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 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
24 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
25
26 struct memblock memblock __initdata_memblock = {
27 .memory.regions = memblock_memory_init_regions,
28 .memory.cnt = 1, /* empty dummy entry */
29 .memory.max = INIT_MEMBLOCK_REGIONS,
30
31 .reserved.regions = memblock_reserved_init_regions,
32 .reserved.cnt = 1, /* empty dummy entry */
33 .reserved.max = INIT_MEMBLOCK_REGIONS,
34
35 .current_limit = MEMBLOCK_ALLOC_ANYWHERE,
36 };
37
38 int memblock_debug __initdata_memblock;
39 static int memblock_can_resize __initdata_memblock;
40 static int memblock_memory_in_slab __initdata_memblock = 0;
41 static int memblock_reserved_in_slab __initdata_memblock = 0;
42
43 /* inline so we don't get a warning when pr_debug is compiled out */
memblock_type_name(struct memblock_type * type)44 static inline const char *memblock_type_name(struct memblock_type *type)
45 {
46 if (type == &memblock.memory)
47 return "memory";
48 else if (type == &memblock.reserved)
49 return "reserved";
50 else
51 return "unknown";
52 }
53
54 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
memblock_cap_size(phys_addr_t base,phys_addr_t * size)55 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
56 {
57 return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
58 }
59
60 /*
61 * Address comparison utilities
62 */
memblock_addrs_overlap(phys_addr_t base1,phys_addr_t size1,phys_addr_t base2,phys_addr_t size2)63 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
64 phys_addr_t base2, phys_addr_t size2)
65 {
66 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
67 }
68
memblock_overlaps_region(struct memblock_type * type,phys_addr_t base,phys_addr_t size)69 static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
70 phys_addr_t base, phys_addr_t size)
71 {
72 unsigned long i;
73
74 for (i = 0; i < type->cnt; i++) {
75 phys_addr_t rgnbase = type->regions[i].base;
76 phys_addr_t rgnsize = type->regions[i].size;
77 if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
78 break;
79 }
80
81 return (i < type->cnt) ? i : -1;
82 }
83
84 /**
85 * memblock_find_in_range_node - find free area in given range and node
86 * @start: start of candidate range
87 * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
88 * @size: size of free area to find
89 * @align: alignment of free area to find
90 * @nid: nid of the free area to find, %MAX_NUMNODES for any node
91 *
92 * Find @size free area aligned to @align in the specified range and node.
93 *
94 * RETURNS:
95 * Found address on success, %0 on failure.
96 */
memblock_find_in_range_node(phys_addr_t start,phys_addr_t end,phys_addr_t size,phys_addr_t align,int nid)97 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t start,
98 phys_addr_t end, phys_addr_t size,
99 phys_addr_t align, int nid)
100 {
101 phys_addr_t this_start, this_end, cand;
102 u64 i;
103
104 /* pump up @end */
105 if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
106 end = memblock.current_limit;
107
108 /* avoid allocating the first page */
109 start = max_t(phys_addr_t, start, PAGE_SIZE);
110 end = max(start, end);
111
112 for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) {
113 this_start = clamp(this_start, start, end);
114 this_end = clamp(this_end, start, end);
115
116 if (this_end < size)
117 continue;
118
119 cand = round_down(this_end - size, align);
120 if (cand >= this_start)
121 return cand;
122 }
123 return 0;
124 }
125
126 /**
127 * memblock_find_in_range - find free area in given range
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 *
133 * Find @size free area aligned to @align in the specified range.
134 *
135 * RETURNS:
136 * Found address on success, %0 on failure.
137 */
memblock_find_in_range(phys_addr_t start,phys_addr_t end,phys_addr_t size,phys_addr_t align)138 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
139 phys_addr_t end, phys_addr_t size,
140 phys_addr_t align)
141 {
142 return memblock_find_in_range_node(start, end, size, align,
143 MAX_NUMNODES);
144 }
145
memblock_remove_region(struct memblock_type * type,unsigned long r)146 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
147 {
148 type->total_size -= type->regions[r].size;
149 memmove(&type->regions[r], &type->regions[r + 1],
150 (type->cnt - (r + 1)) * sizeof(type->regions[r]));
151 type->cnt--;
152
153 /* Special case for empty arrays */
154 if (type->cnt == 0) {
155 WARN_ON(type->total_size != 0);
156 type->cnt = 1;
157 type->regions[0].base = 0;
158 type->regions[0].size = 0;
159 memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
160 }
161 }
162
get_allocated_memblock_reserved_regions_info(phys_addr_t * addr)163 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
164 phys_addr_t *addr)
165 {
166 if (memblock.reserved.regions == memblock_reserved_init_regions)
167 return 0;
168
169 *addr = __pa(memblock.reserved.regions);
170
171 return PAGE_ALIGN(sizeof(struct memblock_region) *
172 memblock.reserved.max);
173 }
174
175 /**
176 * memblock_double_array - double the size of the memblock regions array
177 * @type: memblock type of the regions array being doubled
178 * @new_area_start: starting address of memory range to avoid overlap with
179 * @new_area_size: size of memory range to avoid overlap with
180 *
181 * Double the size of the @type regions array. If memblock is being used to
182 * allocate memory for a new reserved regions array and there is a previously
183 * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
184 * waiting to be reserved, ensure the memory used by the new array does
185 * not overlap.
186 *
187 * RETURNS:
188 * 0 on success, -1 on failure.
189 */
memblock_double_array(struct memblock_type * type,phys_addr_t new_area_start,phys_addr_t new_area_size)190 static int __init_memblock memblock_double_array(struct memblock_type *type,
191 phys_addr_t new_area_start,
192 phys_addr_t new_area_size)
193 {
194 struct memblock_region *new_array, *old_array;
195 phys_addr_t old_alloc_size, new_alloc_size;
196 phys_addr_t old_size, new_size, addr;
197 int use_slab = slab_is_available();
198 int *in_slab;
199
200 /* We don't allow resizing until we know about the reserved regions
201 * of memory that aren't suitable for allocation
202 */
203 if (!memblock_can_resize)
204 return -1;
205
206 /* Calculate new doubled size */
207 old_size = type->max * sizeof(struct memblock_region);
208 new_size = old_size << 1;
209 /*
210 * We need to allocated new one align to PAGE_SIZE,
211 * so we can free them completely later.
212 */
213 old_alloc_size = PAGE_ALIGN(old_size);
214 new_alloc_size = PAGE_ALIGN(new_size);
215
216 /* Retrieve the slab flag */
217 if (type == &memblock.memory)
218 in_slab = &memblock_memory_in_slab;
219 else
220 in_slab = &memblock_reserved_in_slab;
221
222 /* Try to find some space for it.
223 *
224 * WARNING: We assume that either slab_is_available() and we use it or
225 * we use MEMBLOCK for allocations. That means that this is unsafe to use
226 * when bootmem is currently active (unless bootmem itself is implemented
227 * on top of MEMBLOCK which isn't the case yet)
228 *
229 * This should however not be an issue for now, as we currently only
230 * call into MEMBLOCK while it's still active, or much later when slab is
231 * active for memory hotplug operations
232 */
233 if (use_slab) {
234 new_array = kmalloc(new_size, GFP_KERNEL);
235 addr = new_array ? __pa(new_array) : 0;
236 } else {
237 /* only exclude range when trying to double reserved.regions */
238 if (type != &memblock.reserved)
239 new_area_start = new_area_size = 0;
240
241 addr = memblock_find_in_range(new_area_start + new_area_size,
242 memblock.current_limit,
243 new_alloc_size, PAGE_SIZE);
244 if (!addr && new_area_size)
245 addr = memblock_find_in_range(0,
246 min(new_area_start, memblock.current_limit),
247 new_alloc_size, PAGE_SIZE);
248
249 new_array = addr ? __va(addr) : 0;
250 }
251 if (!addr) {
252 pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
253 memblock_type_name(type), type->max, type->max * 2);
254 return -1;
255 }
256
257 memblock_dbg("memblock: %s array is doubled to %ld at [%#010llx-%#010llx]",
258 memblock_type_name(type), type->max * 2, (u64)addr, (u64)addr + new_size - 1);
259
260 /* Found space, we now need to move the array over before
261 * we add the reserved region since it may be our reserved
262 * array itself that is full.
263 */
264 memcpy(new_array, type->regions, old_size);
265 memset(new_array + type->max, 0, old_size);
266 old_array = type->regions;
267 type->regions = new_array;
268 type->max <<= 1;
269
270 /* Free old array. We needn't free it if the array is the
271 * static one
272 */
273 if (*in_slab)
274 kfree(old_array);
275 else if (old_array != memblock_memory_init_regions &&
276 old_array != memblock_reserved_init_regions)
277 memblock_free(__pa(old_array), old_alloc_size);
278
279 /* Reserve the new array if that comes from the memblock.
280 * Otherwise, we needn't do it
281 */
282 if (!use_slab)
283 BUG_ON(memblock_reserve(addr, new_alloc_size));
284
285 /* Update slab flag */
286 *in_slab = use_slab;
287
288 return 0;
289 }
290
291 /**
292 * memblock_merge_regions - merge neighboring compatible regions
293 * @type: memblock type to scan
294 *
295 * Scan @type and merge neighboring compatible regions.
296 */
memblock_merge_regions(struct memblock_type * type)297 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
298 {
299 int i = 0;
300
301 /* cnt never goes below 1 */
302 while (i < type->cnt - 1) {
303 struct memblock_region *this = &type->regions[i];
304 struct memblock_region *next = &type->regions[i + 1];
305
306 if (this->base + this->size != next->base ||
307 memblock_get_region_node(this) !=
308 memblock_get_region_node(next)) {
309 BUG_ON(this->base + this->size > next->base);
310 i++;
311 continue;
312 }
313
314 this->size += next->size;
315 memmove(next, next + 1, (type->cnt - (i + 1)) * sizeof(*next));
316 type->cnt--;
317 }
318 }
319
320 /**
321 * memblock_insert_region - insert new memblock region
322 * @type: memblock type to insert into
323 * @idx: index for the insertion point
324 * @base: base address of the new region
325 * @size: size of the new region
326 *
327 * Insert new memblock region [@base,@base+@size) into @type at @idx.
328 * @type must already have extra room to accomodate the new region.
329 */
memblock_insert_region(struct memblock_type * type,int idx,phys_addr_t base,phys_addr_t size,int nid)330 static void __init_memblock memblock_insert_region(struct memblock_type *type,
331 int idx, phys_addr_t base,
332 phys_addr_t size, int nid)
333 {
334 struct memblock_region *rgn = &type->regions[idx];
335
336 BUG_ON(type->cnt >= type->max);
337 memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
338 rgn->base = base;
339 rgn->size = size;
340 memblock_set_region_node(rgn, nid);
341 type->cnt++;
342 type->total_size += size;
343 }
344
345 /**
346 * memblock_add_region - add new memblock region
347 * @type: memblock type to add new region into
348 * @base: base address of the new region
349 * @size: size of the new region
350 * @nid: nid of the new region
351 *
352 * Add new memblock region [@base,@base+@size) into @type. The new region
353 * is allowed to overlap with existing ones - overlaps don't affect already
354 * existing regions. @type is guaranteed to be minimal (all neighbouring
355 * compatible regions are merged) after the addition.
356 *
357 * RETURNS:
358 * 0 on success, -errno on failure.
359 */
memblock_add_region(struct memblock_type * type,phys_addr_t base,phys_addr_t size,int nid)360 static int __init_memblock memblock_add_region(struct memblock_type *type,
361 phys_addr_t base, phys_addr_t size, int nid)
362 {
363 bool insert = false;
364 phys_addr_t obase = base;
365 phys_addr_t end = base + memblock_cap_size(base, &size);
366 int i, nr_new;
367
368 if (!size)
369 return 0;
370
371 /* special case for empty array */
372 if (type->regions[0].size == 0) {
373 WARN_ON(type->cnt != 1 || type->total_size);
374 type->regions[0].base = base;
375 type->regions[0].size = size;
376 memblock_set_region_node(&type->regions[0], nid);
377 type->total_size = size;
378 return 0;
379 }
380 repeat:
381 /*
382 * The following is executed twice. Once with %false @insert and
383 * then with %true. The first counts the number of regions needed
384 * to accomodate the new area. The second actually inserts them.
385 */
386 base = obase;
387 nr_new = 0;
388
389 for (i = 0; i < type->cnt; i++) {
390 struct memblock_region *rgn = &type->regions[i];
391 phys_addr_t rbase = rgn->base;
392 phys_addr_t rend = rbase + rgn->size;
393
394 if (rbase >= end)
395 break;
396 if (rend <= base)
397 continue;
398 /*
399 * @rgn overlaps. If it separates the lower part of new
400 * area, insert that portion.
401 */
402 if (rbase > base) {
403 nr_new++;
404 if (insert)
405 memblock_insert_region(type, i++, base,
406 rbase - base, nid);
407 }
408 /* area below @rend is dealt with, forget about it */
409 base = min(rend, end);
410 }
411
412 /* insert the remaining portion */
413 if (base < end) {
414 nr_new++;
415 if (insert)
416 memblock_insert_region(type, i, base, end - base, nid);
417 }
418
419 /*
420 * If this was the first round, resize array and repeat for actual
421 * insertions; otherwise, merge and return.
422 */
423 if (!insert) {
424 while (type->cnt + nr_new > type->max)
425 if (memblock_double_array(type, obase, size) < 0)
426 return -ENOMEM;
427 insert = true;
428 goto repeat;
429 } else {
430 memblock_merge_regions(type);
431 return 0;
432 }
433 }
434
memblock_add_node(phys_addr_t base,phys_addr_t size,int nid)435 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
436 int nid)
437 {
438 return memblock_add_region(&memblock.memory, base, size, nid);
439 }
440
memblock_add(phys_addr_t base,phys_addr_t size)441 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
442 {
443 return memblock_add_region(&memblock.memory, base, size, MAX_NUMNODES);
444 }
445
446 /**
447 * memblock_isolate_range - isolate given range into disjoint memblocks
448 * @type: memblock type to isolate range for
449 * @base: base of range to isolate
450 * @size: size of range to isolate
451 * @start_rgn: out parameter for the start of isolated region
452 * @end_rgn: out parameter for the end of isolated region
453 *
454 * Walk @type and ensure that regions don't cross the boundaries defined by
455 * [@base,@base+@size). Crossing regions are split at the boundaries,
456 * which may create at most two more regions. The index of the first
457 * region inside the range is returned in *@start_rgn and end in *@end_rgn.
458 *
459 * RETURNS:
460 * 0 on success, -errno on failure.
461 */
memblock_isolate_range(struct memblock_type * type,phys_addr_t base,phys_addr_t size,int * start_rgn,int * end_rgn)462 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
463 phys_addr_t base, phys_addr_t size,
464 int *start_rgn, int *end_rgn)
465 {
466 phys_addr_t end = base + memblock_cap_size(base, &size);
467 int i;
468
469 *start_rgn = *end_rgn = 0;
470
471 if (!size)
472 return 0;
473
474 /* we'll create at most two more regions */
475 while (type->cnt + 2 > type->max)
476 if (memblock_double_array(type, base, size) < 0)
477 return -ENOMEM;
478
479 for (i = 0; i < type->cnt; i++) {
480 struct memblock_region *rgn = &type->regions[i];
481 phys_addr_t rbase = rgn->base;
482 phys_addr_t rend = rbase + rgn->size;
483
484 if (rbase >= end)
485 break;
486 if (rend <= base)
487 continue;
488
489 if (rbase < base) {
490 /*
491 * @rgn intersects from below. Split and continue
492 * to process the next region - the new top half.
493 */
494 rgn->base = base;
495 rgn->size -= base - rbase;
496 type->total_size -= base - rbase;
497 memblock_insert_region(type, i, rbase, base - rbase,
498 memblock_get_region_node(rgn));
499 } else if (rend > end) {
500 /*
501 * @rgn intersects from above. Split and redo the
502 * current region - the new bottom half.
503 */
504 rgn->base = end;
505 rgn->size -= end - rbase;
506 type->total_size -= end - rbase;
507 memblock_insert_region(type, i--, rbase, end - rbase,
508 memblock_get_region_node(rgn));
509 } else {
510 /* @rgn is fully contained, record it */
511 if (!*end_rgn)
512 *start_rgn = i;
513 *end_rgn = i + 1;
514 }
515 }
516
517 return 0;
518 }
519
__memblock_remove(struct memblock_type * type,phys_addr_t base,phys_addr_t size)520 static int __init_memblock __memblock_remove(struct memblock_type *type,
521 phys_addr_t base, phys_addr_t size)
522 {
523 int start_rgn, end_rgn;
524 int i, ret;
525
526 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
527 if (ret)
528 return ret;
529
530 for (i = end_rgn - 1; i >= start_rgn; i--)
531 memblock_remove_region(type, i);
532 return 0;
533 }
534
memblock_remove(phys_addr_t base,phys_addr_t size)535 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
536 {
537 return __memblock_remove(&memblock.memory, base, size);
538 }
539
memblock_free(phys_addr_t base,phys_addr_t size)540 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
541 {
542 memblock_dbg(" memblock_free: [%#016llx-%#016llx] %pF\n",
543 (unsigned long long)base,
544 (unsigned long long)base + size,
545 (void *)_RET_IP_);
546
547 return __memblock_remove(&memblock.reserved, base, size);
548 }
549
memblock_reserve(phys_addr_t base,phys_addr_t size)550 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
551 {
552 struct memblock_type *_rgn = &memblock.reserved;
553
554 memblock_dbg("memblock_reserve: [%#016llx-%#016llx] %pF\n",
555 (unsigned long long)base,
556 (unsigned long long)base + size,
557 (void *)_RET_IP_);
558
559 return memblock_add_region(_rgn, base, size, MAX_NUMNODES);
560 }
561
562 /**
563 * __next_free_mem_range - next function for for_each_free_mem_range()
564 * @idx: pointer to u64 loop variable
565 * @nid: nid: node selector, %MAX_NUMNODES for all nodes
566 * @p_start: ptr to phys_addr_t for start address of the range, can be %NULL
567 * @p_end: ptr to phys_addr_t for end address of the range, can be %NULL
568 * @p_nid: ptr to int for nid of the range, can be %NULL
569 *
570 * Find the first free area from *@idx which matches @nid, fill the out
571 * parameters, and update *@idx for the next iteration. The lower 32bit of
572 * *@idx contains index into memory region and the upper 32bit indexes the
573 * areas before each reserved region. For example, if reserved regions
574 * look like the following,
575 *
576 * 0:[0-16), 1:[32-48), 2:[128-130)
577 *
578 * The upper 32bit indexes the following regions.
579 *
580 * 0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
581 *
582 * As both region arrays are sorted, the function advances the two indices
583 * in lockstep and returns each intersection.
584 */
__next_free_mem_range(u64 * idx,int nid,phys_addr_t * out_start,phys_addr_t * out_end,int * out_nid)585 void __init_memblock __next_free_mem_range(u64 *idx, int nid,
586 phys_addr_t *out_start,
587 phys_addr_t *out_end, int *out_nid)
588 {
589 struct memblock_type *mem = &memblock.memory;
590 struct memblock_type *rsv = &memblock.reserved;
591 int mi = *idx & 0xffffffff;
592 int ri = *idx >> 32;
593
594 for ( ; mi < mem->cnt; mi++) {
595 struct memblock_region *m = &mem->regions[mi];
596 phys_addr_t m_start = m->base;
597 phys_addr_t m_end = m->base + m->size;
598
599 /* only memory regions are associated with nodes, check it */
600 if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m))
601 continue;
602
603 /* scan areas before each reservation for intersection */
604 for ( ; ri < rsv->cnt + 1; ri++) {
605 struct memblock_region *r = &rsv->regions[ri];
606 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
607 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
608
609 /* if ri advanced past mi, break out to advance mi */
610 if (r_start >= m_end)
611 break;
612 /* if the two regions intersect, we're done */
613 if (m_start < r_end) {
614 if (out_start)
615 *out_start = max(m_start, r_start);
616 if (out_end)
617 *out_end = min(m_end, r_end);
618 if (out_nid)
619 *out_nid = memblock_get_region_node(m);
620 /*
621 * The region which ends first is advanced
622 * for the next iteration.
623 */
624 if (m_end <= r_end)
625 mi++;
626 else
627 ri++;
628 *idx = (u32)mi | (u64)ri << 32;
629 return;
630 }
631 }
632 }
633
634 /* signal end of iteration */
635 *idx = ULLONG_MAX;
636 }
637
638 /**
639 * __next_free_mem_range_rev - next function for for_each_free_mem_range_reverse()
640 * @idx: pointer to u64 loop variable
641 * @nid: nid: node selector, %MAX_NUMNODES for all nodes
642 * @p_start: ptr to phys_addr_t for start address of the range, can be %NULL
643 * @p_end: ptr to phys_addr_t for end address of the range, can be %NULL
644 * @p_nid: ptr to int for nid of the range, can be %NULL
645 *
646 * Reverse of __next_free_mem_range().
647 */
__next_free_mem_range_rev(u64 * idx,int nid,phys_addr_t * out_start,phys_addr_t * out_end,int * out_nid)648 void __init_memblock __next_free_mem_range_rev(u64 *idx, int nid,
649 phys_addr_t *out_start,
650 phys_addr_t *out_end, int *out_nid)
651 {
652 struct memblock_type *mem = &memblock.memory;
653 struct memblock_type *rsv = &memblock.reserved;
654 int mi = *idx & 0xffffffff;
655 int ri = *idx >> 32;
656
657 if (*idx == (u64)ULLONG_MAX) {
658 mi = mem->cnt - 1;
659 ri = rsv->cnt;
660 }
661
662 for ( ; mi >= 0; mi--) {
663 struct memblock_region *m = &mem->regions[mi];
664 phys_addr_t m_start = m->base;
665 phys_addr_t m_end = m->base + m->size;
666
667 /* only memory regions are associated with nodes, check it */
668 if (nid != MAX_NUMNODES && nid != memblock_get_region_node(m))
669 continue;
670
671 /* scan areas before each reservation for intersection */
672 for ( ; ri >= 0; ri--) {
673 struct memblock_region *r = &rsv->regions[ri];
674 phys_addr_t r_start = ri ? r[-1].base + r[-1].size : 0;
675 phys_addr_t r_end = ri < rsv->cnt ? r->base : ULLONG_MAX;
676
677 /* if ri advanced past mi, break out to advance mi */
678 if (r_end <= m_start)
679 break;
680 /* if the two regions intersect, we're done */
681 if (m_end > r_start) {
682 if (out_start)
683 *out_start = max(m_start, r_start);
684 if (out_end)
685 *out_end = min(m_end, r_end);
686 if (out_nid)
687 *out_nid = memblock_get_region_node(m);
688
689 if (m_start >= r_start)
690 mi--;
691 else
692 ri--;
693 *idx = (u32)mi | (u64)ri << 32;
694 return;
695 }
696 }
697 }
698
699 *idx = ULLONG_MAX;
700 }
701
702 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
703 /*
704 * Common iterator interface used to define for_each_mem_range().
705 */
__next_mem_pfn_range(int * idx,int nid,unsigned long * out_start_pfn,unsigned long * out_end_pfn,int * out_nid)706 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
707 unsigned long *out_start_pfn,
708 unsigned long *out_end_pfn, int *out_nid)
709 {
710 struct memblock_type *type = &memblock.memory;
711 struct memblock_region *r;
712
713 while (++*idx < type->cnt) {
714 r = &type->regions[*idx];
715
716 if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
717 continue;
718 if (nid == MAX_NUMNODES || nid == r->nid)
719 break;
720 }
721 if (*idx >= type->cnt) {
722 *idx = -1;
723 return;
724 }
725
726 if (out_start_pfn)
727 *out_start_pfn = PFN_UP(r->base);
728 if (out_end_pfn)
729 *out_end_pfn = PFN_DOWN(r->base + r->size);
730 if (out_nid)
731 *out_nid = r->nid;
732 }
733
734 /**
735 * memblock_set_node - set node ID on memblock regions
736 * @base: base of area to set node ID for
737 * @size: size of area to set node ID for
738 * @nid: node ID to set
739 *
740 * Set the nid of memblock memory regions in [@base,@base+@size) to @nid.
741 * Regions which cross the area boundaries are split as necessary.
742 *
743 * RETURNS:
744 * 0 on success, -errno on failure.
745 */
memblock_set_node(phys_addr_t base,phys_addr_t size,int nid)746 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
747 int nid)
748 {
749 struct memblock_type *type = &memblock.memory;
750 int start_rgn, end_rgn;
751 int i, ret;
752
753 ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
754 if (ret)
755 return ret;
756
757 for (i = start_rgn; i < end_rgn; i++)
758 type->regions[i].nid = nid;
759
760 memblock_merge_regions(type);
761 return 0;
762 }
763 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
764
memblock_alloc_base_nid(phys_addr_t size,phys_addr_t align,phys_addr_t max_addr,int nid)765 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
766 phys_addr_t align, phys_addr_t max_addr,
767 int nid)
768 {
769 phys_addr_t found;
770
771 /* align @size to avoid excessive fragmentation on reserved array */
772 size = round_up(size, align);
773
774 found = memblock_find_in_range_node(0, max_addr, size, align, nid);
775 if (found && !memblock_reserve(found, size))
776 return found;
777
778 return 0;
779 }
780
memblock_alloc_nid(phys_addr_t size,phys_addr_t align,int nid)781 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
782 {
783 return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
784 }
785
__memblock_alloc_base(phys_addr_t size,phys_addr_t align,phys_addr_t max_addr)786 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
787 {
788 return memblock_alloc_base_nid(size, align, max_addr, MAX_NUMNODES);
789 }
790
memblock_alloc_base(phys_addr_t size,phys_addr_t align,phys_addr_t max_addr)791 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
792 {
793 phys_addr_t alloc;
794
795 alloc = __memblock_alloc_base(size, align, max_addr);
796
797 if (alloc == 0)
798 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
799 (unsigned long long) size, (unsigned long long) max_addr);
800
801 return alloc;
802 }
803
memblock_alloc(phys_addr_t size,phys_addr_t align)804 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
805 {
806 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
807 }
808
memblock_alloc_try_nid(phys_addr_t size,phys_addr_t align,int nid)809 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
810 {
811 phys_addr_t res = memblock_alloc_nid(size, align, nid);
812
813 if (res)
814 return res;
815 return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
816 }
817
818
819 /*
820 * Remaining API functions
821 */
822
memblock_phys_mem_size(void)823 phys_addr_t __init memblock_phys_mem_size(void)
824 {
825 return memblock.memory.total_size;
826 }
827
828 /* lowest address */
memblock_start_of_DRAM(void)829 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
830 {
831 return memblock.memory.regions[0].base;
832 }
833
memblock_end_of_DRAM(void)834 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
835 {
836 int idx = memblock.memory.cnt - 1;
837
838 return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
839 }
840
memblock_enforce_memory_limit(phys_addr_t limit)841 void __init memblock_enforce_memory_limit(phys_addr_t limit)
842 {
843 unsigned long i;
844 phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
845
846 if (!limit)
847 return;
848
849 /* find out max address */
850 for (i = 0; i < memblock.memory.cnt; i++) {
851 struct memblock_region *r = &memblock.memory.regions[i];
852
853 if (limit <= r->size) {
854 max_addr = r->base + limit;
855 break;
856 }
857 limit -= r->size;
858 }
859
860 /* truncate both memory and reserved regions */
861 __memblock_remove(&memblock.memory, max_addr, (phys_addr_t)ULLONG_MAX);
862 __memblock_remove(&memblock.reserved, max_addr, (phys_addr_t)ULLONG_MAX);
863 }
864
memblock_search(struct memblock_type * type,phys_addr_t addr)865 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
866 {
867 unsigned int left = 0, right = type->cnt;
868
869 do {
870 unsigned int mid = (right + left) / 2;
871
872 if (addr < type->regions[mid].base)
873 right = mid;
874 else if (addr >= (type->regions[mid].base +
875 type->regions[mid].size))
876 left = mid + 1;
877 else
878 return mid;
879 } while (left < right);
880 return -1;
881 }
882
memblock_is_reserved(phys_addr_t addr)883 int __init memblock_is_reserved(phys_addr_t addr)
884 {
885 return memblock_search(&memblock.reserved, addr) != -1;
886 }
887
memblock_is_memory(phys_addr_t addr)888 int __init_memblock memblock_is_memory(phys_addr_t addr)
889 {
890 return memblock_search(&memblock.memory, addr) != -1;
891 }
892
memblock_is_region_memory(phys_addr_t base,phys_addr_t size)893 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
894 {
895 int idx = memblock_search(&memblock.memory, base);
896 phys_addr_t end = base + memblock_cap_size(base, &size);
897
898 if (idx == -1)
899 return 0;
900 return memblock.memory.regions[idx].base <= base &&
901 (memblock.memory.regions[idx].base +
902 memblock.memory.regions[idx].size) >= end;
903 }
904
memblock_is_region_reserved(phys_addr_t base,phys_addr_t size)905 int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
906 {
907 memblock_cap_size(base, &size);
908 return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
909 }
910
memblock_trim_memory(phys_addr_t align)911 void __init_memblock memblock_trim_memory(phys_addr_t align)
912 {
913 int i;
914 phys_addr_t start, end, orig_start, orig_end;
915 struct memblock_type *mem = &memblock.memory;
916
917 for (i = 0; i < mem->cnt; i++) {
918 orig_start = mem->regions[i].base;
919 orig_end = mem->regions[i].base + mem->regions[i].size;
920 start = round_up(orig_start, align);
921 end = round_down(orig_end, align);
922
923 if (start == orig_start && end == orig_end)
924 continue;
925
926 if (start < end) {
927 mem->regions[i].base = start;
928 mem->regions[i].size = end - start;
929 } else {
930 memblock_remove_region(mem, i);
931 i--;
932 }
933 }
934 }
935
memblock_set_current_limit(phys_addr_t limit)936 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
937 {
938 memblock.current_limit = limit;
939 }
940
memblock_dump(struct memblock_type * type,char * name)941 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
942 {
943 unsigned long long base, size;
944 int i;
945
946 pr_info(" %s.cnt = 0x%lx\n", name, type->cnt);
947
948 for (i = 0; i < type->cnt; i++) {
949 struct memblock_region *rgn = &type->regions[i];
950 char nid_buf[32] = "";
951
952 base = rgn->base;
953 size = rgn->size;
954 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
955 if (memblock_get_region_node(rgn) != MAX_NUMNODES)
956 snprintf(nid_buf, sizeof(nid_buf), " on node %d",
957 memblock_get_region_node(rgn));
958 #endif
959 pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s\n",
960 name, i, base, base + size - 1, size, nid_buf);
961 }
962 }
963
__memblock_dump_all(void)964 void __init_memblock __memblock_dump_all(void)
965 {
966 pr_info("MEMBLOCK configuration:\n");
967 pr_info(" memory size = %#llx reserved size = %#llx\n",
968 (unsigned long long)memblock.memory.total_size,
969 (unsigned long long)memblock.reserved.total_size);
970
971 memblock_dump(&memblock.memory, "memory");
972 memblock_dump(&memblock.reserved, "reserved");
973 }
974
memblock_allow_resize(void)975 void __init memblock_allow_resize(void)
976 {
977 memblock_can_resize = 1;
978 }
979
early_memblock(char * p)980 static int __init early_memblock(char *p)
981 {
982 if (p && strstr(p, "debug"))
983 memblock_debug = 1;
984 return 0;
985 }
986 early_param("memblock", early_memblock);
987
988 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
989
memblock_debug_show(struct seq_file * m,void * private)990 static int memblock_debug_show(struct seq_file *m, void *private)
991 {
992 struct memblock_type *type = m->private;
993 struct memblock_region *reg;
994 int i;
995
996 for (i = 0; i < type->cnt; i++) {
997 reg = &type->regions[i];
998 seq_printf(m, "%4d: ", i);
999 if (sizeof(phys_addr_t) == 4)
1000 seq_printf(m, "0x%08lx..0x%08lx\n",
1001 (unsigned long)reg->base,
1002 (unsigned long)(reg->base + reg->size - 1));
1003 else
1004 seq_printf(m, "0x%016llx..0x%016llx\n",
1005 (unsigned long long)reg->base,
1006 (unsigned long long)(reg->base + reg->size - 1));
1007
1008 }
1009 return 0;
1010 }
1011
memblock_debug_open(struct inode * inode,struct file * file)1012 static int memblock_debug_open(struct inode *inode, struct file *file)
1013 {
1014 return single_open(file, memblock_debug_show, inode->i_private);
1015 }
1016
1017 static const struct file_operations memblock_debug_fops = {
1018 .open = memblock_debug_open,
1019 .read = seq_read,
1020 .llseek = seq_lseek,
1021 .release = single_release,
1022 };
1023
memblock_init_debugfs(void)1024 static int __init memblock_init_debugfs(void)
1025 {
1026 struct dentry *root = debugfs_create_dir("memblock", NULL);
1027 if (!root)
1028 return -ENXIO;
1029 debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1030 debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1031
1032 return 0;
1033 }
1034 __initcall(memblock_init_debugfs);
1035
1036 #endif /* CONFIG_DEBUG_FS */
1037