• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Procedures for maintaining information about logical memory blocks.
3  *
4  * Peter Bergner, IBM Corp.	June 2001.
5  * Copyright (C) 2001 Peter Bergner.
6  *
7  *      This program is free software; you can redistribute it and/or
8  *      modify it under the terms of the GNU General Public License
9  *      as published by the Free Software Foundation; either version
10  *      2 of the License, or (at your option) any later version.
11  */
12 
13 #include <linux/kernel.h>
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/bitops.h>
17 #include <linux/poison.h>
18 #include <linux/pfn.h>
19 #include <linux/debugfs.h>
20 #include <linux/seq_file.h>
21 #include <linux/memblock.h>
22 
23 #include <asm-generic/sections.h>
24 #include <linux/io.h>
25 
26 #include "internal.h"
27 
28 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
29 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
30 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
31 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
32 #endif
33 
34 struct memblock memblock __initdata_memblock = {
35 	.memory.regions		= memblock_memory_init_regions,
36 	.memory.cnt		= 1,	/* empty dummy entry */
37 	.memory.max		= INIT_MEMBLOCK_REGIONS,
38 
39 	.reserved.regions	= memblock_reserved_init_regions,
40 	.reserved.cnt		= 1,	/* empty dummy entry */
41 	.reserved.max		= INIT_MEMBLOCK_REGIONS,
42 
43 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
44 	.physmem.regions	= memblock_physmem_init_regions,
45 	.physmem.cnt		= 1,	/* empty dummy entry */
46 	.physmem.max		= INIT_PHYSMEM_REGIONS,
47 #endif
48 
49 	.bottom_up		= false,
50 	.current_limit		= MEMBLOCK_ALLOC_ANYWHERE,
51 };
52 
53 int memblock_debug __initdata_memblock;
54 #ifdef CONFIG_MOVABLE_NODE
55 bool movable_node_enabled __initdata_memblock = false;
56 #endif
57 static bool system_has_some_mirror __initdata_memblock = false;
58 static int memblock_can_resize __initdata_memblock;
59 static int memblock_memory_in_slab __initdata_memblock = 0;
60 static int memblock_reserved_in_slab __initdata_memblock = 0;
61 
choose_memblock_flags(void)62 ulong __init_memblock choose_memblock_flags(void)
63 {
64 	return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
65 }
66 
67 /* inline so we don't get a warning when pr_debug is compiled out */
68 static __init_memblock const char *
memblock_type_name(struct memblock_type * type)69 memblock_type_name(struct memblock_type *type)
70 {
71 	if (type == &memblock.memory)
72 		return "memory";
73 	else if (type == &memblock.reserved)
74 		return "reserved";
75 	else
76 		return "unknown";
77 }
78 
79 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
memblock_cap_size(phys_addr_t base,phys_addr_t * size)80 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
81 {
82 	return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
83 }
84 
85 /*
86  * Address comparison utilities
87  */
memblock_addrs_overlap(phys_addr_t base1,phys_addr_t size1,phys_addr_t base2,phys_addr_t size2)88 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
89 				       phys_addr_t base2, phys_addr_t size2)
90 {
91 	return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
92 }
93 
memblock_overlaps_region(struct memblock_type * type,phys_addr_t base,phys_addr_t size)94 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
95 					phys_addr_t base, phys_addr_t size)
96 {
97 	unsigned long i;
98 
99 	for (i = 0; i < type->cnt; i++) {
100 		phys_addr_t rgnbase = type->regions[i].base;
101 		phys_addr_t rgnsize = type->regions[i].size;
102 		if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
103 			break;
104 	}
105 
106 	return i < type->cnt;
107 }
108 
109 /*
110  * __memblock_find_range_bottom_up - find free area utility in bottom-up
111  * @start: start of candidate range
112  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
113  * @size: size of free area to find
114  * @align: alignment of free area to find
115  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
116  * @flags: pick from blocks based on memory attributes
117  *
118  * Utility called from memblock_find_in_range_node(), find free area bottom-up.
119  *
120  * RETURNS:
121  * Found address on success, 0 on failure.
122  */
123 static phys_addr_t __init_memblock
__memblock_find_range_bottom_up(phys_addr_t start,phys_addr_t end,phys_addr_t size,phys_addr_t align,int nid,ulong flags)124 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
125 				phys_addr_t size, phys_addr_t align, int nid,
126 				ulong flags)
127 {
128 	phys_addr_t this_start, this_end, cand;
129 	u64 i;
130 
131 	for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
132 		this_start = clamp(this_start, start, end);
133 		this_end = clamp(this_end, start, end);
134 
135 		cand = round_up(this_start, align);
136 		if (cand < this_end && this_end - cand >= size)
137 			return cand;
138 	}
139 
140 	return 0;
141 }
142 
143 /**
144  * __memblock_find_range_top_down - find free area utility, in top-down
145  * @start: start of candidate range
146  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
147  * @size: size of free area to find
148  * @align: alignment of free area to find
149  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
150  * @flags: pick from blocks based on memory attributes
151  *
152  * Utility called from memblock_find_in_range_node(), find free area top-down.
153  *
154  * RETURNS:
155  * Found address on success, 0 on failure.
156  */
157 static phys_addr_t __init_memblock
__memblock_find_range_top_down(phys_addr_t start,phys_addr_t end,phys_addr_t size,phys_addr_t align,int nid,ulong flags)158 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
159 			       phys_addr_t size, phys_addr_t align, int nid,
160 			       ulong flags)
161 {
162 	phys_addr_t this_start, this_end, cand;
163 	u64 i;
164 
165 	for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
166 					NULL) {
167 		this_start = clamp(this_start, start, end);
168 		this_end = clamp(this_end, start, end);
169 
170 		if (this_end < size)
171 			continue;
172 
173 		cand = round_down(this_end - size, align);
174 		if (cand >= this_start)
175 			return cand;
176 	}
177 
178 	return 0;
179 }
180 
181 /**
182  * memblock_find_in_range_node - find free area in given range and node
183  * @size: size of free area to find
184  * @align: alignment of free area to find
185  * @start: start of candidate range
186  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
187  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
188  * @flags: pick from blocks based on memory attributes
189  *
190  * Find @size free area aligned to @align in the specified range and node.
191  *
192  * RETURNS:
193  * Found address on success, 0 on failure.
194  */
memblock_find_in_range_node(phys_addr_t size,phys_addr_t align,phys_addr_t start,phys_addr_t end,int nid,ulong flags)195 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
196 					phys_addr_t align, phys_addr_t start,
197 					phys_addr_t end, int nid, ulong flags)
198 {
199 	/* pump up @end */
200 	if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
201 		end = memblock.current_limit;
202 
203 	/* avoid allocating the first page */
204 	start = max_t(phys_addr_t, start, PAGE_SIZE);
205 	end = max(start, end);
206 
207 	if (memblock_bottom_up())
208 		return __memblock_find_range_bottom_up(start, end, size, align,
209 						       nid, flags);
210 	else
211 		return __memblock_find_range_top_down(start, end, size, align,
212 						      nid, flags);
213 }
214 
215 /**
216  * memblock_find_in_range - find free area in given range
217  * @start: start of candidate range
218  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
219  * @size: size of free area to find
220  * @align: alignment of free area to find
221  *
222  * Find @size free area aligned to @align in the specified range.
223  *
224  * RETURNS:
225  * Found address on success, 0 on failure.
226  */
memblock_find_in_range(phys_addr_t start,phys_addr_t end,phys_addr_t size,phys_addr_t align)227 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
228 					phys_addr_t end, phys_addr_t size,
229 					phys_addr_t align)
230 {
231 	phys_addr_t ret;
232 	ulong flags = choose_memblock_flags();
233 
234 again:
235 	ret = memblock_find_in_range_node(size, align, start, end,
236 					    NUMA_NO_NODE, flags);
237 
238 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
239 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
240 			&size);
241 		flags &= ~MEMBLOCK_MIRROR;
242 		goto again;
243 	}
244 
245 	return ret;
246 }
247 
memblock_remove_region(struct memblock_type * type,unsigned long r)248 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
249 {
250 	type->total_size -= type->regions[r].size;
251 	memmove(&type->regions[r], &type->regions[r + 1],
252 		(type->cnt - (r + 1)) * sizeof(type->regions[r]));
253 	type->cnt--;
254 
255 	/* Special case for empty arrays */
256 	if (type->cnt == 0) {
257 		WARN_ON(type->total_size != 0);
258 		type->cnt = 1;
259 		type->regions[0].base = 0;
260 		type->regions[0].size = 0;
261 		type->regions[0].flags = 0;
262 		memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
263 	}
264 }
265 
266 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
267 
get_allocated_memblock_reserved_regions_info(phys_addr_t * addr)268 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
269 					phys_addr_t *addr)
270 {
271 	if (memblock.reserved.regions == memblock_reserved_init_regions)
272 		return 0;
273 
274 	*addr = __pa(memblock.reserved.regions);
275 
276 	return PAGE_ALIGN(sizeof(struct memblock_region) *
277 			  memblock.reserved.max);
278 }
279 
get_allocated_memblock_memory_regions_info(phys_addr_t * addr)280 phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
281 					phys_addr_t *addr)
282 {
283 	if (memblock.memory.regions == memblock_memory_init_regions)
284 		return 0;
285 
286 	*addr = __pa(memblock.memory.regions);
287 
288 	return PAGE_ALIGN(sizeof(struct memblock_region) *
289 			  memblock.memory.max);
290 }
291 
292 #endif
293 
294 /**
295  * memblock_double_array - double the size of the memblock regions array
296  * @type: memblock type of the regions array being doubled
297  * @new_area_start: starting address of memory range to avoid overlap with
298  * @new_area_size: size of memory range to avoid overlap with
299  *
300  * Double the size of the @type regions array. If memblock is being used to
301  * allocate memory for a new reserved regions array and there is a previously
302  * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
303  * waiting to be reserved, ensure the memory used by the new array does
304  * not overlap.
305  *
306  * RETURNS:
307  * 0 on success, -1 on failure.
308  */
memblock_double_array(struct memblock_type * type,phys_addr_t new_area_start,phys_addr_t new_area_size)309 static int __init_memblock memblock_double_array(struct memblock_type *type,
310 						phys_addr_t new_area_start,
311 						phys_addr_t new_area_size)
312 {
313 	struct memblock_region *new_array, *old_array;
314 	phys_addr_t old_alloc_size, new_alloc_size;
315 	phys_addr_t old_size, new_size, addr;
316 	int use_slab = slab_is_available();
317 	int *in_slab;
318 
319 	/* We don't allow resizing until we know about the reserved regions
320 	 * of memory that aren't suitable for allocation
321 	 */
322 	if (!memblock_can_resize)
323 		return -1;
324 
325 	/* Calculate new doubled size */
326 	old_size = type->max * sizeof(struct memblock_region);
327 	new_size = old_size << 1;
328 	/*
329 	 * We need to allocated new one align to PAGE_SIZE,
330 	 *   so we can free them completely later.
331 	 */
332 	old_alloc_size = PAGE_ALIGN(old_size);
333 	new_alloc_size = PAGE_ALIGN(new_size);
334 
335 	/* Retrieve the slab flag */
336 	if (type == &memblock.memory)
337 		in_slab = &memblock_memory_in_slab;
338 	else
339 		in_slab = &memblock_reserved_in_slab;
340 
341 	/* Try to find some space for it.
342 	 *
343 	 * WARNING: We assume that either slab_is_available() and we use it or
344 	 * we use MEMBLOCK for allocations. That means that this is unsafe to
345 	 * use when bootmem is currently active (unless bootmem itself is
346 	 * implemented on top of MEMBLOCK which isn't the case yet)
347 	 *
348 	 * This should however not be an issue for now, as we currently only
349 	 * call into MEMBLOCK while it's still active, or much later when slab
350 	 * is active for memory hotplug operations
351 	 */
352 	if (use_slab) {
353 		new_array = kmalloc(new_size, GFP_KERNEL);
354 		addr = new_array ? __pa(new_array) : 0;
355 	} else {
356 		/* only exclude range when trying to double reserved.regions */
357 		if (type != &memblock.reserved)
358 			new_area_start = new_area_size = 0;
359 
360 		addr = memblock_find_in_range(new_area_start + new_area_size,
361 						memblock.current_limit,
362 						new_alloc_size, PAGE_SIZE);
363 		if (!addr && new_area_size)
364 			addr = memblock_find_in_range(0,
365 				min(new_area_start, memblock.current_limit),
366 				new_alloc_size, PAGE_SIZE);
367 
368 		new_array = addr ? __va(addr) : NULL;
369 	}
370 	if (!addr) {
371 		pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
372 		       memblock_type_name(type), type->max, type->max * 2);
373 		return -1;
374 	}
375 
376 	memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
377 			memblock_type_name(type), type->max * 2, (u64)addr,
378 			(u64)addr + new_size - 1);
379 
380 	/*
381 	 * Found space, we now need to move the array over before we add the
382 	 * reserved region since it may be our reserved array itself that is
383 	 * full.
384 	 */
385 	memcpy(new_array, type->regions, old_size);
386 	memset(new_array + type->max, 0, old_size);
387 	old_array = type->regions;
388 	type->regions = new_array;
389 	type->max <<= 1;
390 
391 	/* Free old array. We needn't free it if the array is the static one */
392 	if (*in_slab)
393 		kfree(old_array);
394 	else if (old_array != memblock_memory_init_regions &&
395 		 old_array != memblock_reserved_init_regions)
396 		memblock_free(__pa(old_array), old_alloc_size);
397 
398 	/*
399 	 * Reserve the new array if that comes from the memblock.  Otherwise, we
400 	 * needn't do it
401 	 */
402 	if (!use_slab)
403 		BUG_ON(memblock_reserve(addr, new_alloc_size));
404 
405 	/* Update slab flag */
406 	*in_slab = use_slab;
407 
408 	return 0;
409 }
410 
411 /**
412  * memblock_merge_regions - merge neighboring compatible regions
413  * @type: memblock type to scan
414  *
415  * Scan @type and merge neighboring compatible regions.
416  */
memblock_merge_regions(struct memblock_type * type)417 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
418 {
419 	int i = 0;
420 
421 	/* cnt never goes below 1 */
422 	while (i < type->cnt - 1) {
423 		struct memblock_region *this = &type->regions[i];
424 		struct memblock_region *next = &type->regions[i + 1];
425 
426 		if (this->base + this->size != next->base ||
427 		    memblock_get_region_node(this) !=
428 		    memblock_get_region_node(next) ||
429 		    this->flags != next->flags) {
430 			BUG_ON(this->base + this->size > next->base);
431 			i++;
432 			continue;
433 		}
434 
435 		this->size += next->size;
436 		/* move forward from next + 1, index of which is i + 2 */
437 		memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
438 		type->cnt--;
439 	}
440 }
441 
442 /**
443  * memblock_insert_region - insert new memblock region
444  * @type:	memblock type to insert into
445  * @idx:	index for the insertion point
446  * @base:	base address of the new region
447  * @size:	size of the new region
448  * @nid:	node id of the new region
449  * @flags:	flags of the new region
450  *
451  * Insert new memblock region [@base,@base+@size) into @type at @idx.
452  * @type must already have extra room to accomodate the new region.
453  */
memblock_insert_region(struct memblock_type * type,int idx,phys_addr_t base,phys_addr_t size,int nid,unsigned long flags)454 static void __init_memblock memblock_insert_region(struct memblock_type *type,
455 						   int idx, phys_addr_t base,
456 						   phys_addr_t size,
457 						   int nid, unsigned long flags)
458 {
459 	struct memblock_region *rgn = &type->regions[idx];
460 
461 	BUG_ON(type->cnt >= type->max);
462 	memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
463 	rgn->base = base;
464 	rgn->size = size;
465 	rgn->flags = flags;
466 	memblock_set_region_node(rgn, nid);
467 	type->cnt++;
468 	type->total_size += size;
469 }
470 
471 /**
472  * memblock_add_range - add new memblock region
473  * @type: memblock type to add new region into
474  * @base: base address of the new region
475  * @size: size of the new region
476  * @nid: nid of the new region
477  * @flags: flags of the new region
478  *
479  * Add new memblock region [@base,@base+@size) into @type.  The new region
480  * is allowed to overlap with existing ones - overlaps don't affect already
481  * existing regions.  @type is guaranteed to be minimal (all neighbouring
482  * compatible regions are merged) after the addition.
483  *
484  * RETURNS:
485  * 0 on success, -errno on failure.
486  */
memblock_add_range(struct memblock_type * type,phys_addr_t base,phys_addr_t size,int nid,unsigned long flags)487 int __init_memblock memblock_add_range(struct memblock_type *type,
488 				phys_addr_t base, phys_addr_t size,
489 				int nid, unsigned long flags)
490 {
491 	bool insert = false;
492 	phys_addr_t obase = base;
493 	phys_addr_t end = base + memblock_cap_size(base, &size);
494 	int i, nr_new;
495 
496 	if (!size)
497 		return 0;
498 
499 	/* special case for empty array */
500 	if (type->regions[0].size == 0) {
501 		WARN_ON(type->cnt != 1 || type->total_size);
502 		type->regions[0].base = base;
503 		type->regions[0].size = size;
504 		type->regions[0].flags = flags;
505 		memblock_set_region_node(&type->regions[0], nid);
506 		type->total_size = size;
507 		return 0;
508 	}
509 repeat:
510 	/*
511 	 * The following is executed twice.  Once with %false @insert and
512 	 * then with %true.  The first counts the number of regions needed
513 	 * to accomodate the new area.  The second actually inserts them.
514 	 */
515 	base = obase;
516 	nr_new = 0;
517 
518 	for (i = 0; i < type->cnt; i++) {
519 		struct memblock_region *rgn = &type->regions[i];
520 		phys_addr_t rbase = rgn->base;
521 		phys_addr_t rend = rbase + rgn->size;
522 
523 		if (rbase >= end)
524 			break;
525 		if (rend <= base)
526 			continue;
527 		/*
528 		 * @rgn overlaps.  If it separates the lower part of new
529 		 * area, insert that portion.
530 		 */
531 		if (rbase > base) {
532 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
533 			WARN_ON(nid != memblock_get_region_node(rgn));
534 #endif
535 			WARN_ON(flags != rgn->flags);
536 			nr_new++;
537 			if (insert)
538 				memblock_insert_region(type, i++, base,
539 						       rbase - base, nid,
540 						       flags);
541 		}
542 		/* area below @rend is dealt with, forget about it */
543 		base = min(rend, end);
544 	}
545 
546 	/* insert the remaining portion */
547 	if (base < end) {
548 		nr_new++;
549 		if (insert)
550 			memblock_insert_region(type, i, base, end - base,
551 					       nid, flags);
552 	}
553 
554 	/*
555 	 * If this was the first round, resize array and repeat for actual
556 	 * insertions; otherwise, merge and return.
557 	 */
558 	if (!insert) {
559 		while (type->cnt + nr_new > type->max)
560 			if (memblock_double_array(type, obase, size) < 0)
561 				return -ENOMEM;
562 		insert = true;
563 		goto repeat;
564 	} else {
565 		memblock_merge_regions(type);
566 		return 0;
567 	}
568 }
569 
memblock_add_node(phys_addr_t base,phys_addr_t size,int nid)570 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
571 				       int nid)
572 {
573 	return memblock_add_range(&memblock.memory, base, size, nid, 0);
574 }
575 
memblock_add_region(phys_addr_t base,phys_addr_t size,int nid,unsigned long flags)576 static int __init_memblock memblock_add_region(phys_addr_t base,
577 						phys_addr_t size,
578 						int nid,
579 						unsigned long flags)
580 {
581 	struct memblock_type *type = &memblock.memory;
582 
583 	memblock_dbg("memblock_add: [%#016llx-%#016llx] flags %#02lx %pF\n",
584 		     (unsigned long long)base,
585 		     (unsigned long long)base + size - 1,
586 		     flags, (void *)_RET_IP_);
587 
588 	return memblock_add_range(type, base, size, nid, flags);
589 }
590 
memblock_add(phys_addr_t base,phys_addr_t size)591 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
592 {
593 	return memblock_add_region(base, size, MAX_NUMNODES, 0);
594 }
595 
596 /**
597  * memblock_isolate_range - isolate given range into disjoint memblocks
598  * @type: memblock type to isolate range for
599  * @base: base of range to isolate
600  * @size: size of range to isolate
601  * @start_rgn: out parameter for the start of isolated region
602  * @end_rgn: out parameter for the end of isolated region
603  *
604  * Walk @type and ensure that regions don't cross the boundaries defined by
605  * [@base,@base+@size).  Crossing regions are split at the boundaries,
606  * which may create at most two more regions.  The index of the first
607  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
608  *
609  * RETURNS:
610  * 0 on success, -errno on failure.
611  */
memblock_isolate_range(struct memblock_type * type,phys_addr_t base,phys_addr_t size,int * start_rgn,int * end_rgn)612 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
613 					phys_addr_t base, phys_addr_t size,
614 					int *start_rgn, int *end_rgn)
615 {
616 	phys_addr_t end = base + memblock_cap_size(base, &size);
617 	int i;
618 
619 	*start_rgn = *end_rgn = 0;
620 
621 	if (!size)
622 		return 0;
623 
624 	/* we'll create at most two more regions */
625 	while (type->cnt + 2 > type->max)
626 		if (memblock_double_array(type, base, size) < 0)
627 			return -ENOMEM;
628 
629 	for (i = 0; i < type->cnt; i++) {
630 		struct memblock_region *rgn = &type->regions[i];
631 		phys_addr_t rbase = rgn->base;
632 		phys_addr_t rend = rbase + rgn->size;
633 
634 		if (rbase >= end)
635 			break;
636 		if (rend <= base)
637 			continue;
638 
639 		if (rbase < base) {
640 			/*
641 			 * @rgn intersects from below.  Split and continue
642 			 * to process the next region - the new top half.
643 			 */
644 			rgn->base = base;
645 			rgn->size -= base - rbase;
646 			type->total_size -= base - rbase;
647 			memblock_insert_region(type, i, rbase, base - rbase,
648 					       memblock_get_region_node(rgn),
649 					       rgn->flags);
650 		} else if (rend > end) {
651 			/*
652 			 * @rgn intersects from above.  Split and redo the
653 			 * current region - the new bottom half.
654 			 */
655 			rgn->base = end;
656 			rgn->size -= end - rbase;
657 			type->total_size -= end - rbase;
658 			memblock_insert_region(type, i--, rbase, end - rbase,
659 					       memblock_get_region_node(rgn),
660 					       rgn->flags);
661 		} else {
662 			/* @rgn is fully contained, record it */
663 			if (!*end_rgn)
664 				*start_rgn = i;
665 			*end_rgn = i + 1;
666 		}
667 	}
668 
669 	return 0;
670 }
671 
memblock_remove_range(struct memblock_type * type,phys_addr_t base,phys_addr_t size)672 static int __init_memblock memblock_remove_range(struct memblock_type *type,
673 					  phys_addr_t base, phys_addr_t size)
674 {
675 	int start_rgn, end_rgn;
676 	int i, ret;
677 
678 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
679 	if (ret)
680 		return ret;
681 
682 	for (i = end_rgn - 1; i >= start_rgn; i--)
683 		memblock_remove_region(type, i);
684 	return 0;
685 }
686 
memblock_remove(phys_addr_t base,phys_addr_t size)687 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
688 {
689 	return memblock_remove_range(&memblock.memory, base, size);
690 }
691 
692 
memblock_free(phys_addr_t base,phys_addr_t size)693 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
694 {
695 	memblock_dbg("   memblock_free: [%#016llx-%#016llx] %pF\n",
696 		     (unsigned long long)base,
697 		     (unsigned long long)base + size - 1,
698 		     (void *)_RET_IP_);
699 
700 	kmemleak_free_part(__va(base), size);
701 	return memblock_remove_range(&memblock.reserved, base, size);
702 }
703 
memblock_reserve_region(phys_addr_t base,phys_addr_t size,int nid,unsigned long flags)704 static int __init_memblock memblock_reserve_region(phys_addr_t base,
705 						   phys_addr_t size,
706 						   int nid,
707 						   unsigned long flags)
708 {
709 	struct memblock_type *type = &memblock.reserved;
710 
711 	memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
712 		     (unsigned long long)base,
713 		     (unsigned long long)base + size - 1,
714 		     flags, (void *)_RET_IP_);
715 
716 	return memblock_add_range(type, base, size, nid, flags);
717 }
718 
memblock_reserve(phys_addr_t base,phys_addr_t size)719 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
720 {
721 	return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
722 }
723 
724 /**
725  *
726  * This function isolates region [@base, @base + @size), and sets/clears flag
727  *
728  * Return 0 on success, -errno on failure.
729  */
memblock_setclr_flag(phys_addr_t base,phys_addr_t size,int set,int flag)730 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
731 				phys_addr_t size, int set, int flag)
732 {
733 	struct memblock_type *type = &memblock.memory;
734 	int i, ret, start_rgn, end_rgn;
735 
736 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
737 	if (ret)
738 		return ret;
739 
740 	for (i = start_rgn; i < end_rgn; i++)
741 		if (set)
742 			memblock_set_region_flags(&type->regions[i], flag);
743 		else
744 			memblock_clear_region_flags(&type->regions[i], flag);
745 
746 	memblock_merge_regions(type);
747 	return 0;
748 }
749 
750 /**
751  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
752  * @base: the base phys addr of the region
753  * @size: the size of the region
754  *
755  * Return 0 on success, -errno on failure.
756  */
memblock_mark_hotplug(phys_addr_t base,phys_addr_t size)757 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
758 {
759 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
760 }
761 
762 /**
763  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
764  * @base: the base phys addr of the region
765  * @size: the size of the region
766  *
767  * Return 0 on success, -errno on failure.
768  */
memblock_clear_hotplug(phys_addr_t base,phys_addr_t size)769 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
770 {
771 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
772 }
773 
774 /**
775  * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
776  * @base: the base phys addr of the region
777  * @size: the size of the region
778  *
779  * Return 0 on success, -errno on failure.
780  */
memblock_mark_mirror(phys_addr_t base,phys_addr_t size)781 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
782 {
783 	system_has_some_mirror = true;
784 
785 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
786 }
787 
788 /**
789  * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
790  * @base: the base phys addr of the region
791  * @size: the size of the region
792  *
793  * Return 0 on success, -errno on failure.
794  */
memblock_mark_nomap(phys_addr_t base,phys_addr_t size)795 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
796 {
797 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
798 }
799 
800 /**
801  * __next_reserved_mem_region - next function for for_each_reserved_region()
802  * @idx: pointer to u64 loop variable
803  * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
804  * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
805  *
806  * Iterate over all reserved memory regions.
807  */
__next_reserved_mem_region(u64 * idx,phys_addr_t * out_start,phys_addr_t * out_end)808 void __init_memblock __next_reserved_mem_region(u64 *idx,
809 					   phys_addr_t *out_start,
810 					   phys_addr_t *out_end)
811 {
812 	struct memblock_type *type = &memblock.reserved;
813 
814 	if (*idx >= 0 && *idx < type->cnt) {
815 		struct memblock_region *r = &type->regions[*idx];
816 		phys_addr_t base = r->base;
817 		phys_addr_t size = r->size;
818 
819 		if (out_start)
820 			*out_start = base;
821 		if (out_end)
822 			*out_end = base + size - 1;
823 
824 		*idx += 1;
825 		return;
826 	}
827 
828 	/* signal end of iteration */
829 	*idx = ULLONG_MAX;
830 }
831 
832 /**
833  * __next__mem_range - next function for for_each_free_mem_range() etc.
834  * @idx: pointer to u64 loop variable
835  * @nid: node selector, %NUMA_NO_NODE for all nodes
836  * @flags: pick from blocks based on memory attributes
837  * @type_a: pointer to memblock_type from where the range is taken
838  * @type_b: pointer to memblock_type which excludes memory from being taken
839  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
840  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
841  * @out_nid: ptr to int for nid of the range, can be %NULL
842  *
843  * Find the first area from *@idx which matches @nid, fill the out
844  * parameters, and update *@idx for the next iteration.  The lower 32bit of
845  * *@idx contains index into type_a and the upper 32bit indexes the
846  * areas before each region in type_b.	For example, if type_b regions
847  * look like the following,
848  *
849  *	0:[0-16), 1:[32-48), 2:[128-130)
850  *
851  * The upper 32bit indexes the following regions.
852  *
853  *	0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
854  *
855  * As both region arrays are sorted, the function advances the two indices
856  * in lockstep and returns each intersection.
857  */
__next_mem_range(u64 * idx,int nid,ulong flags,struct memblock_type * type_a,struct memblock_type * type_b,phys_addr_t * out_start,phys_addr_t * out_end,int * out_nid)858 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
859 				      struct memblock_type *type_a,
860 				      struct memblock_type *type_b,
861 				      phys_addr_t *out_start,
862 				      phys_addr_t *out_end, int *out_nid)
863 {
864 	int idx_a = *idx & 0xffffffff;
865 	int idx_b = *idx >> 32;
866 
867 	if (WARN_ONCE(nid == MAX_NUMNODES,
868 	"Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
869 		nid = NUMA_NO_NODE;
870 
871 	for (; idx_a < type_a->cnt; idx_a++) {
872 		struct memblock_region *m = &type_a->regions[idx_a];
873 
874 		phys_addr_t m_start = m->base;
875 		phys_addr_t m_end = m->base + m->size;
876 		int	    m_nid = memblock_get_region_node(m);
877 
878 		/* only memory regions are associated with nodes, check it */
879 		if (nid != NUMA_NO_NODE && nid != m_nid)
880 			continue;
881 
882 		/* skip hotpluggable memory regions if needed */
883 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
884 			continue;
885 
886 		/* if we want mirror memory skip non-mirror memory regions */
887 		if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
888 			continue;
889 
890 		/* skip nomap memory unless we were asked for it explicitly */
891 		if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
892 			continue;
893 
894 		if (!type_b) {
895 			if (out_start)
896 				*out_start = m_start;
897 			if (out_end)
898 				*out_end = m_end;
899 			if (out_nid)
900 				*out_nid = m_nid;
901 			idx_a++;
902 			*idx = (u32)idx_a | (u64)idx_b << 32;
903 			return;
904 		}
905 
906 		/* scan areas before each reservation */
907 		for (; idx_b < type_b->cnt + 1; idx_b++) {
908 			struct memblock_region *r;
909 			phys_addr_t r_start;
910 			phys_addr_t r_end;
911 
912 			r = &type_b->regions[idx_b];
913 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
914 			r_end = idx_b < type_b->cnt ?
915 				r->base : ULLONG_MAX;
916 
917 			/*
918 			 * if idx_b advanced past idx_a,
919 			 * break out to advance idx_a
920 			 */
921 			if (r_start >= m_end)
922 				break;
923 			/* if the two regions intersect, we're done */
924 			if (m_start < r_end) {
925 				if (out_start)
926 					*out_start =
927 						max(m_start, r_start);
928 				if (out_end)
929 					*out_end = min(m_end, r_end);
930 				if (out_nid)
931 					*out_nid = m_nid;
932 				/*
933 				 * The region which ends first is
934 				 * advanced for the next iteration.
935 				 */
936 				if (m_end <= r_end)
937 					idx_a++;
938 				else
939 					idx_b++;
940 				*idx = (u32)idx_a | (u64)idx_b << 32;
941 				return;
942 			}
943 		}
944 	}
945 
946 	/* signal end of iteration */
947 	*idx = ULLONG_MAX;
948 }
949 
950 /**
951  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
952  *
953  * Finds the next range from type_a which is not marked as unsuitable
954  * in type_b.
955  *
956  * @idx: pointer to u64 loop variable
957  * @nid: node selector, %NUMA_NO_NODE for all nodes
958  * @flags: pick from blocks based on memory attributes
959  * @type_a: pointer to memblock_type from where the range is taken
960  * @type_b: pointer to memblock_type which excludes memory from being taken
961  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
962  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
963  * @out_nid: ptr to int for nid of the range, can be %NULL
964  *
965  * Reverse of __next_mem_range().
966  */
__next_mem_range_rev(u64 * idx,int nid,ulong flags,struct memblock_type * type_a,struct memblock_type * type_b,phys_addr_t * out_start,phys_addr_t * out_end,int * out_nid)967 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
968 					  struct memblock_type *type_a,
969 					  struct memblock_type *type_b,
970 					  phys_addr_t *out_start,
971 					  phys_addr_t *out_end, int *out_nid)
972 {
973 	int idx_a = *idx & 0xffffffff;
974 	int idx_b = *idx >> 32;
975 
976 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
977 		nid = NUMA_NO_NODE;
978 
979 	if (*idx == (u64)ULLONG_MAX) {
980 		idx_a = type_a->cnt - 1;
981 		idx_b = type_b->cnt;
982 	}
983 
984 	for (; idx_a >= 0; idx_a--) {
985 		struct memblock_region *m = &type_a->regions[idx_a];
986 
987 		phys_addr_t m_start = m->base;
988 		phys_addr_t m_end = m->base + m->size;
989 		int m_nid = memblock_get_region_node(m);
990 
991 		/* only memory regions are associated with nodes, check it */
992 		if (nid != NUMA_NO_NODE && nid != m_nid)
993 			continue;
994 
995 		/* skip hotpluggable memory regions if needed */
996 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
997 			continue;
998 
999 		/* if we want mirror memory skip non-mirror memory regions */
1000 		if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1001 			continue;
1002 
1003 		/* skip nomap memory unless we were asked for it explicitly */
1004 		if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1005 			continue;
1006 
1007 		if (!type_b) {
1008 			if (out_start)
1009 				*out_start = m_start;
1010 			if (out_end)
1011 				*out_end = m_end;
1012 			if (out_nid)
1013 				*out_nid = m_nid;
1014 			idx_a++;
1015 			*idx = (u32)idx_a | (u64)idx_b << 32;
1016 			return;
1017 		}
1018 
1019 		/* scan areas before each reservation */
1020 		for (; idx_b >= 0; idx_b--) {
1021 			struct memblock_region *r;
1022 			phys_addr_t r_start;
1023 			phys_addr_t r_end;
1024 
1025 			r = &type_b->regions[idx_b];
1026 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1027 			r_end = idx_b < type_b->cnt ?
1028 				r->base : ULLONG_MAX;
1029 			/*
1030 			 * if idx_b advanced past idx_a,
1031 			 * break out to advance idx_a
1032 			 */
1033 
1034 			if (r_end <= m_start)
1035 				break;
1036 			/* if the two regions intersect, we're done */
1037 			if (m_end > r_start) {
1038 				if (out_start)
1039 					*out_start = max(m_start, r_start);
1040 				if (out_end)
1041 					*out_end = min(m_end, r_end);
1042 				if (out_nid)
1043 					*out_nid = m_nid;
1044 				if (m_start >= r_start)
1045 					idx_a--;
1046 				else
1047 					idx_b--;
1048 				*idx = (u32)idx_a | (u64)idx_b << 32;
1049 				return;
1050 			}
1051 		}
1052 	}
1053 	/* signal end of iteration */
1054 	*idx = ULLONG_MAX;
1055 }
1056 
1057 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1058 /*
1059  * Common iterator interface used to define for_each_mem_range().
1060  */
__next_mem_pfn_range(int * idx,int nid,unsigned long * out_start_pfn,unsigned long * out_end_pfn,int * out_nid)1061 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1062 				unsigned long *out_start_pfn,
1063 				unsigned long *out_end_pfn, int *out_nid)
1064 {
1065 	struct memblock_type *type = &memblock.memory;
1066 	struct memblock_region *r;
1067 
1068 	while (++*idx < type->cnt) {
1069 		r = &type->regions[*idx];
1070 
1071 		if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1072 			continue;
1073 		if (nid == MAX_NUMNODES || nid == r->nid)
1074 			break;
1075 	}
1076 	if (*idx >= type->cnt) {
1077 		*idx = -1;
1078 		return;
1079 	}
1080 
1081 	if (out_start_pfn)
1082 		*out_start_pfn = PFN_UP(r->base);
1083 	if (out_end_pfn)
1084 		*out_end_pfn = PFN_DOWN(r->base + r->size);
1085 	if (out_nid)
1086 		*out_nid = r->nid;
1087 }
1088 
1089 /**
1090  * memblock_set_node - set node ID on memblock regions
1091  * @base: base of area to set node ID for
1092  * @size: size of area to set node ID for
1093  * @type: memblock type to set node ID for
1094  * @nid: node ID to set
1095  *
1096  * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1097  * Regions which cross the area boundaries are split as necessary.
1098  *
1099  * RETURNS:
1100  * 0 on success, -errno on failure.
1101  */
memblock_set_node(phys_addr_t base,phys_addr_t size,struct memblock_type * type,int nid)1102 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1103 				      struct memblock_type *type, int nid)
1104 {
1105 	int start_rgn, end_rgn;
1106 	int i, ret;
1107 
1108 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1109 	if (ret)
1110 		return ret;
1111 
1112 	for (i = start_rgn; i < end_rgn; i++)
1113 		memblock_set_region_node(&type->regions[i], nid);
1114 
1115 	memblock_merge_regions(type);
1116 	return 0;
1117 }
1118 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1119 
memblock_alloc_range_nid(phys_addr_t size,phys_addr_t align,phys_addr_t start,phys_addr_t end,int nid,ulong flags)1120 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1121 					phys_addr_t align, phys_addr_t start,
1122 					phys_addr_t end, int nid, ulong flags)
1123 {
1124 	phys_addr_t found;
1125 
1126 	if (!align)
1127 		align = SMP_CACHE_BYTES;
1128 
1129 	found = memblock_find_in_range_node(size, align, start, end, nid,
1130 					    flags);
1131 	if (found && !memblock_reserve(found, size)) {
1132 		/*
1133 		 * The min_count is set to 0 so that memblock allocations are
1134 		 * never reported as leaks.
1135 		 */
1136 		kmemleak_alloc(__va(found), size, 0, 0);
1137 		return found;
1138 	}
1139 	return 0;
1140 }
1141 
memblock_alloc_range(phys_addr_t size,phys_addr_t align,phys_addr_t start,phys_addr_t end,ulong flags)1142 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1143 					phys_addr_t start, phys_addr_t end,
1144 					ulong flags)
1145 {
1146 	return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1147 					flags);
1148 }
1149 
memblock_alloc_base_nid(phys_addr_t size,phys_addr_t align,phys_addr_t max_addr,int nid,ulong flags)1150 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1151 					phys_addr_t align, phys_addr_t max_addr,
1152 					int nid, ulong flags)
1153 {
1154 	return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1155 }
1156 
memblock_alloc_nid(phys_addr_t size,phys_addr_t align,int nid)1157 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1158 {
1159 	ulong flags = choose_memblock_flags();
1160 	phys_addr_t ret;
1161 
1162 again:
1163 	ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1164 				      nid, flags);
1165 
1166 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
1167 		flags &= ~MEMBLOCK_MIRROR;
1168 		goto again;
1169 	}
1170 	return ret;
1171 }
1172 
__memblock_alloc_base(phys_addr_t size,phys_addr_t align,phys_addr_t max_addr)1173 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1174 {
1175 	return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1176 				       MEMBLOCK_NONE);
1177 }
1178 
memblock_alloc_base(phys_addr_t size,phys_addr_t align,phys_addr_t max_addr)1179 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1180 {
1181 	phys_addr_t alloc;
1182 
1183 	alloc = __memblock_alloc_base(size, align, max_addr);
1184 
1185 	if (alloc == 0)
1186 		panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1187 		      (unsigned long long) size, (unsigned long long) max_addr);
1188 
1189 	return alloc;
1190 }
1191 
memblock_alloc(phys_addr_t size,phys_addr_t align)1192 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1193 {
1194 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1195 }
1196 
memblock_alloc_try_nid(phys_addr_t size,phys_addr_t align,int nid)1197 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1198 {
1199 	phys_addr_t res = memblock_alloc_nid(size, align, nid);
1200 
1201 	if (res)
1202 		return res;
1203 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1204 }
1205 
1206 /**
1207  * memblock_virt_alloc_internal - allocate boot memory block
1208  * @size: size of memory block to be allocated in bytes
1209  * @align: alignment of the region and block's size
1210  * @min_addr: the lower bound of the memory region to allocate (phys address)
1211  * @max_addr: the upper bound of the memory region to allocate (phys address)
1212  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1213  *
1214  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1215  * will fall back to memory below @min_addr. Also, allocation may fall back
1216  * to any node in the system if the specified node can not
1217  * hold the requested memory.
1218  *
1219  * The allocation is performed from memory region limited by
1220  * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1221  *
1222  * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1223  *
1224  * The phys address of allocated boot memory block is converted to virtual and
1225  * allocated memory is reset to 0.
1226  *
1227  * In addition, function sets the min_count to 0 using kmemleak_alloc for
1228  * allocated boot memory block, so that it is never reported as leaks.
1229  *
1230  * RETURNS:
1231  * Virtual address of allocated memory block on success, NULL on failure.
1232  */
memblock_virt_alloc_internal(phys_addr_t size,phys_addr_t align,phys_addr_t min_addr,phys_addr_t max_addr,int nid)1233 static void * __init memblock_virt_alloc_internal(
1234 				phys_addr_t size, phys_addr_t align,
1235 				phys_addr_t min_addr, phys_addr_t max_addr,
1236 				int nid)
1237 {
1238 	phys_addr_t alloc;
1239 	void *ptr;
1240 	ulong flags = choose_memblock_flags();
1241 
1242 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1243 		nid = NUMA_NO_NODE;
1244 
1245 	/*
1246 	 * Detect any accidental use of these APIs after slab is ready, as at
1247 	 * this moment memblock may be deinitialized already and its
1248 	 * internal data may be destroyed (after execution of free_all_bootmem)
1249 	 */
1250 	if (WARN_ON_ONCE(slab_is_available()))
1251 		return kzalloc_node(size, GFP_NOWAIT, nid);
1252 
1253 	if (!align)
1254 		align = SMP_CACHE_BYTES;
1255 
1256 	if (max_addr > memblock.current_limit)
1257 		max_addr = memblock.current_limit;
1258 
1259 again:
1260 	alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1261 					    nid, flags);
1262 	if (alloc)
1263 		goto done;
1264 
1265 	if (nid != NUMA_NO_NODE) {
1266 		alloc = memblock_find_in_range_node(size, align, min_addr,
1267 						    max_addr, NUMA_NO_NODE,
1268 						    flags);
1269 		if (alloc)
1270 			goto done;
1271 	}
1272 
1273 	if (min_addr) {
1274 		min_addr = 0;
1275 		goto again;
1276 	}
1277 
1278 	if (flags & MEMBLOCK_MIRROR) {
1279 		flags &= ~MEMBLOCK_MIRROR;
1280 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1281 			&size);
1282 		goto again;
1283 	}
1284 
1285 	return NULL;
1286 done:
1287 	memblock_reserve(alloc, size);
1288 	ptr = phys_to_virt(alloc);
1289 	memset(ptr, 0, size);
1290 
1291 	/*
1292 	 * The min_count is set to 0 so that bootmem allocated blocks
1293 	 * are never reported as leaks. This is because many of these blocks
1294 	 * are only referred via the physical address which is not
1295 	 * looked up by kmemleak.
1296 	 */
1297 	kmemleak_alloc(ptr, size, 0, 0);
1298 
1299 	return ptr;
1300 }
1301 
1302 /**
1303  * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1304  * @size: size of memory block to be allocated in bytes
1305  * @align: alignment of the region and block's size
1306  * @min_addr: the lower bound of the memory region from where the allocation
1307  *	  is preferred (phys address)
1308  * @max_addr: the upper bound of the memory region from where the allocation
1309  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1310  *	      allocate only from memory limited by memblock.current_limit value
1311  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1312  *
1313  * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1314  * additional debug information (including caller info), if enabled.
1315  *
1316  * RETURNS:
1317  * Virtual address of allocated memory block on success, NULL on failure.
1318  */
memblock_virt_alloc_try_nid_nopanic(phys_addr_t size,phys_addr_t align,phys_addr_t min_addr,phys_addr_t max_addr,int nid)1319 void * __init memblock_virt_alloc_try_nid_nopanic(
1320 				phys_addr_t size, phys_addr_t align,
1321 				phys_addr_t min_addr, phys_addr_t max_addr,
1322 				int nid)
1323 {
1324 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1325 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1326 		     (u64)max_addr, (void *)_RET_IP_);
1327 	return memblock_virt_alloc_internal(size, align, min_addr,
1328 					     max_addr, nid);
1329 }
1330 
1331 /**
1332  * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1333  * @size: size of memory block to be allocated in bytes
1334  * @align: alignment of the region and block's size
1335  * @min_addr: the lower bound of the memory region from where the allocation
1336  *	  is preferred (phys address)
1337  * @max_addr: the upper bound of the memory region from where the allocation
1338  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1339  *	      allocate only from memory limited by memblock.current_limit value
1340  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1341  *
1342  * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1343  * which provides debug information (including caller info), if enabled,
1344  * and panics if the request can not be satisfied.
1345  *
1346  * RETURNS:
1347  * Virtual address of allocated memory block on success, NULL on failure.
1348  */
memblock_virt_alloc_try_nid(phys_addr_t size,phys_addr_t align,phys_addr_t min_addr,phys_addr_t max_addr,int nid)1349 void * __init memblock_virt_alloc_try_nid(
1350 			phys_addr_t size, phys_addr_t align,
1351 			phys_addr_t min_addr, phys_addr_t max_addr,
1352 			int nid)
1353 {
1354 	void *ptr;
1355 
1356 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1357 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1358 		     (u64)max_addr, (void *)_RET_IP_);
1359 	ptr = memblock_virt_alloc_internal(size, align,
1360 					   min_addr, max_addr, nid);
1361 	if (ptr)
1362 		return ptr;
1363 
1364 	panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1365 	      __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1366 	      (u64)max_addr);
1367 	return NULL;
1368 }
1369 
1370 /**
1371  * __memblock_free_early - free boot memory block
1372  * @base: phys starting address of the  boot memory block
1373  * @size: size of the boot memory block in bytes
1374  *
1375  * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1376  * The freeing memory will not be released to the buddy allocator.
1377  */
__memblock_free_early(phys_addr_t base,phys_addr_t size)1378 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1379 {
1380 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1381 		     __func__, (u64)base, (u64)base + size - 1,
1382 		     (void *)_RET_IP_);
1383 	kmemleak_free_part(__va(base), size);
1384 	memblock_remove_range(&memblock.reserved, base, size);
1385 }
1386 
1387 /*
1388  * __memblock_free_late - free bootmem block pages directly to buddy allocator
1389  * @addr: phys starting address of the  boot memory block
1390  * @size: size of the boot memory block in bytes
1391  *
1392  * This is only useful when the bootmem allocator has already been torn
1393  * down, but we are still initializing the system.  Pages are released directly
1394  * to the buddy allocator, no bootmem metadata is updated because it is gone.
1395  */
__memblock_free_late(phys_addr_t base,phys_addr_t size)1396 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1397 {
1398 	u64 cursor, end;
1399 
1400 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1401 		     __func__, (u64)base, (u64)base + size - 1,
1402 		     (void *)_RET_IP_);
1403 	kmemleak_free_part(__va(base), size);
1404 	cursor = PFN_UP(base);
1405 	end = PFN_DOWN(base + size);
1406 
1407 	for (; cursor < end; cursor++) {
1408 		__free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1409 		totalram_pages++;
1410 	}
1411 }
1412 
1413 /*
1414  * Remaining API functions
1415  */
1416 
memblock_phys_mem_size(void)1417 phys_addr_t __init memblock_phys_mem_size(void)
1418 {
1419 	return memblock.memory.total_size;
1420 }
1421 
memblock_mem_size(unsigned long limit_pfn)1422 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1423 {
1424 	unsigned long pages = 0;
1425 	struct memblock_region *r;
1426 	unsigned long start_pfn, end_pfn;
1427 
1428 	for_each_memblock(memory, r) {
1429 		start_pfn = memblock_region_memory_base_pfn(r);
1430 		end_pfn = memblock_region_memory_end_pfn(r);
1431 		start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1432 		end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1433 		pages += end_pfn - start_pfn;
1434 	}
1435 
1436 	return PFN_PHYS(pages);
1437 }
1438 
1439 /* lowest address */
memblock_start_of_DRAM(void)1440 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1441 {
1442 	return memblock.memory.regions[0].base;
1443 }
1444 
memblock_end_of_DRAM(void)1445 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1446 {
1447 	int idx = memblock.memory.cnt - 1;
1448 
1449 	return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1450 }
1451 
memblock_enforce_memory_limit(phys_addr_t limit)1452 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1453 {
1454 	phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1455 	struct memblock_region *r;
1456 
1457 	if (!limit)
1458 		return;
1459 
1460 	/* find out max address */
1461 	for_each_memblock(memory, r) {
1462 		if (limit <= r->size) {
1463 			max_addr = r->base + limit;
1464 			break;
1465 		}
1466 		limit -= r->size;
1467 	}
1468 
1469 	/* truncate both memory and reserved regions */
1470 	memblock_remove_range(&memblock.memory, max_addr,
1471 			      (phys_addr_t)ULLONG_MAX);
1472 	memblock_remove_range(&memblock.reserved, max_addr,
1473 			      (phys_addr_t)ULLONG_MAX);
1474 }
1475 
memblock_search(struct memblock_type * type,phys_addr_t addr)1476 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1477 {
1478 	unsigned int left = 0, right = type->cnt;
1479 
1480 	do {
1481 		unsigned int mid = (right + left) / 2;
1482 
1483 		if (addr < type->regions[mid].base)
1484 			right = mid;
1485 		else if (addr >= (type->regions[mid].base +
1486 				  type->regions[mid].size))
1487 			left = mid + 1;
1488 		else
1489 			return mid;
1490 	} while (left < right);
1491 	return -1;
1492 }
1493 
memblock_is_reserved(phys_addr_t addr)1494 int __init memblock_is_reserved(phys_addr_t addr)
1495 {
1496 	return memblock_search(&memblock.reserved, addr) != -1;
1497 }
1498 
memblock_is_memory(phys_addr_t addr)1499 int __init_memblock memblock_is_memory(phys_addr_t addr)
1500 {
1501 	return memblock_search(&memblock.memory, addr) != -1;
1502 }
1503 
memblock_is_map_memory(phys_addr_t addr)1504 int __init_memblock memblock_is_map_memory(phys_addr_t addr)
1505 {
1506 	int i = memblock_search(&memblock.memory, addr);
1507 
1508 	if (i == -1)
1509 		return false;
1510 	return !memblock_is_nomap(&memblock.memory.regions[i]);
1511 }
1512 
1513 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
memblock_search_pfn_nid(unsigned long pfn,unsigned long * start_pfn,unsigned long * end_pfn)1514 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1515 			 unsigned long *start_pfn, unsigned long *end_pfn)
1516 {
1517 	struct memblock_type *type = &memblock.memory;
1518 	int mid = memblock_search(type, PFN_PHYS(pfn));
1519 
1520 	if (mid == -1)
1521 		return -1;
1522 
1523 	*start_pfn = PFN_DOWN(type->regions[mid].base);
1524 	*end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1525 
1526 	return type->regions[mid].nid;
1527 }
1528 #endif
1529 
1530 /**
1531  * memblock_is_region_memory - check if a region is a subset of memory
1532  * @base: base of region to check
1533  * @size: size of region to check
1534  *
1535  * Check if the region [@base, @base+@size) is a subset of a memory block.
1536  *
1537  * RETURNS:
1538  * 0 if false, non-zero if true
1539  */
memblock_is_region_memory(phys_addr_t base,phys_addr_t size)1540 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1541 {
1542 	int idx = memblock_search(&memblock.memory, base);
1543 	phys_addr_t end = base + memblock_cap_size(base, &size);
1544 
1545 	if (idx == -1)
1546 		return 0;
1547 	return memblock.memory.regions[idx].base <= base &&
1548 		(memblock.memory.regions[idx].base +
1549 		 memblock.memory.regions[idx].size) >= end;
1550 }
1551 
1552 /**
1553  * memblock_is_region_reserved - check if a region intersects reserved memory
1554  * @base: base of region to check
1555  * @size: size of region to check
1556  *
1557  * Check if the region [@base, @base+@size) intersects a reserved memory block.
1558  *
1559  * RETURNS:
1560  * True if they intersect, false if not.
1561  */
memblock_is_region_reserved(phys_addr_t base,phys_addr_t size)1562 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1563 {
1564 	memblock_cap_size(base, &size);
1565 	return memblock_overlaps_region(&memblock.reserved, base, size);
1566 }
1567 
memblock_trim_memory(phys_addr_t align)1568 void __init_memblock memblock_trim_memory(phys_addr_t align)
1569 {
1570 	phys_addr_t start, end, orig_start, orig_end;
1571 	struct memblock_region *r;
1572 
1573 	for_each_memblock(memory, r) {
1574 		orig_start = r->base;
1575 		orig_end = r->base + r->size;
1576 		start = round_up(orig_start, align);
1577 		end = round_down(orig_end, align);
1578 
1579 		if (start == orig_start && end == orig_end)
1580 			continue;
1581 
1582 		if (start < end) {
1583 			r->base = start;
1584 			r->size = end - start;
1585 		} else {
1586 			memblock_remove_region(&memblock.memory,
1587 					       r - memblock.memory.regions);
1588 			r--;
1589 		}
1590 	}
1591 }
1592 
memblock_set_current_limit(phys_addr_t limit)1593 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1594 {
1595 	memblock.current_limit = limit;
1596 }
1597 
memblock_get_current_limit(void)1598 phys_addr_t __init_memblock memblock_get_current_limit(void)
1599 {
1600 	return memblock.current_limit;
1601 }
1602 
memblock_dump(struct memblock_type * type,char * name)1603 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1604 {
1605 	unsigned long long base, size;
1606 	unsigned long flags;
1607 	int i;
1608 
1609 	pr_info(" %s.cnt  = 0x%lx\n", name, type->cnt);
1610 
1611 	for (i = 0; i < type->cnt; i++) {
1612 		struct memblock_region *rgn = &type->regions[i];
1613 		char nid_buf[32] = "";
1614 
1615 		base = rgn->base;
1616 		size = rgn->size;
1617 		flags = rgn->flags;
1618 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1619 		if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1620 			snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1621 				 memblock_get_region_node(rgn));
1622 #endif
1623 		pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1624 			name, i, base, base + size - 1, size, nid_buf, flags);
1625 	}
1626 }
1627 
1628 extern unsigned long __init_memblock
memblock_reserved_memory_within(phys_addr_t start_addr,phys_addr_t end_addr)1629 memblock_reserved_memory_within(phys_addr_t start_addr, phys_addr_t end_addr)
1630 {
1631 	struct memblock_type *type = &memblock.reserved;
1632 	unsigned long size = 0;
1633 	int idx;
1634 
1635 	for (idx = 0; idx < type->cnt; idx++) {
1636 		struct memblock_region *rgn = &type->regions[idx];
1637 		phys_addr_t start, end;
1638 
1639 		if (rgn->base + rgn->size < start_addr)
1640 			continue;
1641 		if (rgn->base > end_addr)
1642 			continue;
1643 
1644 		start = rgn->base;
1645 		end = start + rgn->size;
1646 		size += end - start;
1647 	}
1648 
1649 	return size;
1650 }
1651 
__memblock_dump_all(void)1652 void __init_memblock __memblock_dump_all(void)
1653 {
1654 	pr_info("MEMBLOCK configuration:\n");
1655 	pr_info(" memory size = %#llx reserved size = %#llx\n",
1656 		(unsigned long long)memblock.memory.total_size,
1657 		(unsigned long long)memblock.reserved.total_size);
1658 
1659 	memblock_dump(&memblock.memory, "memory");
1660 	memblock_dump(&memblock.reserved, "reserved");
1661 }
1662 
memblock_allow_resize(void)1663 void __init memblock_allow_resize(void)
1664 {
1665 	memblock_can_resize = 1;
1666 }
1667 
early_memblock(char * p)1668 static int __init early_memblock(char *p)
1669 {
1670 	if (p && strstr(p, "debug"))
1671 		memblock_debug = 1;
1672 	return 0;
1673 }
1674 early_param("memblock", early_memblock);
1675 
1676 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1677 
memblock_debug_show(struct seq_file * m,void * private)1678 static int memblock_debug_show(struct seq_file *m, void *private)
1679 {
1680 	struct memblock_type *type = m->private;
1681 	struct memblock_region *reg;
1682 	int i;
1683 
1684 	for (i = 0; i < type->cnt; i++) {
1685 		reg = &type->regions[i];
1686 		seq_printf(m, "%4d: ", i);
1687 		if (sizeof(phys_addr_t) == 4)
1688 			seq_printf(m, "0x%08lx..0x%08lx\n",
1689 				   (unsigned long)reg->base,
1690 				   (unsigned long)(reg->base + reg->size - 1));
1691 		else
1692 			seq_printf(m, "0x%016llx..0x%016llx\n",
1693 				   (unsigned long long)reg->base,
1694 				   (unsigned long long)(reg->base + reg->size - 1));
1695 
1696 	}
1697 	return 0;
1698 }
1699 
memblock_debug_open(struct inode * inode,struct file * file)1700 static int memblock_debug_open(struct inode *inode, struct file *file)
1701 {
1702 	return single_open(file, memblock_debug_show, inode->i_private);
1703 }
1704 
1705 static const struct file_operations memblock_debug_fops = {
1706 	.open = memblock_debug_open,
1707 	.read = seq_read,
1708 	.llseek = seq_lseek,
1709 	.release = single_release,
1710 };
1711 
memblock_init_debugfs(void)1712 static int __init memblock_init_debugfs(void)
1713 {
1714 	struct dentry *root = debugfs_create_dir("memblock", NULL);
1715 	if (!root)
1716 		return -ENXIO;
1717 	debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1718 	debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1719 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1720 	debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1721 #endif
1722 
1723 	return 0;
1724 }
1725 __initcall(memblock_init_debugfs);
1726 
1727 #endif /* CONFIG_DEBUG_FS */
1728