xref: /openbmc/linux/mm/memblock.c (revision 976fa9a3)
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 int memblock_can_resize __initdata_memblock;
58 static int memblock_memory_in_slab __initdata_memblock = 0;
59 static int memblock_reserved_in_slab __initdata_memblock = 0;
60 
61 /* inline so we don't get a warning when pr_debug is compiled out */
62 static __init_memblock const char *
63 memblock_type_name(struct memblock_type *type)
64 {
65 	if (type == &memblock.memory)
66 		return "memory";
67 	else if (type == &memblock.reserved)
68 		return "reserved";
69 	else
70 		return "unknown";
71 }
72 
73 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
74 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
75 {
76 	return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
77 }
78 
79 /*
80  * Address comparison utilities
81  */
82 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
83 				       phys_addr_t base2, phys_addr_t size2)
84 {
85 	return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
86 }
87 
88 static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
89 					phys_addr_t base, phys_addr_t size)
90 {
91 	unsigned long i;
92 
93 	for (i = 0; i < type->cnt; i++) {
94 		phys_addr_t rgnbase = type->regions[i].base;
95 		phys_addr_t rgnsize = type->regions[i].size;
96 		if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
97 			break;
98 	}
99 
100 	return (i < type->cnt) ? i : -1;
101 }
102 
103 /*
104  * __memblock_find_range_bottom_up - find free area utility in bottom-up
105  * @start: start of candidate range
106  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
107  * @size: size of free area to find
108  * @align: alignment of free area to find
109  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
110  *
111  * Utility called from memblock_find_in_range_node(), find free area bottom-up.
112  *
113  * RETURNS:
114  * Found address on success, 0 on failure.
115  */
116 static phys_addr_t __init_memblock
117 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
118 				phys_addr_t size, phys_addr_t align, int nid)
119 {
120 	phys_addr_t this_start, this_end, cand;
121 	u64 i;
122 
123 	for_each_free_mem_range(i, nid, &this_start, &this_end, NULL) {
124 		this_start = clamp(this_start, start, end);
125 		this_end = clamp(this_end, start, end);
126 
127 		cand = round_up(this_start, align);
128 		if (cand < this_end && this_end - cand >= size)
129 			return cand;
130 	}
131 
132 	return 0;
133 }
134 
135 /**
136  * __memblock_find_range_top_down - find free area utility, in top-down
137  * @start: start of candidate range
138  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
139  * @size: size of free area to find
140  * @align: alignment of free area to find
141  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
142  *
143  * Utility called from memblock_find_in_range_node(), find free area top-down.
144  *
145  * RETURNS:
146  * Found address on success, 0 on failure.
147  */
148 static phys_addr_t __init_memblock
149 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
150 			       phys_addr_t size, phys_addr_t align, int nid)
151 {
152 	phys_addr_t this_start, this_end, cand;
153 	u64 i;
154 
155 	for_each_free_mem_range_reverse(i, nid, &this_start, &this_end, NULL) {
156 		this_start = clamp(this_start, start, end);
157 		this_end = clamp(this_end, start, end);
158 
159 		if (this_end < size)
160 			continue;
161 
162 		cand = round_down(this_end - size, align);
163 		if (cand >= this_start)
164 			return cand;
165 	}
166 
167 	return 0;
168 }
169 
170 /**
171  * memblock_find_in_range_node - find free area in given range and node
172  * @size: size of free area to find
173  * @align: alignment of free area to find
174  * @start: start of candidate range
175  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
176  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
177  *
178  * Find @size free area aligned to @align in the specified range and node.
179  *
180  * When allocation direction is bottom-up, the @start should be greater
181  * than the end of the kernel image. Otherwise, it will be trimmed. The
182  * reason is that we want the bottom-up allocation just near the kernel
183  * image so it is highly likely that the allocated memory and the kernel
184  * will reside in the same node.
185  *
186  * If bottom-up allocation failed, will try to allocate memory top-down.
187  *
188  * RETURNS:
189  * Found address on success, 0 on failure.
190  */
191 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
192 					phys_addr_t align, phys_addr_t start,
193 					phys_addr_t end, int nid)
194 {
195 	phys_addr_t kernel_end, ret;
196 
197 	/* pump up @end */
198 	if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
199 		end = memblock.current_limit;
200 
201 	/* avoid allocating the first page */
202 	start = max_t(phys_addr_t, start, PAGE_SIZE);
203 	end = max(start, end);
204 	kernel_end = __pa_symbol(_end);
205 
206 	/*
207 	 * try bottom-up allocation only when bottom-up mode
208 	 * is set and @end is above the kernel image.
209 	 */
210 	if (memblock_bottom_up() && end > kernel_end) {
211 		phys_addr_t bottom_up_start;
212 
213 		/* make sure we will allocate above the kernel */
214 		bottom_up_start = max(start, kernel_end);
215 
216 		/* ok, try bottom-up allocation first */
217 		ret = __memblock_find_range_bottom_up(bottom_up_start, end,
218 						      size, align, nid);
219 		if (ret)
220 			return ret;
221 
222 		/*
223 		 * we always limit bottom-up allocation above the kernel,
224 		 * but top-down allocation doesn't have the limit, so
225 		 * retrying top-down allocation may succeed when bottom-up
226 		 * allocation failed.
227 		 *
228 		 * bottom-up allocation is expected to be fail very rarely,
229 		 * so we use WARN_ONCE() here to see the stack trace if
230 		 * fail happens.
231 		 */
232 		WARN_ONCE(1, "memblock: bottom-up allocation failed, "
233 			     "memory hotunplug may be affected\n");
234 	}
235 
236 	return __memblock_find_range_top_down(start, end, size, align, nid);
237 }
238 
239 /**
240  * memblock_find_in_range - find free area in given range
241  * @start: start of candidate range
242  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
243  * @size: size of free area to find
244  * @align: alignment of free area to find
245  *
246  * Find @size free area aligned to @align in the specified range.
247  *
248  * RETURNS:
249  * Found address on success, 0 on failure.
250  */
251 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
252 					phys_addr_t end, phys_addr_t size,
253 					phys_addr_t align)
254 {
255 	return memblock_find_in_range_node(size, align, start, end,
256 					    NUMA_NO_NODE);
257 }
258 
259 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
260 {
261 	type->total_size -= type->regions[r].size;
262 	memmove(&type->regions[r], &type->regions[r + 1],
263 		(type->cnt - (r + 1)) * sizeof(type->regions[r]));
264 	type->cnt--;
265 
266 	/* Special case for empty arrays */
267 	if (type->cnt == 0) {
268 		WARN_ON(type->total_size != 0);
269 		type->cnt = 1;
270 		type->regions[0].base = 0;
271 		type->regions[0].size = 0;
272 		type->regions[0].flags = 0;
273 		memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
274 	}
275 }
276 
277 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
278 
279 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
280 					phys_addr_t *addr)
281 {
282 	if (memblock.reserved.regions == memblock_reserved_init_regions)
283 		return 0;
284 
285 	*addr = __pa(memblock.reserved.regions);
286 
287 	return PAGE_ALIGN(sizeof(struct memblock_region) *
288 			  memblock.reserved.max);
289 }
290 
291 phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
292 					phys_addr_t *addr)
293 {
294 	if (memblock.memory.regions == memblock_memory_init_regions)
295 		return 0;
296 
297 	*addr = __pa(memblock.memory.regions);
298 
299 	return PAGE_ALIGN(sizeof(struct memblock_region) *
300 			  memblock.memory.max);
301 }
302 
303 #endif
304 
305 /**
306  * memblock_double_array - double the size of the memblock regions array
307  * @type: memblock type of the regions array being doubled
308  * @new_area_start: starting address of memory range to avoid overlap with
309  * @new_area_size: size of memory range to avoid overlap with
310  *
311  * Double the size of the @type regions array. If memblock is being used to
312  * allocate memory for a new reserved regions array and there is a previously
313  * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
314  * waiting to be reserved, ensure the memory used by the new array does
315  * not overlap.
316  *
317  * RETURNS:
318  * 0 on success, -1 on failure.
319  */
320 static int __init_memblock memblock_double_array(struct memblock_type *type,
321 						phys_addr_t new_area_start,
322 						phys_addr_t new_area_size)
323 {
324 	struct memblock_region *new_array, *old_array;
325 	phys_addr_t old_alloc_size, new_alloc_size;
326 	phys_addr_t old_size, new_size, addr;
327 	int use_slab = slab_is_available();
328 	int *in_slab;
329 
330 	/* We don't allow resizing until we know about the reserved regions
331 	 * of memory that aren't suitable for allocation
332 	 */
333 	if (!memblock_can_resize)
334 		return -1;
335 
336 	/* Calculate new doubled size */
337 	old_size = type->max * sizeof(struct memblock_region);
338 	new_size = old_size << 1;
339 	/*
340 	 * We need to allocated new one align to PAGE_SIZE,
341 	 *   so we can free them completely later.
342 	 */
343 	old_alloc_size = PAGE_ALIGN(old_size);
344 	new_alloc_size = PAGE_ALIGN(new_size);
345 
346 	/* Retrieve the slab flag */
347 	if (type == &memblock.memory)
348 		in_slab = &memblock_memory_in_slab;
349 	else
350 		in_slab = &memblock_reserved_in_slab;
351 
352 	/* Try to find some space for it.
353 	 *
354 	 * WARNING: We assume that either slab_is_available() and we use it or
355 	 * we use MEMBLOCK for allocations. That means that this is unsafe to
356 	 * use when bootmem is currently active (unless bootmem itself is
357 	 * implemented on top of MEMBLOCK which isn't the case yet)
358 	 *
359 	 * This should however not be an issue for now, as we currently only
360 	 * call into MEMBLOCK while it's still active, or much later when slab
361 	 * is active for memory hotplug operations
362 	 */
363 	if (use_slab) {
364 		new_array = kmalloc(new_size, GFP_KERNEL);
365 		addr = new_array ? __pa(new_array) : 0;
366 	} else {
367 		/* only exclude range when trying to double reserved.regions */
368 		if (type != &memblock.reserved)
369 			new_area_start = new_area_size = 0;
370 
371 		addr = memblock_find_in_range(new_area_start + new_area_size,
372 						memblock.current_limit,
373 						new_alloc_size, PAGE_SIZE);
374 		if (!addr && new_area_size)
375 			addr = memblock_find_in_range(0,
376 				min(new_area_start, memblock.current_limit),
377 				new_alloc_size, PAGE_SIZE);
378 
379 		new_array = addr ? __va(addr) : NULL;
380 	}
381 	if (!addr) {
382 		pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
383 		       memblock_type_name(type), type->max, type->max * 2);
384 		return -1;
385 	}
386 
387 	memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
388 			memblock_type_name(type), type->max * 2, (u64)addr,
389 			(u64)addr + new_size - 1);
390 
391 	/*
392 	 * Found space, we now need to move the array over before we add the
393 	 * reserved region since it may be our reserved array itself that is
394 	 * full.
395 	 */
396 	memcpy(new_array, type->regions, old_size);
397 	memset(new_array + type->max, 0, old_size);
398 	old_array = type->regions;
399 	type->regions = new_array;
400 	type->max <<= 1;
401 
402 	/* Free old array. We needn't free it if the array is the static one */
403 	if (*in_slab)
404 		kfree(old_array);
405 	else if (old_array != memblock_memory_init_regions &&
406 		 old_array != memblock_reserved_init_regions)
407 		memblock_free(__pa(old_array), old_alloc_size);
408 
409 	/*
410 	 * Reserve the new array if that comes from the memblock.  Otherwise, we
411 	 * needn't do it
412 	 */
413 	if (!use_slab)
414 		BUG_ON(memblock_reserve(addr, new_alloc_size));
415 
416 	/* Update slab flag */
417 	*in_slab = use_slab;
418 
419 	return 0;
420 }
421 
422 /**
423  * memblock_merge_regions - merge neighboring compatible regions
424  * @type: memblock type to scan
425  *
426  * Scan @type and merge neighboring compatible regions.
427  */
428 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
429 {
430 	int i = 0;
431 
432 	/* cnt never goes below 1 */
433 	while (i < type->cnt - 1) {
434 		struct memblock_region *this = &type->regions[i];
435 		struct memblock_region *next = &type->regions[i + 1];
436 
437 		if (this->base + this->size != next->base ||
438 		    memblock_get_region_node(this) !=
439 		    memblock_get_region_node(next) ||
440 		    this->flags != next->flags) {
441 			BUG_ON(this->base + this->size > next->base);
442 			i++;
443 			continue;
444 		}
445 
446 		this->size += next->size;
447 		/* move forward from next + 1, index of which is i + 2 */
448 		memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
449 		type->cnt--;
450 	}
451 }
452 
453 /**
454  * memblock_insert_region - insert new memblock region
455  * @type:	memblock type to insert into
456  * @idx:	index for the insertion point
457  * @base:	base address of the new region
458  * @size:	size of the new region
459  * @nid:	node id of the new region
460  * @flags:	flags of the new region
461  *
462  * Insert new memblock region [@base,@base+@size) into @type at @idx.
463  * @type must already have extra room to accomodate the new region.
464  */
465 static void __init_memblock memblock_insert_region(struct memblock_type *type,
466 						   int idx, phys_addr_t base,
467 						   phys_addr_t size,
468 						   int nid, unsigned long flags)
469 {
470 	struct memblock_region *rgn = &type->regions[idx];
471 
472 	BUG_ON(type->cnt >= type->max);
473 	memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
474 	rgn->base = base;
475 	rgn->size = size;
476 	rgn->flags = flags;
477 	memblock_set_region_node(rgn, nid);
478 	type->cnt++;
479 	type->total_size += size;
480 }
481 
482 /**
483  * memblock_add_range - add new memblock region
484  * @type: memblock type to add new region into
485  * @base: base address of the new region
486  * @size: size of the new region
487  * @nid: nid of the new region
488  * @flags: flags of the new region
489  *
490  * Add new memblock region [@base,@base+@size) into @type.  The new region
491  * is allowed to overlap with existing ones - overlaps don't affect already
492  * existing regions.  @type is guaranteed to be minimal (all neighbouring
493  * compatible regions are merged) after the addition.
494  *
495  * RETURNS:
496  * 0 on success, -errno on failure.
497  */
498 int __init_memblock memblock_add_range(struct memblock_type *type,
499 				phys_addr_t base, phys_addr_t size,
500 				int nid, unsigned long flags)
501 {
502 	bool insert = false;
503 	phys_addr_t obase = base;
504 	phys_addr_t end = base + memblock_cap_size(base, &size);
505 	int i, nr_new;
506 
507 	if (!size)
508 		return 0;
509 
510 	/* special case for empty array */
511 	if (type->regions[0].size == 0) {
512 		WARN_ON(type->cnt != 1 || type->total_size);
513 		type->regions[0].base = base;
514 		type->regions[0].size = size;
515 		type->regions[0].flags = flags;
516 		memblock_set_region_node(&type->regions[0], nid);
517 		type->total_size = size;
518 		return 0;
519 	}
520 repeat:
521 	/*
522 	 * The following is executed twice.  Once with %false @insert and
523 	 * then with %true.  The first counts the number of regions needed
524 	 * to accomodate the new area.  The second actually inserts them.
525 	 */
526 	base = obase;
527 	nr_new = 0;
528 
529 	for (i = 0; i < type->cnt; i++) {
530 		struct memblock_region *rgn = &type->regions[i];
531 		phys_addr_t rbase = rgn->base;
532 		phys_addr_t rend = rbase + rgn->size;
533 
534 		if (rbase >= end)
535 			break;
536 		if (rend <= base)
537 			continue;
538 		/*
539 		 * @rgn overlaps.  If it separates the lower part of new
540 		 * area, insert that portion.
541 		 */
542 		if (rbase > base) {
543 			nr_new++;
544 			if (insert)
545 				memblock_insert_region(type, i++, base,
546 						       rbase - base, nid,
547 						       flags);
548 		}
549 		/* area below @rend is dealt with, forget about it */
550 		base = min(rend, end);
551 	}
552 
553 	/* insert the remaining portion */
554 	if (base < end) {
555 		nr_new++;
556 		if (insert)
557 			memblock_insert_region(type, i, base, end - base,
558 					       nid, flags);
559 	}
560 
561 	/*
562 	 * If this was the first round, resize array and repeat for actual
563 	 * insertions; otherwise, merge and return.
564 	 */
565 	if (!insert) {
566 		while (type->cnt + nr_new > type->max)
567 			if (memblock_double_array(type, obase, size) < 0)
568 				return -ENOMEM;
569 		insert = true;
570 		goto repeat;
571 	} else {
572 		memblock_merge_regions(type);
573 		return 0;
574 	}
575 }
576 
577 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
578 				       int nid)
579 {
580 	return memblock_add_range(&memblock.memory, base, size, nid, 0);
581 }
582 
583 static int __init_memblock memblock_add_region(phys_addr_t base,
584 						phys_addr_t size,
585 						int nid,
586 						unsigned long flags)
587 {
588 	struct memblock_type *_rgn = &memblock.memory;
589 
590 	memblock_dbg("memblock_add: [%#016llx-%#016llx] flags %#02lx %pF\n",
591 		     (unsigned long long)base,
592 		     (unsigned long long)base + size - 1,
593 		     flags, (void *)_RET_IP_);
594 
595 	return memblock_add_range(_rgn, base, size, nid, flags);
596 }
597 
598 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
599 {
600 	return memblock_add_region(base, size, MAX_NUMNODES, 0);
601 }
602 
603 /**
604  * memblock_isolate_range - isolate given range into disjoint memblocks
605  * @type: memblock type to isolate range for
606  * @base: base of range to isolate
607  * @size: size of range to isolate
608  * @start_rgn: out parameter for the start of isolated region
609  * @end_rgn: out parameter for the end of isolated region
610  *
611  * Walk @type and ensure that regions don't cross the boundaries defined by
612  * [@base,@base+@size).  Crossing regions are split at the boundaries,
613  * which may create at most two more regions.  The index of the first
614  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
615  *
616  * RETURNS:
617  * 0 on success, -errno on failure.
618  */
619 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
620 					phys_addr_t base, phys_addr_t size,
621 					int *start_rgn, int *end_rgn)
622 {
623 	phys_addr_t end = base + memblock_cap_size(base, &size);
624 	int i;
625 
626 	*start_rgn = *end_rgn = 0;
627 
628 	if (!size)
629 		return 0;
630 
631 	/* we'll create at most two more regions */
632 	while (type->cnt + 2 > type->max)
633 		if (memblock_double_array(type, base, size) < 0)
634 			return -ENOMEM;
635 
636 	for (i = 0; i < type->cnt; i++) {
637 		struct memblock_region *rgn = &type->regions[i];
638 		phys_addr_t rbase = rgn->base;
639 		phys_addr_t rend = rbase + rgn->size;
640 
641 		if (rbase >= end)
642 			break;
643 		if (rend <= base)
644 			continue;
645 
646 		if (rbase < base) {
647 			/*
648 			 * @rgn intersects from below.  Split and continue
649 			 * to process the next region - the new top half.
650 			 */
651 			rgn->base = base;
652 			rgn->size -= base - rbase;
653 			type->total_size -= base - rbase;
654 			memblock_insert_region(type, i, rbase, base - rbase,
655 					       memblock_get_region_node(rgn),
656 					       rgn->flags);
657 		} else if (rend > end) {
658 			/*
659 			 * @rgn intersects from above.  Split and redo the
660 			 * current region - the new bottom half.
661 			 */
662 			rgn->base = end;
663 			rgn->size -= end - rbase;
664 			type->total_size -= end - rbase;
665 			memblock_insert_region(type, i--, rbase, end - rbase,
666 					       memblock_get_region_node(rgn),
667 					       rgn->flags);
668 		} else {
669 			/* @rgn is fully contained, record it */
670 			if (!*end_rgn)
671 				*start_rgn = i;
672 			*end_rgn = i + 1;
673 		}
674 	}
675 
676 	return 0;
677 }
678 
679 int __init_memblock memblock_remove_range(struct memblock_type *type,
680 					  phys_addr_t base, phys_addr_t size)
681 {
682 	int start_rgn, end_rgn;
683 	int i, ret;
684 
685 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
686 	if (ret)
687 		return ret;
688 
689 	for (i = end_rgn - 1; i >= start_rgn; i--)
690 		memblock_remove_region(type, i);
691 	return 0;
692 }
693 
694 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
695 {
696 	return memblock_remove_range(&memblock.memory, base, size);
697 }
698 
699 
700 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
701 {
702 	memblock_dbg("   memblock_free: [%#016llx-%#016llx] %pF\n",
703 		     (unsigned long long)base,
704 		     (unsigned long long)base + size - 1,
705 		     (void *)_RET_IP_);
706 
707 	kmemleak_free_part(__va(base), size);
708 	return memblock_remove_range(&memblock.reserved, base, size);
709 }
710 
711 static int __init_memblock memblock_reserve_region(phys_addr_t base,
712 						   phys_addr_t size,
713 						   int nid,
714 						   unsigned long flags)
715 {
716 	struct memblock_type *type = &memblock.reserved;
717 
718 	memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
719 		     (unsigned long long)base,
720 		     (unsigned long long)base + size - 1,
721 		     flags, (void *)_RET_IP_);
722 
723 	return memblock_add_range(type, base, size, nid, flags);
724 }
725 
726 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
727 {
728 	return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
729 }
730 
731 /**
732  *
733  * This function isolates region [@base, @base + @size), and sets/clears flag
734  *
735  * Return 0 on succees, -errno on failure.
736  */
737 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
738 				phys_addr_t size, int set, int flag)
739 {
740 	struct memblock_type *type = &memblock.memory;
741 	int i, ret, start_rgn, end_rgn;
742 
743 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
744 	if (ret)
745 		return ret;
746 
747 	for (i = start_rgn; i < end_rgn; i++)
748 		if (set)
749 			memblock_set_region_flags(&type->regions[i], flag);
750 		else
751 			memblock_clear_region_flags(&type->regions[i], flag);
752 
753 	memblock_merge_regions(type);
754 	return 0;
755 }
756 
757 /**
758  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
759  * @base: the base phys addr of the region
760  * @size: the size of the region
761  *
762  * Return 0 on succees, -errno on failure.
763  */
764 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
765 {
766 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
767 }
768 
769 /**
770  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
771  * @base: the base phys addr of the region
772  * @size: the size of the region
773  *
774  * Return 0 on succees, -errno on failure.
775  */
776 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
777 {
778 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
779 }
780 
781 /**
782  * __next__mem_range - next function for for_each_free_mem_range() etc.
783  * @idx: pointer to u64 loop variable
784  * @nid: node selector, %NUMA_NO_NODE for all nodes
785  * @type_a: pointer to memblock_type from where the range is taken
786  * @type_b: pointer to memblock_type which excludes memory from being taken
787  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
788  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
789  * @out_nid: ptr to int for nid of the range, can be %NULL
790  *
791  * Find the first area from *@idx which matches @nid, fill the out
792  * parameters, and update *@idx for the next iteration.  The lower 32bit of
793  * *@idx contains index into type_a and the upper 32bit indexes the
794  * areas before each region in type_b.	For example, if type_b regions
795  * look like the following,
796  *
797  *	0:[0-16), 1:[32-48), 2:[128-130)
798  *
799  * The upper 32bit indexes the following regions.
800  *
801  *	0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
802  *
803  * As both region arrays are sorted, the function advances the two indices
804  * in lockstep and returns each intersection.
805  */
806 void __init_memblock __next_mem_range(u64 *idx, int nid,
807 				      struct memblock_type *type_a,
808 				      struct memblock_type *type_b,
809 				      phys_addr_t *out_start,
810 				      phys_addr_t *out_end, int *out_nid)
811 {
812 	int idx_a = *idx & 0xffffffff;
813 	int idx_b = *idx >> 32;
814 
815 	if (WARN_ONCE(nid == MAX_NUMNODES,
816 	"Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
817 		nid = NUMA_NO_NODE;
818 
819 	for (; idx_a < type_a->cnt; idx_a++) {
820 		struct memblock_region *m = &type_a->regions[idx_a];
821 
822 		phys_addr_t m_start = m->base;
823 		phys_addr_t m_end = m->base + m->size;
824 		int	    m_nid = memblock_get_region_node(m);
825 
826 		/* only memory regions are associated with nodes, check it */
827 		if (nid != NUMA_NO_NODE && nid != m_nid)
828 			continue;
829 
830 		/* skip hotpluggable memory regions if needed */
831 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
832 			continue;
833 
834 		if (!type_b) {
835 			if (out_start)
836 				*out_start = m_start;
837 			if (out_end)
838 				*out_end = m_end;
839 			if (out_nid)
840 				*out_nid = m_nid;
841 			idx_a++;
842 			*idx = (u32)idx_a | (u64)idx_b << 32;
843 			return;
844 		}
845 
846 		/* scan areas before each reservation */
847 		for (; idx_b < type_b->cnt + 1; idx_b++) {
848 			struct memblock_region *r;
849 			phys_addr_t r_start;
850 			phys_addr_t r_end;
851 
852 			r = &type_b->regions[idx_b];
853 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
854 			r_end = idx_b < type_b->cnt ?
855 				r->base : ULLONG_MAX;
856 
857 			/*
858 			 * if idx_b advanced past idx_a,
859 			 * break out to advance idx_a
860 			 */
861 			if (r_start >= m_end)
862 				break;
863 			/* if the two regions intersect, we're done */
864 			if (m_start < r_end) {
865 				if (out_start)
866 					*out_start =
867 						max(m_start, r_start);
868 				if (out_end)
869 					*out_end = min(m_end, r_end);
870 				if (out_nid)
871 					*out_nid = m_nid;
872 				/*
873 				 * The region which ends first is
874 				 * advanced for the next iteration.
875 				 */
876 				if (m_end <= r_end)
877 					idx_a++;
878 				else
879 					idx_b++;
880 				*idx = (u32)idx_a | (u64)idx_b << 32;
881 				return;
882 			}
883 		}
884 	}
885 
886 	/* signal end of iteration */
887 	*idx = ULLONG_MAX;
888 }
889 
890 /**
891  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
892  *
893  * Finds the next range from type_a which is not marked as unsuitable
894  * in type_b.
895  *
896  * @idx: pointer to u64 loop variable
897  * @nid: nid: node selector, %NUMA_NO_NODE for all nodes
898  * @type_a: pointer to memblock_type from where the range is taken
899  * @type_b: pointer to memblock_type which excludes memory from being taken
900  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
901  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
902  * @out_nid: ptr to int for nid of the range, can be %NULL
903  *
904  * Reverse of __next_mem_range().
905  */
906 void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
907 					  struct memblock_type *type_a,
908 					  struct memblock_type *type_b,
909 					  phys_addr_t *out_start,
910 					  phys_addr_t *out_end, int *out_nid)
911 {
912 	int idx_a = *idx & 0xffffffff;
913 	int idx_b = *idx >> 32;
914 
915 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
916 		nid = NUMA_NO_NODE;
917 
918 	if (*idx == (u64)ULLONG_MAX) {
919 		idx_a = type_a->cnt - 1;
920 		idx_b = type_b->cnt;
921 	}
922 
923 	for (; idx_a >= 0; idx_a--) {
924 		struct memblock_region *m = &type_a->regions[idx_a];
925 
926 		phys_addr_t m_start = m->base;
927 		phys_addr_t m_end = m->base + m->size;
928 		int m_nid = memblock_get_region_node(m);
929 
930 		/* only memory regions are associated with nodes, check it */
931 		if (nid != NUMA_NO_NODE && nid != m_nid)
932 			continue;
933 
934 		/* skip hotpluggable memory regions if needed */
935 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
936 			continue;
937 
938 		if (!type_b) {
939 			if (out_start)
940 				*out_start = m_start;
941 			if (out_end)
942 				*out_end = m_end;
943 			if (out_nid)
944 				*out_nid = m_nid;
945 			idx_a++;
946 			*idx = (u32)idx_a | (u64)idx_b << 32;
947 			return;
948 		}
949 
950 		/* scan areas before each reservation */
951 		for (; idx_b >= 0; idx_b--) {
952 			struct memblock_region *r;
953 			phys_addr_t r_start;
954 			phys_addr_t r_end;
955 
956 			r = &type_b->regions[idx_b];
957 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
958 			r_end = idx_b < type_b->cnt ?
959 				r->base : ULLONG_MAX;
960 			/*
961 			 * if idx_b advanced past idx_a,
962 			 * break out to advance idx_a
963 			 */
964 
965 			if (r_end <= m_start)
966 				break;
967 			/* if the two regions intersect, we're done */
968 			if (m_end > r_start) {
969 				if (out_start)
970 					*out_start = max(m_start, r_start);
971 				if (out_end)
972 					*out_end = min(m_end, r_end);
973 				if (out_nid)
974 					*out_nid = m_nid;
975 				if (m_start >= r_start)
976 					idx_a--;
977 				else
978 					idx_b--;
979 				*idx = (u32)idx_a | (u64)idx_b << 32;
980 				return;
981 			}
982 		}
983 	}
984 	/* signal end of iteration */
985 	*idx = ULLONG_MAX;
986 }
987 
988 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
989 /*
990  * Common iterator interface used to define for_each_mem_range().
991  */
992 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
993 				unsigned long *out_start_pfn,
994 				unsigned long *out_end_pfn, int *out_nid)
995 {
996 	struct memblock_type *type = &memblock.memory;
997 	struct memblock_region *r;
998 
999 	while (++*idx < type->cnt) {
1000 		r = &type->regions[*idx];
1001 
1002 		if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1003 			continue;
1004 		if (nid == MAX_NUMNODES || nid == r->nid)
1005 			break;
1006 	}
1007 	if (*idx >= type->cnt) {
1008 		*idx = -1;
1009 		return;
1010 	}
1011 
1012 	if (out_start_pfn)
1013 		*out_start_pfn = PFN_UP(r->base);
1014 	if (out_end_pfn)
1015 		*out_end_pfn = PFN_DOWN(r->base + r->size);
1016 	if (out_nid)
1017 		*out_nid = r->nid;
1018 }
1019 
1020 /**
1021  * memblock_set_node - set node ID on memblock regions
1022  * @base: base of area to set node ID for
1023  * @size: size of area to set node ID for
1024  * @type: memblock type to set node ID for
1025  * @nid: node ID to set
1026  *
1027  * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1028  * Regions which cross the area boundaries are split as necessary.
1029  *
1030  * RETURNS:
1031  * 0 on success, -errno on failure.
1032  */
1033 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1034 				      struct memblock_type *type, int nid)
1035 {
1036 	int start_rgn, end_rgn;
1037 	int i, ret;
1038 
1039 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1040 	if (ret)
1041 		return ret;
1042 
1043 	for (i = start_rgn; i < end_rgn; i++)
1044 		memblock_set_region_node(&type->regions[i], nid);
1045 
1046 	memblock_merge_regions(type);
1047 	return 0;
1048 }
1049 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1050 
1051 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1052 					phys_addr_t align, phys_addr_t start,
1053 					phys_addr_t end, int nid)
1054 {
1055 	phys_addr_t found;
1056 
1057 	if (!align)
1058 		align = SMP_CACHE_BYTES;
1059 
1060 	found = memblock_find_in_range_node(size, align, start, end, nid);
1061 	if (found && !memblock_reserve(found, size)) {
1062 		/*
1063 		 * The min_count is set to 0 so that memblock allocations are
1064 		 * never reported as leaks.
1065 		 */
1066 		kmemleak_alloc(__va(found), size, 0, 0);
1067 		return found;
1068 	}
1069 	return 0;
1070 }
1071 
1072 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1073 					phys_addr_t start, phys_addr_t end)
1074 {
1075 	return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE);
1076 }
1077 
1078 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1079 					phys_addr_t align, phys_addr_t max_addr,
1080 					int nid)
1081 {
1082 	return memblock_alloc_range_nid(size, align, 0, max_addr, nid);
1083 }
1084 
1085 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1086 {
1087 	return memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
1088 }
1089 
1090 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1091 {
1092 	return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE);
1093 }
1094 
1095 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1096 {
1097 	phys_addr_t alloc;
1098 
1099 	alloc = __memblock_alloc_base(size, align, max_addr);
1100 
1101 	if (alloc == 0)
1102 		panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1103 		      (unsigned long long) size, (unsigned long long) max_addr);
1104 
1105 	return alloc;
1106 }
1107 
1108 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1109 {
1110 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1111 }
1112 
1113 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1114 {
1115 	phys_addr_t res = memblock_alloc_nid(size, align, nid);
1116 
1117 	if (res)
1118 		return res;
1119 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1120 }
1121 
1122 /**
1123  * memblock_virt_alloc_internal - allocate boot memory block
1124  * @size: size of memory block to be allocated in bytes
1125  * @align: alignment of the region and block's size
1126  * @min_addr: the lower bound of the memory region to allocate (phys address)
1127  * @max_addr: the upper bound of the memory region to allocate (phys address)
1128  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1129  *
1130  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1131  * will fall back to memory below @min_addr. Also, allocation may fall back
1132  * to any node in the system if the specified node can not
1133  * hold the requested memory.
1134  *
1135  * The allocation is performed from memory region limited by
1136  * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1137  *
1138  * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1139  *
1140  * The phys address of allocated boot memory block is converted to virtual and
1141  * allocated memory is reset to 0.
1142  *
1143  * In addition, function sets the min_count to 0 using kmemleak_alloc for
1144  * allocated boot memory block, so that it is never reported as leaks.
1145  *
1146  * RETURNS:
1147  * Virtual address of allocated memory block on success, NULL on failure.
1148  */
1149 static void * __init memblock_virt_alloc_internal(
1150 				phys_addr_t size, phys_addr_t align,
1151 				phys_addr_t min_addr, phys_addr_t max_addr,
1152 				int nid)
1153 {
1154 	phys_addr_t alloc;
1155 	void *ptr;
1156 
1157 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1158 		nid = NUMA_NO_NODE;
1159 
1160 	/*
1161 	 * Detect any accidental use of these APIs after slab is ready, as at
1162 	 * this moment memblock may be deinitialized already and its
1163 	 * internal data may be destroyed (after execution of free_all_bootmem)
1164 	 */
1165 	if (WARN_ON_ONCE(slab_is_available()))
1166 		return kzalloc_node(size, GFP_NOWAIT, nid);
1167 
1168 	if (!align)
1169 		align = SMP_CACHE_BYTES;
1170 
1171 	if (max_addr > memblock.current_limit)
1172 		max_addr = memblock.current_limit;
1173 
1174 again:
1175 	alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1176 					    nid);
1177 	if (alloc)
1178 		goto done;
1179 
1180 	if (nid != NUMA_NO_NODE) {
1181 		alloc = memblock_find_in_range_node(size, align, min_addr,
1182 						    max_addr,  NUMA_NO_NODE);
1183 		if (alloc)
1184 			goto done;
1185 	}
1186 
1187 	if (min_addr) {
1188 		min_addr = 0;
1189 		goto again;
1190 	} else {
1191 		goto error;
1192 	}
1193 
1194 done:
1195 	memblock_reserve(alloc, size);
1196 	ptr = phys_to_virt(alloc);
1197 	memset(ptr, 0, size);
1198 
1199 	/*
1200 	 * The min_count is set to 0 so that bootmem allocated blocks
1201 	 * are never reported as leaks. This is because many of these blocks
1202 	 * are only referred via the physical address which is not
1203 	 * looked up by kmemleak.
1204 	 */
1205 	kmemleak_alloc(ptr, size, 0, 0);
1206 
1207 	return ptr;
1208 
1209 error:
1210 	return NULL;
1211 }
1212 
1213 /**
1214  * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1215  * @size: size of memory block to be allocated in bytes
1216  * @align: alignment of the region and block's size
1217  * @min_addr: the lower bound of the memory region from where the allocation
1218  *	  is preferred (phys address)
1219  * @max_addr: the upper bound of the memory region from where the allocation
1220  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1221  *	      allocate only from memory limited by memblock.current_limit value
1222  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1223  *
1224  * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1225  * additional debug information (including caller info), if enabled.
1226  *
1227  * RETURNS:
1228  * Virtual address of allocated memory block on success, NULL on failure.
1229  */
1230 void * __init memblock_virt_alloc_try_nid_nopanic(
1231 				phys_addr_t size, phys_addr_t align,
1232 				phys_addr_t min_addr, phys_addr_t max_addr,
1233 				int nid)
1234 {
1235 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1236 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1237 		     (u64)max_addr, (void *)_RET_IP_);
1238 	return memblock_virt_alloc_internal(size, align, min_addr,
1239 					     max_addr, nid);
1240 }
1241 
1242 /**
1243  * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1244  * @size: size of memory block to be allocated in bytes
1245  * @align: alignment of the region and block's size
1246  * @min_addr: the lower bound of the memory region from where the allocation
1247  *	  is preferred (phys address)
1248  * @max_addr: the upper bound of the memory region from where the allocation
1249  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1250  *	      allocate only from memory limited by memblock.current_limit value
1251  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1252  *
1253  * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1254  * which provides debug information (including caller info), if enabled,
1255  * and panics if the request can not be satisfied.
1256  *
1257  * RETURNS:
1258  * Virtual address of allocated memory block on success, NULL on failure.
1259  */
1260 void * __init memblock_virt_alloc_try_nid(
1261 			phys_addr_t size, phys_addr_t align,
1262 			phys_addr_t min_addr, phys_addr_t max_addr,
1263 			int nid)
1264 {
1265 	void *ptr;
1266 
1267 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1268 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1269 		     (u64)max_addr, (void *)_RET_IP_);
1270 	ptr = memblock_virt_alloc_internal(size, align,
1271 					   min_addr, max_addr, nid);
1272 	if (ptr)
1273 		return ptr;
1274 
1275 	panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1276 	      __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1277 	      (u64)max_addr);
1278 	return NULL;
1279 }
1280 
1281 /**
1282  * __memblock_free_early - free boot memory block
1283  * @base: phys starting address of the  boot memory block
1284  * @size: size of the boot memory block in bytes
1285  *
1286  * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1287  * The freeing memory will not be released to the buddy allocator.
1288  */
1289 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1290 {
1291 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1292 		     __func__, (u64)base, (u64)base + size - 1,
1293 		     (void *)_RET_IP_);
1294 	kmemleak_free_part(__va(base), size);
1295 	memblock_remove_range(&memblock.reserved, base, size);
1296 }
1297 
1298 /*
1299  * __memblock_free_late - free bootmem block pages directly to buddy allocator
1300  * @addr: phys starting address of the  boot memory block
1301  * @size: size of the boot memory block in bytes
1302  *
1303  * This is only useful when the bootmem allocator has already been torn
1304  * down, but we are still initializing the system.  Pages are released directly
1305  * to the buddy allocator, no bootmem metadata is updated because it is gone.
1306  */
1307 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1308 {
1309 	u64 cursor, end;
1310 
1311 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1312 		     __func__, (u64)base, (u64)base + size - 1,
1313 		     (void *)_RET_IP_);
1314 	kmemleak_free_part(__va(base), size);
1315 	cursor = PFN_UP(base);
1316 	end = PFN_DOWN(base + size);
1317 
1318 	for (; cursor < end; cursor++) {
1319 		__free_pages_bootmem(pfn_to_page(cursor), 0);
1320 		totalram_pages++;
1321 	}
1322 }
1323 
1324 /*
1325  * Remaining API functions
1326  */
1327 
1328 phys_addr_t __init memblock_phys_mem_size(void)
1329 {
1330 	return memblock.memory.total_size;
1331 }
1332 
1333 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1334 {
1335 	unsigned long pages = 0;
1336 	struct memblock_region *r;
1337 	unsigned long start_pfn, end_pfn;
1338 
1339 	for_each_memblock(memory, r) {
1340 		start_pfn = memblock_region_memory_base_pfn(r);
1341 		end_pfn = memblock_region_memory_end_pfn(r);
1342 		start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1343 		end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1344 		pages += end_pfn - start_pfn;
1345 	}
1346 
1347 	return PFN_PHYS(pages);
1348 }
1349 
1350 /* lowest address */
1351 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1352 {
1353 	return memblock.memory.regions[0].base;
1354 }
1355 
1356 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1357 {
1358 	int idx = memblock.memory.cnt - 1;
1359 
1360 	return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1361 }
1362 
1363 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1364 {
1365 	phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1366 	struct memblock_region *r;
1367 
1368 	if (!limit)
1369 		return;
1370 
1371 	/* find out max address */
1372 	for_each_memblock(memory, r) {
1373 		if (limit <= r->size) {
1374 			max_addr = r->base + limit;
1375 			break;
1376 		}
1377 		limit -= r->size;
1378 	}
1379 
1380 	/* truncate both memory and reserved regions */
1381 	memblock_remove_range(&memblock.memory, max_addr,
1382 			      (phys_addr_t)ULLONG_MAX);
1383 	memblock_remove_range(&memblock.reserved, max_addr,
1384 			      (phys_addr_t)ULLONG_MAX);
1385 }
1386 
1387 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1388 {
1389 	unsigned int left = 0, right = type->cnt;
1390 
1391 	do {
1392 		unsigned int mid = (right + left) / 2;
1393 
1394 		if (addr < type->regions[mid].base)
1395 			right = mid;
1396 		else if (addr >= (type->regions[mid].base +
1397 				  type->regions[mid].size))
1398 			left = mid + 1;
1399 		else
1400 			return mid;
1401 	} while (left < right);
1402 	return -1;
1403 }
1404 
1405 int __init memblock_is_reserved(phys_addr_t addr)
1406 {
1407 	return memblock_search(&memblock.reserved, addr) != -1;
1408 }
1409 
1410 int __init_memblock memblock_is_memory(phys_addr_t addr)
1411 {
1412 	return memblock_search(&memblock.memory, addr) != -1;
1413 }
1414 
1415 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1416 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1417 			 unsigned long *start_pfn, unsigned long *end_pfn)
1418 {
1419 	struct memblock_type *type = &memblock.memory;
1420 	int mid = memblock_search(type, PFN_PHYS(pfn));
1421 
1422 	if (mid == -1)
1423 		return -1;
1424 
1425 	*start_pfn = PFN_DOWN(type->regions[mid].base);
1426 	*end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1427 
1428 	return type->regions[mid].nid;
1429 }
1430 #endif
1431 
1432 /**
1433  * memblock_is_region_memory - check if a region is a subset of memory
1434  * @base: base of region to check
1435  * @size: size of region to check
1436  *
1437  * Check if the region [@base, @base+@size) is a subset of a memory block.
1438  *
1439  * RETURNS:
1440  * 0 if false, non-zero if true
1441  */
1442 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1443 {
1444 	int idx = memblock_search(&memblock.memory, base);
1445 	phys_addr_t end = base + memblock_cap_size(base, &size);
1446 
1447 	if (idx == -1)
1448 		return 0;
1449 	return memblock.memory.regions[idx].base <= base &&
1450 		(memblock.memory.regions[idx].base +
1451 		 memblock.memory.regions[idx].size) >= end;
1452 }
1453 
1454 /**
1455  * memblock_is_region_reserved - check if a region intersects reserved memory
1456  * @base: base of region to check
1457  * @size: size of region to check
1458  *
1459  * Check if the region [@base, @base+@size) intersects a reserved memory block.
1460  *
1461  * RETURNS:
1462  * 0 if false, non-zero if true
1463  */
1464 int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1465 {
1466 	memblock_cap_size(base, &size);
1467 	return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
1468 }
1469 
1470 void __init_memblock memblock_trim_memory(phys_addr_t align)
1471 {
1472 	phys_addr_t start, end, orig_start, orig_end;
1473 	struct memblock_region *r;
1474 
1475 	for_each_memblock(memory, r) {
1476 		orig_start = r->base;
1477 		orig_end = r->base + r->size;
1478 		start = round_up(orig_start, align);
1479 		end = round_down(orig_end, align);
1480 
1481 		if (start == orig_start && end == orig_end)
1482 			continue;
1483 
1484 		if (start < end) {
1485 			r->base = start;
1486 			r->size = end - start;
1487 		} else {
1488 			memblock_remove_region(&memblock.memory,
1489 					       r - memblock.memory.regions);
1490 			r--;
1491 		}
1492 	}
1493 }
1494 
1495 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1496 {
1497 	memblock.current_limit = limit;
1498 }
1499 
1500 phys_addr_t __init_memblock memblock_get_current_limit(void)
1501 {
1502 	return memblock.current_limit;
1503 }
1504 
1505 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1506 {
1507 	unsigned long long base, size;
1508 	unsigned long flags;
1509 	int i;
1510 
1511 	pr_info(" %s.cnt  = 0x%lx\n", name, type->cnt);
1512 
1513 	for (i = 0; i < type->cnt; i++) {
1514 		struct memblock_region *rgn = &type->regions[i];
1515 		char nid_buf[32] = "";
1516 
1517 		base = rgn->base;
1518 		size = rgn->size;
1519 		flags = rgn->flags;
1520 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1521 		if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1522 			snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1523 				 memblock_get_region_node(rgn));
1524 #endif
1525 		pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1526 			name, i, base, base + size - 1, size, nid_buf, flags);
1527 	}
1528 }
1529 
1530 void __init_memblock __memblock_dump_all(void)
1531 {
1532 	pr_info("MEMBLOCK configuration:\n");
1533 	pr_info(" memory size = %#llx reserved size = %#llx\n",
1534 		(unsigned long long)memblock.memory.total_size,
1535 		(unsigned long long)memblock.reserved.total_size);
1536 
1537 	memblock_dump(&memblock.memory, "memory");
1538 	memblock_dump(&memblock.reserved, "reserved");
1539 }
1540 
1541 void __init memblock_allow_resize(void)
1542 {
1543 	memblock_can_resize = 1;
1544 }
1545 
1546 static int __init early_memblock(char *p)
1547 {
1548 	if (p && strstr(p, "debug"))
1549 		memblock_debug = 1;
1550 	return 0;
1551 }
1552 early_param("memblock", early_memblock);
1553 
1554 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1555 
1556 static int memblock_debug_show(struct seq_file *m, void *private)
1557 {
1558 	struct memblock_type *type = m->private;
1559 	struct memblock_region *reg;
1560 	int i;
1561 
1562 	for (i = 0; i < type->cnt; i++) {
1563 		reg = &type->regions[i];
1564 		seq_printf(m, "%4d: ", i);
1565 		if (sizeof(phys_addr_t) == 4)
1566 			seq_printf(m, "0x%08lx..0x%08lx\n",
1567 				   (unsigned long)reg->base,
1568 				   (unsigned long)(reg->base + reg->size - 1));
1569 		else
1570 			seq_printf(m, "0x%016llx..0x%016llx\n",
1571 				   (unsigned long long)reg->base,
1572 				   (unsigned long long)(reg->base + reg->size - 1));
1573 
1574 	}
1575 	return 0;
1576 }
1577 
1578 static int memblock_debug_open(struct inode *inode, struct file *file)
1579 {
1580 	return single_open(file, memblock_debug_show, inode->i_private);
1581 }
1582 
1583 static const struct file_operations memblock_debug_fops = {
1584 	.open = memblock_debug_open,
1585 	.read = seq_read,
1586 	.llseek = seq_lseek,
1587 	.release = single_release,
1588 };
1589 
1590 static int __init memblock_init_debugfs(void)
1591 {
1592 	struct dentry *root = debugfs_create_dir("memblock", NULL);
1593 	if (!root)
1594 		return -ENXIO;
1595 	debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1596 	debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1597 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1598 	debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1599 #endif
1600 
1601 	return 0;
1602 }
1603 __initcall(memblock_init_debugfs);
1604 
1605 #endif /* CONFIG_DEBUG_FS */
1606