xref: /openbmc/linux/mm/memblock.c (revision 110e6f26)
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 
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 *
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 */
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  */
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 
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 		if (memblock_addrs_overlap(base, size, type->regions[i].base,
101 					   type->regions[i].size))
102 			break;
103 	return i < type->cnt;
104 }
105 
106 /*
107  * __memblock_find_range_bottom_up - find free area utility in bottom-up
108  * @start: start of candidate range
109  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
110  * @size: size of free area to find
111  * @align: alignment of free area to find
112  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
113  * @flags: pick from blocks based on memory attributes
114  *
115  * Utility called from memblock_find_in_range_node(), find free area bottom-up.
116  *
117  * RETURNS:
118  * Found address on success, 0 on failure.
119  */
120 static phys_addr_t __init_memblock
121 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
122 				phys_addr_t size, phys_addr_t align, int nid,
123 				ulong flags)
124 {
125 	phys_addr_t this_start, this_end, cand;
126 	u64 i;
127 
128 	for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
129 		this_start = clamp(this_start, start, end);
130 		this_end = clamp(this_end, start, end);
131 
132 		cand = round_up(this_start, align);
133 		if (cand < this_end && this_end - cand >= size)
134 			return cand;
135 	}
136 
137 	return 0;
138 }
139 
140 /**
141  * __memblock_find_range_top_down - find free area utility, in top-down
142  * @start: start of candidate range
143  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
144  * @size: size of free area to find
145  * @align: alignment of free area to find
146  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
147  * @flags: pick from blocks based on memory attributes
148  *
149  * Utility called from memblock_find_in_range_node(), find free area top-down.
150  *
151  * RETURNS:
152  * Found address on success, 0 on failure.
153  */
154 static phys_addr_t __init_memblock
155 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
156 			       phys_addr_t size, phys_addr_t align, int nid,
157 			       ulong flags)
158 {
159 	phys_addr_t this_start, this_end, cand;
160 	u64 i;
161 
162 	for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
163 					NULL) {
164 		this_start = clamp(this_start, start, end);
165 		this_end = clamp(this_end, start, end);
166 
167 		if (this_end < size)
168 			continue;
169 
170 		cand = round_down(this_end - size, align);
171 		if (cand >= this_start)
172 			return cand;
173 	}
174 
175 	return 0;
176 }
177 
178 /**
179  * memblock_find_in_range_node - find free area in given range and node
180  * @size: size of free area to find
181  * @align: alignment of free area to find
182  * @start: start of candidate range
183  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
184  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
185  * @flags: pick from blocks based on memory attributes
186  *
187  * Find @size free area aligned to @align in the specified range and node.
188  *
189  * When allocation direction is bottom-up, the @start should be greater
190  * than the end of the kernel image. Otherwise, it will be trimmed. The
191  * reason is that we want the bottom-up allocation just near the kernel
192  * image so it is highly likely that the allocated memory and the kernel
193  * will reside in the same node.
194  *
195  * If bottom-up allocation failed, will try to allocate memory top-down.
196  *
197  * RETURNS:
198  * Found address on success, 0 on failure.
199  */
200 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
201 					phys_addr_t align, phys_addr_t start,
202 					phys_addr_t end, int nid, ulong flags)
203 {
204 	phys_addr_t kernel_end, ret;
205 
206 	/* pump up @end */
207 	if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
208 		end = memblock.current_limit;
209 
210 	/* avoid allocating the first page */
211 	start = max_t(phys_addr_t, start, PAGE_SIZE);
212 	end = max(start, end);
213 	kernel_end = __pa_symbol(_end);
214 
215 	/*
216 	 * try bottom-up allocation only when bottom-up mode
217 	 * is set and @end is above the kernel image.
218 	 */
219 	if (memblock_bottom_up() && end > kernel_end) {
220 		phys_addr_t bottom_up_start;
221 
222 		/* make sure we will allocate above the kernel */
223 		bottom_up_start = max(start, kernel_end);
224 
225 		/* ok, try bottom-up allocation first */
226 		ret = __memblock_find_range_bottom_up(bottom_up_start, end,
227 						      size, align, nid, flags);
228 		if (ret)
229 			return ret;
230 
231 		/*
232 		 * we always limit bottom-up allocation above the kernel,
233 		 * but top-down allocation doesn't have the limit, so
234 		 * retrying top-down allocation may succeed when bottom-up
235 		 * allocation failed.
236 		 *
237 		 * bottom-up allocation is expected to be fail very rarely,
238 		 * so we use WARN_ONCE() here to see the stack trace if
239 		 * fail happens.
240 		 */
241 		WARN_ONCE(1, "memblock: bottom-up allocation failed, memory hotunplug may be affected\n");
242 	}
243 
244 	return __memblock_find_range_top_down(start, end, size, align, nid,
245 					      flags);
246 }
247 
248 /**
249  * memblock_find_in_range - find free area in given range
250  * @start: start of candidate range
251  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
252  * @size: size of free area to find
253  * @align: alignment of free area to find
254  *
255  * Find @size free area aligned to @align in the specified range.
256  *
257  * RETURNS:
258  * Found address on success, 0 on failure.
259  */
260 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
261 					phys_addr_t end, phys_addr_t size,
262 					phys_addr_t align)
263 {
264 	phys_addr_t ret;
265 	ulong flags = choose_memblock_flags();
266 
267 again:
268 	ret = memblock_find_in_range_node(size, align, start, end,
269 					    NUMA_NO_NODE, flags);
270 
271 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
272 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
273 			&size);
274 		flags &= ~MEMBLOCK_MIRROR;
275 		goto again;
276 	}
277 
278 	return ret;
279 }
280 
281 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
282 {
283 	type->total_size -= type->regions[r].size;
284 	memmove(&type->regions[r], &type->regions[r + 1],
285 		(type->cnt - (r + 1)) * sizeof(type->regions[r]));
286 	type->cnt--;
287 
288 	/* Special case for empty arrays */
289 	if (type->cnt == 0) {
290 		WARN_ON(type->total_size != 0);
291 		type->cnt = 1;
292 		type->regions[0].base = 0;
293 		type->regions[0].size = 0;
294 		type->regions[0].flags = 0;
295 		memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
296 	}
297 }
298 
299 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
300 
301 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
302 					phys_addr_t *addr)
303 {
304 	if (memblock.reserved.regions == memblock_reserved_init_regions)
305 		return 0;
306 
307 	*addr = __pa(memblock.reserved.regions);
308 
309 	return PAGE_ALIGN(sizeof(struct memblock_region) *
310 			  memblock.reserved.max);
311 }
312 
313 phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
314 					phys_addr_t *addr)
315 {
316 	if (memblock.memory.regions == memblock_memory_init_regions)
317 		return 0;
318 
319 	*addr = __pa(memblock.memory.regions);
320 
321 	return PAGE_ALIGN(sizeof(struct memblock_region) *
322 			  memblock.memory.max);
323 }
324 
325 #endif
326 
327 /**
328  * memblock_double_array - double the size of the memblock regions array
329  * @type: memblock type of the regions array being doubled
330  * @new_area_start: starting address of memory range to avoid overlap with
331  * @new_area_size: size of memory range to avoid overlap with
332  *
333  * Double the size of the @type regions array. If memblock is being used to
334  * allocate memory for a new reserved regions array and there is a previously
335  * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
336  * waiting to be reserved, ensure the memory used by the new array does
337  * not overlap.
338  *
339  * RETURNS:
340  * 0 on success, -1 on failure.
341  */
342 static int __init_memblock memblock_double_array(struct memblock_type *type,
343 						phys_addr_t new_area_start,
344 						phys_addr_t new_area_size)
345 {
346 	struct memblock_region *new_array, *old_array;
347 	phys_addr_t old_alloc_size, new_alloc_size;
348 	phys_addr_t old_size, new_size, addr;
349 	int use_slab = slab_is_available();
350 	int *in_slab;
351 
352 	/* We don't allow resizing until we know about the reserved regions
353 	 * of memory that aren't suitable for allocation
354 	 */
355 	if (!memblock_can_resize)
356 		return -1;
357 
358 	/* Calculate new doubled size */
359 	old_size = type->max * sizeof(struct memblock_region);
360 	new_size = old_size << 1;
361 	/*
362 	 * We need to allocated new one align to PAGE_SIZE,
363 	 *   so we can free them completely later.
364 	 */
365 	old_alloc_size = PAGE_ALIGN(old_size);
366 	new_alloc_size = PAGE_ALIGN(new_size);
367 
368 	/* Retrieve the slab flag */
369 	if (type == &memblock.memory)
370 		in_slab = &memblock_memory_in_slab;
371 	else
372 		in_slab = &memblock_reserved_in_slab;
373 
374 	/* Try to find some space for it.
375 	 *
376 	 * WARNING: We assume that either slab_is_available() and we use it or
377 	 * we use MEMBLOCK for allocations. That means that this is unsafe to
378 	 * use when bootmem is currently active (unless bootmem itself is
379 	 * implemented on top of MEMBLOCK which isn't the case yet)
380 	 *
381 	 * This should however not be an issue for now, as we currently only
382 	 * call into MEMBLOCK while it's still active, or much later when slab
383 	 * is active for memory hotplug operations
384 	 */
385 	if (use_slab) {
386 		new_array = kmalloc(new_size, GFP_KERNEL);
387 		addr = new_array ? __pa(new_array) : 0;
388 	} else {
389 		/* only exclude range when trying to double reserved.regions */
390 		if (type != &memblock.reserved)
391 			new_area_start = new_area_size = 0;
392 
393 		addr = memblock_find_in_range(new_area_start + new_area_size,
394 						memblock.current_limit,
395 						new_alloc_size, PAGE_SIZE);
396 		if (!addr && new_area_size)
397 			addr = memblock_find_in_range(0,
398 				min(new_area_start, memblock.current_limit),
399 				new_alloc_size, PAGE_SIZE);
400 
401 		new_array = addr ? __va(addr) : NULL;
402 	}
403 	if (!addr) {
404 		pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
405 		       memblock_type_name(type), type->max, type->max * 2);
406 		return -1;
407 	}
408 
409 	memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
410 			memblock_type_name(type), type->max * 2, (u64)addr,
411 			(u64)addr + new_size - 1);
412 
413 	/*
414 	 * Found space, we now need to move the array over before we add the
415 	 * reserved region since it may be our reserved array itself that is
416 	 * full.
417 	 */
418 	memcpy(new_array, type->regions, old_size);
419 	memset(new_array + type->max, 0, old_size);
420 	old_array = type->regions;
421 	type->regions = new_array;
422 	type->max <<= 1;
423 
424 	/* Free old array. We needn't free it if the array is the static one */
425 	if (*in_slab)
426 		kfree(old_array);
427 	else if (old_array != memblock_memory_init_regions &&
428 		 old_array != memblock_reserved_init_regions)
429 		memblock_free(__pa(old_array), old_alloc_size);
430 
431 	/*
432 	 * Reserve the new array if that comes from the memblock.  Otherwise, we
433 	 * needn't do it
434 	 */
435 	if (!use_slab)
436 		BUG_ON(memblock_reserve(addr, new_alloc_size));
437 
438 	/* Update slab flag */
439 	*in_slab = use_slab;
440 
441 	return 0;
442 }
443 
444 /**
445  * memblock_merge_regions - merge neighboring compatible regions
446  * @type: memblock type to scan
447  *
448  * Scan @type and merge neighboring compatible regions.
449  */
450 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
451 {
452 	int i = 0;
453 
454 	/* cnt never goes below 1 */
455 	while (i < type->cnt - 1) {
456 		struct memblock_region *this = &type->regions[i];
457 		struct memblock_region *next = &type->regions[i + 1];
458 
459 		if (this->base + this->size != next->base ||
460 		    memblock_get_region_node(this) !=
461 		    memblock_get_region_node(next) ||
462 		    this->flags != next->flags) {
463 			BUG_ON(this->base + this->size > next->base);
464 			i++;
465 			continue;
466 		}
467 
468 		this->size += next->size;
469 		/* move forward from next + 1, index of which is i + 2 */
470 		memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
471 		type->cnt--;
472 	}
473 }
474 
475 /**
476  * memblock_insert_region - insert new memblock region
477  * @type:	memblock type to insert into
478  * @idx:	index for the insertion point
479  * @base:	base address of the new region
480  * @size:	size of the new region
481  * @nid:	node id of the new region
482  * @flags:	flags of the new region
483  *
484  * Insert new memblock region [@base,@base+@size) into @type at @idx.
485  * @type must already have extra room to accomodate the new region.
486  */
487 static void __init_memblock memblock_insert_region(struct memblock_type *type,
488 						   int idx, phys_addr_t base,
489 						   phys_addr_t size,
490 						   int nid, unsigned long flags)
491 {
492 	struct memblock_region *rgn = &type->regions[idx];
493 
494 	BUG_ON(type->cnt >= type->max);
495 	memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
496 	rgn->base = base;
497 	rgn->size = size;
498 	rgn->flags = flags;
499 	memblock_set_region_node(rgn, nid);
500 	type->cnt++;
501 	type->total_size += size;
502 }
503 
504 /**
505  * memblock_add_range - add new memblock region
506  * @type: memblock type to add new region into
507  * @base: base address of the new region
508  * @size: size of the new region
509  * @nid: nid of the new region
510  * @flags: flags of the new region
511  *
512  * Add new memblock region [@base,@base+@size) into @type.  The new region
513  * is allowed to overlap with existing ones - overlaps don't affect already
514  * existing regions.  @type is guaranteed to be minimal (all neighbouring
515  * compatible regions are merged) after the addition.
516  *
517  * RETURNS:
518  * 0 on success, -errno on failure.
519  */
520 int __init_memblock memblock_add_range(struct memblock_type *type,
521 				phys_addr_t base, phys_addr_t size,
522 				int nid, unsigned long flags)
523 {
524 	bool insert = false;
525 	phys_addr_t obase = base;
526 	phys_addr_t end = base + memblock_cap_size(base, &size);
527 	int idx, nr_new;
528 	struct memblock_region *rgn;
529 
530 	if (!size)
531 		return 0;
532 
533 	/* special case for empty array */
534 	if (type->regions[0].size == 0) {
535 		WARN_ON(type->cnt != 1 || type->total_size);
536 		type->regions[0].base = base;
537 		type->regions[0].size = size;
538 		type->regions[0].flags = flags;
539 		memblock_set_region_node(&type->regions[0], nid);
540 		type->total_size = size;
541 		return 0;
542 	}
543 repeat:
544 	/*
545 	 * The following is executed twice.  Once with %false @insert and
546 	 * then with %true.  The first counts the number of regions needed
547 	 * to accomodate the new area.  The second actually inserts them.
548 	 */
549 	base = obase;
550 	nr_new = 0;
551 
552 	for_each_memblock_type(type, rgn) {
553 		phys_addr_t rbase = rgn->base;
554 		phys_addr_t rend = rbase + rgn->size;
555 
556 		if (rbase >= end)
557 			break;
558 		if (rend <= base)
559 			continue;
560 		/*
561 		 * @rgn overlaps.  If it separates the lower part of new
562 		 * area, insert that portion.
563 		 */
564 		if (rbase > base) {
565 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
566 			WARN_ON(nid != memblock_get_region_node(rgn));
567 #endif
568 			WARN_ON(flags != rgn->flags);
569 			nr_new++;
570 			if (insert)
571 				memblock_insert_region(type, idx++, base,
572 						       rbase - base, nid,
573 						       flags);
574 		}
575 		/* area below @rend is dealt with, forget about it */
576 		base = min(rend, end);
577 	}
578 
579 	/* insert the remaining portion */
580 	if (base < end) {
581 		nr_new++;
582 		if (insert)
583 			memblock_insert_region(type, idx, base, end - base,
584 					       nid, flags);
585 	}
586 
587 	/*
588 	 * If this was the first round, resize array and repeat for actual
589 	 * insertions; otherwise, merge and return.
590 	 */
591 	if (!insert) {
592 		while (type->cnt + nr_new > type->max)
593 			if (memblock_double_array(type, obase, size) < 0)
594 				return -ENOMEM;
595 		insert = true;
596 		goto repeat;
597 	} else {
598 		memblock_merge_regions(type);
599 		return 0;
600 	}
601 }
602 
603 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
604 				       int nid)
605 {
606 	return memblock_add_range(&memblock.memory, base, size, nid, 0);
607 }
608 
609 static int __init_memblock memblock_add_region(phys_addr_t base,
610 						phys_addr_t size,
611 						int nid,
612 						unsigned long flags)
613 {
614 	memblock_dbg("memblock_add: [%#016llx-%#016llx] flags %#02lx %pF\n",
615 		     (unsigned long long)base,
616 		     (unsigned long long)base + size - 1,
617 		     flags, (void *)_RET_IP_);
618 
619 	return memblock_add_range(&memblock.memory, base, size, nid, flags);
620 }
621 
622 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
623 {
624 	return memblock_add_region(base, size, MAX_NUMNODES, 0);
625 }
626 
627 /**
628  * memblock_isolate_range - isolate given range into disjoint memblocks
629  * @type: memblock type to isolate range for
630  * @base: base of range to isolate
631  * @size: size of range to isolate
632  * @start_rgn: out parameter for the start of isolated region
633  * @end_rgn: out parameter for the end of isolated region
634  *
635  * Walk @type and ensure that regions don't cross the boundaries defined by
636  * [@base,@base+@size).  Crossing regions are split at the boundaries,
637  * which may create at most two more regions.  The index of the first
638  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
639  *
640  * RETURNS:
641  * 0 on success, -errno on failure.
642  */
643 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
644 					phys_addr_t base, phys_addr_t size,
645 					int *start_rgn, int *end_rgn)
646 {
647 	phys_addr_t end = base + memblock_cap_size(base, &size);
648 	int idx;
649 	struct memblock_region *rgn;
650 
651 	*start_rgn = *end_rgn = 0;
652 
653 	if (!size)
654 		return 0;
655 
656 	/* we'll create at most two more regions */
657 	while (type->cnt + 2 > type->max)
658 		if (memblock_double_array(type, base, size) < 0)
659 			return -ENOMEM;
660 
661 	for_each_memblock_type(type, rgn) {
662 		phys_addr_t rbase = rgn->base;
663 		phys_addr_t rend = rbase + rgn->size;
664 
665 		if (rbase >= end)
666 			break;
667 		if (rend <= base)
668 			continue;
669 
670 		if (rbase < base) {
671 			/*
672 			 * @rgn intersects from below.  Split and continue
673 			 * to process the next region - the new top half.
674 			 */
675 			rgn->base = base;
676 			rgn->size -= base - rbase;
677 			type->total_size -= base - rbase;
678 			memblock_insert_region(type, idx, rbase, base - rbase,
679 					       memblock_get_region_node(rgn),
680 					       rgn->flags);
681 		} else if (rend > end) {
682 			/*
683 			 * @rgn intersects from above.  Split and redo the
684 			 * current region - the new bottom half.
685 			 */
686 			rgn->base = end;
687 			rgn->size -= end - rbase;
688 			type->total_size -= end - rbase;
689 			memblock_insert_region(type, idx--, rbase, end - rbase,
690 					       memblock_get_region_node(rgn),
691 					       rgn->flags);
692 		} else {
693 			/* @rgn is fully contained, record it */
694 			if (!*end_rgn)
695 				*start_rgn = idx;
696 			*end_rgn = idx + 1;
697 		}
698 	}
699 
700 	return 0;
701 }
702 
703 static int __init_memblock memblock_remove_range(struct memblock_type *type,
704 					  phys_addr_t base, phys_addr_t size)
705 {
706 	int start_rgn, end_rgn;
707 	int i, ret;
708 
709 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
710 	if (ret)
711 		return ret;
712 
713 	for (i = end_rgn - 1; i >= start_rgn; i--)
714 		memblock_remove_region(type, i);
715 	return 0;
716 }
717 
718 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
719 {
720 	return memblock_remove_range(&memblock.memory, base, size);
721 }
722 
723 
724 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
725 {
726 	memblock_dbg("   memblock_free: [%#016llx-%#016llx] %pF\n",
727 		     (unsigned long long)base,
728 		     (unsigned long long)base + size - 1,
729 		     (void *)_RET_IP_);
730 
731 	kmemleak_free_part(__va(base), size);
732 	return memblock_remove_range(&memblock.reserved, base, size);
733 }
734 
735 static int __init_memblock memblock_reserve_region(phys_addr_t base,
736 						   phys_addr_t size,
737 						   int nid,
738 						   unsigned long flags)
739 {
740 	memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
741 		     (unsigned long long)base,
742 		     (unsigned long long)base + size - 1,
743 		     flags, (void *)_RET_IP_);
744 
745 	return memblock_add_range(&memblock.reserved, base, size, nid, flags);
746 }
747 
748 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
749 {
750 	return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
751 }
752 
753 /**
754  *
755  * This function isolates region [@base, @base + @size), and sets/clears flag
756  *
757  * Return 0 on success, -errno on failure.
758  */
759 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
760 				phys_addr_t size, int set, int flag)
761 {
762 	struct memblock_type *type = &memblock.memory;
763 	int i, ret, start_rgn, end_rgn;
764 
765 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
766 	if (ret)
767 		return ret;
768 
769 	for (i = start_rgn; i < end_rgn; i++)
770 		if (set)
771 			memblock_set_region_flags(&type->regions[i], flag);
772 		else
773 			memblock_clear_region_flags(&type->regions[i], flag);
774 
775 	memblock_merge_regions(type);
776 	return 0;
777 }
778 
779 /**
780  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
781  * @base: the base phys addr of the region
782  * @size: the size of the region
783  *
784  * Return 0 on success, -errno on failure.
785  */
786 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
787 {
788 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
789 }
790 
791 /**
792  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
793  * @base: the base phys addr of the region
794  * @size: the size of the region
795  *
796  * Return 0 on success, -errno on failure.
797  */
798 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
799 {
800 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
801 }
802 
803 /**
804  * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
805  * @base: the base phys addr of the region
806  * @size: the size of the region
807  *
808  * Return 0 on success, -errno on failure.
809  */
810 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
811 {
812 	system_has_some_mirror = true;
813 
814 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
815 }
816 
817 /**
818  * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
819  * @base: the base phys addr of the region
820  * @size: the size of the region
821  *
822  * Return 0 on success, -errno on failure.
823  */
824 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
825 {
826 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
827 }
828 
829 /**
830  * __next_reserved_mem_region - next function for for_each_reserved_region()
831  * @idx: pointer to u64 loop variable
832  * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
833  * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
834  *
835  * Iterate over all reserved memory regions.
836  */
837 void __init_memblock __next_reserved_mem_region(u64 *idx,
838 					   phys_addr_t *out_start,
839 					   phys_addr_t *out_end)
840 {
841 	struct memblock_type *type = &memblock.reserved;
842 
843 	if (*idx >= 0 && *idx < type->cnt) {
844 		struct memblock_region *r = &type->regions[*idx];
845 		phys_addr_t base = r->base;
846 		phys_addr_t size = r->size;
847 
848 		if (out_start)
849 			*out_start = base;
850 		if (out_end)
851 			*out_end = base + size - 1;
852 
853 		*idx += 1;
854 		return;
855 	}
856 
857 	/* signal end of iteration */
858 	*idx = ULLONG_MAX;
859 }
860 
861 /**
862  * __next__mem_range - next function for for_each_free_mem_range() etc.
863  * @idx: pointer to u64 loop variable
864  * @nid: node selector, %NUMA_NO_NODE for all nodes
865  * @flags: pick from blocks based on memory attributes
866  * @type_a: pointer to memblock_type from where the range is taken
867  * @type_b: pointer to memblock_type which excludes memory from being taken
868  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
869  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
870  * @out_nid: ptr to int for nid of the range, can be %NULL
871  *
872  * Find the first area from *@idx which matches @nid, fill the out
873  * parameters, and update *@idx for the next iteration.  The lower 32bit of
874  * *@idx contains index into type_a and the upper 32bit indexes the
875  * areas before each region in type_b.	For example, if type_b regions
876  * look like the following,
877  *
878  *	0:[0-16), 1:[32-48), 2:[128-130)
879  *
880  * The upper 32bit indexes the following regions.
881  *
882  *	0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
883  *
884  * As both region arrays are sorted, the function advances the two indices
885  * in lockstep and returns each intersection.
886  */
887 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
888 				      struct memblock_type *type_a,
889 				      struct memblock_type *type_b,
890 				      phys_addr_t *out_start,
891 				      phys_addr_t *out_end, int *out_nid)
892 {
893 	int idx_a = *idx & 0xffffffff;
894 	int idx_b = *idx >> 32;
895 
896 	if (WARN_ONCE(nid == MAX_NUMNODES,
897 	"Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
898 		nid = NUMA_NO_NODE;
899 
900 	for (; idx_a < type_a->cnt; idx_a++) {
901 		struct memblock_region *m = &type_a->regions[idx_a];
902 
903 		phys_addr_t m_start = m->base;
904 		phys_addr_t m_end = m->base + m->size;
905 		int	    m_nid = memblock_get_region_node(m);
906 
907 		/* only memory regions are associated with nodes, check it */
908 		if (nid != NUMA_NO_NODE && nid != m_nid)
909 			continue;
910 
911 		/* skip hotpluggable memory regions if needed */
912 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
913 			continue;
914 
915 		/* if we want mirror memory skip non-mirror memory regions */
916 		if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
917 			continue;
918 
919 		/* skip nomap memory unless we were asked for it explicitly */
920 		if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
921 			continue;
922 
923 		if (!type_b) {
924 			if (out_start)
925 				*out_start = m_start;
926 			if (out_end)
927 				*out_end = m_end;
928 			if (out_nid)
929 				*out_nid = m_nid;
930 			idx_a++;
931 			*idx = (u32)idx_a | (u64)idx_b << 32;
932 			return;
933 		}
934 
935 		/* scan areas before each reservation */
936 		for (; idx_b < type_b->cnt + 1; idx_b++) {
937 			struct memblock_region *r;
938 			phys_addr_t r_start;
939 			phys_addr_t r_end;
940 
941 			r = &type_b->regions[idx_b];
942 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
943 			r_end = idx_b < type_b->cnt ?
944 				r->base : ULLONG_MAX;
945 
946 			/*
947 			 * if idx_b advanced past idx_a,
948 			 * break out to advance idx_a
949 			 */
950 			if (r_start >= m_end)
951 				break;
952 			/* if the two regions intersect, we're done */
953 			if (m_start < r_end) {
954 				if (out_start)
955 					*out_start =
956 						max(m_start, r_start);
957 				if (out_end)
958 					*out_end = min(m_end, r_end);
959 				if (out_nid)
960 					*out_nid = m_nid;
961 				/*
962 				 * The region which ends first is
963 				 * advanced for the next iteration.
964 				 */
965 				if (m_end <= r_end)
966 					idx_a++;
967 				else
968 					idx_b++;
969 				*idx = (u32)idx_a | (u64)idx_b << 32;
970 				return;
971 			}
972 		}
973 	}
974 
975 	/* signal end of iteration */
976 	*idx = ULLONG_MAX;
977 }
978 
979 /**
980  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
981  *
982  * Finds the next range from type_a which is not marked as unsuitable
983  * in type_b.
984  *
985  * @idx: pointer to u64 loop variable
986  * @nid: node selector, %NUMA_NO_NODE for all nodes
987  * @flags: pick from blocks based on memory attributes
988  * @type_a: pointer to memblock_type from where the range is taken
989  * @type_b: pointer to memblock_type which excludes memory from being taken
990  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
991  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
992  * @out_nid: ptr to int for nid of the range, can be %NULL
993  *
994  * Reverse of __next_mem_range().
995  */
996 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
997 					  struct memblock_type *type_a,
998 					  struct memblock_type *type_b,
999 					  phys_addr_t *out_start,
1000 					  phys_addr_t *out_end, int *out_nid)
1001 {
1002 	int idx_a = *idx & 0xffffffff;
1003 	int idx_b = *idx >> 32;
1004 
1005 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1006 		nid = NUMA_NO_NODE;
1007 
1008 	if (*idx == (u64)ULLONG_MAX) {
1009 		idx_a = type_a->cnt - 1;
1010 		idx_b = type_b->cnt;
1011 	}
1012 
1013 	for (; idx_a >= 0; idx_a--) {
1014 		struct memblock_region *m = &type_a->regions[idx_a];
1015 
1016 		phys_addr_t m_start = m->base;
1017 		phys_addr_t m_end = m->base + m->size;
1018 		int m_nid = memblock_get_region_node(m);
1019 
1020 		/* only memory regions are associated with nodes, check it */
1021 		if (nid != NUMA_NO_NODE && nid != m_nid)
1022 			continue;
1023 
1024 		/* skip hotpluggable memory regions if needed */
1025 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1026 			continue;
1027 
1028 		/* if we want mirror memory skip non-mirror memory regions */
1029 		if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1030 			continue;
1031 
1032 		/* skip nomap memory unless we were asked for it explicitly */
1033 		if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1034 			continue;
1035 
1036 		if (!type_b) {
1037 			if (out_start)
1038 				*out_start = m_start;
1039 			if (out_end)
1040 				*out_end = m_end;
1041 			if (out_nid)
1042 				*out_nid = m_nid;
1043 			idx_a++;
1044 			*idx = (u32)idx_a | (u64)idx_b << 32;
1045 			return;
1046 		}
1047 
1048 		/* scan areas before each reservation */
1049 		for (; idx_b >= 0; idx_b--) {
1050 			struct memblock_region *r;
1051 			phys_addr_t r_start;
1052 			phys_addr_t r_end;
1053 
1054 			r = &type_b->regions[idx_b];
1055 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1056 			r_end = idx_b < type_b->cnt ?
1057 				r->base : ULLONG_MAX;
1058 			/*
1059 			 * if idx_b advanced past idx_a,
1060 			 * break out to advance idx_a
1061 			 */
1062 
1063 			if (r_end <= m_start)
1064 				break;
1065 			/* if the two regions intersect, we're done */
1066 			if (m_end > r_start) {
1067 				if (out_start)
1068 					*out_start = max(m_start, r_start);
1069 				if (out_end)
1070 					*out_end = min(m_end, r_end);
1071 				if (out_nid)
1072 					*out_nid = m_nid;
1073 				if (m_start >= r_start)
1074 					idx_a--;
1075 				else
1076 					idx_b--;
1077 				*idx = (u32)idx_a | (u64)idx_b << 32;
1078 				return;
1079 			}
1080 		}
1081 	}
1082 	/* signal end of iteration */
1083 	*idx = ULLONG_MAX;
1084 }
1085 
1086 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1087 /*
1088  * Common iterator interface used to define for_each_mem_range().
1089  */
1090 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1091 				unsigned long *out_start_pfn,
1092 				unsigned long *out_end_pfn, int *out_nid)
1093 {
1094 	struct memblock_type *type = &memblock.memory;
1095 	struct memblock_region *r;
1096 
1097 	while (++*idx < type->cnt) {
1098 		r = &type->regions[*idx];
1099 
1100 		if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1101 			continue;
1102 		if (nid == MAX_NUMNODES || nid == r->nid)
1103 			break;
1104 	}
1105 	if (*idx >= type->cnt) {
1106 		*idx = -1;
1107 		return;
1108 	}
1109 
1110 	if (out_start_pfn)
1111 		*out_start_pfn = PFN_UP(r->base);
1112 	if (out_end_pfn)
1113 		*out_end_pfn = PFN_DOWN(r->base + r->size);
1114 	if (out_nid)
1115 		*out_nid = r->nid;
1116 }
1117 
1118 /**
1119  * memblock_set_node - set node ID on memblock regions
1120  * @base: base of area to set node ID for
1121  * @size: size of area to set node ID for
1122  * @type: memblock type to set node ID for
1123  * @nid: node ID to set
1124  *
1125  * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1126  * Regions which cross the area boundaries are split as necessary.
1127  *
1128  * RETURNS:
1129  * 0 on success, -errno on failure.
1130  */
1131 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1132 				      struct memblock_type *type, int nid)
1133 {
1134 	int start_rgn, end_rgn;
1135 	int i, ret;
1136 
1137 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1138 	if (ret)
1139 		return ret;
1140 
1141 	for (i = start_rgn; i < end_rgn; i++)
1142 		memblock_set_region_node(&type->regions[i], nid);
1143 
1144 	memblock_merge_regions(type);
1145 	return 0;
1146 }
1147 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1148 
1149 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1150 					phys_addr_t align, phys_addr_t start,
1151 					phys_addr_t end, int nid, ulong flags)
1152 {
1153 	phys_addr_t found;
1154 
1155 	if (!align)
1156 		align = SMP_CACHE_BYTES;
1157 
1158 	found = memblock_find_in_range_node(size, align, start, end, nid,
1159 					    flags);
1160 	if (found && !memblock_reserve(found, size)) {
1161 		/*
1162 		 * The min_count is set to 0 so that memblock allocations are
1163 		 * never reported as leaks.
1164 		 */
1165 		kmemleak_alloc(__va(found), size, 0, 0);
1166 		return found;
1167 	}
1168 	return 0;
1169 }
1170 
1171 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1172 					phys_addr_t start, phys_addr_t end,
1173 					ulong flags)
1174 {
1175 	return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1176 					flags);
1177 }
1178 
1179 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1180 					phys_addr_t align, phys_addr_t max_addr,
1181 					int nid, ulong flags)
1182 {
1183 	return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1184 }
1185 
1186 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1187 {
1188 	ulong flags = choose_memblock_flags();
1189 	phys_addr_t ret;
1190 
1191 again:
1192 	ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1193 				      nid, flags);
1194 
1195 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
1196 		flags &= ~MEMBLOCK_MIRROR;
1197 		goto again;
1198 	}
1199 	return ret;
1200 }
1201 
1202 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1203 {
1204 	return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1205 				       MEMBLOCK_NONE);
1206 }
1207 
1208 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1209 {
1210 	phys_addr_t alloc;
1211 
1212 	alloc = __memblock_alloc_base(size, align, max_addr);
1213 
1214 	if (alloc == 0)
1215 		panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1216 		      (unsigned long long) size, (unsigned long long) max_addr);
1217 
1218 	return alloc;
1219 }
1220 
1221 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1222 {
1223 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1224 }
1225 
1226 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1227 {
1228 	phys_addr_t res = memblock_alloc_nid(size, align, nid);
1229 
1230 	if (res)
1231 		return res;
1232 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1233 }
1234 
1235 /**
1236  * memblock_virt_alloc_internal - allocate boot memory block
1237  * @size: size of memory block to be allocated in bytes
1238  * @align: alignment of the region and block's size
1239  * @min_addr: the lower bound of the memory region to allocate (phys address)
1240  * @max_addr: the upper bound of the memory region to allocate (phys address)
1241  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1242  *
1243  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1244  * will fall back to memory below @min_addr. Also, allocation may fall back
1245  * to any node in the system if the specified node can not
1246  * hold the requested memory.
1247  *
1248  * The allocation is performed from memory region limited by
1249  * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1250  *
1251  * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1252  *
1253  * The phys address of allocated boot memory block is converted to virtual and
1254  * allocated memory is reset to 0.
1255  *
1256  * In addition, function sets the min_count to 0 using kmemleak_alloc for
1257  * allocated boot memory block, so that it is never reported as leaks.
1258  *
1259  * RETURNS:
1260  * Virtual address of allocated memory block on success, NULL on failure.
1261  */
1262 static void * __init memblock_virt_alloc_internal(
1263 				phys_addr_t size, phys_addr_t align,
1264 				phys_addr_t min_addr, phys_addr_t max_addr,
1265 				int nid)
1266 {
1267 	phys_addr_t alloc;
1268 	void *ptr;
1269 	ulong flags = choose_memblock_flags();
1270 
1271 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1272 		nid = NUMA_NO_NODE;
1273 
1274 	/*
1275 	 * Detect any accidental use of these APIs after slab is ready, as at
1276 	 * this moment memblock may be deinitialized already and its
1277 	 * internal data may be destroyed (after execution of free_all_bootmem)
1278 	 */
1279 	if (WARN_ON_ONCE(slab_is_available()))
1280 		return kzalloc_node(size, GFP_NOWAIT, nid);
1281 
1282 	if (!align)
1283 		align = SMP_CACHE_BYTES;
1284 
1285 	if (max_addr > memblock.current_limit)
1286 		max_addr = memblock.current_limit;
1287 
1288 again:
1289 	alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1290 					    nid, flags);
1291 	if (alloc)
1292 		goto done;
1293 
1294 	if (nid != NUMA_NO_NODE) {
1295 		alloc = memblock_find_in_range_node(size, align, min_addr,
1296 						    max_addr, NUMA_NO_NODE,
1297 						    flags);
1298 		if (alloc)
1299 			goto done;
1300 	}
1301 
1302 	if (min_addr) {
1303 		min_addr = 0;
1304 		goto again;
1305 	}
1306 
1307 	if (flags & MEMBLOCK_MIRROR) {
1308 		flags &= ~MEMBLOCK_MIRROR;
1309 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1310 			&size);
1311 		goto again;
1312 	}
1313 
1314 	return NULL;
1315 done:
1316 	memblock_reserve(alloc, size);
1317 	ptr = phys_to_virt(alloc);
1318 	memset(ptr, 0, size);
1319 
1320 	/*
1321 	 * The min_count is set to 0 so that bootmem allocated blocks
1322 	 * are never reported as leaks. This is because many of these blocks
1323 	 * are only referred via the physical address which is not
1324 	 * looked up by kmemleak.
1325 	 */
1326 	kmemleak_alloc(ptr, size, 0, 0);
1327 
1328 	return ptr;
1329 }
1330 
1331 /**
1332  * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
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 version of _memblock_virt_alloc_try_nid_nopanic() which provides
1343  * additional debug information (including caller info), if enabled.
1344  *
1345  * RETURNS:
1346  * Virtual address of allocated memory block on success, NULL on failure.
1347  */
1348 void * __init memblock_virt_alloc_try_nid_nopanic(
1349 				phys_addr_t size, phys_addr_t align,
1350 				phys_addr_t min_addr, phys_addr_t max_addr,
1351 				int nid)
1352 {
1353 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1354 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1355 		     (u64)max_addr, (void *)_RET_IP_);
1356 	return memblock_virt_alloc_internal(size, align, min_addr,
1357 					     max_addr, nid);
1358 }
1359 
1360 /**
1361  * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1362  * @size: size of memory block to be allocated in bytes
1363  * @align: alignment of the region and block's size
1364  * @min_addr: the lower bound of the memory region from where the allocation
1365  *	  is preferred (phys address)
1366  * @max_addr: the upper bound of the memory region from where the allocation
1367  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1368  *	      allocate only from memory limited by memblock.current_limit value
1369  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1370  *
1371  * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1372  * which provides debug information (including caller info), if enabled,
1373  * and panics if the request can not be satisfied.
1374  *
1375  * RETURNS:
1376  * Virtual address of allocated memory block on success, NULL on failure.
1377  */
1378 void * __init memblock_virt_alloc_try_nid(
1379 			phys_addr_t size, phys_addr_t align,
1380 			phys_addr_t min_addr, phys_addr_t max_addr,
1381 			int nid)
1382 {
1383 	void *ptr;
1384 
1385 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1386 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1387 		     (u64)max_addr, (void *)_RET_IP_);
1388 	ptr = memblock_virt_alloc_internal(size, align,
1389 					   min_addr, max_addr, nid);
1390 	if (ptr)
1391 		return ptr;
1392 
1393 	panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1394 	      __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1395 	      (u64)max_addr);
1396 	return NULL;
1397 }
1398 
1399 /**
1400  * __memblock_free_early - free boot memory block
1401  * @base: phys starting address of the  boot memory block
1402  * @size: size of the boot memory block in bytes
1403  *
1404  * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1405  * The freeing memory will not be released to the buddy allocator.
1406  */
1407 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1408 {
1409 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1410 		     __func__, (u64)base, (u64)base + size - 1,
1411 		     (void *)_RET_IP_);
1412 	kmemleak_free_part(__va(base), size);
1413 	memblock_remove_range(&memblock.reserved, base, size);
1414 }
1415 
1416 /*
1417  * __memblock_free_late - free bootmem block pages directly to buddy allocator
1418  * @addr: phys starting address of the  boot memory block
1419  * @size: size of the boot memory block in bytes
1420  *
1421  * This is only useful when the bootmem allocator has already been torn
1422  * down, but we are still initializing the system.  Pages are released directly
1423  * to the buddy allocator, no bootmem metadata is updated because it is gone.
1424  */
1425 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1426 {
1427 	u64 cursor, end;
1428 
1429 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1430 		     __func__, (u64)base, (u64)base + size - 1,
1431 		     (void *)_RET_IP_);
1432 	kmemleak_free_part(__va(base), size);
1433 	cursor = PFN_UP(base);
1434 	end = PFN_DOWN(base + size);
1435 
1436 	for (; cursor < end; cursor++) {
1437 		__free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1438 		totalram_pages++;
1439 	}
1440 }
1441 
1442 /*
1443  * Remaining API functions
1444  */
1445 
1446 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1447 {
1448 	return memblock.memory.total_size;
1449 }
1450 
1451 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1452 {
1453 	unsigned long pages = 0;
1454 	struct memblock_region *r;
1455 	unsigned long start_pfn, end_pfn;
1456 
1457 	for_each_memblock(memory, r) {
1458 		start_pfn = memblock_region_memory_base_pfn(r);
1459 		end_pfn = memblock_region_memory_end_pfn(r);
1460 		start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1461 		end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1462 		pages += end_pfn - start_pfn;
1463 	}
1464 
1465 	return PFN_PHYS(pages);
1466 }
1467 
1468 /* lowest address */
1469 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1470 {
1471 	return memblock.memory.regions[0].base;
1472 }
1473 
1474 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1475 {
1476 	int idx = memblock.memory.cnt - 1;
1477 
1478 	return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1479 }
1480 
1481 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1482 {
1483 	phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1484 	struct memblock_region *r;
1485 
1486 	if (!limit)
1487 		return;
1488 
1489 	/* find out max address */
1490 	for_each_memblock(memory, r) {
1491 		if (limit <= r->size) {
1492 			max_addr = r->base + limit;
1493 			break;
1494 		}
1495 		limit -= r->size;
1496 	}
1497 
1498 	/* truncate both memory and reserved regions */
1499 	memblock_remove_range(&memblock.memory, max_addr,
1500 			      (phys_addr_t)ULLONG_MAX);
1501 	memblock_remove_range(&memblock.reserved, max_addr,
1502 			      (phys_addr_t)ULLONG_MAX);
1503 }
1504 
1505 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1506 {
1507 	unsigned int left = 0, right = type->cnt;
1508 
1509 	do {
1510 		unsigned int mid = (right + left) / 2;
1511 
1512 		if (addr < type->regions[mid].base)
1513 			right = mid;
1514 		else if (addr >= (type->regions[mid].base +
1515 				  type->regions[mid].size))
1516 			left = mid + 1;
1517 		else
1518 			return mid;
1519 	} while (left < right);
1520 	return -1;
1521 }
1522 
1523 bool __init memblock_is_reserved(phys_addr_t addr)
1524 {
1525 	return memblock_search(&memblock.reserved, addr) != -1;
1526 }
1527 
1528 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1529 {
1530 	return memblock_search(&memblock.memory, addr) != -1;
1531 }
1532 
1533 int __init_memblock memblock_is_map_memory(phys_addr_t addr)
1534 {
1535 	int i = memblock_search(&memblock.memory, addr);
1536 
1537 	if (i == -1)
1538 		return false;
1539 	return !memblock_is_nomap(&memblock.memory.regions[i]);
1540 }
1541 
1542 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1543 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1544 			 unsigned long *start_pfn, unsigned long *end_pfn)
1545 {
1546 	struct memblock_type *type = &memblock.memory;
1547 	int mid = memblock_search(type, PFN_PHYS(pfn));
1548 
1549 	if (mid == -1)
1550 		return -1;
1551 
1552 	*start_pfn = PFN_DOWN(type->regions[mid].base);
1553 	*end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1554 
1555 	return type->regions[mid].nid;
1556 }
1557 #endif
1558 
1559 /**
1560  * memblock_is_region_memory - check if a region is a subset of memory
1561  * @base: base of region to check
1562  * @size: size of region to check
1563  *
1564  * Check if the region [@base, @base+@size) is a subset of a memory block.
1565  *
1566  * RETURNS:
1567  * 0 if false, non-zero if true
1568  */
1569 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1570 {
1571 	int idx = memblock_search(&memblock.memory, base);
1572 	phys_addr_t end = base + memblock_cap_size(base, &size);
1573 
1574 	if (idx == -1)
1575 		return 0;
1576 	return memblock.memory.regions[idx].base <= base &&
1577 		(memblock.memory.regions[idx].base +
1578 		 memblock.memory.regions[idx].size) >= end;
1579 }
1580 
1581 /**
1582  * memblock_is_region_reserved - check if a region intersects reserved memory
1583  * @base: base of region to check
1584  * @size: size of region to check
1585  *
1586  * Check if the region [@base, @base+@size) intersects a reserved memory block.
1587  *
1588  * RETURNS:
1589  * True if they intersect, false if not.
1590  */
1591 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1592 {
1593 	memblock_cap_size(base, &size);
1594 	return memblock_overlaps_region(&memblock.reserved, base, size);
1595 }
1596 
1597 void __init_memblock memblock_trim_memory(phys_addr_t align)
1598 {
1599 	phys_addr_t start, end, orig_start, orig_end;
1600 	struct memblock_region *r;
1601 
1602 	for_each_memblock(memory, r) {
1603 		orig_start = r->base;
1604 		orig_end = r->base + r->size;
1605 		start = round_up(orig_start, align);
1606 		end = round_down(orig_end, align);
1607 
1608 		if (start == orig_start && end == orig_end)
1609 			continue;
1610 
1611 		if (start < end) {
1612 			r->base = start;
1613 			r->size = end - start;
1614 		} else {
1615 			memblock_remove_region(&memblock.memory,
1616 					       r - memblock.memory.regions);
1617 			r--;
1618 		}
1619 	}
1620 }
1621 
1622 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1623 {
1624 	memblock.current_limit = limit;
1625 }
1626 
1627 phys_addr_t __init_memblock memblock_get_current_limit(void)
1628 {
1629 	return memblock.current_limit;
1630 }
1631 
1632 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1633 {
1634 	unsigned long long base, size;
1635 	unsigned long flags;
1636 	int idx;
1637 	struct memblock_region *rgn;
1638 
1639 	pr_info(" %s.cnt  = 0x%lx\n", name, type->cnt);
1640 
1641 	for_each_memblock_type(type, rgn) {
1642 		char nid_buf[32] = "";
1643 
1644 		base = rgn->base;
1645 		size = rgn->size;
1646 		flags = rgn->flags;
1647 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1648 		if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1649 			snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1650 				 memblock_get_region_node(rgn));
1651 #endif
1652 		pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1653 			name, idx, base, base + size - 1, size, nid_buf, flags);
1654 	}
1655 }
1656 
1657 void __init_memblock __memblock_dump_all(void)
1658 {
1659 	pr_info("MEMBLOCK configuration:\n");
1660 	pr_info(" memory size = %#llx reserved size = %#llx\n",
1661 		(unsigned long long)memblock.memory.total_size,
1662 		(unsigned long long)memblock.reserved.total_size);
1663 
1664 	memblock_dump(&memblock.memory, "memory");
1665 	memblock_dump(&memblock.reserved, "reserved");
1666 }
1667 
1668 void __init memblock_allow_resize(void)
1669 {
1670 	memblock_can_resize = 1;
1671 }
1672 
1673 static int __init early_memblock(char *p)
1674 {
1675 	if (p && strstr(p, "debug"))
1676 		memblock_debug = 1;
1677 	return 0;
1678 }
1679 early_param("memblock", early_memblock);
1680 
1681 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1682 
1683 static int memblock_debug_show(struct seq_file *m, void *private)
1684 {
1685 	struct memblock_type *type = m->private;
1686 	struct memblock_region *reg;
1687 	int i;
1688 
1689 	for (i = 0; i < type->cnt; i++) {
1690 		reg = &type->regions[i];
1691 		seq_printf(m, "%4d: ", i);
1692 		if (sizeof(phys_addr_t) == 4)
1693 			seq_printf(m, "0x%08lx..0x%08lx\n",
1694 				   (unsigned long)reg->base,
1695 				   (unsigned long)(reg->base + reg->size - 1));
1696 		else
1697 			seq_printf(m, "0x%016llx..0x%016llx\n",
1698 				   (unsigned long long)reg->base,
1699 				   (unsigned long long)(reg->base + reg->size - 1));
1700 
1701 	}
1702 	return 0;
1703 }
1704 
1705 static int memblock_debug_open(struct inode *inode, struct file *file)
1706 {
1707 	return single_open(file, memblock_debug_show, inode->i_private);
1708 }
1709 
1710 static const struct file_operations memblock_debug_fops = {
1711 	.open = memblock_debug_open,
1712 	.read = seq_read,
1713 	.llseek = seq_lseek,
1714 	.release = single_release,
1715 };
1716 
1717 static int __init memblock_init_debugfs(void)
1718 {
1719 	struct dentry *root = debugfs_create_dir("memblock", NULL);
1720 	if (!root)
1721 		return -ENXIO;
1722 	debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1723 	debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1724 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1725 	debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1726 #endif
1727 
1728 	return 0;
1729 }
1730 __initcall(memblock_init_debugfs);
1731 
1732 #endif /* CONFIG_DEBUG_FS */
1733