xref: /openbmc/linux/mm/memblock.c (revision 7587eb18)
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 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
610 {
611 	memblock_dbg("memblock_add: [%#016llx-%#016llx] flags %#02lx %pF\n",
612 		     (unsigned long long)base,
613 		     (unsigned long long)base + size - 1,
614 		     0UL, (void *)_RET_IP_);
615 
616 	return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
617 }
618 
619 /**
620  * memblock_isolate_range - isolate given range into disjoint memblocks
621  * @type: memblock type to isolate range for
622  * @base: base of range to isolate
623  * @size: size of range to isolate
624  * @start_rgn: out parameter for the start of isolated region
625  * @end_rgn: out parameter for the end of isolated region
626  *
627  * Walk @type and ensure that regions don't cross the boundaries defined by
628  * [@base,@base+@size).  Crossing regions are split at the boundaries,
629  * which may create at most two more regions.  The index of the first
630  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
631  *
632  * RETURNS:
633  * 0 on success, -errno on failure.
634  */
635 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
636 					phys_addr_t base, phys_addr_t size,
637 					int *start_rgn, int *end_rgn)
638 {
639 	phys_addr_t end = base + memblock_cap_size(base, &size);
640 	int idx;
641 	struct memblock_region *rgn;
642 
643 	*start_rgn = *end_rgn = 0;
644 
645 	if (!size)
646 		return 0;
647 
648 	/* we'll create at most two more regions */
649 	while (type->cnt + 2 > type->max)
650 		if (memblock_double_array(type, base, size) < 0)
651 			return -ENOMEM;
652 
653 	for_each_memblock_type(type, rgn) {
654 		phys_addr_t rbase = rgn->base;
655 		phys_addr_t rend = rbase + rgn->size;
656 
657 		if (rbase >= end)
658 			break;
659 		if (rend <= base)
660 			continue;
661 
662 		if (rbase < base) {
663 			/*
664 			 * @rgn intersects from below.  Split and continue
665 			 * to process the next region - the new top half.
666 			 */
667 			rgn->base = base;
668 			rgn->size -= base - rbase;
669 			type->total_size -= base - rbase;
670 			memblock_insert_region(type, idx, rbase, base - rbase,
671 					       memblock_get_region_node(rgn),
672 					       rgn->flags);
673 		} else if (rend > end) {
674 			/*
675 			 * @rgn intersects from above.  Split and redo the
676 			 * current region - the new bottom half.
677 			 */
678 			rgn->base = end;
679 			rgn->size -= end - rbase;
680 			type->total_size -= end - rbase;
681 			memblock_insert_region(type, idx--, rbase, end - rbase,
682 					       memblock_get_region_node(rgn),
683 					       rgn->flags);
684 		} else {
685 			/* @rgn is fully contained, record it */
686 			if (!*end_rgn)
687 				*start_rgn = idx;
688 			*end_rgn = idx + 1;
689 		}
690 	}
691 
692 	return 0;
693 }
694 
695 static int __init_memblock memblock_remove_range(struct memblock_type *type,
696 					  phys_addr_t base, phys_addr_t size)
697 {
698 	int start_rgn, end_rgn;
699 	int i, ret;
700 
701 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
702 	if (ret)
703 		return ret;
704 
705 	for (i = end_rgn - 1; i >= start_rgn; i--)
706 		memblock_remove_region(type, i);
707 	return 0;
708 }
709 
710 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
711 {
712 	return memblock_remove_range(&memblock.memory, base, size);
713 }
714 
715 
716 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
717 {
718 	memblock_dbg("   memblock_free: [%#016llx-%#016llx] %pF\n",
719 		     (unsigned long long)base,
720 		     (unsigned long long)base + size - 1,
721 		     (void *)_RET_IP_);
722 
723 	kmemleak_free_part(__va(base), size);
724 	return memblock_remove_range(&memblock.reserved, base, size);
725 }
726 
727 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
728 {
729 	memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
730 		     (unsigned long long)base,
731 		     (unsigned long long)base + size - 1,
732 		     0UL, (void *)_RET_IP_);
733 
734 	return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
735 }
736 
737 /**
738  *
739  * This function isolates region [@base, @base + @size), and sets/clears flag
740  *
741  * Return 0 on success, -errno on failure.
742  */
743 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
744 				phys_addr_t size, int set, int flag)
745 {
746 	struct memblock_type *type = &memblock.memory;
747 	int i, ret, start_rgn, end_rgn;
748 
749 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
750 	if (ret)
751 		return ret;
752 
753 	for (i = start_rgn; i < end_rgn; i++)
754 		if (set)
755 			memblock_set_region_flags(&type->regions[i], flag);
756 		else
757 			memblock_clear_region_flags(&type->regions[i], flag);
758 
759 	memblock_merge_regions(type);
760 	return 0;
761 }
762 
763 /**
764  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
765  * @base: the base phys addr of the region
766  * @size: the size of the region
767  *
768  * Return 0 on success, -errno on failure.
769  */
770 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
771 {
772 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
773 }
774 
775 /**
776  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
777  * @base: the base phys addr of the region
778  * @size: the size of the region
779  *
780  * Return 0 on success, -errno on failure.
781  */
782 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
783 {
784 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
785 }
786 
787 /**
788  * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
789  * @base: the base phys addr of the region
790  * @size: the size of the region
791  *
792  * Return 0 on success, -errno on failure.
793  */
794 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
795 {
796 	system_has_some_mirror = true;
797 
798 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
799 }
800 
801 /**
802  * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
803  * @base: the base phys addr of the region
804  * @size: the size of the region
805  *
806  * Return 0 on success, -errno on failure.
807  */
808 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
809 {
810 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
811 }
812 
813 /**
814  * __next_reserved_mem_region - next function for for_each_reserved_region()
815  * @idx: pointer to u64 loop variable
816  * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
817  * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
818  *
819  * Iterate over all reserved memory regions.
820  */
821 void __init_memblock __next_reserved_mem_region(u64 *idx,
822 					   phys_addr_t *out_start,
823 					   phys_addr_t *out_end)
824 {
825 	struct memblock_type *type = &memblock.reserved;
826 
827 	if (*idx < type->cnt) {
828 		struct memblock_region *r = &type->regions[*idx];
829 		phys_addr_t base = r->base;
830 		phys_addr_t size = r->size;
831 
832 		if (out_start)
833 			*out_start = base;
834 		if (out_end)
835 			*out_end = base + size - 1;
836 
837 		*idx += 1;
838 		return;
839 	}
840 
841 	/* signal end of iteration */
842 	*idx = ULLONG_MAX;
843 }
844 
845 /**
846  * __next__mem_range - next function for for_each_free_mem_range() etc.
847  * @idx: pointer to u64 loop variable
848  * @nid: node selector, %NUMA_NO_NODE for all nodes
849  * @flags: pick from blocks based on memory attributes
850  * @type_a: pointer to memblock_type from where the range is taken
851  * @type_b: pointer to memblock_type which excludes memory from being taken
852  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
853  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
854  * @out_nid: ptr to int for nid of the range, can be %NULL
855  *
856  * Find the first area from *@idx which matches @nid, fill the out
857  * parameters, and update *@idx for the next iteration.  The lower 32bit of
858  * *@idx contains index into type_a and the upper 32bit indexes the
859  * areas before each region in type_b.	For example, if type_b regions
860  * look like the following,
861  *
862  *	0:[0-16), 1:[32-48), 2:[128-130)
863  *
864  * The upper 32bit indexes the following regions.
865  *
866  *	0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
867  *
868  * As both region arrays are sorted, the function advances the two indices
869  * in lockstep and returns each intersection.
870  */
871 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
872 				      struct memblock_type *type_a,
873 				      struct memblock_type *type_b,
874 				      phys_addr_t *out_start,
875 				      phys_addr_t *out_end, int *out_nid)
876 {
877 	int idx_a = *idx & 0xffffffff;
878 	int idx_b = *idx >> 32;
879 
880 	if (WARN_ONCE(nid == MAX_NUMNODES,
881 	"Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
882 		nid = NUMA_NO_NODE;
883 
884 	for (; idx_a < type_a->cnt; idx_a++) {
885 		struct memblock_region *m = &type_a->regions[idx_a];
886 
887 		phys_addr_t m_start = m->base;
888 		phys_addr_t m_end = m->base + m->size;
889 		int	    m_nid = memblock_get_region_node(m);
890 
891 		/* only memory regions are associated with nodes, check it */
892 		if (nid != NUMA_NO_NODE && nid != m_nid)
893 			continue;
894 
895 		/* skip hotpluggable memory regions if needed */
896 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
897 			continue;
898 
899 		/* if we want mirror memory skip non-mirror memory regions */
900 		if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
901 			continue;
902 
903 		/* skip nomap memory unless we were asked for it explicitly */
904 		if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
905 			continue;
906 
907 		if (!type_b) {
908 			if (out_start)
909 				*out_start = m_start;
910 			if (out_end)
911 				*out_end = m_end;
912 			if (out_nid)
913 				*out_nid = m_nid;
914 			idx_a++;
915 			*idx = (u32)idx_a | (u64)idx_b << 32;
916 			return;
917 		}
918 
919 		/* scan areas before each reservation */
920 		for (; idx_b < type_b->cnt + 1; idx_b++) {
921 			struct memblock_region *r;
922 			phys_addr_t r_start;
923 			phys_addr_t r_end;
924 
925 			r = &type_b->regions[idx_b];
926 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
927 			r_end = idx_b < type_b->cnt ?
928 				r->base : ULLONG_MAX;
929 
930 			/*
931 			 * if idx_b advanced past idx_a,
932 			 * break out to advance idx_a
933 			 */
934 			if (r_start >= m_end)
935 				break;
936 			/* if the two regions intersect, we're done */
937 			if (m_start < r_end) {
938 				if (out_start)
939 					*out_start =
940 						max(m_start, r_start);
941 				if (out_end)
942 					*out_end = min(m_end, r_end);
943 				if (out_nid)
944 					*out_nid = m_nid;
945 				/*
946 				 * The region which ends first is
947 				 * advanced for the next iteration.
948 				 */
949 				if (m_end <= r_end)
950 					idx_a++;
951 				else
952 					idx_b++;
953 				*idx = (u32)idx_a | (u64)idx_b << 32;
954 				return;
955 			}
956 		}
957 	}
958 
959 	/* signal end of iteration */
960 	*idx = ULLONG_MAX;
961 }
962 
963 /**
964  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
965  *
966  * Finds the next range from type_a which is not marked as unsuitable
967  * in type_b.
968  *
969  * @idx: pointer to u64 loop variable
970  * @nid: node selector, %NUMA_NO_NODE for all nodes
971  * @flags: pick from blocks based on memory attributes
972  * @type_a: pointer to memblock_type from where the range is taken
973  * @type_b: pointer to memblock_type which excludes memory from being taken
974  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
975  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
976  * @out_nid: ptr to int for nid of the range, can be %NULL
977  *
978  * Reverse of __next_mem_range().
979  */
980 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
981 					  struct memblock_type *type_a,
982 					  struct memblock_type *type_b,
983 					  phys_addr_t *out_start,
984 					  phys_addr_t *out_end, int *out_nid)
985 {
986 	int idx_a = *idx & 0xffffffff;
987 	int idx_b = *idx >> 32;
988 
989 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
990 		nid = NUMA_NO_NODE;
991 
992 	if (*idx == (u64)ULLONG_MAX) {
993 		idx_a = type_a->cnt - 1;
994 		idx_b = type_b->cnt;
995 	}
996 
997 	for (; idx_a >= 0; idx_a--) {
998 		struct memblock_region *m = &type_a->regions[idx_a];
999 
1000 		phys_addr_t m_start = m->base;
1001 		phys_addr_t m_end = m->base + m->size;
1002 		int m_nid = memblock_get_region_node(m);
1003 
1004 		/* only memory regions are associated with nodes, check it */
1005 		if (nid != NUMA_NO_NODE && nid != m_nid)
1006 			continue;
1007 
1008 		/* skip hotpluggable memory regions if needed */
1009 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1010 			continue;
1011 
1012 		/* if we want mirror memory skip non-mirror memory regions */
1013 		if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1014 			continue;
1015 
1016 		/* skip nomap memory unless we were asked for it explicitly */
1017 		if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1018 			continue;
1019 
1020 		if (!type_b) {
1021 			if (out_start)
1022 				*out_start = m_start;
1023 			if (out_end)
1024 				*out_end = m_end;
1025 			if (out_nid)
1026 				*out_nid = m_nid;
1027 			idx_a++;
1028 			*idx = (u32)idx_a | (u64)idx_b << 32;
1029 			return;
1030 		}
1031 
1032 		/* scan areas before each reservation */
1033 		for (; idx_b >= 0; idx_b--) {
1034 			struct memblock_region *r;
1035 			phys_addr_t r_start;
1036 			phys_addr_t r_end;
1037 
1038 			r = &type_b->regions[idx_b];
1039 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1040 			r_end = idx_b < type_b->cnt ?
1041 				r->base : ULLONG_MAX;
1042 			/*
1043 			 * if idx_b advanced past idx_a,
1044 			 * break out to advance idx_a
1045 			 */
1046 
1047 			if (r_end <= m_start)
1048 				break;
1049 			/* if the two regions intersect, we're done */
1050 			if (m_end > r_start) {
1051 				if (out_start)
1052 					*out_start = max(m_start, r_start);
1053 				if (out_end)
1054 					*out_end = min(m_end, r_end);
1055 				if (out_nid)
1056 					*out_nid = m_nid;
1057 				if (m_start >= r_start)
1058 					idx_a--;
1059 				else
1060 					idx_b--;
1061 				*idx = (u32)idx_a | (u64)idx_b << 32;
1062 				return;
1063 			}
1064 		}
1065 	}
1066 	/* signal end of iteration */
1067 	*idx = ULLONG_MAX;
1068 }
1069 
1070 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1071 /*
1072  * Common iterator interface used to define for_each_mem_range().
1073  */
1074 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1075 				unsigned long *out_start_pfn,
1076 				unsigned long *out_end_pfn, int *out_nid)
1077 {
1078 	struct memblock_type *type = &memblock.memory;
1079 	struct memblock_region *r;
1080 
1081 	while (++*idx < type->cnt) {
1082 		r = &type->regions[*idx];
1083 
1084 		if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1085 			continue;
1086 		if (nid == MAX_NUMNODES || nid == r->nid)
1087 			break;
1088 	}
1089 	if (*idx >= type->cnt) {
1090 		*idx = -1;
1091 		return;
1092 	}
1093 
1094 	if (out_start_pfn)
1095 		*out_start_pfn = PFN_UP(r->base);
1096 	if (out_end_pfn)
1097 		*out_end_pfn = PFN_DOWN(r->base + r->size);
1098 	if (out_nid)
1099 		*out_nid = r->nid;
1100 }
1101 
1102 /**
1103  * memblock_set_node - set node ID on memblock regions
1104  * @base: base of area to set node ID for
1105  * @size: size of area to set node ID for
1106  * @type: memblock type to set node ID for
1107  * @nid: node ID to set
1108  *
1109  * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1110  * Regions which cross the area boundaries are split as necessary.
1111  *
1112  * RETURNS:
1113  * 0 on success, -errno on failure.
1114  */
1115 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1116 				      struct memblock_type *type, int nid)
1117 {
1118 	int start_rgn, end_rgn;
1119 	int i, ret;
1120 
1121 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1122 	if (ret)
1123 		return ret;
1124 
1125 	for (i = start_rgn; i < end_rgn; i++)
1126 		memblock_set_region_node(&type->regions[i], nid);
1127 
1128 	memblock_merge_regions(type);
1129 	return 0;
1130 }
1131 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1132 
1133 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1134 					phys_addr_t align, phys_addr_t start,
1135 					phys_addr_t end, int nid, ulong flags)
1136 {
1137 	phys_addr_t found;
1138 
1139 	if (!align)
1140 		align = SMP_CACHE_BYTES;
1141 
1142 	found = memblock_find_in_range_node(size, align, start, end, nid,
1143 					    flags);
1144 	if (found && !memblock_reserve(found, size)) {
1145 		/*
1146 		 * The min_count is set to 0 so that memblock allocations are
1147 		 * never reported as leaks.
1148 		 */
1149 		kmemleak_alloc(__va(found), size, 0, 0);
1150 		return found;
1151 	}
1152 	return 0;
1153 }
1154 
1155 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1156 					phys_addr_t start, phys_addr_t end,
1157 					ulong flags)
1158 {
1159 	return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1160 					flags);
1161 }
1162 
1163 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1164 					phys_addr_t align, phys_addr_t max_addr,
1165 					int nid, ulong flags)
1166 {
1167 	return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1168 }
1169 
1170 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1171 {
1172 	ulong flags = choose_memblock_flags();
1173 	phys_addr_t ret;
1174 
1175 again:
1176 	ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1177 				      nid, flags);
1178 
1179 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
1180 		flags &= ~MEMBLOCK_MIRROR;
1181 		goto again;
1182 	}
1183 	return ret;
1184 }
1185 
1186 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1187 {
1188 	return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1189 				       MEMBLOCK_NONE);
1190 }
1191 
1192 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1193 {
1194 	phys_addr_t alloc;
1195 
1196 	alloc = __memblock_alloc_base(size, align, max_addr);
1197 
1198 	if (alloc == 0)
1199 		panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1200 		      (unsigned long long) size, (unsigned long long) max_addr);
1201 
1202 	return alloc;
1203 }
1204 
1205 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1206 {
1207 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1208 }
1209 
1210 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1211 {
1212 	phys_addr_t res = memblock_alloc_nid(size, align, nid);
1213 
1214 	if (res)
1215 		return res;
1216 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1217 }
1218 
1219 /**
1220  * memblock_virt_alloc_internal - allocate boot memory block
1221  * @size: size of memory block to be allocated in bytes
1222  * @align: alignment of the region and block's size
1223  * @min_addr: the lower bound of the memory region to allocate (phys address)
1224  * @max_addr: the upper bound of the memory region to allocate (phys address)
1225  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1226  *
1227  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1228  * will fall back to memory below @min_addr. Also, allocation may fall back
1229  * to any node in the system if the specified node can not
1230  * hold the requested memory.
1231  *
1232  * The allocation is performed from memory region limited by
1233  * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1234  *
1235  * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1236  *
1237  * The phys address of allocated boot memory block is converted to virtual and
1238  * allocated memory is reset to 0.
1239  *
1240  * In addition, function sets the min_count to 0 using kmemleak_alloc for
1241  * allocated boot memory block, so that it is never reported as leaks.
1242  *
1243  * RETURNS:
1244  * Virtual address of allocated memory block on success, NULL on failure.
1245  */
1246 static void * __init memblock_virt_alloc_internal(
1247 				phys_addr_t size, phys_addr_t align,
1248 				phys_addr_t min_addr, phys_addr_t max_addr,
1249 				int nid)
1250 {
1251 	phys_addr_t alloc;
1252 	void *ptr;
1253 	ulong flags = choose_memblock_flags();
1254 
1255 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1256 		nid = NUMA_NO_NODE;
1257 
1258 	/*
1259 	 * Detect any accidental use of these APIs after slab is ready, as at
1260 	 * this moment memblock may be deinitialized already and its
1261 	 * internal data may be destroyed (after execution of free_all_bootmem)
1262 	 */
1263 	if (WARN_ON_ONCE(slab_is_available()))
1264 		return kzalloc_node(size, GFP_NOWAIT, nid);
1265 
1266 	if (!align)
1267 		align = SMP_CACHE_BYTES;
1268 
1269 	if (max_addr > memblock.current_limit)
1270 		max_addr = memblock.current_limit;
1271 
1272 again:
1273 	alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1274 					    nid, flags);
1275 	if (alloc)
1276 		goto done;
1277 
1278 	if (nid != NUMA_NO_NODE) {
1279 		alloc = memblock_find_in_range_node(size, align, min_addr,
1280 						    max_addr, NUMA_NO_NODE,
1281 						    flags);
1282 		if (alloc)
1283 			goto done;
1284 	}
1285 
1286 	if (min_addr) {
1287 		min_addr = 0;
1288 		goto again;
1289 	}
1290 
1291 	if (flags & MEMBLOCK_MIRROR) {
1292 		flags &= ~MEMBLOCK_MIRROR;
1293 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1294 			&size);
1295 		goto again;
1296 	}
1297 
1298 	return NULL;
1299 done:
1300 	memblock_reserve(alloc, size);
1301 	ptr = phys_to_virt(alloc);
1302 	memset(ptr, 0, size);
1303 
1304 	/*
1305 	 * The min_count is set to 0 so that bootmem allocated blocks
1306 	 * are never reported as leaks. This is because many of these blocks
1307 	 * are only referred via the physical address which is not
1308 	 * looked up by kmemleak.
1309 	 */
1310 	kmemleak_alloc(ptr, size, 0, 0);
1311 
1312 	return ptr;
1313 }
1314 
1315 /**
1316  * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1317  * @size: size of memory block to be allocated in bytes
1318  * @align: alignment of the region and block's size
1319  * @min_addr: the lower bound of the memory region from where the allocation
1320  *	  is preferred (phys address)
1321  * @max_addr: the upper bound of the memory region from where the allocation
1322  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1323  *	      allocate only from memory limited by memblock.current_limit value
1324  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1325  *
1326  * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1327  * additional debug information (including caller info), if enabled.
1328  *
1329  * RETURNS:
1330  * Virtual address of allocated memory block on success, NULL on failure.
1331  */
1332 void * __init memblock_virt_alloc_try_nid_nopanic(
1333 				phys_addr_t size, phys_addr_t align,
1334 				phys_addr_t min_addr, phys_addr_t max_addr,
1335 				int nid)
1336 {
1337 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1338 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1339 		     (u64)max_addr, (void *)_RET_IP_);
1340 	return memblock_virt_alloc_internal(size, align, min_addr,
1341 					     max_addr, nid);
1342 }
1343 
1344 /**
1345  * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1346  * @size: size of memory block to be allocated in bytes
1347  * @align: alignment of the region and block's size
1348  * @min_addr: the lower bound of the memory region from where the allocation
1349  *	  is preferred (phys address)
1350  * @max_addr: the upper bound of the memory region from where the allocation
1351  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1352  *	      allocate only from memory limited by memblock.current_limit value
1353  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1354  *
1355  * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1356  * which provides debug information (including caller info), if enabled,
1357  * and panics if the request can not be satisfied.
1358  *
1359  * RETURNS:
1360  * Virtual address of allocated memory block on success, NULL on failure.
1361  */
1362 void * __init memblock_virt_alloc_try_nid(
1363 			phys_addr_t size, phys_addr_t align,
1364 			phys_addr_t min_addr, phys_addr_t max_addr,
1365 			int nid)
1366 {
1367 	void *ptr;
1368 
1369 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1370 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1371 		     (u64)max_addr, (void *)_RET_IP_);
1372 	ptr = memblock_virt_alloc_internal(size, align,
1373 					   min_addr, max_addr, nid);
1374 	if (ptr)
1375 		return ptr;
1376 
1377 	panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1378 	      __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1379 	      (u64)max_addr);
1380 	return NULL;
1381 }
1382 
1383 /**
1384  * __memblock_free_early - free boot memory block
1385  * @base: phys starting address of the  boot memory block
1386  * @size: size of the boot memory block in bytes
1387  *
1388  * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1389  * The freeing memory will not be released to the buddy allocator.
1390  */
1391 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1392 {
1393 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1394 		     __func__, (u64)base, (u64)base + size - 1,
1395 		     (void *)_RET_IP_);
1396 	kmemleak_free_part(__va(base), size);
1397 	memblock_remove_range(&memblock.reserved, base, size);
1398 }
1399 
1400 /*
1401  * __memblock_free_late - free bootmem block pages directly to buddy allocator
1402  * @addr: phys starting address of the  boot memory block
1403  * @size: size of the boot memory block in bytes
1404  *
1405  * This is only useful when the bootmem allocator has already been torn
1406  * down, but we are still initializing the system.  Pages are released directly
1407  * to the buddy allocator, no bootmem metadata is updated because it is gone.
1408  */
1409 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1410 {
1411 	u64 cursor, end;
1412 
1413 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1414 		     __func__, (u64)base, (u64)base + size - 1,
1415 		     (void *)_RET_IP_);
1416 	kmemleak_free_part(__va(base), size);
1417 	cursor = PFN_UP(base);
1418 	end = PFN_DOWN(base + size);
1419 
1420 	for (; cursor < end; cursor++) {
1421 		__free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1422 		totalram_pages++;
1423 	}
1424 }
1425 
1426 /*
1427  * Remaining API functions
1428  */
1429 
1430 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1431 {
1432 	return memblock.memory.total_size;
1433 }
1434 
1435 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1436 {
1437 	unsigned long pages = 0;
1438 	struct memblock_region *r;
1439 	unsigned long start_pfn, end_pfn;
1440 
1441 	for_each_memblock(memory, r) {
1442 		start_pfn = memblock_region_memory_base_pfn(r);
1443 		end_pfn = memblock_region_memory_end_pfn(r);
1444 		start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1445 		end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1446 		pages += end_pfn - start_pfn;
1447 	}
1448 
1449 	return PFN_PHYS(pages);
1450 }
1451 
1452 /* lowest address */
1453 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1454 {
1455 	return memblock.memory.regions[0].base;
1456 }
1457 
1458 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1459 {
1460 	int idx = memblock.memory.cnt - 1;
1461 
1462 	return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1463 }
1464 
1465 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1466 {
1467 	phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1468 	struct memblock_region *r;
1469 
1470 	if (!limit)
1471 		return;
1472 
1473 	/* find out max address */
1474 	for_each_memblock(memory, r) {
1475 		if (limit <= r->size) {
1476 			max_addr = r->base + limit;
1477 			break;
1478 		}
1479 		limit -= r->size;
1480 	}
1481 
1482 	/* truncate both memory and reserved regions */
1483 	memblock_remove_range(&memblock.memory, max_addr,
1484 			      (phys_addr_t)ULLONG_MAX);
1485 	memblock_remove_range(&memblock.reserved, max_addr,
1486 			      (phys_addr_t)ULLONG_MAX);
1487 }
1488 
1489 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1490 {
1491 	unsigned int left = 0, right = type->cnt;
1492 
1493 	do {
1494 		unsigned int mid = (right + left) / 2;
1495 
1496 		if (addr < type->regions[mid].base)
1497 			right = mid;
1498 		else if (addr >= (type->regions[mid].base +
1499 				  type->regions[mid].size))
1500 			left = mid + 1;
1501 		else
1502 			return mid;
1503 	} while (left < right);
1504 	return -1;
1505 }
1506 
1507 bool __init memblock_is_reserved(phys_addr_t addr)
1508 {
1509 	return memblock_search(&memblock.reserved, addr) != -1;
1510 }
1511 
1512 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1513 {
1514 	return memblock_search(&memblock.memory, addr) != -1;
1515 }
1516 
1517 int __init_memblock memblock_is_map_memory(phys_addr_t addr)
1518 {
1519 	int i = memblock_search(&memblock.memory, addr);
1520 
1521 	if (i == -1)
1522 		return false;
1523 	return !memblock_is_nomap(&memblock.memory.regions[i]);
1524 }
1525 
1526 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1527 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1528 			 unsigned long *start_pfn, unsigned long *end_pfn)
1529 {
1530 	struct memblock_type *type = &memblock.memory;
1531 	int mid = memblock_search(type, PFN_PHYS(pfn));
1532 
1533 	if (mid == -1)
1534 		return -1;
1535 
1536 	*start_pfn = PFN_DOWN(type->regions[mid].base);
1537 	*end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1538 
1539 	return type->regions[mid].nid;
1540 }
1541 #endif
1542 
1543 /**
1544  * memblock_is_region_memory - check if a region is a subset of memory
1545  * @base: base of region to check
1546  * @size: size of region to check
1547  *
1548  * Check if the region [@base, @base+@size) is a subset of a memory block.
1549  *
1550  * RETURNS:
1551  * 0 if false, non-zero if true
1552  */
1553 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1554 {
1555 	int idx = memblock_search(&memblock.memory, base);
1556 	phys_addr_t end = base + memblock_cap_size(base, &size);
1557 
1558 	if (idx == -1)
1559 		return 0;
1560 	return memblock.memory.regions[idx].base <= base &&
1561 		(memblock.memory.regions[idx].base +
1562 		 memblock.memory.regions[idx].size) >= end;
1563 }
1564 
1565 /**
1566  * memblock_is_region_reserved - check if a region intersects reserved memory
1567  * @base: base of region to check
1568  * @size: size of region to check
1569  *
1570  * Check if the region [@base, @base+@size) intersects a reserved memory block.
1571  *
1572  * RETURNS:
1573  * True if they intersect, false if not.
1574  */
1575 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1576 {
1577 	memblock_cap_size(base, &size);
1578 	return memblock_overlaps_region(&memblock.reserved, base, size);
1579 }
1580 
1581 void __init_memblock memblock_trim_memory(phys_addr_t align)
1582 {
1583 	phys_addr_t start, end, orig_start, orig_end;
1584 	struct memblock_region *r;
1585 
1586 	for_each_memblock(memory, r) {
1587 		orig_start = r->base;
1588 		orig_end = r->base + r->size;
1589 		start = round_up(orig_start, align);
1590 		end = round_down(orig_end, align);
1591 
1592 		if (start == orig_start && end == orig_end)
1593 			continue;
1594 
1595 		if (start < end) {
1596 			r->base = start;
1597 			r->size = end - start;
1598 		} else {
1599 			memblock_remove_region(&memblock.memory,
1600 					       r - memblock.memory.regions);
1601 			r--;
1602 		}
1603 	}
1604 }
1605 
1606 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1607 {
1608 	memblock.current_limit = limit;
1609 }
1610 
1611 phys_addr_t __init_memblock memblock_get_current_limit(void)
1612 {
1613 	return memblock.current_limit;
1614 }
1615 
1616 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1617 {
1618 	unsigned long long base, size;
1619 	unsigned long flags;
1620 	int idx;
1621 	struct memblock_region *rgn;
1622 
1623 	pr_info(" %s.cnt  = 0x%lx\n", name, type->cnt);
1624 
1625 	for_each_memblock_type(type, rgn) {
1626 		char nid_buf[32] = "";
1627 
1628 		base = rgn->base;
1629 		size = rgn->size;
1630 		flags = rgn->flags;
1631 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1632 		if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1633 			snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1634 				 memblock_get_region_node(rgn));
1635 #endif
1636 		pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1637 			name, idx, base, base + size - 1, size, nid_buf, flags);
1638 	}
1639 }
1640 
1641 void __init_memblock __memblock_dump_all(void)
1642 {
1643 	pr_info("MEMBLOCK configuration:\n");
1644 	pr_info(" memory size = %#llx reserved size = %#llx\n",
1645 		(unsigned long long)memblock.memory.total_size,
1646 		(unsigned long long)memblock.reserved.total_size);
1647 
1648 	memblock_dump(&memblock.memory, "memory");
1649 	memblock_dump(&memblock.reserved, "reserved");
1650 }
1651 
1652 void __init memblock_allow_resize(void)
1653 {
1654 	memblock_can_resize = 1;
1655 }
1656 
1657 static int __init early_memblock(char *p)
1658 {
1659 	if (p && strstr(p, "debug"))
1660 		memblock_debug = 1;
1661 	return 0;
1662 }
1663 early_param("memblock", early_memblock);
1664 
1665 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1666 
1667 static int memblock_debug_show(struct seq_file *m, void *private)
1668 {
1669 	struct memblock_type *type = m->private;
1670 	struct memblock_region *reg;
1671 	int i;
1672 
1673 	for (i = 0; i < type->cnt; i++) {
1674 		reg = &type->regions[i];
1675 		seq_printf(m, "%4d: ", i);
1676 		if (sizeof(phys_addr_t) == 4)
1677 			seq_printf(m, "0x%08lx..0x%08lx\n",
1678 				   (unsigned long)reg->base,
1679 				   (unsigned long)(reg->base + reg->size - 1));
1680 		else
1681 			seq_printf(m, "0x%016llx..0x%016llx\n",
1682 				   (unsigned long long)reg->base,
1683 				   (unsigned long long)(reg->base + reg->size - 1));
1684 
1685 	}
1686 	return 0;
1687 }
1688 
1689 static int memblock_debug_open(struct inode *inode, struct file *file)
1690 {
1691 	return single_open(file, memblock_debug_show, inode->i_private);
1692 }
1693 
1694 static const struct file_operations memblock_debug_fops = {
1695 	.open = memblock_debug_open,
1696 	.read = seq_read,
1697 	.llseek = seq_lseek,
1698 	.release = single_release,
1699 };
1700 
1701 static int __init memblock_init_debugfs(void)
1702 {
1703 	struct dentry *root = debugfs_create_dir("memblock", NULL);
1704 	if (!root)
1705 		return -ENXIO;
1706 	debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1707 	debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1708 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1709 	debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1710 #endif
1711 
1712 	return 0;
1713 }
1714 __initcall(memblock_init_debugfs);
1715 
1716 #endif /* CONFIG_DEBUG_FS */
1717