]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - mm/memblock.c
memblock, x86: Replace memblock_x86_reserve/free_range() with generic ones
[mirror_ubuntu-zesty-kernel.git] / mm / memblock.c
CommitLineData
95f72d1e
YL
1/*
2 * Procedures for maintaining information about logical memory blocks.
3 *
4 * Peter Bergner, IBM Corp. June 2001.
5 * Copyright (C) 2001 Peter Bergner.
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 */
12
13#include <linux/kernel.h>
142b45a7 14#include <linux/slab.h>
95f72d1e
YL
15#include <linux/init.h>
16#include <linux/bitops.h>
449e8df3 17#include <linux/poison.h>
c196f76f 18#include <linux/pfn.h>
6d03b885
BH
19#include <linux/debugfs.h>
20#include <linux/seq_file.h>
95f72d1e
YL
21#include <linux/memblock.h>
22
10d06439 23struct memblock memblock __initdata_memblock;
95f72d1e 24
10d06439
YL
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;
95f72d1e 29
142b45a7
BH
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
6ed311b2
BH
41/*
42 * Address comparison utilities
43 */
10d06439 44static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
2898cc4c 45 phys_addr_t base2, phys_addr_t size2)
95f72d1e
YL
46{
47 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
48}
49
10d06439 50long __init_memblock memblock_overlaps_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size)
6ed311b2
BH
51{
52 unsigned long i;
53
54 for (i = 0; i < type->cnt; i++) {
55 phys_addr_t rgnbase = type->regions[i].base;
56 phys_addr_t rgnsize = type->regions[i].size;
57 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
58 break;
59 }
60
61 return (i < type->cnt) ? i : -1;
62}
63
64/*
65 * Find, allocate, deallocate or reserve unreserved regions. All allocations
66 * are top-down.
67 */
68
cd79481d 69static phys_addr_t __init_memblock memblock_find_region(phys_addr_t start, phys_addr_t end,
6ed311b2
BH
70 phys_addr_t size, phys_addr_t align)
71{
72 phys_addr_t base, res_base;
73 long j;
74
f1af98c7
YL
75 /* In case, huge size is requested */
76 if (end < size)
1f5026a7 77 return 0;
f1af98c7 78
348968eb 79 base = round_down(end - size, align);
f1af98c7 80
25818f0f
BH
81 /* Prevent allocations returning 0 as it's also used to
82 * indicate an allocation failure
83 */
84 if (start == 0)
85 start = PAGE_SIZE;
86
6ed311b2
BH
87 while (start <= base) {
88 j = memblock_overlaps_region(&memblock.reserved, base, size);
89 if (j < 0)
90 return base;
91 res_base = memblock.reserved.regions[j].base;
92 if (res_base < size)
93 break;
348968eb 94 base = round_down(res_base - size, align);
6ed311b2
BH
95 }
96
1f5026a7 97 return 0;
6ed311b2
BH
98}
99
fc769a8e
TH
100/*
101 * Find a free area with specified alignment in a specific range.
102 */
103phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start, phys_addr_t end,
104 phys_addr_t size, phys_addr_t align)
6ed311b2
BH
105{
106 long i;
6ed311b2
BH
107
108 BUG_ON(0 == size);
109
6ed311b2 110 /* Pump up max_addr */
fef501d4
BH
111 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
112 end = memblock.current_limit;
6ed311b2
BH
113
114 /* We do a top-down search, this tends to limit memory
115 * fragmentation by keeping early boot allocs near the
116 * top of memory
117 */
118 for (i = memblock.memory.cnt - 1; i >= 0; i--) {
119 phys_addr_t memblockbase = memblock.memory.regions[i].base;
120 phys_addr_t memblocksize = memblock.memory.regions[i].size;
fef501d4 121 phys_addr_t bottom, top, found;
6ed311b2
BH
122
123 if (memblocksize < size)
124 continue;
fef501d4
BH
125 if ((memblockbase + memblocksize) <= start)
126 break;
127 bottom = max(memblockbase, start);
128 top = min(memblockbase + memblocksize, end);
129 if (bottom >= top)
130 continue;
131 found = memblock_find_region(bottom, top, size, align);
1f5026a7 132 if (found)
fef501d4 133 return found;
6ed311b2 134 }
1f5026a7 135 return 0;
6ed311b2
BH
136}
137
7950c407
YL
138/*
139 * Free memblock.reserved.regions
140 */
141int __init_memblock memblock_free_reserved_regions(void)
142{
143 if (memblock.reserved.regions == memblock_reserved_init_regions)
144 return 0;
145
146 return memblock_free(__pa(memblock.reserved.regions),
147 sizeof(struct memblock_region) * memblock.reserved.max);
148}
149
150/*
151 * Reserve memblock.reserved.regions
152 */
153int __init_memblock memblock_reserve_reserved_regions(void)
154{
155 if (memblock.reserved.regions == memblock_reserved_init_regions)
156 return 0;
157
158 return memblock_reserve(__pa(memblock.reserved.regions),
159 sizeof(struct memblock_region) * memblock.reserved.max);
160}
161
10d06439 162static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
95f72d1e 163{
7c0caeb8
TH
164 memmove(&type->regions[r], &type->regions[r + 1],
165 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
e3239ff9 166 type->cnt--;
95f72d1e 167
8f7a6605
BH
168 /* Special case for empty arrays */
169 if (type->cnt == 0) {
170 type->cnt = 1;
171 type->regions[0].base = 0;
172 type->regions[0].size = 0;
7c0caeb8 173 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
8f7a6605 174 }
95f72d1e
YL
175}
176
142b45a7
BH
177/* Defined below but needed now */
178static long memblock_add_region(struct memblock_type *type, phys_addr_t base, phys_addr_t size);
179
10d06439 180static int __init_memblock memblock_double_array(struct memblock_type *type)
142b45a7
BH
181{
182 struct memblock_region *new_array, *old_array;
183 phys_addr_t old_size, new_size, addr;
184 int use_slab = slab_is_available();
185
186 /* We don't allow resizing until we know about the reserved regions
187 * of memory that aren't suitable for allocation
188 */
189 if (!memblock_can_resize)
190 return -1;
191
142b45a7
BH
192 /* Calculate new doubled size */
193 old_size = type->max * sizeof(struct memblock_region);
194 new_size = old_size << 1;
195
196 /* Try to find some space for it.
197 *
198 * WARNING: We assume that either slab_is_available() and we use it or
199 * we use MEMBLOCK for allocations. That means that this is unsafe to use
200 * when bootmem is currently active (unless bootmem itself is implemented
201 * on top of MEMBLOCK which isn't the case yet)
202 *
203 * This should however not be an issue for now, as we currently only
204 * call into MEMBLOCK while it's still active, or much later when slab is
205 * active for memory hotplug operations
206 */
207 if (use_slab) {
208 new_array = kmalloc(new_size, GFP_KERNEL);
1f5026a7 209 addr = new_array ? __pa(new_array) : 0;
142b45a7 210 } else
fc769a8e 211 addr = memblock_find_in_range(0, MEMBLOCK_ALLOC_ACCESSIBLE, new_size, sizeof(phys_addr_t));
1f5026a7 212 if (!addr) {
142b45a7
BH
213 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
214 memblock_type_name(type), type->max, type->max * 2);
215 return -1;
216 }
217 new_array = __va(addr);
218
ea9e4376
YL
219 memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]",
220 memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1);
221
142b45a7
BH
222 /* Found space, we now need to move the array over before
223 * we add the reserved region since it may be our reserved
224 * array itself that is full.
225 */
226 memcpy(new_array, type->regions, old_size);
227 memset(new_array + type->max, 0, old_size);
228 old_array = type->regions;
229 type->regions = new_array;
230 type->max <<= 1;
231
232 /* If we use SLAB that's it, we are done */
233 if (use_slab)
234 return 0;
235
236 /* Add the new reserved region now. Should not fail ! */
8f7a6605 237 BUG_ON(memblock_add_region(&memblock.reserved, addr, new_size));
142b45a7
BH
238
239 /* If the array wasn't our static init one, then free it. We only do
240 * that before SLAB is available as later on, we don't know whether
241 * to use kfree or free_bootmem_pages(). Shouldn't be a big deal
242 * anyways
243 */
244 if (old_array != memblock_memory_init_regions &&
245 old_array != memblock_reserved_init_regions)
246 memblock_free(__pa(old_array), old_size);
247
248 return 0;
249}
250
784656f9
TH
251/**
252 * memblock_merge_regions - merge neighboring compatible regions
253 * @type: memblock type to scan
254 *
255 * Scan @type and merge neighboring compatible regions.
256 */
257static void __init_memblock memblock_merge_regions(struct memblock_type *type)
95f72d1e 258{
784656f9 259 int i = 0;
95f72d1e 260
784656f9
TH
261 /* cnt never goes below 1 */
262 while (i < type->cnt - 1) {
263 struct memblock_region *this = &type->regions[i];
264 struct memblock_region *next = &type->regions[i + 1];
95f72d1e 265
7c0caeb8
TH
266 if (this->base + this->size != next->base ||
267 memblock_get_region_node(this) !=
268 memblock_get_region_node(next)) {
784656f9
TH
269 BUG_ON(this->base + this->size > next->base);
270 i++;
271 continue;
8f7a6605
BH
272 }
273
784656f9
TH
274 this->size += next->size;
275 memmove(next, next + 1, (type->cnt - (i + 1)) * sizeof(*next));
276 type->cnt--;
95f72d1e 277 }
784656f9 278}
95f72d1e 279
784656f9
TH
280/**
281 * memblock_insert_region - insert new memblock region
282 * @type: memblock type to insert into
283 * @idx: index for the insertion point
284 * @base: base address of the new region
285 * @size: size of the new region
286 *
287 * Insert new memblock region [@base,@base+@size) into @type at @idx.
288 * @type must already have extra room to accomodate the new region.
289 */
290static void __init_memblock memblock_insert_region(struct memblock_type *type,
291 int idx, phys_addr_t base,
7c0caeb8 292 phys_addr_t size, int nid)
784656f9
TH
293{
294 struct memblock_region *rgn = &type->regions[idx];
295
296 BUG_ON(type->cnt >= type->max);
297 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
298 rgn->base = base;
299 rgn->size = size;
7c0caeb8 300 memblock_set_region_node(rgn, nid);
784656f9
TH
301 type->cnt++;
302}
303
304/**
305 * memblock_add_region - add new memblock region
306 * @type: memblock type to add new region into
307 * @base: base address of the new region
308 * @size: size of the new region
309 *
310 * Add new memblock region [@base,@base+@size) into @type. The new region
311 * is allowed to overlap with existing ones - overlaps don't affect already
312 * existing regions. @type is guaranteed to be minimal (all neighbouring
313 * compatible regions are merged) after the addition.
314 *
315 * RETURNS:
316 * 0 on success, -errno on failure.
317 */
318static long __init_memblock memblock_add_region(struct memblock_type *type,
319 phys_addr_t base, phys_addr_t size)
320{
321 bool insert = false;
322 phys_addr_t obase = base, end = base + size;
323 int i, nr_new;
324
325 /* special case for empty array */
326 if (type->regions[0].size == 0) {
327 WARN_ON(type->cnt != 1);
8f7a6605
BH
328 type->regions[0].base = base;
329 type->regions[0].size = size;
7c0caeb8 330 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
8f7a6605 331 return 0;
95f72d1e 332 }
784656f9
TH
333repeat:
334 /*
335 * The following is executed twice. Once with %false @insert and
336 * then with %true. The first counts the number of regions needed
337 * to accomodate the new area. The second actually inserts them.
142b45a7 338 */
784656f9
TH
339 base = obase;
340 nr_new = 0;
95f72d1e 341
784656f9
TH
342 for (i = 0; i < type->cnt; i++) {
343 struct memblock_region *rgn = &type->regions[i];
344 phys_addr_t rbase = rgn->base;
345 phys_addr_t rend = rbase + rgn->size;
346
347 if (rbase >= end)
95f72d1e 348 break;
784656f9
TH
349 if (rend <= base)
350 continue;
351 /*
352 * @rgn overlaps. If it separates the lower part of new
353 * area, insert that portion.
354 */
355 if (rbase > base) {
356 nr_new++;
357 if (insert)
358 memblock_insert_region(type, i++, base,
7c0caeb8 359 rbase - base, MAX_NUMNODES);
95f72d1e 360 }
784656f9
TH
361 /* area below @rend is dealt with, forget about it */
362 base = min(rend, end);
95f72d1e 363 }
784656f9
TH
364
365 /* insert the remaining portion */
366 if (base < end) {
367 nr_new++;
368 if (insert)
7c0caeb8
TH
369 memblock_insert_region(type, i, base, end - base,
370 MAX_NUMNODES);
95f72d1e 371 }
95f72d1e 372
784656f9
TH
373 /*
374 * If this was the first round, resize array and repeat for actual
375 * insertions; otherwise, merge and return.
142b45a7 376 */
784656f9
TH
377 if (!insert) {
378 while (type->cnt + nr_new > type->max)
379 if (memblock_double_array(type) < 0)
380 return -ENOMEM;
381 insert = true;
382 goto repeat;
383 } else {
384 memblock_merge_regions(type);
385 return 0;
142b45a7 386 }
95f72d1e
YL
387}
388
10d06439 389long __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
95f72d1e 390{
e3239ff9 391 return memblock_add_region(&memblock.memory, base, size);
95f72d1e
YL
392}
393
8f7a6605
BH
394static long __init_memblock __memblock_remove(struct memblock_type *type,
395 phys_addr_t base, phys_addr_t size)
95f72d1e 396{
2898cc4c 397 phys_addr_t end = base + size;
95f72d1e
YL
398 int i;
399
8f7a6605
BH
400 /* Walk through the array for collisions */
401 for (i = 0; i < type->cnt; i++) {
402 struct memblock_region *rgn = &type->regions[i];
403 phys_addr_t rend = rgn->base + rgn->size;
95f72d1e 404
8f7a6605
BH
405 /* Nothing more to do, exit */
406 if (rgn->base > end || rgn->size == 0)
95f72d1e 407 break;
95f72d1e 408
8f7a6605
BH
409 /* If we fully enclose the block, drop it */
410 if (base <= rgn->base && end >= rend) {
411 memblock_remove_region(type, i--);
412 continue;
413 }
95f72d1e 414
8f7a6605
BH
415 /* If we are fully enclosed within a block
416 * then we need to split it and we are done
417 */
418 if (base > rgn->base && end < rend) {
419 rgn->size = base - rgn->base;
420 if (!memblock_add_region(type, end, rend - end))
421 return 0;
422 /* Failure to split is bad, we at least
423 * restore the block before erroring
424 */
425 rgn->size = rend - rgn->base;
426 WARN_ON(1);
427 return -1;
428 }
95f72d1e 429
8f7a6605
BH
430 /* Check if we need to trim the bottom of a block */
431 if (rgn->base < end && rend > end) {
432 rgn->size -= end - rgn->base;
433 rgn->base = end;
434 break;
435 }
95f72d1e 436
8f7a6605
BH
437 /* And check if we need to trim the top of a block */
438 if (base < rend)
439 rgn->size -= rend - base;
95f72d1e 440
8f7a6605
BH
441 }
442 return 0;
95f72d1e
YL
443}
444
10d06439 445long __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
95f72d1e
YL
446{
447 return __memblock_remove(&memblock.memory, base, size);
448}
449
3661ca66 450long __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
95f72d1e 451{
24aa0788
TH
452 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
453 base, base + size, (void *)_RET_IP_);
454
95f72d1e
YL
455 return __memblock_remove(&memblock.reserved, base, size);
456}
457
3661ca66 458long __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
95f72d1e 459{
e3239ff9 460 struct memblock_type *_rgn = &memblock.reserved;
95f72d1e 461
24aa0788
TH
462 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] %pF\n",
463 base, base + size, (void *)_RET_IP_);
95f72d1e
YL
464 BUG_ON(0 == size);
465
466 return memblock_add_region(_rgn, base, size);
467}
468
35fd0808
TH
469/**
470 * __next_free_mem_range - next function for for_each_free_mem_range()
471 * @idx: pointer to u64 loop variable
472 * @nid: nid: node selector, %MAX_NUMNODES for all nodes
473 * @p_start: ptr to phys_addr_t for start address of the range, can be %NULL
474 * @p_end: ptr to phys_addr_t for end address of the range, can be %NULL
475 * @p_nid: ptr to int for nid of the range, can be %NULL
476 *
477 * Find the first free area from *@idx which matches @nid, fill the out
478 * parameters, and update *@idx for the next iteration. The lower 32bit of
479 * *@idx contains index into memory region and the upper 32bit indexes the
480 * areas before each reserved region. For example, if reserved regions
481 * look like the following,
482 *
483 * 0:[0-16), 1:[32-48), 2:[128-130)
484 *
485 * The upper 32bit indexes the following regions.
486 *
487 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
488 *
489 * As both region arrays are sorted, the function advances the two indices
490 * in lockstep and returns each intersection.
491 */
492void __init_memblock __next_free_mem_range(u64 *idx, int nid,
493 phys_addr_t *out_start,
494 phys_addr_t *out_end, int *out_nid)
495{
496 struct memblock_type *mem = &memblock.memory;
497 struct memblock_type *rsv = &memblock.reserved;
498 int mi = *idx & 0xffffffff;
499 int ri = *idx >> 32;
500
501 for ( ; mi < mem->cnt; mi++) {
502 struct memblock_region *m = &mem->regions[mi];
503 phys_addr_t m_start = m->base;
504 phys_addr_t m_end = m->base + m->size;
505
506 /* only memory regions are associated with nodes, check it */
507 if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m))
508 continue;
509
510 /* scan areas before each reservation for intersection */
511 for ( ; ri < rsv->cnt + 1; ri++) {
512 struct memblock_region *r = &rsv->regions[ri];
513 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
514 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
515
516 /* if ri advanced past mi, break out to advance mi */
517 if (r_start >= m_end)
518 break;
519 /* if the two regions intersect, we're done */
520 if (m_start < r_end) {
521 if (out_start)
522 *out_start = max(m_start, r_start);
523 if (out_end)
524 *out_end = min(m_end, r_end);
525 if (out_nid)
526 *out_nid = memblock_get_region_node(m);
527 /*
528 * The region which ends first is advanced
529 * for the next iteration.
530 */
531 if (m_end <= r_end)
532 mi++;
533 else
534 ri++;
535 *idx = (u32)mi | (u64)ri << 32;
536 return;
537 }
538 }
539 }
540
541 /* signal end of iteration */
542 *idx = ULLONG_MAX;
543}
544
7c0caeb8
TH
545#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
546/*
547 * Common iterator interface used to define for_each_mem_range().
548 */
549void __init_memblock __next_mem_pfn_range(int *idx, int nid,
550 unsigned long *out_start_pfn,
551 unsigned long *out_end_pfn, int *out_nid)
552{
553 struct memblock_type *type = &memblock.memory;
554 struct memblock_region *r;
555
556 while (++*idx < type->cnt) {
557 r = &type->regions[*idx];
558
559 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
560 continue;
561 if (nid == MAX_NUMNODES || nid == r->nid)
562 break;
563 }
564 if (*idx >= type->cnt) {
565 *idx = -1;
566 return;
567 }
568
569 if (out_start_pfn)
570 *out_start_pfn = PFN_UP(r->base);
571 if (out_end_pfn)
572 *out_end_pfn = PFN_DOWN(r->base + r->size);
573 if (out_nid)
574 *out_nid = r->nid;
575}
576
577/**
578 * memblock_set_node - set node ID on memblock regions
579 * @base: base of area to set node ID for
580 * @size: size of area to set node ID for
581 * @nid: node ID to set
582 *
583 * Set the nid of memblock memory regions in [@base,@base+@size) to @nid.
584 * Regions which cross the area boundaries are split as necessary.
585 *
586 * RETURNS:
587 * 0 on success, -errno on failure.
588 */
589int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
590 int nid)
591{
592 struct memblock_type *type = &memblock.memory;
593 phys_addr_t end = base + size;
594 int i;
595
596 /* we'll create at most two more regions */
597 while (type->cnt + 2 > type->max)
598 if (memblock_double_array(type) < 0)
599 return -ENOMEM;
600
601 for (i = 0; i < type->cnt; i++) {
602 struct memblock_region *rgn = &type->regions[i];
603 phys_addr_t rbase = rgn->base;
604 phys_addr_t rend = rbase + rgn->size;
605
606 if (rbase >= end)
607 break;
608 if (rend <= base)
609 continue;
610
611 if (rbase < base) {
612 /*
613 * @rgn intersects from below. Split and continue
614 * to process the next region - the new top half.
615 */
616 rgn->base = base;
617 rgn->size = rend - rgn->base;
618 memblock_insert_region(type, i, rbase, base - rbase,
619 rgn->nid);
620 } else if (rend > end) {
621 /*
622 * @rgn intersects from above. Split and redo the
623 * current region - the new bottom half.
624 */
625 rgn->base = end;
626 rgn->size = rend - rgn->base;
627 memblock_insert_region(type, i--, rbase, end - rbase,
628 rgn->nid);
629 } else {
630 /* @rgn is fully contained, set ->nid */
631 rgn->nid = nid;
632 }
633 }
634
635 memblock_merge_regions(type);
636 return 0;
637}
638#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
639
6ed311b2 640phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
95f72d1e 641{
6ed311b2 642 phys_addr_t found;
95f72d1e 643
6ed311b2
BH
644 /* We align the size to limit fragmentation. Without this, a lot of
645 * small allocs quickly eat up the whole reserve array on sparc
646 */
348968eb 647 size = round_up(size, align);
95f72d1e 648
fc769a8e 649 found = memblock_find_in_range(0, max_addr, size, align);
1f5026a7 650 if (found && !memblock_add_region(&memblock.reserved, found, size))
6ed311b2 651 return found;
95f72d1e 652
6ed311b2 653 return 0;
95f72d1e
YL
654}
655
6ed311b2 656phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
95f72d1e 657{
6ed311b2
BH
658 phys_addr_t alloc;
659
660 alloc = __memblock_alloc_base(size, align, max_addr);
661
662 if (alloc == 0)
663 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
664 (unsigned long long) size, (unsigned long long) max_addr);
665
666 return alloc;
95f72d1e
YL
667}
668
6ed311b2 669phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
95f72d1e 670{
6ed311b2
BH
671 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
672}
95f72d1e 673
95f72d1e 674
6ed311b2 675/*
34e18455 676 * Additional node-local top-down allocators.
c196f76f
BH
677 *
678 * WARNING: Only available after early_node_map[] has been populated,
679 * on some architectures, that is after all the calls to add_active_range()
680 * have been done to populate it.
6ed311b2 681 */
95f72d1e 682
34e18455
TH
683static phys_addr_t __init memblock_nid_range_rev(phys_addr_t start,
684 phys_addr_t end, int *nid)
c3f72b57 685{
c196f76f 686#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
c196f76f
BH
687 unsigned long start_pfn, end_pfn;
688 int i;
689
b2fea988 690 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, nid)
34e18455
TH
691 if (end > PFN_PHYS(start_pfn) && end <= PFN_PHYS(end_pfn))
692 return max(start, PFN_PHYS(start_pfn));
c196f76f 693#endif
c3f72b57 694 *nid = 0;
34e18455 695 return start;
c3f72b57
BH
696}
697
e6498040
TH
698phys_addr_t __init memblock_find_in_range_node(phys_addr_t start,
699 phys_addr_t end,
2898cc4c
BH
700 phys_addr_t size,
701 phys_addr_t align, int nid)
95f72d1e 702{
e6498040
TH
703 struct memblock_type *mem = &memblock.memory;
704 int i;
95f72d1e 705
e6498040 706 BUG_ON(0 == size);
95f72d1e 707
e6498040
TH
708 /* Pump up max_addr */
709 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
710 end = memblock.current_limit;
95f72d1e 711
e6498040
TH
712 for (i = mem->cnt - 1; i >= 0; i--) {
713 struct memblock_region *r = &mem->regions[i];
714 phys_addr_t base = max(start, r->base);
715 phys_addr_t top = min(end, r->base + r->size);
716
717 while (base < top) {
718 phys_addr_t tbase, ret;
719 int tnid;
720
721 tbase = memblock_nid_range_rev(base, top, &tnid);
722 if (nid == MAX_NUMNODES || tnid == nid) {
723 ret = memblock_find_region(tbase, top, size, align);
724 if (ret)
725 return ret;
726 }
727 top = tbase;
95f72d1e 728 }
95f72d1e 729 }
e6498040 730
1f5026a7 731 return 0;
95f72d1e
YL
732}
733
2898cc4c 734phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
95f72d1e 735{
e6498040 736 phys_addr_t found;
95f72d1e 737
e6498040
TH
738 /*
739 * We align the size to limit fragmentation. Without this, a lot of
7f219c73
BH
740 * small allocs quickly eat up the whole reserve array on sparc
741 */
348968eb 742 size = round_up(size, align);
7f219c73 743
e6498040
TH
744 found = memblock_find_in_range_node(0, MEMBLOCK_ALLOC_ACCESSIBLE,
745 size, align, nid);
746 if (found && !memblock_add_region(&memblock.reserved, found, size))
747 return found;
95f72d1e 748
9d1e2492
BH
749 return 0;
750}
751
752phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
753{
754 phys_addr_t res = memblock_alloc_nid(size, align, nid);
755
756 if (res)
757 return res;
15fb0972 758 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
95f72d1e
YL
759}
760
9d1e2492
BH
761
762/*
763 * Remaining API functions
764 */
765
95f72d1e 766/* You must call memblock_analyze() before this. */
2898cc4c 767phys_addr_t __init memblock_phys_mem_size(void)
95f72d1e 768{
4734b594 769 return memblock.memory_size;
95f72d1e
YL
770}
771
10d06439 772phys_addr_t __init_memblock memblock_end_of_DRAM(void)
95f72d1e
YL
773{
774 int idx = memblock.memory.cnt - 1;
775
e3239ff9 776 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
95f72d1e
YL
777}
778
779/* You must call memblock_analyze() after this. */
2898cc4c 780void __init memblock_enforce_memory_limit(phys_addr_t memory_limit)
95f72d1e
YL
781{
782 unsigned long i;
2898cc4c 783 phys_addr_t limit;
e3239ff9 784 struct memblock_region *p;
95f72d1e
YL
785
786 if (!memory_limit)
787 return;
788
789 /* Truncate the memblock regions to satisfy the memory limit. */
790 limit = memory_limit;
791 for (i = 0; i < memblock.memory.cnt; i++) {
e3239ff9
BH
792 if (limit > memblock.memory.regions[i].size) {
793 limit -= memblock.memory.regions[i].size;
95f72d1e
YL
794 continue;
795 }
796
e3239ff9 797 memblock.memory.regions[i].size = limit;
95f72d1e
YL
798 memblock.memory.cnt = i + 1;
799 break;
800 }
801
95f72d1e
YL
802 memory_limit = memblock_end_of_DRAM();
803
804 /* And truncate any reserves above the limit also. */
805 for (i = 0; i < memblock.reserved.cnt; i++) {
e3239ff9 806 p = &memblock.reserved.regions[i];
95f72d1e
YL
807
808 if (p->base > memory_limit)
809 p->size = 0;
810 else if ((p->base + p->size) > memory_limit)
811 p->size = memory_limit - p->base;
812
813 if (p->size == 0) {
814 memblock_remove_region(&memblock.reserved, i);
815 i--;
816 }
817 }
818}
819
cd79481d 820static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
72d4b0b4
BH
821{
822 unsigned int left = 0, right = type->cnt;
823
824 do {
825 unsigned int mid = (right + left) / 2;
826
827 if (addr < type->regions[mid].base)
828 right = mid;
829 else if (addr >= (type->regions[mid].base +
830 type->regions[mid].size))
831 left = mid + 1;
832 else
833 return mid;
834 } while (left < right);
835 return -1;
836}
837
2898cc4c 838int __init memblock_is_reserved(phys_addr_t addr)
95f72d1e 839{
72d4b0b4
BH
840 return memblock_search(&memblock.reserved, addr) != -1;
841}
95f72d1e 842
3661ca66 843int __init_memblock memblock_is_memory(phys_addr_t addr)
72d4b0b4
BH
844{
845 return memblock_search(&memblock.memory, addr) != -1;
846}
847
3661ca66 848int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
72d4b0b4 849{
abb65272 850 int idx = memblock_search(&memblock.memory, base);
72d4b0b4
BH
851
852 if (idx == -1)
853 return 0;
abb65272
TV
854 return memblock.memory.regions[idx].base <= base &&
855 (memblock.memory.regions[idx].base +
856 memblock.memory.regions[idx].size) >= (base + size);
95f72d1e
YL
857}
858
10d06439 859int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
95f72d1e 860{
f1c2c19c 861 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
95f72d1e
YL
862}
863
e63075a3 864
3661ca66 865void __init_memblock memblock_set_current_limit(phys_addr_t limit)
e63075a3
BH
866{
867 memblock.current_limit = limit;
868}
869
7c0caeb8 870static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
6ed311b2
BH
871{
872 unsigned long long base, size;
873 int i;
874
7c0caeb8 875 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
6ed311b2 876
7c0caeb8
TH
877 for (i = 0; i < type->cnt; i++) {
878 struct memblock_region *rgn = &type->regions[i];
879 char nid_buf[32] = "";
880
881 base = rgn->base;
882 size = rgn->size;
883#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
884 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
885 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
886 memblock_get_region_node(rgn));
887#endif
888 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s\n",
889 name, i, base, base + size - 1, size, nid_buf);
6ed311b2
BH
890 }
891}
892
10d06439 893void __init_memblock memblock_dump_all(void)
6ed311b2
BH
894{
895 if (!memblock_debug)
896 return;
897
898 pr_info("MEMBLOCK configuration:\n");
899 pr_info(" memory size = 0x%llx\n", (unsigned long long)memblock.memory_size);
900
901 memblock_dump(&memblock.memory, "memory");
902 memblock_dump(&memblock.reserved, "reserved");
903}
904
905void __init memblock_analyze(void)
906{
907 int i;
908
909 /* Check marker in the unused last array entry */
910 WARN_ON(memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS].base
911 != (phys_addr_t)RED_INACTIVE);
912 WARN_ON(memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS].base
913 != (phys_addr_t)RED_INACTIVE);
914
915 memblock.memory_size = 0;
916
917 for (i = 0; i < memblock.memory.cnt; i++)
918 memblock.memory_size += memblock.memory.regions[i].size;
142b45a7
BH
919
920 /* We allow resizing from there */
921 memblock_can_resize = 1;
6ed311b2
BH
922}
923
7590abe8
BH
924void __init memblock_init(void)
925{
236260b9
JF
926 static int init_done __initdata = 0;
927
928 if (init_done)
929 return;
930 init_done = 1;
931
7590abe8
BH
932 /* Hookup the initial arrays */
933 memblock.memory.regions = memblock_memory_init_regions;
934 memblock.memory.max = INIT_MEMBLOCK_REGIONS;
935 memblock.reserved.regions = memblock_reserved_init_regions;
936 memblock.reserved.max = INIT_MEMBLOCK_REGIONS;
937
938 /* Write a marker in the unused last array entry */
939 memblock.memory.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
940 memblock.reserved.regions[INIT_MEMBLOCK_REGIONS].base = (phys_addr_t)RED_INACTIVE;
941
942 /* Create a dummy zero size MEMBLOCK which will get coalesced away later.
943 * This simplifies the memblock_add() code below...
944 */
945 memblock.memory.regions[0].base = 0;
946 memblock.memory.regions[0].size = 0;
7c0caeb8 947 memblock_set_region_node(&memblock.memory.regions[0], MAX_NUMNODES);
7590abe8
BH
948 memblock.memory.cnt = 1;
949
950 /* Ditto. */
951 memblock.reserved.regions[0].base = 0;
952 memblock.reserved.regions[0].size = 0;
7c0caeb8 953 memblock_set_region_node(&memblock.reserved.regions[0], MAX_NUMNODES);
7590abe8
BH
954 memblock.reserved.cnt = 1;
955
956 memblock.current_limit = MEMBLOCK_ALLOC_ANYWHERE;
957}
958
6ed311b2
BH
959static int __init early_memblock(char *p)
960{
961 if (p && strstr(p, "debug"))
962 memblock_debug = 1;
963 return 0;
964}
965early_param("memblock", early_memblock);
966
c378ddd5 967#if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
6d03b885
BH
968
969static int memblock_debug_show(struct seq_file *m, void *private)
970{
971 struct memblock_type *type = m->private;
972 struct memblock_region *reg;
973 int i;
974
975 for (i = 0; i < type->cnt; i++) {
976 reg = &type->regions[i];
977 seq_printf(m, "%4d: ", i);
978 if (sizeof(phys_addr_t) == 4)
979 seq_printf(m, "0x%08lx..0x%08lx\n",
980 (unsigned long)reg->base,
981 (unsigned long)(reg->base + reg->size - 1));
982 else
983 seq_printf(m, "0x%016llx..0x%016llx\n",
984 (unsigned long long)reg->base,
985 (unsigned long long)(reg->base + reg->size - 1));
986
987 }
988 return 0;
989}
990
991static int memblock_debug_open(struct inode *inode, struct file *file)
992{
993 return single_open(file, memblock_debug_show, inode->i_private);
994}
995
996static const struct file_operations memblock_debug_fops = {
997 .open = memblock_debug_open,
998 .read = seq_read,
999 .llseek = seq_lseek,
1000 .release = single_release,
1001};
1002
1003static int __init memblock_init_debugfs(void)
1004{
1005 struct dentry *root = debugfs_create_dir("memblock", NULL);
1006 if (!root)
1007 return -ENXIO;
1008 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1009 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1010
1011 return 0;
1012}
1013__initcall(memblock_init_debugfs);
1014
1015#endif /* CONFIG_DEBUG_FS */