757 int i;
758
759 /* Check marker in the unused last array entry */
760 WARN_ON(memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS].base
761 != MEMBLOCK_INACTIVE);
762 WARN_ON(memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS].base
763 != MEMBLOCK_INACTIVE);
764
765 memblock.memory_size = 0;
766
767 for (i = 0; i < memblock.memory.cnt; i++)
768 memblock.memory_size += memblock.memory.regions[i].size;
769
770 /* We allow resizing from there */
771 memblock_can_resize = 1;
772}
773
774void __init memblock_init(void)
775{
776 static int init_done __initdata = 0;
777
778 if (init_done)
779 return;
780 init_done = 1;
781
782 /* Hookup the initial arrays */
783 memblock.memory.regions = memblock_memory_init_regions;
784 memblock.memory.max = INIT_MEMBLOCK_REGIONS;
785 memblock.reserved.regions = memblock_reserved_init_regions;
786 memblock.reserved.max = INIT_MEMBLOCK_REGIONS;
787
788 /* Write a marker in the unused last array entry */
789 memblock.memory.regions[INIT_MEMBLOCK_REGIONS].base = MEMBLOCK_INACTIVE;
790 memblock.reserved.regions[INIT_MEMBLOCK_REGIONS].base = MEMBLOCK_INACTIVE;
791
792 /* Create a dummy zero size MEMBLOCK which will get coalesced away later.
793 * This simplifies the memblock_add() code below...
794 */
795 memblock.memory.regions[0].base = 0;
796 memblock.memory.regions[0].size = 0;
797 memblock.memory.cnt = 1;
798
799 /* Ditto. */
800 memblock.reserved.regions[0].base = 0;
801 memblock.reserved.regions[0].size = 0;
802 memblock.reserved.cnt = 1;
803
804 memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE;
805}
806
807static int __init early_memblock(char *p)
808{
809 if (p && strstr(p, "debug"))
810 memblock_debug = 1;
811 return 0;
812}
813early_param("memblock", early_memblock);
814
815#if defined(CONFIG_DEBUG_FS) && !defined(ARCH_DISCARD_MEMBLOCK)
816
817static int memblock_debug_show(struct seq_file *m, void *private)
818{
819 struct memblock_type *type = m->private;
820 struct memblock_region *reg;
821 int i;
822
823 for (i = 0; i < type->cnt; i++) {
824 reg = &type->regions[i];
825 seq_printf(m, "%4d: ", i);
826 if (sizeof(phys_addr_t) == 4)
827 seq_printf(m, "0x%08lx..0x%08lx\n",
828 (unsigned long)reg->base,
829 (unsigned long)(reg->base + reg->size - 1));
830 else
831 seq_printf(m, "0x%016llx..0x%016llx\n",
832 (unsigned long long)reg->base,
833 (unsigned long long)(reg->base + reg->size - 1));
834
835 }
836 return 0;
837}
838
839static int memblock_debug_open(struct inode *inode, struct file *file)
840{
841 return single_open(file, memblock_debug_show, inode->i_private);
842}
843
844static const struct file_operations memblock_debug_fops = {
845 .open = memblock_debug_open,
846 .read = seq_read,
847 .llseek = seq_lseek,
848 .release = single_release,
849};
850
851static int __init memblock_init_debugfs(void)
852{
853 struct dentry *root = debugfs_create_dir("memblock", NULL);
854 if (!root)
855 return -ENXIO;
856 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
857 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
858
859 return 0;
860}
861__initcall(memblock_init_debugfs);
862
863#endif /* CONFIG_DEBUG_FS */
1/*
2 * Procedures for maintaining information about logical memory blocks.
3 *
4 * Peter Bergner, IBM Corp. June 2001.
5 * Copyright (C) 2001 Peter Bergner.
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 */
12
13#include <linux/kernel.h>
14#include <linux/slab.h>
15#include <linux/init.h>
16#include <linux/bitops.h>
17#include <linux/poison.h>
18#include <linux/pfn.h>
19#include <linux/debugfs.h>
20#include <linux/seq_file.h>
21#include <linux/memblock.h>
22
23#include <asm-generic/sections.h>
24#include <linux/io.h>
25
26#include "internal.h"
27
28static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
29static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
30
31struct memblock memblock __initdata_memblock = {
32 .memory.regions = memblock_memory_init_regions,
33 .memory.cnt = 1, /* empty dummy entry */
34 .memory.max = INIT_MEMBLOCK_REGIONS,
35
36 .reserved.regions = memblock_reserved_init_regions,
37 .reserved.cnt = 1, /* empty dummy entry */
38 .reserved.max = INIT_MEMBLOCK_REGIONS,
39
40 .bottom_up = false,
41 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
42};
43
44int memblock_debug __initdata_memblock;
45#ifdef CONFIG_MOVABLE_NODE
46bool movable_node_enabled __initdata_memblock = false;
47#endif
48static int memblock_can_resize __initdata_memblock;
49static int memblock_memory_in_slab __initdata_memblock = 0;
50static int memblock_reserved_in_slab __initdata_memblock = 0;
51
52/* inline so we don't get a warning when pr_debug is compiled out */
53static __init_memblock const char *
54memblock_type_name(struct memblock_type *type)
55{
56 if (type == &memblock.memory)
57 return "memory";
58 else if (type == &memblock.reserved)
59 return "reserved";
60 else
61 return "unknown";
62}
63
64/* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
65static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
66{
67 return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
68}
69
70/*
71 * Address comparison utilities
72 */
73static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
74 phys_addr_t base2, phys_addr_t size2)
75{
76 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
77}
78
79static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
80 phys_addr_t base, phys_addr_t size)
81{
82 unsigned long i;
83
84 for (i = 0; i < type->cnt; i++) {
85 phys_addr_t rgnbase = type->regions[i].base;
86 phys_addr_t rgnsize = type->regions[i].size;
87 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
88 break;
89 }
90
91 return (i < type->cnt) ? i : -1;
92}
93
94/*
95 * __memblock_find_range_bottom_up - find free area utility in bottom-up
96 * @start: start of candidate range
97 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
98 * @size: size of free area to find
99 * @align: alignment of free area to find
100 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
101 *
102 * Utility called from memblock_find_in_range_node(), find free area bottom-up.
103 *
104 * RETURNS:
105 * Found address on success, 0 on failure.
106 */
107static phys_addr_t __init_memblock
108__memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
109 phys_addr_t size, phys_addr_t align, int nid)
110{
111 phys_addr_t this_start, this_end, cand;
112 u64 i;
113
114 for_each_free_mem_range(i, nid, &this_start, &this_end, NULL) {
115 this_start = clamp(this_start, start, end);
116 this_end = clamp(this_end, start, end);
117
118 cand = round_up(this_start, align);
119 if (cand < this_end && this_end - cand >= size)
120 return cand;
121 }
122
123 return 0;
124}
125
126/**
127 * __memblock_find_range_top_down - find free area utility, in top-down
128 * @start: start of candidate range
129 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
130 * @size: size of free area to find
131 * @align: alignment of free area to find
132 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
133 *
134 * Utility called from memblock_find_in_range_node(), find free area top-down.
135 *
136 * RETURNS:
137 * Found address on success, 0 on failure.
138 */
139static phys_addr_t __init_memblock
140__memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
141 phys_addr_t size, phys_addr_t align, int nid)
142{
143 phys_addr_t this_start, this_end, cand;
144 u64 i;
145
146 for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) {
147 this_start = clamp(this_start, start, end);
148 this_end = clamp(this_end, start, end);
149
150 if (this_end < size)
151 continue;
152
153 cand = round_down(this_end - size, align);
154 if (cand >= this_start)
155 return cand;
156 }
157
158 return 0;
159}
160
161/**
162 * memblock_find_in_range_node - find free area in given range and node
163 * @size: size of free area to find
164 * @align: alignment of free area to find
165 * @start: start of candidate range
166 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
167 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
168 *
169 * Find @size free area aligned to @align in the specified range and node.
170 *
171 * When allocation direction is bottom-up, the @start should be greater
172 * than the end of the kernel image. Otherwise, it will be trimmed. The
173 * reason is that we want the bottom-up allocation just near the kernel
174 * image so it is highly likely that the allocated memory and the kernel
175 * will reside in the same node.
176 *
177 * If bottom-up allocation failed, will try to allocate memory top-down.
178 *
179 * RETURNS:
180 * Found address on success, 0 on failure.
181 */
182phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
183 phys_addr_t align, phys_addr_t start,
184 phys_addr_t end, int nid)
185{
186 int ret;
187 phys_addr_t kernel_end;
188
189 /* pump up @end */
190 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
191 end = memblock.current_limit;
192
193 /* avoid allocating the first page */
194 start = max_t(phys_addr_t, start, PAGE_SIZE);
195 end = max(start, end);
196 kernel_end = __pa_symbol(_end);
197
198 /*
199 * try bottom-up allocation only when bottom-up mode
200 * is set and @end is above the kernel image.
201 */
202 if (memblock_bottom_up() && end > kernel_end) {
203 phys_addr_t bottom_up_start;
204
205 /* make sure we will allocate above the kernel */
206 bottom_up_start = max(start, kernel_end);
207
208 /* ok, try bottom-up allocation first */
209 ret = __memblock_find_range_bottom_up(bottom_up_start, end,
210 size, align, nid);
211 if (ret)
212 return ret;
213
214 /*
215 * we always limit bottom-up allocation above the kernel,
216 * but top-down allocation doesn't have the limit, so
217 * retrying top-down allocation may succeed when bottom-up
218 * allocation failed.
219 *
220 * bottom-up allocation is expected to be fail very rarely,
221 * so we use WARN_ONCE() here to see the stack trace if
222 * fail happens.
223 */
224 WARN_ONCE(1, "memblock: bottom-up allocation failed, "
225 "memory hotunplug may be affected\n");
226 }
227
228 return __memblock_find_range_top_down(start, end, size, align, nid);
229}
230
231/**
232 * memblock_find_in_range - find free area in given range
233 * @start: start of candidate range
234 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
235 * @size: size of free area to find
236 * @align: alignment of free area to find
237 *
238 * Find @size free area aligned to @align in the specified range.
239 *
240 * RETURNS:
241 * Found address on success, 0 on failure.
242 */
243phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
244 phys_addr_t end, phys_addr_t size,
245 phys_addr_t align)
246{
247 return memblock_find_in_range_node(size, align, start, end,
248 NUMA_NO_NODE);
249}
250
251static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
252{
253 type->total_size -= type->regions[r].size;
254 memmove(&type->regions[r], &type->regions[r + 1],
255 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
256 type->cnt--;
257
258 /* Special case for empty arrays */
259 if (type->cnt == 0) {
260 WARN_ON(type->total_size != 0);
261 type->cnt = 1;
262 type->regions[0].base = 0;
263 type->regions[0].size = 0;
264 type->regions[0].flags = 0;
265 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
266 }
267}
268
269#ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
270
271phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
272 phys_addr_t *addr)
273{
274 if (memblock.reserved.regions == memblock_reserved_init_regions)
275 return 0;
276
277 *addr = __pa(memblock.reserved.regions);
278
279 return PAGE_ALIGN(sizeof(struct memblock_region) *
280 memblock.reserved.max);
281}
282
283phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
284 phys_addr_t *addr)
285{
286 if (memblock.memory.regions == memblock_memory_init_regions)
287 return 0;
288
289 *addr = __pa(memblock.memory.regions);
290
291 return PAGE_ALIGN(sizeof(struct memblock_region) *
292 memblock.memory.max);
293}
294
295#endif
296
297/**
298 * memblock_double_array - double the size of the memblock regions array
299 * @type: memblock type of the regions array being doubled
300 * @new_area_start: starting address of memory range to avoid overlap with
301 * @new_area_size: size of memory range to avoid overlap with
302 *
303 * Double the size of the @type regions array. If memblock is being used to
304 * allocate memory for a new reserved regions array and there is a previously
305 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
306 * waiting to be reserved, ensure the memory used by the new array does
307 * not overlap.
308 *
309 * RETURNS:
310 * 0 on success, -1 on failure.
311 */
312static int __init_memblock memblock_double_array(struct memblock_type *type,
313 phys_addr_t new_area_start,
314 phys_addr_t new_area_size)
315{
316 struct memblock_region *new_array, *old_array;
317 phys_addr_t old_alloc_size, new_alloc_size;
318 phys_addr_t old_size, new_size, addr;
319 int use_slab = slab_is_available();
320 int *in_slab;
321
322 /* We don't allow resizing until we know about the reserved regions
323 * of memory that aren't suitable for allocation
324 */
325 if (!memblock_can_resize)
326 return -1;
327
328 /* Calculate new doubled size */
329 old_size = type->max * sizeof(struct memblock_region);
330 new_size = old_size << 1;
331 /*
332 * We need to allocated new one align to PAGE_SIZE,
333 * so we can free them completely later.
334 */
335 old_alloc_size = PAGE_ALIGN(old_size);
336 new_alloc_size = PAGE_ALIGN(new_size);
337
338 /* Retrieve the slab flag */
339 if (type == &memblock.memory)
340 in_slab = &memblock_memory_in_slab;
341 else
342 in_slab = &memblock_reserved_in_slab;
343
344 /* Try to find some space for it.
345 *
346 * WARNING: We assume that either slab_is_available() and we use it or
347 * we use MEMBLOCK for allocations. That means that this is unsafe to
348 * use when bootmem is currently active (unless bootmem itself is
349 * implemented on top of MEMBLOCK which isn't the case yet)
350 *
351 * This should however not be an issue for now, as we currently only
352 * call into MEMBLOCK while it's still active, or much later when slab
353 * is active for memory hotplug operations
354 */
355 if (use_slab) {
356 new_array = kmalloc(new_size, GFP_KERNEL);
357 addr = new_array ? __pa(new_array) : 0;
358 } else {
359 /* only exclude range when trying to double reserved.regions */
360 if (type != &memblock.reserved)
361 new_area_start = new_area_size = 0;
362
363 addr = memblock_find_in_range(new_area_start + new_area_size,
364 memblock.current_limit,
365 new_alloc_size, PAGE_SIZE);
366 if (!addr && new_area_size)
367 addr = memblock_find_in_range(0,
368 min(new_area_start, memblock.current_limit),
369 new_alloc_size, PAGE_SIZE);
370
371 new_array = addr ? __va(addr) : NULL;
372 }
373 if (!addr) {
374 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
375 memblock_type_name(type), type->max, type->max * 2);
376 return -1;
377 }
378
379 memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
380 memblock_type_name(type), type->max * 2, (u64)addr,
381 (u64)addr + new_size - 1);
382
383 /*
384 * Found space, we now need to move the array over before we add the
385 * reserved region since it may be our reserved array itself that is
386 * full.
387 */
388 memcpy(new_array, type->regions, old_size);
389 memset(new_array + type->max, 0, old_size);
390 old_array = type->regions;
391 type->regions = new_array;
392 type->max <<= 1;
393
394 /* Free old array. We needn't free it if the array is the static one */
395 if (*in_slab)
396 kfree(old_array);
397 else if (old_array != memblock_memory_init_regions &&
398 old_array != memblock_reserved_init_regions)
399 memblock_free(__pa(old_array), old_alloc_size);
400
401 /*
402 * Reserve the new array if that comes from the memblock. Otherwise, we
403 * needn't do it
404 */
405 if (!use_slab)
406 BUG_ON(memblock_reserve(addr, new_alloc_size));
407
408 /* Update slab flag */
409 *in_slab = use_slab;
410
411 return 0;
412}
413
414/**
415 * memblock_merge_regions - merge neighboring compatible regions
416 * @type: memblock type to scan
417 *
418 * Scan @type and merge neighboring compatible regions.
419 */
420static void __init_memblock memblock_merge_regions(struct memblock_type *type)
421{
422 int i = 0;
423
424 /* cnt never goes below 1 */
425 while (i < type->cnt - 1) {
426 struct memblock_region *this = &type->regions[i];
427 struct memblock_region *next = &type->regions[i + 1];
428
429 if (this->base + this->size != next->base ||
430 memblock_get_region_node(this) !=
431 memblock_get_region_node(next) ||
432 this->flags != next->flags) {
433 BUG_ON(this->base + this->size > next->base);
434 i++;
435 continue;
436 }
437
438 this->size += next->size;
439 /* move forward from next + 1, index of which is i + 2 */
440 memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
441 type->cnt--;
442 }
443}
444
445/**
446 * memblock_insert_region - insert new memblock region
447 * @type: memblock type to insert into
448 * @idx: index for the insertion point
449 * @base: base address of the new region
450 * @size: size of the new region
451 * @nid: node id of the new region
452 * @flags: flags of the new region
453 *
454 * Insert new memblock region [@base,@base+@size) into @type at @idx.
455 * @type must already have extra room to accomodate the new region.
456 */
457static void __init_memblock memblock_insert_region(struct memblock_type *type,
458 int idx, phys_addr_t base,
459 phys_addr_t size,
460 int nid, unsigned long flags)
461{
462 struct memblock_region *rgn = &type->regions[idx];
463
464 BUG_ON(type->cnt >= type->max);
465 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
466 rgn->base = base;
467 rgn->size = size;
468 rgn->flags = flags;
469 memblock_set_region_node(rgn, nid);
470 type->cnt++;
471 type->total_size += size;
472}
473
474/**
475 * memblock_add_region - add new memblock region
476 * @type: memblock type to add new region into
477 * @base: base address of the new region
478 * @size: size of the new region
479 * @nid: nid of the new region
480 * @flags: flags of the new region
481 *
482 * Add new memblock region [@base,@base+@size) into @type. The new region
483 * is allowed to overlap with existing ones - overlaps don't affect already
484 * existing regions. @type is guaranteed to be minimal (all neighbouring
485 * compatible regions are merged) after the addition.
486 *
487 * RETURNS:
488 * 0 on success, -errno on failure.
489 */
490static int __init_memblock memblock_add_region(struct memblock_type *type,
491 phys_addr_t base, phys_addr_t size,
492 int nid, unsigned long flags)
493{
494 bool insert = false;
495 phys_addr_t obase = base;
496 phys_addr_t end = base + memblock_cap_size(base, &size);
497 int i, nr_new;
498
499 if (!size)
500 return 0;
501
502 /* special case for empty array */
503 if (type->regions[0].size == 0) {
504 WARN_ON(type->cnt != 1 || type->total_size);
505 type->regions[0].base = base;
506 type->regions[0].size = size;
507 type->regions[0].flags = flags;
508 memblock_set_region_node(&type->regions[0], nid);
509 type->total_size = size;
510 return 0;
511 }
512repeat:
513 /*
514 * The following is executed twice. Once with %false @insert and
515 * then with %true. The first counts the number of regions needed
516 * to accomodate the new area. The second actually inserts them.
517 */
518 base = obase;
519 nr_new = 0;
520
521 for (i = 0; i < type->cnt; i++) {
522 struct memblock_region *rgn = &type->regions[i];
523 phys_addr_t rbase = rgn->base;
524 phys_addr_t rend = rbase + rgn->size;
525
526 if (rbase >= end)
527 break;
528 if (rend <= base)
529 continue;
530 /*
531 * @rgn overlaps. If it separates the lower part of new
532 * area, insert that portion.
533 */
534 if (rbase > base) {
535 nr_new++;
536 if (insert)
537 memblock_insert_region(type, i++, base,
538 rbase - base, nid,
539 flags);
540 }
541 /* area below @rend is dealt with, forget about it */
542 base = min(rend, end);
543 }
544
545 /* insert the remaining portion */
546 if (base < end) {
547 nr_new++;
548 if (insert)
549 memblock_insert_region(type, i, base, end - base,
550 nid, flags);
551 }
552
553 /*
554 * If this was the first round, resize array and repeat for actual
555 * insertions; otherwise, merge and return.
556 */
557 if (!insert) {
558 while (type->cnt + nr_new > type->max)
559 if (memblock_double_array(type, obase, size) < 0)
560 return -ENOMEM;
561 insert = true;
562 goto repeat;
563 } else {
564 memblock_merge_regions(type);
565 return 0;
566 }
567}
568
569int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
570 int nid)
571{
572 return memblock_add_region(&memblock.memory, base, size, nid, 0);
573}
574
575int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
576{
577 return memblock_add_region(&memblock.memory, base, size,
578 MAX_NUMNODES, 0);
579}
580
581/**
582 * memblock_isolate_range - isolate given range into disjoint memblocks
583 * @type: memblock type to isolate range for
584 * @base: base of range to isolate
585 * @size: size of range to isolate
586 * @start_rgn: out parameter for the start of isolated region
587 * @end_rgn: out parameter for the end of isolated region
588 *
589 * Walk @type and ensure that regions don't cross the boundaries defined by
590 * [@base,@base+@size). Crossing regions are split at the boundaries,
591 * which may create at most two more regions. The index of the first
592 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
593 *
594 * RETURNS:
595 * 0 on success, -errno on failure.
596 */
597static int __init_memblock memblock_isolate_range(struct memblock_type *type,
598 phys_addr_t base, phys_addr_t size,
599 int *start_rgn, int *end_rgn)
600{
601 phys_addr_t end = base + memblock_cap_size(base, &size);
602 int i;
603
604 *start_rgn = *end_rgn = 0;
605
606 if (!size)
607 return 0;
608
609 /* we'll create at most two more regions */
610 while (type->cnt + 2 > type->max)
611 if (memblock_double_array(type, base, size) < 0)
612 return -ENOMEM;
613
614 for (i = 0; i < type->cnt; i++) {
615 struct memblock_region *rgn = &type->regions[i];
616 phys_addr_t rbase = rgn->base;
617 phys_addr_t rend = rbase + rgn->size;
618
619 if (rbase >= end)
620 break;
621 if (rend <= base)
622 continue;
623
624 if (rbase < base) {
625 /*
626 * @rgn intersects from below. Split and continue
627 * to process the next region - the new top half.
628 */
629 rgn->base = base;
630 rgn->size -= base - rbase;
631 type->total_size -= base - rbase;
632 memblock_insert_region(type, i, rbase, base - rbase,
633 memblock_get_region_node(rgn),
634 rgn->flags);
635 } else if (rend > end) {
636 /*
637 * @rgn intersects from above. Split and redo the
638 * current region - the new bottom half.
639 */
640 rgn->base = end;
641 rgn->size -= end - rbase;
642 type->total_size -= end - rbase;
643 memblock_insert_region(type, i--, rbase, end - rbase,
644 memblock_get_region_node(rgn),
645 rgn->flags);
646 } else {
647 /* @rgn is fully contained, record it */
648 if (!*end_rgn)
649 *start_rgn = i;
650 *end_rgn = i + 1;
651 }
652 }
653
654 return 0;
655}
656
657static int __init_memblock __memblock_remove(struct memblock_type *type,
658 phys_addr_t base, phys_addr_t size)
659{
660 int start_rgn, end_rgn;
661 int i, ret;
662
663 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
664 if (ret)
665 return ret;
666
667 for (i = end_rgn - 1; i >= start_rgn; i--)
668 memblock_remove_region(type, i);
669 return 0;
670}
671
672int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
673{
674 return __memblock_remove(&memblock.memory, base, size);
675}
676
677int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
678{
679 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
680 (unsigned long long)base,
681 (unsigned long long)base + size - 1,
682 (void *)_RET_IP_);
683
684 return __memblock_remove(&memblock.reserved, base, size);
685}
686
687static int __init_memblock memblock_reserve_region(phys_addr_t base,
688 phys_addr_t size,
689 int nid,
690 unsigned long flags)
691{
692 struct memblock_type *_rgn = &memblock.reserved;
693
694 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
695 (unsigned long long)base,
696 (unsigned long long)base + size - 1,
697 flags, (void *)_RET_IP_);
698
699 return memblock_add_region(_rgn, base, size, nid, flags);
700}
701
702int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
703{
704 return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
705}
706
707/**
708 * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
709 * @base: the base phys addr of the region
710 * @size: the size of the region
711 *
712 * This function isolates region [@base, @base + @size), and mark it with flag
713 * MEMBLOCK_HOTPLUG.
714 *
715 * Return 0 on succees, -errno on failure.
716 */
717int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
718{
719 struct memblock_type *type = &memblock.memory;
720 int i, ret, start_rgn, end_rgn;
721
722 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
723 if (ret)
724 return ret;
725
726 for (i = start_rgn; i < end_rgn; i++)
727 memblock_set_region_flags(&type->regions[i], MEMBLOCK_HOTPLUG);
728
729 memblock_merge_regions(type);
730 return 0;
731}
732
733/**
734 * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
735 * @base: the base phys addr of the region
736 * @size: the size of the region
737 *
738 * This function isolates region [@base, @base + @size), and clear flag
739 * MEMBLOCK_HOTPLUG for the isolated regions.
740 *
741 * Return 0 on succees, -errno on failure.
742 */
743int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
744{
745 struct memblock_type *type = &memblock.memory;
746 int i, ret, start_rgn, end_rgn;
747
748 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
749 if (ret)
750 return ret;
751
752 for (i = start_rgn; i < end_rgn; i++)
753 memblock_clear_region_flags(&type->regions[i],
754 MEMBLOCK_HOTPLUG);
755
756 memblock_merge_regions(type);
757 return 0;
758}
759
760/**
761 * __next_free_mem_range - next function for for_each_free_mem_range()
762 * @idx: pointer to u64 loop variable
763 * @nid: node selector, %NUMA_NO_NODE for all nodes
764 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
765 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
766 * @out_nid: ptr to int for nid of the range, can be %NULL
767 *
768 * Find the first free area from *@idx which matches @nid, fill the out
769 * parameters, and update *@idx for the next iteration. The lower 32bit of
770 * *@idx contains index into memory region and the upper 32bit indexes the
771 * areas before each reserved region. For example, if reserved regions
772 * look like the following,
773 *
774 * 0:[0-16), 1:[32-48), 2:[128-130)
775 *
776 * The upper 32bit indexes the following regions.
777 *
778 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
779 *
780 * As both region arrays are sorted, the function advances the two indices
781 * in lockstep and returns each intersection.
782 */
783void __init_memblock __next_free_mem_range(u64 *idx, int nid,
784 phys_addr_t *out_start,
785 phys_addr_t *out_end, int *out_nid)
786{
787 struct memblock_type *mem = &memblock.memory;
788 struct memblock_type *rsv = &memblock.reserved;
789 int mi = *idx & 0xffffffff;
790 int ri = *idx >> 32;
791
792 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
793 nid = NUMA_NO_NODE;
794
795 for ( ; mi < mem->cnt; mi++) {
796 struct memblock_region *m = &mem->regions[mi];
797 phys_addr_t m_start = m->base;
798 phys_addr_t m_end = m->base + m->size;
799
800 /* only memory regions are associated with nodes, check it */
801 if (nid != NUMA_NO_NODE && nid != memblock_get_region_node(m))
802 continue;
803
804 /* scan areas before each reservation for intersection */
805 for ( ; ri < rsv->cnt + 1; ri++) {
806 struct memblock_region *r = &rsv->regions[ri];
807 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
808 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
809
810 /* if ri advanced past mi, break out to advance mi */
811 if (r_start >= m_end)
812 break;
813 /* if the two regions intersect, we're done */
814 if (m_start < r_end) {
815 if (out_start)
816 *out_start = max(m_start, r_start);
817 if (out_end)
818 *out_end = min(m_end, r_end);
819 if (out_nid)
820 *out_nid = memblock_get_region_node(m);
821 /*
822 * The region which ends first is advanced
823 * for the next iteration.
824 */
825 if (m_end <= r_end)
826 mi++;
827 else
828 ri++;
829 *idx = (u32)mi | (u64)ri << 32;
830 return;
831 }
832 }
833 }
834
835 /* signal end of iteration */
836 *idx = ULLONG_MAX;
837}
838
839/**
840 * __next_free_mem_range_rev - next function for for_each_free_mem_range_reverse()
841 * @idx: pointer to u64 loop variable
842 * @nid: nid: node selector, %NUMA_NO_NODE for all nodes
843 * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
844 * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
845 * @out_nid: ptr to int for nid of the range, can be %NULL
846 *
847 * Reverse of __next_free_mem_range().
848 *
849 * Linux kernel cannot migrate pages used by itself. Memory hotplug users won't
850 * be able to hot-remove hotpluggable memory used by the kernel. So this
851 * function skip hotpluggable regions if needed when allocating memory for the
852 * kernel.
853 */
854void __init_memblock __next_free_mem_range_rev(u64 *idx, int nid,
855 phys_addr_t *out_start,
856 phys_addr_t *out_end, int *out_nid)
857{
858 struct memblock_type *mem = &memblock.memory;
859 struct memblock_type *rsv = &memblock.reserved;
860 int mi = *idx & 0xffffffff;
861 int ri = *idx >> 32;
862
863 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
864 nid = NUMA_NO_NODE;
865
866 if (*idx == (u64)ULLONG_MAX) {
867 mi = mem->cnt - 1;
868 ri = rsv->cnt;
869 }
870
871 for ( ; mi >= 0; mi--) {
872 struct memblock_region *m = &mem->regions[mi];
873 phys_addr_t m_start = m->base;
874 phys_addr_t m_end = m->base + m->size;
875
876 /* only memory regions are associated with nodes, check it */
877 if (nid != NUMA_NO_NODE && nid != memblock_get_region_node(m))
878 continue;
879
880 /* skip hotpluggable memory regions if needed */
881 if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
882 continue;
883
884 /* scan areas before each reservation for intersection */
885 for ( ; ri >= 0; ri--) {
886 struct memblock_region *r = &rsv->regions[ri];
887 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
888 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
889
890 /* if ri advanced past mi, break out to advance mi */
891 if (r_end <= m_start)
892 break;
893 /* if the two regions intersect, we're done */
894 if (m_end > r_start) {
895 if (out_start)
896 *out_start = max(m_start, r_start);
897 if (out_end)
898 *out_end = min(m_end, r_end);
899 if (out_nid)
900 *out_nid = memblock_get_region_node(m);
901
902 if (m_start >= r_start)
903 mi--;
904 else
905 ri--;
906 *idx = (u32)mi | (u64)ri << 32;
907 return;
908 }
909 }
910 }
911
912 *idx = ULLONG_MAX;
913}
914
915#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
916/*
917 * Common iterator interface used to define for_each_mem_range().
918 */
919void __init_memblock __next_mem_pfn_range(int *idx, int nid,
920 unsigned long *out_start_pfn,
921 unsigned long *out_end_pfn, int *out_nid)
922{
923 struct memblock_type *type = &memblock.memory;
924 struct memblock_region *r;
925
926 while (++*idx < type->cnt) {
927 r = &type->regions[*idx];
928
929 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
930 continue;
931 if (nid == MAX_NUMNODES || nid == r->nid)
932 break;
933 }
934 if (*idx >= type->cnt) {
935 *idx = -1;
936 return;
937 }
938
939 if (out_start_pfn)
940 *out_start_pfn = PFN_UP(r->base);
941 if (out_end_pfn)
942 *out_end_pfn = PFN_DOWN(r->base + r->size);
943 if (out_nid)
944 *out_nid = r->nid;
945}
946
947/**
948 * memblock_set_node - set node ID on memblock regions
949 * @base: base of area to set node ID for
950 * @size: size of area to set node ID for
951 * @type: memblock type to set node ID for
952 * @nid: node ID to set
953 *
954 * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
955 * Regions which cross the area boundaries are split as necessary.
956 *
957 * RETURNS:
958 * 0 on success, -errno on failure.
959 */
960int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
961 struct memblock_type *type, int nid)
962{
963 int start_rgn, end_rgn;
964 int i, ret;
965
966 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
967 if (ret)
968 return ret;
969
970 for (i = start_rgn; i < end_rgn; i++)
971 memblock_set_region_node(&type->regions[i], nid);
972
973 memblock_merge_regions(type);
974 return 0;
975}
976#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
977
978static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
979 phys_addr_t align, phys_addr_t max_addr,
980 int nid)
981{
982 phys_addr_t found;
983
984 if (!align)
985 align = SMP_CACHE_BYTES;
986
987 found = memblock_find_in_range_node(size, align, 0, max_addr, nid);
988 if (found && !memblock_reserve(found, size))
989 return found;
990
991 return 0;
992}
993
994phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
995{
996 return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
997}
998
999phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1000{
1001 return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE);
1002}
1003
1004phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1005{
1006 phys_addr_t alloc;
1007
1008 alloc = __memblock_alloc_base(size, align, max_addr);
1009
1010 if (alloc == 0)
1011 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1012 (unsigned long long) size, (unsigned long long) max_addr);
1013
1014 return alloc;
1015}
1016
1017phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1018{
1019 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1020}
1021
1022phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1023{
1024 phys_addr_t res = memblock_alloc_nid(size, align, nid);
1025
1026 if (res)
1027 return res;
1028 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1029}
1030
1031/**
1032 * memblock_virt_alloc_internal - allocate boot memory block
1033 * @size: size of memory block to be allocated in bytes
1034 * @align: alignment of the region and block's size
1035 * @min_addr: the lower bound of the memory region to allocate (phys address)
1036 * @max_addr: the upper bound of the memory region to allocate (phys address)
1037 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1038 *
1039 * The @min_addr limit is dropped if it can not be satisfied and the allocation
1040 * will fall back to memory below @min_addr. Also, allocation may fall back
1041 * to any node in the system if the specified node can not
1042 * hold the requested memory.
1043 *
1044 * The allocation is performed from memory region limited by
1045 * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1046 *
1047 * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1048 *
1049 * The phys address of allocated boot memory block is converted to virtual and
1050 * allocated memory is reset to 0.
1051 *
1052 * In addition, function sets the min_count to 0 using kmemleak_alloc for
1053 * allocated boot memory block, so that it is never reported as leaks.
1054 *
1055 * RETURNS:
1056 * Virtual address of allocated memory block on success, NULL on failure.
1057 */
1058static void * __init memblock_virt_alloc_internal(
1059 phys_addr_t size, phys_addr_t align,
1060 phys_addr_t min_addr, phys_addr_t max_addr,
1061 int nid)
1062{
1063 phys_addr_t alloc;
1064 void *ptr;
1065
1066 if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1067 nid = NUMA_NO_NODE;
1068
1069 /*
1070 * Detect any accidental use of these APIs after slab is ready, as at
1071 * this moment memblock may be deinitialized already and its
1072 * internal data may be destroyed (after execution of free_all_bootmem)
1073 */
1074 if (WARN_ON_ONCE(slab_is_available()))
1075 return kzalloc_node(size, GFP_NOWAIT, nid);
1076
1077 if (!align)
1078 align = SMP_CACHE_BYTES;
1079
1080 if (max_addr > memblock.current_limit)
1081 max_addr = memblock.current_limit;
1082
1083again:
1084 alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1085 nid);
1086 if (alloc)
1087 goto done;
1088
1089 if (nid != NUMA_NO_NODE) {
1090 alloc = memblock_find_in_range_node(size, align, min_addr,
1091 max_addr, NUMA_NO_NODE);
1092 if (alloc)
1093 goto done;
1094 }
1095
1096 if (min_addr) {
1097 min_addr = 0;
1098 goto again;
1099 } else {
1100 goto error;
1101 }
1102
1103done:
1104 memblock_reserve(alloc, size);
1105 ptr = phys_to_virt(alloc);
1106 memset(ptr, 0, size);
1107
1108 /*
1109 * The min_count is set to 0 so that bootmem allocated blocks
1110 * are never reported as leaks. This is because many of these blocks
1111 * are only referred via the physical address which is not
1112 * looked up by kmemleak.
1113 */
1114 kmemleak_alloc(ptr, size, 0, 0);
1115
1116 return ptr;
1117
1118error:
1119 return NULL;
1120}
1121
1122/**
1123 * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1124 * @size: size of memory block to be allocated in bytes
1125 * @align: alignment of the region and block's size
1126 * @min_addr: the lower bound of the memory region from where the allocation
1127 * is preferred (phys address)
1128 * @max_addr: the upper bound of the memory region from where the allocation
1129 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1130 * allocate only from memory limited by memblock.current_limit value
1131 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1132 *
1133 * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1134 * additional debug information (including caller info), if enabled.
1135 *
1136 * RETURNS:
1137 * Virtual address of allocated memory block on success, NULL on failure.
1138 */
1139void * __init memblock_virt_alloc_try_nid_nopanic(
1140 phys_addr_t size, phys_addr_t align,
1141 phys_addr_t min_addr, phys_addr_t max_addr,
1142 int nid)
1143{
1144 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1145 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1146 (u64)max_addr, (void *)_RET_IP_);
1147 return memblock_virt_alloc_internal(size, align, min_addr,
1148 max_addr, nid);
1149}
1150
1151/**
1152 * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1153 * @size: size of memory block to be allocated in bytes
1154 * @align: alignment of the region and block's size
1155 * @min_addr: the lower bound of the memory region from where the allocation
1156 * is preferred (phys address)
1157 * @max_addr: the upper bound of the memory region from where the allocation
1158 * is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1159 * allocate only from memory limited by memblock.current_limit value
1160 * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1161 *
1162 * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1163 * which provides debug information (including caller info), if enabled,
1164 * and panics if the request can not be satisfied.
1165 *
1166 * RETURNS:
1167 * Virtual address of allocated memory block on success, NULL on failure.
1168 */
1169void * __init memblock_virt_alloc_try_nid(
1170 phys_addr_t size, phys_addr_t align,
1171 phys_addr_t min_addr, phys_addr_t max_addr,
1172 int nid)
1173{
1174 void *ptr;
1175
1176 memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1177 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1178 (u64)max_addr, (void *)_RET_IP_);
1179 ptr = memblock_virt_alloc_internal(size, align,
1180 min_addr, max_addr, nid);
1181 if (ptr)
1182 return ptr;
1183
1184 panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1185 __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1186 (u64)max_addr);
1187 return NULL;
1188}
1189
1190/**
1191 * __memblock_free_early - free boot memory block
1192 * @base: phys starting address of the boot memory block
1193 * @size: size of the boot memory block in bytes
1194 *
1195 * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1196 * The freeing memory will not be released to the buddy allocator.
1197 */
1198void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1199{
1200 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1201 __func__, (u64)base, (u64)base + size - 1,
1202 (void *)_RET_IP_);
1203 kmemleak_free_part(__va(base), size);
1204 __memblock_remove(&memblock.reserved, base, size);
1205}
1206
1207/*
1208 * __memblock_free_late - free bootmem block pages directly to buddy allocator
1209 * @addr: phys starting address of the boot memory block
1210 * @size: size of the boot memory block in bytes
1211 *
1212 * This is only useful when the bootmem allocator has already been torn
1213 * down, but we are still initializing the system. Pages are released directly
1214 * to the buddy allocator, no bootmem metadata is updated because it is gone.
1215 */
1216void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1217{
1218 u64 cursor, end;
1219
1220 memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1221 __func__, (u64)base, (u64)base + size - 1,
1222 (void *)_RET_IP_);
1223 kmemleak_free_part(__va(base), size);
1224 cursor = PFN_UP(base);
1225 end = PFN_DOWN(base + size);
1226
1227 for (; cursor < end; cursor++) {
1228 __free_pages_bootmem(pfn_to_page(cursor), 0);
1229 totalram_pages++;
1230 }
1231}
1232
1233/*
1234 * Remaining API functions
1235 */
1236
1237phys_addr_t __init memblock_phys_mem_size(void)
1238{
1239 return memblock.memory.total_size;
1240}
1241
1242phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1243{
1244 unsigned long pages = 0;
1245 struct memblock_region *r;
1246 unsigned long start_pfn, end_pfn;
1247
1248 for_each_memblock(memory, r) {
1249 start_pfn = memblock_region_memory_base_pfn(r);
1250 end_pfn = memblock_region_memory_end_pfn(r);
1251 start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1252 end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1253 pages += end_pfn - start_pfn;
1254 }
1255
1256 return PFN_PHYS(pages);
1257}
1258
1259/* lowest address */
1260phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1261{
1262 return memblock.memory.regions[0].base;
1263}
1264
1265phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1266{
1267 int idx = memblock.memory.cnt - 1;
1268
1269 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1270}
1271
1272void __init memblock_enforce_memory_limit(phys_addr_t limit)
1273{
1274 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1275 struct memblock_region *r;
1276
1277 if (!limit)
1278 return;
1279
1280 /* find out max address */
1281 for_each_memblock(memory, r) {
1282 if (limit <= r->size) {
1283 max_addr = r->base + limit;
1284 break;
1285 }
1286 limit -= r->size;
1287 }
1288
1289 /* truncate both memory and reserved regions */
1290 __memblock_remove(&memblock.memory, max_addr, (phys_addr_t)ULLONG_MAX);
1291 __memblock_remove(&memblock.reserved, max_addr, (phys_addr_t)ULLONG_MAX);
1292}
1293
1294static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1295{
1296 unsigned int left = 0, right = type->cnt;
1297
1298 do {
1299 unsigned int mid = (right + left) / 2;
1300
1301 if (addr < type->regions[mid].base)
1302 right = mid;
1303 else if (addr >= (type->regions[mid].base +
1304 type->regions[mid].size))
1305 left = mid + 1;
1306 else
1307 return mid;
1308 } while (left < right);
1309 return -1;
1310}
1311
1312int __init memblock_is_reserved(phys_addr_t addr)
1313{
1314 return memblock_search(&memblock.reserved, addr) != -1;
1315}
1316
1317int __init_memblock memblock_is_memory(phys_addr_t addr)
1318{
1319 return memblock_search(&memblock.memory, addr) != -1;
1320}
1321
1322#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1323int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1324 unsigned long *start_pfn, unsigned long *end_pfn)
1325{
1326 struct memblock_type *type = &memblock.memory;
1327 int mid = memblock_search(type, PFN_PHYS(pfn));
1328
1329 if (mid == -1)
1330 return -1;
1331
1332 *start_pfn = type->regions[mid].base >> PAGE_SHIFT;
1333 *end_pfn = (type->regions[mid].base + type->regions[mid].size)
1334 >> PAGE_SHIFT;
1335
1336 return type->regions[mid].nid;
1337}
1338#endif
1339
1340/**
1341 * memblock_is_region_memory - check if a region is a subset of memory
1342 * @base: base of region to check
1343 * @size: size of region to check
1344 *
1345 * Check if the region [@base, @base+@size) is a subset of a memory block.
1346 *
1347 * RETURNS:
1348 * 0 if false, non-zero if true
1349 */
1350int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1351{
1352 int idx = memblock_search(&memblock.memory, base);
1353 phys_addr_t end = base + memblock_cap_size(base, &size);
1354
1355 if (idx == -1)
1356 return 0;
1357 return memblock.memory.regions[idx].base <= base &&
1358 (memblock.memory.regions[idx].base +
1359 memblock.memory.regions[idx].size) >= end;
1360}
1361
1362/**
1363 * memblock_is_region_reserved - check if a region intersects reserved memory
1364 * @base: base of region to check
1365 * @size: size of region to check
1366 *
1367 * Check if the region [@base, @base+@size) intersects a reserved memory block.
1368 *
1369 * RETURNS:
1370 * 0 if false, non-zero if true
1371 */
1372int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1373{
1374 memblock_cap_size(base, &size);
1375 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
1376}
1377
1378void __init_memblock memblock_trim_memory(phys_addr_t align)
1379{
1380 phys_addr_t start, end, orig_start, orig_end;
1381 struct memblock_region *r;
1382
1383 for_each_memblock(memory, r) {
1384 orig_start = r->base;
1385 orig_end = r->base + r->size;
1386 start = round_up(orig_start, align);
1387 end = round_down(orig_end, align);
1388
1389 if (start == orig_start && end == orig_end)
1390 continue;
1391
1392 if (start < end) {
1393 r->base = start;
1394 r->size = end - start;
1395 } else {
1396 memblock_remove_region(&memblock.memory,
1397 r - memblock.memory.regions);
1398 r--;
1399 }
1400 }
1401}
1402
1403void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1404{
1405 memblock.current_limit = limit;
1406}
1407
1408phys_addr_t __init_memblock memblock_get_current_limit(void)
1409{
1410 return memblock.current_limit;
1411}
1412
1413static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1414{
1415 unsigned long long base, size;
1416 unsigned long flags;
1417 int i;
1418
1419 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
1420
1421 for (i = 0; i < type->cnt; i++) {
1422 struct memblock_region *rgn = &type->regions[i];
1423 char nid_buf[32] = "";
1424
1425 base = rgn->base;
1426 size = rgn->size;
1427 flags = rgn->flags;
1428#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1429 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1430 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1431 memblock_get_region_node(rgn));
1432#endif
1433 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1434 name, i, base, base + size - 1, size, nid_buf, flags);
1435 }
1436}
1437
1438void __init_memblock __memblock_dump_all(void)
1439{
1440 pr_info("MEMBLOCK configuration:\n");
1441 pr_info(" memory size = %#llx reserved size = %#llx\n",
1442 (unsigned long long)memblock.memory.total_size,
1443 (unsigned long long)memblock.reserved.total_size);
1444
1445 memblock_dump(&memblock.memory, "memory");
1446 memblock_dump(&memblock.reserved, "reserved");
1447}
1448
1449void __init memblock_allow_resize(void)
1450{
1451 memblock_can_resize = 1;
1452}
1453
1454static int __init early_memblock(char *p)
1455{
1456 if (p && strstr(p, "debug"))
1457 memblock_debug = 1;
1458 return 0;
1459}
1460early_param("memblock", early_memblock);
1461
1462#if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1463
1464static int memblock_debug_show(struct seq_file *m, void *private)
1465{
1466 struct memblock_type *type = m->private;
1467 struct memblock_region *reg;
1468 int i;
1469
1470 for (i = 0; i < type->cnt; i++) {
1471 reg = &type->regions[i];
1472 seq_printf(m, "%4d: ", i);
1473 if (sizeof(phys_addr_t) == 4)
1474 seq_printf(m, "0x%08lx..0x%08lx\n",
1475 (unsigned long)reg->base,
1476 (unsigned long)(reg->base + reg->size - 1));
1477 else
1478 seq_printf(m, "0x%016llx..0x%016llx\n",
1479 (unsigned long long)reg->base,
1480 (unsigned long long)(reg->base + reg->size - 1));
1481
1482 }
1483 return 0;
1484}
1485
1486static int memblock_debug_open(struct inode *inode, struct file *file)
1487{
1488 return single_open(file, memblock_debug_show, inode->i_private);
1489}
1490
1491static const struct file_operations memblock_debug_fops = {
1492 .open = memblock_debug_open,
1493 .read = seq_read,
1494 .llseek = seq_lseek,
1495 .release = single_release,
1496};
1497
1498static int __init memblock_init_debugfs(void)
1499{
1500 struct dentry *root = debugfs_create_dir("memblock", NULL);
1501 if (!root)
1502 return -ENXIO;
1503 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1504 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1505
1506 return 0;
1507}
1508__initcall(memblock_init_debugfs);
1509
1510#endif /* CONFIG_DEBUG_FS */