xref: /openbmc/linux/mm/memblock.c (revision 644d87dc)
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/kmemleak.h>
21 #include <linux/seq_file.h>
22 #include <linux/memblock.h>
23 
24 #include <asm/sections.h>
25 #include <linux/io.h>
26 
27 #include "internal.h"
28 
29 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
30 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
31 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
32 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
33 #endif
34 
35 struct memblock memblock __initdata_memblock = {
36 	.memory.regions		= memblock_memory_init_regions,
37 	.memory.cnt		= 1,	/* empty dummy entry */
38 	.memory.max		= INIT_MEMBLOCK_REGIONS,
39 	.memory.name		= "memory",
40 
41 	.reserved.regions	= memblock_reserved_init_regions,
42 	.reserved.cnt		= 1,	/* empty dummy entry */
43 	.reserved.max		= INIT_MEMBLOCK_REGIONS,
44 	.reserved.name		= "reserved",
45 
46 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
47 	.physmem.regions	= memblock_physmem_init_regions,
48 	.physmem.cnt		= 1,	/* empty dummy entry */
49 	.physmem.max		= INIT_PHYSMEM_REGIONS,
50 	.physmem.name		= "physmem",
51 #endif
52 
53 	.bottom_up		= false,
54 	.current_limit		= MEMBLOCK_ALLOC_ANYWHERE,
55 };
56 
57 int memblock_debug __initdata_memblock;
58 static bool system_has_some_mirror __initdata_memblock = false;
59 static int memblock_can_resize __initdata_memblock;
60 static int memblock_memory_in_slab __initdata_memblock = 0;
61 static int memblock_reserved_in_slab __initdata_memblock = 0;
62 
63 ulong __init_memblock choose_memblock_flags(void)
64 {
65 	return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
66 }
67 
68 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
69 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
70 {
71 	return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
72 }
73 
74 /*
75  * Address comparison utilities
76  */
77 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
78 				       phys_addr_t base2, phys_addr_t size2)
79 {
80 	return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
81 }
82 
83 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
84 					phys_addr_t base, phys_addr_t size)
85 {
86 	unsigned long i;
87 
88 	for (i = 0; i < type->cnt; i++)
89 		if (memblock_addrs_overlap(base, size, type->regions[i].base,
90 					   type->regions[i].size))
91 			break;
92 	return i < type->cnt;
93 }
94 
95 /*
96  * __memblock_find_range_bottom_up - find free area utility in bottom-up
97  * @start: start of candidate range
98  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
99  * @size: size of free area to find
100  * @align: alignment of free area to find
101  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
102  * @flags: pick from blocks based on memory attributes
103  *
104  * Utility called from memblock_find_in_range_node(), find free area bottom-up.
105  *
106  * RETURNS:
107  * Found address on success, 0 on failure.
108  */
109 static phys_addr_t __init_memblock
110 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
111 				phys_addr_t size, phys_addr_t align, int nid,
112 				ulong flags)
113 {
114 	phys_addr_t this_start, this_end, cand;
115 	u64 i;
116 
117 	for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
118 		this_start = clamp(this_start, start, end);
119 		this_end = clamp(this_end, start, end);
120 
121 		cand = round_up(this_start, align);
122 		if (cand < this_end && this_end - cand >= size)
123 			return cand;
124 	}
125 
126 	return 0;
127 }
128 
129 /**
130  * __memblock_find_range_top_down - find free area utility, in top-down
131  * @start: start of candidate range
132  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
133  * @size: size of free area to find
134  * @align: alignment of free area to find
135  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
136  * @flags: pick from blocks based on memory attributes
137  *
138  * Utility called from memblock_find_in_range_node(), find free area top-down.
139  *
140  * RETURNS:
141  * Found address on success, 0 on failure.
142  */
143 static phys_addr_t __init_memblock
144 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
145 			       phys_addr_t size, phys_addr_t align, int nid,
146 			       ulong flags)
147 {
148 	phys_addr_t this_start, this_end, cand;
149 	u64 i;
150 
151 	for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
152 					NULL) {
153 		this_start = clamp(this_start, start, end);
154 		this_end = clamp(this_end, start, end);
155 
156 		if (this_end < size)
157 			continue;
158 
159 		cand = round_down(this_end - size, align);
160 		if (cand >= this_start)
161 			return cand;
162 	}
163 
164 	return 0;
165 }
166 
167 /**
168  * memblock_find_in_range_node - find free area in given range and node
169  * @size: size of free area to find
170  * @align: alignment of free area to find
171  * @start: start of candidate range
172  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
173  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
174  * @flags: pick from blocks based on memory attributes
175  *
176  * Find @size free area aligned to @align in the specified range and node.
177  *
178  * When allocation direction is bottom-up, the @start should be greater
179  * than the end of the kernel image. Otherwise, it will be trimmed. The
180  * reason is that we want the bottom-up allocation just near the kernel
181  * image so it is highly likely that the allocated memory and the kernel
182  * will reside in the same node.
183  *
184  * If bottom-up allocation failed, will try to allocate memory top-down.
185  *
186  * RETURNS:
187  * Found address on success, 0 on failure.
188  */
189 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
190 					phys_addr_t align, phys_addr_t start,
191 					phys_addr_t end, int nid, ulong flags)
192 {
193 	phys_addr_t kernel_end, ret;
194 
195 	/* pump up @end */
196 	if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
197 		end = memblock.current_limit;
198 
199 	/* avoid allocating the first page */
200 	start = max_t(phys_addr_t, start, PAGE_SIZE);
201 	end = max(start, end);
202 	kernel_end = __pa_symbol(_end);
203 
204 	/*
205 	 * try bottom-up allocation only when bottom-up mode
206 	 * is set and @end is above the kernel image.
207 	 */
208 	if (memblock_bottom_up() && end > kernel_end) {
209 		phys_addr_t bottom_up_start;
210 
211 		/* make sure we will allocate above the kernel */
212 		bottom_up_start = max(start, kernel_end);
213 
214 		/* ok, try bottom-up allocation first */
215 		ret = __memblock_find_range_bottom_up(bottom_up_start, end,
216 						      size, align, nid, flags);
217 		if (ret)
218 			return ret;
219 
220 		/*
221 		 * we always limit bottom-up allocation above the kernel,
222 		 * but top-down allocation doesn't have the limit, so
223 		 * retrying top-down allocation may succeed when bottom-up
224 		 * allocation failed.
225 		 *
226 		 * bottom-up allocation is expected to be fail very rarely,
227 		 * so we use WARN_ONCE() here to see the stack trace if
228 		 * fail happens.
229 		 */
230 		WARN_ONCE(1, "memblock: bottom-up allocation failed, memory hotunplug may be affected\n");
231 	}
232 
233 	return __memblock_find_range_top_down(start, end, size, align, nid,
234 					      flags);
235 }
236 
237 /**
238  * memblock_find_in_range - find free area in given range
239  * @start: start of candidate range
240  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
241  * @size: size of free area to find
242  * @align: alignment of free area to find
243  *
244  * Find @size free area aligned to @align in the specified range.
245  *
246  * RETURNS:
247  * Found address on success, 0 on failure.
248  */
249 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
250 					phys_addr_t end, phys_addr_t size,
251 					phys_addr_t align)
252 {
253 	phys_addr_t ret;
254 	ulong flags = choose_memblock_flags();
255 
256 again:
257 	ret = memblock_find_in_range_node(size, align, start, end,
258 					    NUMA_NO_NODE, flags);
259 
260 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
261 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
262 			&size);
263 		flags &= ~MEMBLOCK_MIRROR;
264 		goto again;
265 	}
266 
267 	return ret;
268 }
269 
270 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
271 {
272 	type->total_size -= type->regions[r].size;
273 	memmove(&type->regions[r], &type->regions[r + 1],
274 		(type->cnt - (r + 1)) * sizeof(type->regions[r]));
275 	type->cnt--;
276 
277 	/* Special case for empty arrays */
278 	if (type->cnt == 0) {
279 		WARN_ON(type->total_size != 0);
280 		type->cnt = 1;
281 		type->regions[0].base = 0;
282 		type->regions[0].size = 0;
283 		type->regions[0].flags = 0;
284 		memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
285 	}
286 }
287 
288 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
289 /**
290  * Discard memory and reserved arrays if they were allocated
291  */
292 void __init memblock_discard(void)
293 {
294 	phys_addr_t addr, size;
295 
296 	if (memblock.reserved.regions != memblock_reserved_init_regions) {
297 		addr = __pa(memblock.reserved.regions);
298 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
299 				  memblock.reserved.max);
300 		__memblock_free_late(addr, size);
301 	}
302 
303 	if (memblock.memory.regions != memblock_memory_init_regions) {
304 		addr = __pa(memblock.memory.regions);
305 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
306 				  memblock.memory.max);
307 		__memblock_free_late(addr, size);
308 	}
309 }
310 #endif
311 
312 /**
313  * memblock_double_array - double the size of the memblock regions array
314  * @type: memblock type of the regions array being doubled
315  * @new_area_start: starting address of memory range to avoid overlap with
316  * @new_area_size: size of memory range to avoid overlap with
317  *
318  * Double the size of the @type regions array. If memblock is being used to
319  * allocate memory for a new reserved regions array and there is a previously
320  * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
321  * waiting to be reserved, ensure the memory used by the new array does
322  * not overlap.
323  *
324  * RETURNS:
325  * 0 on success, -1 on failure.
326  */
327 static int __init_memblock memblock_double_array(struct memblock_type *type,
328 						phys_addr_t new_area_start,
329 						phys_addr_t new_area_size)
330 {
331 	struct memblock_region *new_array, *old_array;
332 	phys_addr_t old_alloc_size, new_alloc_size;
333 	phys_addr_t old_size, new_size, addr;
334 	int use_slab = slab_is_available();
335 	int *in_slab;
336 
337 	/* We don't allow resizing until we know about the reserved regions
338 	 * of memory that aren't suitable for allocation
339 	 */
340 	if (!memblock_can_resize)
341 		return -1;
342 
343 	/* Calculate new doubled size */
344 	old_size = type->max * sizeof(struct memblock_region);
345 	new_size = old_size << 1;
346 	/*
347 	 * We need to allocated new one align to PAGE_SIZE,
348 	 *   so we can free them completely later.
349 	 */
350 	old_alloc_size = PAGE_ALIGN(old_size);
351 	new_alloc_size = PAGE_ALIGN(new_size);
352 
353 	/* Retrieve the slab flag */
354 	if (type == &memblock.memory)
355 		in_slab = &memblock_memory_in_slab;
356 	else
357 		in_slab = &memblock_reserved_in_slab;
358 
359 	/* Try to find some space for it.
360 	 *
361 	 * WARNING: We assume that either slab_is_available() and we use it or
362 	 * we use MEMBLOCK for allocations. That means that this is unsafe to
363 	 * use when bootmem is currently active (unless bootmem itself is
364 	 * implemented on top of MEMBLOCK which isn't the case yet)
365 	 *
366 	 * This should however not be an issue for now, as we currently only
367 	 * call into MEMBLOCK while it's still active, or much later when slab
368 	 * is active for memory hotplug operations
369 	 */
370 	if (use_slab) {
371 		new_array = kmalloc(new_size, GFP_KERNEL);
372 		addr = new_array ? __pa(new_array) : 0;
373 	} else {
374 		/* only exclude range when trying to double reserved.regions */
375 		if (type != &memblock.reserved)
376 			new_area_start = new_area_size = 0;
377 
378 		addr = memblock_find_in_range(new_area_start + new_area_size,
379 						memblock.current_limit,
380 						new_alloc_size, PAGE_SIZE);
381 		if (!addr && new_area_size)
382 			addr = memblock_find_in_range(0,
383 				min(new_area_start, memblock.current_limit),
384 				new_alloc_size, PAGE_SIZE);
385 
386 		new_array = addr ? __va(addr) : NULL;
387 	}
388 	if (!addr) {
389 		pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
390 		       type->name, type->max, type->max * 2);
391 		return -1;
392 	}
393 
394 	memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
395 			type->name, type->max * 2, (u64)addr,
396 			(u64)addr + new_size - 1);
397 
398 	/*
399 	 * Found space, we now need to move the array over before we add the
400 	 * reserved region since it may be our reserved array itself that is
401 	 * full.
402 	 */
403 	memcpy(new_array, type->regions, old_size);
404 	memset(new_array + type->max, 0, old_size);
405 	old_array = type->regions;
406 	type->regions = new_array;
407 	type->max <<= 1;
408 
409 	/* Free old array. We needn't free it if the array is the static one */
410 	if (*in_slab)
411 		kfree(old_array);
412 	else if (old_array != memblock_memory_init_regions &&
413 		 old_array != memblock_reserved_init_regions)
414 		memblock_free(__pa(old_array), old_alloc_size);
415 
416 	/*
417 	 * Reserve the new array if that comes from the memblock.  Otherwise, we
418 	 * needn't do it
419 	 */
420 	if (!use_slab)
421 		BUG_ON(memblock_reserve(addr, new_alloc_size));
422 
423 	/* Update slab flag */
424 	*in_slab = use_slab;
425 
426 	return 0;
427 }
428 
429 /**
430  * memblock_merge_regions - merge neighboring compatible regions
431  * @type: memblock type to scan
432  *
433  * Scan @type and merge neighboring compatible regions.
434  */
435 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
436 {
437 	int i = 0;
438 
439 	/* cnt never goes below 1 */
440 	while (i < type->cnt - 1) {
441 		struct memblock_region *this = &type->regions[i];
442 		struct memblock_region *next = &type->regions[i + 1];
443 
444 		if (this->base + this->size != next->base ||
445 		    memblock_get_region_node(this) !=
446 		    memblock_get_region_node(next) ||
447 		    this->flags != next->flags) {
448 			BUG_ON(this->base + this->size > next->base);
449 			i++;
450 			continue;
451 		}
452 
453 		this->size += next->size;
454 		/* move forward from next + 1, index of which is i + 2 */
455 		memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
456 		type->cnt--;
457 	}
458 }
459 
460 /**
461  * memblock_insert_region - insert new memblock region
462  * @type:	memblock type to insert into
463  * @idx:	index for the insertion point
464  * @base:	base address of the new region
465  * @size:	size of the new region
466  * @nid:	node id of the new region
467  * @flags:	flags of the new region
468  *
469  * Insert new memblock region [@base,@base+@size) into @type at @idx.
470  * @type must already have extra room to accommodate the new region.
471  */
472 static void __init_memblock memblock_insert_region(struct memblock_type *type,
473 						   int idx, phys_addr_t base,
474 						   phys_addr_t size,
475 						   int nid, unsigned long flags)
476 {
477 	struct memblock_region *rgn = &type->regions[idx];
478 
479 	BUG_ON(type->cnt >= type->max);
480 	memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
481 	rgn->base = base;
482 	rgn->size = size;
483 	rgn->flags = flags;
484 	memblock_set_region_node(rgn, nid);
485 	type->cnt++;
486 	type->total_size += size;
487 }
488 
489 /**
490  * memblock_add_range - add new memblock region
491  * @type: memblock type to add new region into
492  * @base: base address of the new region
493  * @size: size of the new region
494  * @nid: nid of the new region
495  * @flags: flags of the new region
496  *
497  * Add new memblock region [@base,@base+@size) into @type.  The new region
498  * is allowed to overlap with existing ones - overlaps don't affect already
499  * existing regions.  @type is guaranteed to be minimal (all neighbouring
500  * compatible regions are merged) after the addition.
501  *
502  * RETURNS:
503  * 0 on success, -errno on failure.
504  */
505 int __init_memblock memblock_add_range(struct memblock_type *type,
506 				phys_addr_t base, phys_addr_t size,
507 				int nid, unsigned long flags)
508 {
509 	bool insert = false;
510 	phys_addr_t obase = base;
511 	phys_addr_t end = base + memblock_cap_size(base, &size);
512 	int idx, nr_new;
513 	struct memblock_region *rgn;
514 
515 	if (!size)
516 		return 0;
517 
518 	/* special case for empty array */
519 	if (type->regions[0].size == 0) {
520 		WARN_ON(type->cnt != 1 || type->total_size);
521 		type->regions[0].base = base;
522 		type->regions[0].size = size;
523 		type->regions[0].flags = flags;
524 		memblock_set_region_node(&type->regions[0], nid);
525 		type->total_size = size;
526 		return 0;
527 	}
528 repeat:
529 	/*
530 	 * The following is executed twice.  Once with %false @insert and
531 	 * then with %true.  The first counts the number of regions needed
532 	 * to accommodate the new area.  The second actually inserts them.
533 	 */
534 	base = obase;
535 	nr_new = 0;
536 
537 	for_each_memblock_type(idx, type, rgn) {
538 		phys_addr_t rbase = rgn->base;
539 		phys_addr_t rend = rbase + rgn->size;
540 
541 		if (rbase >= end)
542 			break;
543 		if (rend <= base)
544 			continue;
545 		/*
546 		 * @rgn overlaps.  If it separates the lower part of new
547 		 * area, insert that portion.
548 		 */
549 		if (rbase > base) {
550 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
551 			WARN_ON(nid != memblock_get_region_node(rgn));
552 #endif
553 			WARN_ON(flags != rgn->flags);
554 			nr_new++;
555 			if (insert)
556 				memblock_insert_region(type, idx++, base,
557 						       rbase - base, nid,
558 						       flags);
559 		}
560 		/* area below @rend is dealt with, forget about it */
561 		base = min(rend, end);
562 	}
563 
564 	/* insert the remaining portion */
565 	if (base < end) {
566 		nr_new++;
567 		if (insert)
568 			memblock_insert_region(type, idx, base, end - base,
569 					       nid, flags);
570 	}
571 
572 	if (!nr_new)
573 		return 0;
574 
575 	/*
576 	 * If this was the first round, resize array and repeat for actual
577 	 * insertions; otherwise, merge and return.
578 	 */
579 	if (!insert) {
580 		while (type->cnt + nr_new > type->max)
581 			if (memblock_double_array(type, obase, size) < 0)
582 				return -ENOMEM;
583 		insert = true;
584 		goto repeat;
585 	} else {
586 		memblock_merge_regions(type);
587 		return 0;
588 	}
589 }
590 
591 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
592 				       int nid)
593 {
594 	return memblock_add_range(&memblock.memory, base, size, nid, 0);
595 }
596 
597 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
598 {
599 	phys_addr_t end = base + size - 1;
600 
601 	memblock_dbg("memblock_add: [%pa-%pa] %pF\n",
602 		     &base, &end, (void *)_RET_IP_);
603 
604 	return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
605 }
606 
607 /**
608  * memblock_isolate_range - isolate given range into disjoint memblocks
609  * @type: memblock type to isolate range for
610  * @base: base of range to isolate
611  * @size: size of range to isolate
612  * @start_rgn: out parameter for the start of isolated region
613  * @end_rgn: out parameter for the end of isolated region
614  *
615  * Walk @type and ensure that regions don't cross the boundaries defined by
616  * [@base,@base+@size).  Crossing regions are split at the boundaries,
617  * which may create at most two more regions.  The index of the first
618  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
619  *
620  * RETURNS:
621  * 0 on success, -errno on failure.
622  */
623 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
624 					phys_addr_t base, phys_addr_t size,
625 					int *start_rgn, int *end_rgn)
626 {
627 	phys_addr_t end = base + memblock_cap_size(base, &size);
628 	int idx;
629 	struct memblock_region *rgn;
630 
631 	*start_rgn = *end_rgn = 0;
632 
633 	if (!size)
634 		return 0;
635 
636 	/* we'll create at most two more regions */
637 	while (type->cnt + 2 > type->max)
638 		if (memblock_double_array(type, base, size) < 0)
639 			return -ENOMEM;
640 
641 	for_each_memblock_type(idx, type, rgn) {
642 		phys_addr_t rbase = rgn->base;
643 		phys_addr_t rend = rbase + rgn->size;
644 
645 		if (rbase >= end)
646 			break;
647 		if (rend <= base)
648 			continue;
649 
650 		if (rbase < base) {
651 			/*
652 			 * @rgn intersects from below.  Split and continue
653 			 * to process the next region - the new top half.
654 			 */
655 			rgn->base = base;
656 			rgn->size -= base - rbase;
657 			type->total_size -= base - rbase;
658 			memblock_insert_region(type, idx, rbase, base - rbase,
659 					       memblock_get_region_node(rgn),
660 					       rgn->flags);
661 		} else if (rend > end) {
662 			/*
663 			 * @rgn intersects from above.  Split and redo the
664 			 * current region - the new bottom half.
665 			 */
666 			rgn->base = end;
667 			rgn->size -= end - rbase;
668 			type->total_size -= end - rbase;
669 			memblock_insert_region(type, idx--, rbase, end - rbase,
670 					       memblock_get_region_node(rgn),
671 					       rgn->flags);
672 		} else {
673 			/* @rgn is fully contained, record it */
674 			if (!*end_rgn)
675 				*start_rgn = idx;
676 			*end_rgn = idx + 1;
677 		}
678 	}
679 
680 	return 0;
681 }
682 
683 static int __init_memblock memblock_remove_range(struct memblock_type *type,
684 					  phys_addr_t base, phys_addr_t size)
685 {
686 	int start_rgn, end_rgn;
687 	int i, ret;
688 
689 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
690 	if (ret)
691 		return ret;
692 
693 	for (i = end_rgn - 1; i >= start_rgn; i--)
694 		memblock_remove_region(type, i);
695 	return 0;
696 }
697 
698 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
699 {
700 	return memblock_remove_range(&memblock.memory, base, size);
701 }
702 
703 
704 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
705 {
706 	phys_addr_t end = base + size - 1;
707 
708 	memblock_dbg("   memblock_free: [%pa-%pa] %pF\n",
709 		     &base, &end, (void *)_RET_IP_);
710 
711 	kmemleak_free_part_phys(base, size);
712 	return memblock_remove_range(&memblock.reserved, base, size);
713 }
714 
715 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
716 {
717 	phys_addr_t end = base + size - 1;
718 
719 	memblock_dbg("memblock_reserve: [%pa-%pa] %pF\n",
720 		     &base, &end, (void *)_RET_IP_);
721 
722 	return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
723 }
724 
725 /**
726  *
727  * This function isolates region [@base, @base + @size), and sets/clears flag
728  *
729  * Return 0 on success, -errno on failure.
730  */
731 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
732 				phys_addr_t size, int set, int flag)
733 {
734 	struct memblock_type *type = &memblock.memory;
735 	int i, ret, start_rgn, end_rgn;
736 
737 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
738 	if (ret)
739 		return ret;
740 
741 	for (i = start_rgn; i < end_rgn; i++)
742 		if (set)
743 			memblock_set_region_flags(&type->regions[i], flag);
744 		else
745 			memblock_clear_region_flags(&type->regions[i], flag);
746 
747 	memblock_merge_regions(type);
748 	return 0;
749 }
750 
751 /**
752  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
753  * @base: the base phys addr of the region
754  * @size: the size of the region
755  *
756  * Return 0 on success, -errno on failure.
757  */
758 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
759 {
760 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
761 }
762 
763 /**
764  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
765  * @base: the base phys addr of the region
766  * @size: the size of the region
767  *
768  * Return 0 on success, -errno on failure.
769  */
770 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
771 {
772 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
773 }
774 
775 /**
776  * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
777  * @base: the base phys addr of the region
778  * @size: the size of the region
779  *
780  * Return 0 on success, -errno on failure.
781  */
782 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
783 {
784 	system_has_some_mirror = true;
785 
786 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
787 }
788 
789 /**
790  * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
791  * @base: the base phys addr of the region
792  * @size: the size of the region
793  *
794  * Return 0 on success, -errno on failure.
795  */
796 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
797 {
798 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
799 }
800 
801 /**
802  * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
803  * @base: the base phys addr of the region
804  * @size: the size of the region
805  *
806  * Return 0 on success, -errno on failure.
807  */
808 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
809 {
810 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
811 }
812 
813 /**
814  * __next_reserved_mem_region - next function for for_each_reserved_region()
815  * @idx: pointer to u64 loop variable
816  * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
817  * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
818  *
819  * Iterate over all reserved memory regions.
820  */
821 void __init_memblock __next_reserved_mem_region(u64 *idx,
822 					   phys_addr_t *out_start,
823 					   phys_addr_t *out_end)
824 {
825 	struct memblock_type *type = &memblock.reserved;
826 
827 	if (*idx < type->cnt) {
828 		struct memblock_region *r = &type->regions[*idx];
829 		phys_addr_t base = r->base;
830 		phys_addr_t size = r->size;
831 
832 		if (out_start)
833 			*out_start = base;
834 		if (out_end)
835 			*out_end = base + size - 1;
836 
837 		*idx += 1;
838 		return;
839 	}
840 
841 	/* signal end of iteration */
842 	*idx = ULLONG_MAX;
843 }
844 
845 /**
846  * __next__mem_range - next function for for_each_free_mem_range() etc.
847  * @idx: pointer to u64 loop variable
848  * @nid: node selector, %NUMA_NO_NODE for all nodes
849  * @flags: pick from blocks based on memory attributes
850  * @type_a: pointer to memblock_type from where the range is taken
851  * @type_b: pointer to memblock_type which excludes memory from being taken
852  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
853  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
854  * @out_nid: ptr to int for nid of the range, can be %NULL
855  *
856  * Find the first area from *@idx which matches @nid, fill the out
857  * parameters, and update *@idx for the next iteration.  The lower 32bit of
858  * *@idx contains index into type_a and the upper 32bit indexes the
859  * areas before each region in type_b.	For example, if type_b regions
860  * look like the following,
861  *
862  *	0:[0-16), 1:[32-48), 2:[128-130)
863  *
864  * The upper 32bit indexes the following regions.
865  *
866  *	0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
867  *
868  * As both region arrays are sorted, the function advances the two indices
869  * in lockstep and returns each intersection.
870  */
871 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
872 				      struct memblock_type *type_a,
873 				      struct memblock_type *type_b,
874 				      phys_addr_t *out_start,
875 				      phys_addr_t *out_end, int *out_nid)
876 {
877 	int idx_a = *idx & 0xffffffff;
878 	int idx_b = *idx >> 32;
879 
880 	if (WARN_ONCE(nid == MAX_NUMNODES,
881 	"Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
882 		nid = NUMA_NO_NODE;
883 
884 	for (; idx_a < type_a->cnt; idx_a++) {
885 		struct memblock_region *m = &type_a->regions[idx_a];
886 
887 		phys_addr_t m_start = m->base;
888 		phys_addr_t m_end = m->base + m->size;
889 		int	    m_nid = memblock_get_region_node(m);
890 
891 		/* only memory regions are associated with nodes, check it */
892 		if (nid != NUMA_NO_NODE && nid != m_nid)
893 			continue;
894 
895 		/* skip hotpluggable memory regions if needed */
896 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
897 			continue;
898 
899 		/* if we want mirror memory skip non-mirror memory regions */
900 		if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
901 			continue;
902 
903 		/* skip nomap memory unless we were asked for it explicitly */
904 		if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
905 			continue;
906 
907 		if (!type_b) {
908 			if (out_start)
909 				*out_start = m_start;
910 			if (out_end)
911 				*out_end = m_end;
912 			if (out_nid)
913 				*out_nid = m_nid;
914 			idx_a++;
915 			*idx = (u32)idx_a | (u64)idx_b << 32;
916 			return;
917 		}
918 
919 		/* scan areas before each reservation */
920 		for (; idx_b < type_b->cnt + 1; idx_b++) {
921 			struct memblock_region *r;
922 			phys_addr_t r_start;
923 			phys_addr_t r_end;
924 
925 			r = &type_b->regions[idx_b];
926 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
927 			r_end = idx_b < type_b->cnt ?
928 				r->base : (phys_addr_t)ULLONG_MAX;
929 
930 			/*
931 			 * if idx_b advanced past idx_a,
932 			 * break out to advance idx_a
933 			 */
934 			if (r_start >= m_end)
935 				break;
936 			/* if the two regions intersect, we're done */
937 			if (m_start < r_end) {
938 				if (out_start)
939 					*out_start =
940 						max(m_start, r_start);
941 				if (out_end)
942 					*out_end = min(m_end, r_end);
943 				if (out_nid)
944 					*out_nid = m_nid;
945 				/*
946 				 * The region which ends first is
947 				 * advanced for the next iteration.
948 				 */
949 				if (m_end <= r_end)
950 					idx_a++;
951 				else
952 					idx_b++;
953 				*idx = (u32)idx_a | (u64)idx_b << 32;
954 				return;
955 			}
956 		}
957 	}
958 
959 	/* signal end of iteration */
960 	*idx = ULLONG_MAX;
961 }
962 
963 /**
964  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
965  *
966  * Finds the next range from type_a which is not marked as unsuitable
967  * in type_b.
968  *
969  * @idx: pointer to u64 loop variable
970  * @nid: node selector, %NUMA_NO_NODE for all nodes
971  * @flags: pick from blocks based on memory attributes
972  * @type_a: pointer to memblock_type from where the range is taken
973  * @type_b: pointer to memblock_type which excludes memory from being taken
974  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
975  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
976  * @out_nid: ptr to int for nid of the range, can be %NULL
977  *
978  * Reverse of __next_mem_range().
979  */
980 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
981 					  struct memblock_type *type_a,
982 					  struct memblock_type *type_b,
983 					  phys_addr_t *out_start,
984 					  phys_addr_t *out_end, int *out_nid)
985 {
986 	int idx_a = *idx & 0xffffffff;
987 	int idx_b = *idx >> 32;
988 
989 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
990 		nid = NUMA_NO_NODE;
991 
992 	if (*idx == (u64)ULLONG_MAX) {
993 		idx_a = type_a->cnt - 1;
994 		if (type_b != NULL)
995 			idx_b = type_b->cnt;
996 		else
997 			idx_b = 0;
998 	}
999 
1000 	for (; idx_a >= 0; idx_a--) {
1001 		struct memblock_region *m = &type_a->regions[idx_a];
1002 
1003 		phys_addr_t m_start = m->base;
1004 		phys_addr_t m_end = m->base + m->size;
1005 		int m_nid = memblock_get_region_node(m);
1006 
1007 		/* only memory regions are associated with nodes, check it */
1008 		if (nid != NUMA_NO_NODE && nid != m_nid)
1009 			continue;
1010 
1011 		/* skip hotpluggable memory regions if needed */
1012 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1013 			continue;
1014 
1015 		/* if we want mirror memory skip non-mirror memory regions */
1016 		if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1017 			continue;
1018 
1019 		/* skip nomap memory unless we were asked for it explicitly */
1020 		if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1021 			continue;
1022 
1023 		if (!type_b) {
1024 			if (out_start)
1025 				*out_start = m_start;
1026 			if (out_end)
1027 				*out_end = m_end;
1028 			if (out_nid)
1029 				*out_nid = m_nid;
1030 			idx_a--;
1031 			*idx = (u32)idx_a | (u64)idx_b << 32;
1032 			return;
1033 		}
1034 
1035 		/* scan areas before each reservation */
1036 		for (; idx_b >= 0; idx_b--) {
1037 			struct memblock_region *r;
1038 			phys_addr_t r_start;
1039 			phys_addr_t r_end;
1040 
1041 			r = &type_b->regions[idx_b];
1042 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1043 			r_end = idx_b < type_b->cnt ?
1044 				r->base : (phys_addr_t)ULLONG_MAX;
1045 			/*
1046 			 * if idx_b advanced past idx_a,
1047 			 * break out to advance idx_a
1048 			 */
1049 
1050 			if (r_end <= m_start)
1051 				break;
1052 			/* if the two regions intersect, we're done */
1053 			if (m_end > r_start) {
1054 				if (out_start)
1055 					*out_start = max(m_start, r_start);
1056 				if (out_end)
1057 					*out_end = min(m_end, r_end);
1058 				if (out_nid)
1059 					*out_nid = m_nid;
1060 				if (m_start >= r_start)
1061 					idx_a--;
1062 				else
1063 					idx_b--;
1064 				*idx = (u32)idx_a | (u64)idx_b << 32;
1065 				return;
1066 			}
1067 		}
1068 	}
1069 	/* signal end of iteration */
1070 	*idx = ULLONG_MAX;
1071 }
1072 
1073 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1074 /*
1075  * Common iterator interface used to define for_each_mem_range().
1076  */
1077 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1078 				unsigned long *out_start_pfn,
1079 				unsigned long *out_end_pfn, int *out_nid)
1080 {
1081 	struct memblock_type *type = &memblock.memory;
1082 	struct memblock_region *r;
1083 
1084 	while (++*idx < type->cnt) {
1085 		r = &type->regions[*idx];
1086 
1087 		if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1088 			continue;
1089 		if (nid == MAX_NUMNODES || nid == r->nid)
1090 			break;
1091 	}
1092 	if (*idx >= type->cnt) {
1093 		*idx = -1;
1094 		return;
1095 	}
1096 
1097 	if (out_start_pfn)
1098 		*out_start_pfn = PFN_UP(r->base);
1099 	if (out_end_pfn)
1100 		*out_end_pfn = PFN_DOWN(r->base + r->size);
1101 	if (out_nid)
1102 		*out_nid = r->nid;
1103 }
1104 
1105 /**
1106  * memblock_set_node - set node ID on memblock regions
1107  * @base: base of area to set node ID for
1108  * @size: size of area to set node ID for
1109  * @type: memblock type to set node ID for
1110  * @nid: node ID to set
1111  *
1112  * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1113  * Regions which cross the area boundaries are split as necessary.
1114  *
1115  * RETURNS:
1116  * 0 on success, -errno on failure.
1117  */
1118 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1119 				      struct memblock_type *type, int nid)
1120 {
1121 	int start_rgn, end_rgn;
1122 	int i, ret;
1123 
1124 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1125 	if (ret)
1126 		return ret;
1127 
1128 	for (i = start_rgn; i < end_rgn; i++)
1129 		memblock_set_region_node(&type->regions[i], nid);
1130 
1131 	memblock_merge_regions(type);
1132 	return 0;
1133 }
1134 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1135 
1136 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1137 					phys_addr_t align, phys_addr_t start,
1138 					phys_addr_t end, int nid, ulong flags)
1139 {
1140 	phys_addr_t found;
1141 
1142 	if (!align)
1143 		align = SMP_CACHE_BYTES;
1144 
1145 	found = memblock_find_in_range_node(size, align, start, end, nid,
1146 					    flags);
1147 	if (found && !memblock_reserve(found, size)) {
1148 		/*
1149 		 * The min_count is set to 0 so that memblock allocations are
1150 		 * never reported as leaks.
1151 		 */
1152 		kmemleak_alloc_phys(found, size, 0, 0);
1153 		return found;
1154 	}
1155 	return 0;
1156 }
1157 
1158 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1159 					phys_addr_t start, phys_addr_t end,
1160 					ulong flags)
1161 {
1162 	return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1163 					flags);
1164 }
1165 
1166 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1167 					phys_addr_t align, phys_addr_t max_addr,
1168 					int nid, ulong flags)
1169 {
1170 	return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1171 }
1172 
1173 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1174 {
1175 	ulong flags = choose_memblock_flags();
1176 	phys_addr_t ret;
1177 
1178 again:
1179 	ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1180 				      nid, flags);
1181 
1182 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
1183 		flags &= ~MEMBLOCK_MIRROR;
1184 		goto again;
1185 	}
1186 	return ret;
1187 }
1188 
1189 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1190 {
1191 	return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1192 				       MEMBLOCK_NONE);
1193 }
1194 
1195 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1196 {
1197 	phys_addr_t alloc;
1198 
1199 	alloc = __memblock_alloc_base(size, align, max_addr);
1200 
1201 	if (alloc == 0)
1202 		panic("ERROR: Failed to allocate %pa bytes below %pa.\n",
1203 		      &size, &max_addr);
1204 
1205 	return alloc;
1206 }
1207 
1208 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1209 {
1210 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1211 }
1212 
1213 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1214 {
1215 	phys_addr_t res = memblock_alloc_nid(size, align, nid);
1216 
1217 	if (res)
1218 		return res;
1219 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1220 }
1221 
1222 /**
1223  * memblock_virt_alloc_internal - allocate boot memory block
1224  * @size: size of memory block to be allocated in bytes
1225  * @align: alignment of the region and block's size
1226  * @min_addr: the lower bound of the memory region to allocate (phys address)
1227  * @max_addr: the upper bound of the memory region to allocate (phys address)
1228  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1229  *
1230  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1231  * will fall back to memory below @min_addr. Also, allocation may fall back
1232  * to any node in the system if the specified node can not
1233  * hold the requested memory.
1234  *
1235  * The allocation is performed from memory region limited by
1236  * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1237  *
1238  * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1239  *
1240  * The phys address of allocated boot memory block is converted to virtual and
1241  * allocated memory is reset to 0.
1242  *
1243  * In addition, function sets the min_count to 0 using kmemleak_alloc for
1244  * allocated boot memory block, so that it is never reported as leaks.
1245  *
1246  * RETURNS:
1247  * Virtual address of allocated memory block on success, NULL on failure.
1248  */
1249 static void * __init memblock_virt_alloc_internal(
1250 				phys_addr_t size, phys_addr_t align,
1251 				phys_addr_t min_addr, phys_addr_t max_addr,
1252 				int nid)
1253 {
1254 	phys_addr_t alloc;
1255 	void *ptr;
1256 	ulong flags = choose_memblock_flags();
1257 
1258 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1259 		nid = NUMA_NO_NODE;
1260 
1261 	/*
1262 	 * Detect any accidental use of these APIs after slab is ready, as at
1263 	 * this moment memblock may be deinitialized already and its
1264 	 * internal data may be destroyed (after execution of free_all_bootmem)
1265 	 */
1266 	if (WARN_ON_ONCE(slab_is_available()))
1267 		return kzalloc_node(size, GFP_NOWAIT, nid);
1268 
1269 	if (!align)
1270 		align = SMP_CACHE_BYTES;
1271 
1272 	if (max_addr > memblock.current_limit)
1273 		max_addr = memblock.current_limit;
1274 again:
1275 	alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1276 					    nid, flags);
1277 	if (alloc && !memblock_reserve(alloc, size))
1278 		goto done;
1279 
1280 	if (nid != NUMA_NO_NODE) {
1281 		alloc = memblock_find_in_range_node(size, align, min_addr,
1282 						    max_addr, NUMA_NO_NODE,
1283 						    flags);
1284 		if (alloc && !memblock_reserve(alloc, size))
1285 			goto done;
1286 	}
1287 
1288 	if (min_addr) {
1289 		min_addr = 0;
1290 		goto again;
1291 	}
1292 
1293 	if (flags & MEMBLOCK_MIRROR) {
1294 		flags &= ~MEMBLOCK_MIRROR;
1295 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1296 			&size);
1297 		goto again;
1298 	}
1299 
1300 	return NULL;
1301 done:
1302 	ptr = phys_to_virt(alloc);
1303 
1304 	/*
1305 	 * The min_count is set to 0 so that bootmem allocated blocks
1306 	 * are never reported as leaks. This is because many of these blocks
1307 	 * are only referred via the physical address which is not
1308 	 * looked up by kmemleak.
1309 	 */
1310 	kmemleak_alloc(ptr, size, 0, 0);
1311 
1312 	return ptr;
1313 }
1314 
1315 /**
1316  * memblock_virt_alloc_try_nid_raw - allocate boot memory block without zeroing
1317  * memory and without panicking
1318  * @size: size of memory block to be allocated in bytes
1319  * @align: alignment of the region and block's size
1320  * @min_addr: the lower bound of the memory region from where the allocation
1321  *	  is preferred (phys address)
1322  * @max_addr: the upper bound of the memory region from where the allocation
1323  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1324  *	      allocate only from memory limited by memblock.current_limit value
1325  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1326  *
1327  * Public function, provides additional debug information (including caller
1328  * info), if enabled. Does not zero allocated memory, does not panic if request
1329  * cannot be satisfied.
1330  *
1331  * RETURNS:
1332  * Virtual address of allocated memory block on success, NULL on failure.
1333  */
1334 void * __init memblock_virt_alloc_try_nid_raw(
1335 			phys_addr_t size, phys_addr_t align,
1336 			phys_addr_t min_addr, phys_addr_t max_addr,
1337 			int nid)
1338 {
1339 	void *ptr;
1340 
1341 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1342 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1343 		     (u64)max_addr, (void *)_RET_IP_);
1344 
1345 	ptr = memblock_virt_alloc_internal(size, align,
1346 					   min_addr, max_addr, nid);
1347 #ifdef CONFIG_DEBUG_VM
1348 	if (ptr && size > 0)
1349 		memset(ptr, PAGE_POISON_PATTERN, size);
1350 #endif
1351 	return ptr;
1352 }
1353 
1354 /**
1355  * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1356  * @size: size of memory block to be allocated in bytes
1357  * @align: alignment of the region and block's size
1358  * @min_addr: the lower bound of the memory region from where the allocation
1359  *	  is preferred (phys address)
1360  * @max_addr: the upper bound of the memory region from where the allocation
1361  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1362  *	      allocate only from memory limited by memblock.current_limit value
1363  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1364  *
1365  * Public function, provides additional debug information (including caller
1366  * info), if enabled. This function zeroes the allocated memory.
1367  *
1368  * RETURNS:
1369  * Virtual address of allocated memory block on success, NULL on failure.
1370  */
1371 void * __init memblock_virt_alloc_try_nid_nopanic(
1372 				phys_addr_t size, phys_addr_t align,
1373 				phys_addr_t min_addr, phys_addr_t max_addr,
1374 				int nid)
1375 {
1376 	void *ptr;
1377 
1378 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1379 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1380 		     (u64)max_addr, (void *)_RET_IP_);
1381 
1382 	ptr = memblock_virt_alloc_internal(size, align,
1383 					   min_addr, max_addr, nid);
1384 	if (ptr)
1385 		memset(ptr, 0, size);
1386 	return ptr;
1387 }
1388 
1389 /**
1390  * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1391  * @size: size of memory block to be allocated in bytes
1392  * @align: alignment of the region and block's size
1393  * @min_addr: the lower bound of the memory region from where the allocation
1394  *	  is preferred (phys address)
1395  * @max_addr: the upper bound of the memory region from where the allocation
1396  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1397  *	      allocate only from memory limited by memblock.current_limit value
1398  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1399  *
1400  * Public panicking version of memblock_virt_alloc_try_nid_nopanic()
1401  * which provides debug information (including caller info), if enabled,
1402  * and panics if the request can not be satisfied.
1403  *
1404  * RETURNS:
1405  * Virtual address of allocated memory block on success, NULL on failure.
1406  */
1407 void * __init memblock_virt_alloc_try_nid(
1408 			phys_addr_t size, phys_addr_t align,
1409 			phys_addr_t min_addr, phys_addr_t max_addr,
1410 			int nid)
1411 {
1412 	void *ptr;
1413 
1414 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1415 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1416 		     (u64)max_addr, (void *)_RET_IP_);
1417 	ptr = memblock_virt_alloc_internal(size, align,
1418 					   min_addr, max_addr, nid);
1419 	if (ptr) {
1420 		memset(ptr, 0, size);
1421 		return ptr;
1422 	}
1423 
1424 	panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1425 	      __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1426 	      (u64)max_addr);
1427 	return NULL;
1428 }
1429 
1430 /**
1431  * __memblock_free_early - free boot memory block
1432  * @base: phys starting address of the  boot memory block
1433  * @size: size of the boot memory block in bytes
1434  *
1435  * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1436  * The freeing memory will not be released to the buddy allocator.
1437  */
1438 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1439 {
1440 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1441 		     __func__, (u64)base, (u64)base + size - 1,
1442 		     (void *)_RET_IP_);
1443 	kmemleak_free_part_phys(base, size);
1444 	memblock_remove_range(&memblock.reserved, base, size);
1445 }
1446 
1447 /*
1448  * __memblock_free_late - free bootmem block pages directly to buddy allocator
1449  * @addr: phys starting address of the  boot memory block
1450  * @size: size of the boot memory block in bytes
1451  *
1452  * This is only useful when the bootmem allocator has already been torn
1453  * down, but we are still initializing the system.  Pages are released directly
1454  * to the buddy allocator, no bootmem metadata is updated because it is gone.
1455  */
1456 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1457 {
1458 	u64 cursor, end;
1459 
1460 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1461 		     __func__, (u64)base, (u64)base + size - 1,
1462 		     (void *)_RET_IP_);
1463 	kmemleak_free_part_phys(base, size);
1464 	cursor = PFN_UP(base);
1465 	end = PFN_DOWN(base + size);
1466 
1467 	for (; cursor < end; cursor++) {
1468 		__free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1469 		totalram_pages++;
1470 	}
1471 }
1472 
1473 /*
1474  * Remaining API functions
1475  */
1476 
1477 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1478 {
1479 	return memblock.memory.total_size;
1480 }
1481 
1482 phys_addr_t __init_memblock memblock_reserved_size(void)
1483 {
1484 	return memblock.reserved.total_size;
1485 }
1486 
1487 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1488 {
1489 	unsigned long pages = 0;
1490 	struct memblock_region *r;
1491 	unsigned long start_pfn, end_pfn;
1492 
1493 	for_each_memblock(memory, r) {
1494 		start_pfn = memblock_region_memory_base_pfn(r);
1495 		end_pfn = memblock_region_memory_end_pfn(r);
1496 		start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1497 		end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1498 		pages += end_pfn - start_pfn;
1499 	}
1500 
1501 	return PFN_PHYS(pages);
1502 }
1503 
1504 /* lowest address */
1505 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1506 {
1507 	return memblock.memory.regions[0].base;
1508 }
1509 
1510 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1511 {
1512 	int idx = memblock.memory.cnt - 1;
1513 
1514 	return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1515 }
1516 
1517 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1518 {
1519 	phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1520 	struct memblock_region *r;
1521 
1522 	/*
1523 	 * translate the memory @limit size into the max address within one of
1524 	 * the memory memblock regions, if the @limit exceeds the total size
1525 	 * of those regions, max_addr will keep original value ULLONG_MAX
1526 	 */
1527 	for_each_memblock(memory, r) {
1528 		if (limit <= r->size) {
1529 			max_addr = r->base + limit;
1530 			break;
1531 		}
1532 		limit -= r->size;
1533 	}
1534 
1535 	return max_addr;
1536 }
1537 
1538 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1539 {
1540 	phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1541 
1542 	if (!limit)
1543 		return;
1544 
1545 	max_addr = __find_max_addr(limit);
1546 
1547 	/* @limit exceeds the total size of the memory, do nothing */
1548 	if (max_addr == (phys_addr_t)ULLONG_MAX)
1549 		return;
1550 
1551 	/* truncate both memory and reserved regions */
1552 	memblock_remove_range(&memblock.memory, max_addr,
1553 			      (phys_addr_t)ULLONG_MAX);
1554 	memblock_remove_range(&memblock.reserved, max_addr,
1555 			      (phys_addr_t)ULLONG_MAX);
1556 }
1557 
1558 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1559 {
1560 	int start_rgn, end_rgn;
1561 	int i, ret;
1562 
1563 	if (!size)
1564 		return;
1565 
1566 	ret = memblock_isolate_range(&memblock.memory, base, size,
1567 						&start_rgn, &end_rgn);
1568 	if (ret)
1569 		return;
1570 
1571 	/* remove all the MAP regions */
1572 	for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1573 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1574 			memblock_remove_region(&memblock.memory, i);
1575 
1576 	for (i = start_rgn - 1; i >= 0; i--)
1577 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1578 			memblock_remove_region(&memblock.memory, i);
1579 
1580 	/* truncate the reserved regions */
1581 	memblock_remove_range(&memblock.reserved, 0, base);
1582 	memblock_remove_range(&memblock.reserved,
1583 			base + size, (phys_addr_t)ULLONG_MAX);
1584 }
1585 
1586 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1587 {
1588 	phys_addr_t max_addr;
1589 
1590 	if (!limit)
1591 		return;
1592 
1593 	max_addr = __find_max_addr(limit);
1594 
1595 	/* @limit exceeds the total size of the memory, do nothing */
1596 	if (max_addr == (phys_addr_t)ULLONG_MAX)
1597 		return;
1598 
1599 	memblock_cap_memory_range(0, max_addr);
1600 }
1601 
1602 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1603 {
1604 	unsigned int left = 0, right = type->cnt;
1605 
1606 	do {
1607 		unsigned int mid = (right + left) / 2;
1608 
1609 		if (addr < type->regions[mid].base)
1610 			right = mid;
1611 		else if (addr >= (type->regions[mid].base +
1612 				  type->regions[mid].size))
1613 			left = mid + 1;
1614 		else
1615 			return mid;
1616 	} while (left < right);
1617 	return -1;
1618 }
1619 
1620 bool __init memblock_is_reserved(phys_addr_t addr)
1621 {
1622 	return memblock_search(&memblock.reserved, addr) != -1;
1623 }
1624 
1625 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1626 {
1627 	return memblock_search(&memblock.memory, addr) != -1;
1628 }
1629 
1630 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1631 {
1632 	int i = memblock_search(&memblock.memory, addr);
1633 
1634 	if (i == -1)
1635 		return false;
1636 	return !memblock_is_nomap(&memblock.memory.regions[i]);
1637 }
1638 
1639 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1640 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1641 			 unsigned long *start_pfn, unsigned long *end_pfn)
1642 {
1643 	struct memblock_type *type = &memblock.memory;
1644 	int mid = memblock_search(type, PFN_PHYS(pfn));
1645 
1646 	if (mid == -1)
1647 		return -1;
1648 
1649 	*start_pfn = PFN_DOWN(type->regions[mid].base);
1650 	*end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1651 
1652 	return type->regions[mid].nid;
1653 }
1654 #endif
1655 
1656 /**
1657  * memblock_is_region_memory - check if a region is a subset of memory
1658  * @base: base of region to check
1659  * @size: size of region to check
1660  *
1661  * Check if the region [@base, @base+@size) is a subset of a memory block.
1662  *
1663  * RETURNS:
1664  * 0 if false, non-zero if true
1665  */
1666 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1667 {
1668 	int idx = memblock_search(&memblock.memory, base);
1669 	phys_addr_t end = base + memblock_cap_size(base, &size);
1670 
1671 	if (idx == -1)
1672 		return false;
1673 	return (memblock.memory.regions[idx].base +
1674 		 memblock.memory.regions[idx].size) >= end;
1675 }
1676 
1677 /**
1678  * memblock_is_region_reserved - check if a region intersects reserved memory
1679  * @base: base of region to check
1680  * @size: size of region to check
1681  *
1682  * Check if the region [@base, @base+@size) intersects a reserved memory block.
1683  *
1684  * RETURNS:
1685  * True if they intersect, false if not.
1686  */
1687 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1688 {
1689 	memblock_cap_size(base, &size);
1690 	return memblock_overlaps_region(&memblock.reserved, base, size);
1691 }
1692 
1693 void __init_memblock memblock_trim_memory(phys_addr_t align)
1694 {
1695 	phys_addr_t start, end, orig_start, orig_end;
1696 	struct memblock_region *r;
1697 
1698 	for_each_memblock(memory, r) {
1699 		orig_start = r->base;
1700 		orig_end = r->base + r->size;
1701 		start = round_up(orig_start, align);
1702 		end = round_down(orig_end, align);
1703 
1704 		if (start == orig_start && end == orig_end)
1705 			continue;
1706 
1707 		if (start < end) {
1708 			r->base = start;
1709 			r->size = end - start;
1710 		} else {
1711 			memblock_remove_region(&memblock.memory,
1712 					       r - memblock.memory.regions);
1713 			r--;
1714 		}
1715 	}
1716 }
1717 
1718 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1719 {
1720 	memblock.current_limit = limit;
1721 }
1722 
1723 phys_addr_t __init_memblock memblock_get_current_limit(void)
1724 {
1725 	return memblock.current_limit;
1726 }
1727 
1728 static void __init_memblock memblock_dump(struct memblock_type *type)
1729 {
1730 	phys_addr_t base, end, size;
1731 	unsigned long flags;
1732 	int idx;
1733 	struct memblock_region *rgn;
1734 
1735 	pr_info(" %s.cnt  = 0x%lx\n", type->name, type->cnt);
1736 
1737 	for_each_memblock_type(idx, type, rgn) {
1738 		char nid_buf[32] = "";
1739 
1740 		base = rgn->base;
1741 		size = rgn->size;
1742 		end = base + size - 1;
1743 		flags = rgn->flags;
1744 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1745 		if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1746 			snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1747 				 memblock_get_region_node(rgn));
1748 #endif
1749 		pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#lx\n",
1750 			type->name, idx, &base, &end, &size, nid_buf, flags);
1751 	}
1752 }
1753 
1754 void __init_memblock __memblock_dump_all(void)
1755 {
1756 	pr_info("MEMBLOCK configuration:\n");
1757 	pr_info(" memory size = %pa reserved size = %pa\n",
1758 		&memblock.memory.total_size,
1759 		&memblock.reserved.total_size);
1760 
1761 	memblock_dump(&memblock.memory);
1762 	memblock_dump(&memblock.reserved);
1763 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1764 	memblock_dump(&memblock.physmem);
1765 #endif
1766 }
1767 
1768 void __init memblock_allow_resize(void)
1769 {
1770 	memblock_can_resize = 1;
1771 }
1772 
1773 static int __init early_memblock(char *p)
1774 {
1775 	if (p && strstr(p, "debug"))
1776 		memblock_debug = 1;
1777 	return 0;
1778 }
1779 early_param("memblock", early_memblock);
1780 
1781 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1782 
1783 static int memblock_debug_show(struct seq_file *m, void *private)
1784 {
1785 	struct memblock_type *type = m->private;
1786 	struct memblock_region *reg;
1787 	int i;
1788 	phys_addr_t end;
1789 
1790 	for (i = 0; i < type->cnt; i++) {
1791 		reg = &type->regions[i];
1792 		end = reg->base + reg->size - 1;
1793 
1794 		seq_printf(m, "%4d: ", i);
1795 		seq_printf(m, "%pa..%pa\n", &reg->base, &end);
1796 	}
1797 	return 0;
1798 }
1799 DEFINE_SHOW_ATTRIBUTE(memblock_debug);
1800 
1801 static int __init memblock_init_debugfs(void)
1802 {
1803 	struct dentry *root = debugfs_create_dir("memblock", NULL);
1804 	if (!root)
1805 		return -ENXIO;
1806 	debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1807 	debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1808 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1809 	debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1810 #endif
1811 
1812 	return 0;
1813 }
1814 __initcall(memblock_init_debugfs);
1815 
1816 #endif /* CONFIG_DEBUG_FS */
1817