xref: /openbmc/linux/mm/memblock.c (revision c0e297dc)
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 static long __init_memblock memblock_overlaps_region(struct memblock_type *type,
95 					phys_addr_t base, phys_addr_t size)
96 {
97 	unsigned long i;
98 
99 	for (i = 0; i < type->cnt; i++) {
100 		phys_addr_t rgnbase = type->regions[i].base;
101 		phys_addr_t rgnsize = type->regions[i].size;
102 		if (memblock_addrs_overlap(base, size, rgnbase, rgnsize))
103 			break;
104 	}
105 
106 	return (i < type->cnt) ? i : -1;
107 }
108 
109 /*
110  * __memblock_find_range_bottom_up - find free area utility in bottom-up
111  * @start: start of candidate range
112  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
113  * @size: size of free area to find
114  * @align: alignment of free area to find
115  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
116  * @flags: pick from blocks based on memory attributes
117  *
118  * Utility called from memblock_find_in_range_node(), find free area bottom-up.
119  *
120  * RETURNS:
121  * Found address on success, 0 on failure.
122  */
123 static phys_addr_t __init_memblock
124 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
125 				phys_addr_t size, phys_addr_t align, int nid,
126 				ulong flags)
127 {
128 	phys_addr_t this_start, this_end, cand;
129 	u64 i;
130 
131 	for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
132 		this_start = clamp(this_start, start, end);
133 		this_end = clamp(this_end, start, end);
134 
135 		cand = round_up(this_start, align);
136 		if (cand < this_end && this_end - cand >= size)
137 			return cand;
138 	}
139 
140 	return 0;
141 }
142 
143 /**
144  * __memblock_find_range_top_down - find free area utility, in top-down
145  * @start: start of candidate range
146  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
147  * @size: size of free area to find
148  * @align: alignment of free area to find
149  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
150  * @flags: pick from blocks based on memory attributes
151  *
152  * Utility called from memblock_find_in_range_node(), find free area top-down.
153  *
154  * RETURNS:
155  * Found address on success, 0 on failure.
156  */
157 static phys_addr_t __init_memblock
158 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
159 			       phys_addr_t size, phys_addr_t align, int nid,
160 			       ulong flags)
161 {
162 	phys_addr_t this_start, this_end, cand;
163 	u64 i;
164 
165 	for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
166 					NULL) {
167 		this_start = clamp(this_start, start, end);
168 		this_end = clamp(this_end, start, end);
169 
170 		if (this_end < size)
171 			continue;
172 
173 		cand = round_down(this_end - size, align);
174 		if (cand >= this_start)
175 			return cand;
176 	}
177 
178 	return 0;
179 }
180 
181 /**
182  * memblock_find_in_range_node - find free area in given range and node
183  * @size: size of free area to find
184  * @align: alignment of free area to find
185  * @start: start of candidate range
186  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
187  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
188  * @flags: pick from blocks based on memory attributes
189  *
190  * Find @size free area aligned to @align in the specified range and node.
191  *
192  * When allocation direction is bottom-up, the @start should be greater
193  * than the end of the kernel image. Otherwise, it will be trimmed. The
194  * reason is that we want the bottom-up allocation just near the kernel
195  * image so it is highly likely that the allocated memory and the kernel
196  * will reside in the same node.
197  *
198  * If bottom-up allocation failed, will try to allocate memory top-down.
199  *
200  * RETURNS:
201  * Found address on success, 0 on failure.
202  */
203 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
204 					phys_addr_t align, phys_addr_t start,
205 					phys_addr_t end, int nid, ulong flags)
206 {
207 	phys_addr_t kernel_end, ret;
208 
209 	/* pump up @end */
210 	if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
211 		end = memblock.current_limit;
212 
213 	/* avoid allocating the first page */
214 	start = max_t(phys_addr_t, start, PAGE_SIZE);
215 	end = max(start, end);
216 	kernel_end = __pa_symbol(_end);
217 
218 	/*
219 	 * try bottom-up allocation only when bottom-up mode
220 	 * is set and @end is above the kernel image.
221 	 */
222 	if (memblock_bottom_up() && end > kernel_end) {
223 		phys_addr_t bottom_up_start;
224 
225 		/* make sure we will allocate above the kernel */
226 		bottom_up_start = max(start, kernel_end);
227 
228 		/* ok, try bottom-up allocation first */
229 		ret = __memblock_find_range_bottom_up(bottom_up_start, end,
230 						      size, align, nid, flags);
231 		if (ret)
232 			return ret;
233 
234 		/*
235 		 * we always limit bottom-up allocation above the kernel,
236 		 * but top-down allocation doesn't have the limit, so
237 		 * retrying top-down allocation may succeed when bottom-up
238 		 * allocation failed.
239 		 *
240 		 * bottom-up allocation is expected to be fail very rarely,
241 		 * so we use WARN_ONCE() here to see the stack trace if
242 		 * fail happens.
243 		 */
244 		WARN_ONCE(1, "memblock: bottom-up allocation failed, "
245 			     "memory hotunplug may be affected\n");
246 	}
247 
248 	return __memblock_find_range_top_down(start, end, size, align, nid,
249 					      flags);
250 }
251 
252 /**
253  * memblock_find_in_range - find free area in given range
254  * @start: start of candidate range
255  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
256  * @size: size of free area to find
257  * @align: alignment of free area to find
258  *
259  * Find @size free area aligned to @align in the specified range.
260  *
261  * RETURNS:
262  * Found address on success, 0 on failure.
263  */
264 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
265 					phys_addr_t end, phys_addr_t size,
266 					phys_addr_t align)
267 {
268 	phys_addr_t ret;
269 	ulong flags = choose_memblock_flags();
270 
271 again:
272 	ret = memblock_find_in_range_node(size, align, start, end,
273 					    NUMA_NO_NODE, flags);
274 
275 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
276 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
277 			&size);
278 		flags &= ~MEMBLOCK_MIRROR;
279 		goto again;
280 	}
281 
282 	return ret;
283 }
284 
285 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
286 {
287 	type->total_size -= type->regions[r].size;
288 	memmove(&type->regions[r], &type->regions[r + 1],
289 		(type->cnt - (r + 1)) * sizeof(type->regions[r]));
290 	type->cnt--;
291 
292 	/* Special case for empty arrays */
293 	if (type->cnt == 0) {
294 		WARN_ON(type->total_size != 0);
295 		type->cnt = 1;
296 		type->regions[0].base = 0;
297 		type->regions[0].size = 0;
298 		type->regions[0].flags = 0;
299 		memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
300 	}
301 }
302 
303 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
304 
305 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
306 					phys_addr_t *addr)
307 {
308 	if (memblock.reserved.regions == memblock_reserved_init_regions)
309 		return 0;
310 
311 	*addr = __pa(memblock.reserved.regions);
312 
313 	return PAGE_ALIGN(sizeof(struct memblock_region) *
314 			  memblock.reserved.max);
315 }
316 
317 phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
318 					phys_addr_t *addr)
319 {
320 	if (memblock.memory.regions == memblock_memory_init_regions)
321 		return 0;
322 
323 	*addr = __pa(memblock.memory.regions);
324 
325 	return PAGE_ALIGN(sizeof(struct memblock_region) *
326 			  memblock.memory.max);
327 }
328 
329 #endif
330 
331 /**
332  * memblock_double_array - double the size of the memblock regions array
333  * @type: memblock type of the regions array being doubled
334  * @new_area_start: starting address of memory range to avoid overlap with
335  * @new_area_size: size of memory range to avoid overlap with
336  *
337  * Double the size of the @type regions array. If memblock is being used to
338  * allocate memory for a new reserved regions array and there is a previously
339  * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
340  * waiting to be reserved, ensure the memory used by the new array does
341  * not overlap.
342  *
343  * RETURNS:
344  * 0 on success, -1 on failure.
345  */
346 static int __init_memblock memblock_double_array(struct memblock_type *type,
347 						phys_addr_t new_area_start,
348 						phys_addr_t new_area_size)
349 {
350 	struct memblock_region *new_array, *old_array;
351 	phys_addr_t old_alloc_size, new_alloc_size;
352 	phys_addr_t old_size, new_size, addr;
353 	int use_slab = slab_is_available();
354 	int *in_slab;
355 
356 	/* We don't allow resizing until we know about the reserved regions
357 	 * of memory that aren't suitable for allocation
358 	 */
359 	if (!memblock_can_resize)
360 		return -1;
361 
362 	/* Calculate new doubled size */
363 	old_size = type->max * sizeof(struct memblock_region);
364 	new_size = old_size << 1;
365 	/*
366 	 * We need to allocated new one align to PAGE_SIZE,
367 	 *   so we can free them completely later.
368 	 */
369 	old_alloc_size = PAGE_ALIGN(old_size);
370 	new_alloc_size = PAGE_ALIGN(new_size);
371 
372 	/* Retrieve the slab flag */
373 	if (type == &memblock.memory)
374 		in_slab = &memblock_memory_in_slab;
375 	else
376 		in_slab = &memblock_reserved_in_slab;
377 
378 	/* Try to find some space for it.
379 	 *
380 	 * WARNING: We assume that either slab_is_available() and we use it or
381 	 * we use MEMBLOCK for allocations. That means that this is unsafe to
382 	 * use when bootmem is currently active (unless bootmem itself is
383 	 * implemented on top of MEMBLOCK which isn't the case yet)
384 	 *
385 	 * This should however not be an issue for now, as we currently only
386 	 * call into MEMBLOCK while it's still active, or much later when slab
387 	 * is active for memory hotplug operations
388 	 */
389 	if (use_slab) {
390 		new_array = kmalloc(new_size, GFP_KERNEL);
391 		addr = new_array ? __pa(new_array) : 0;
392 	} else {
393 		/* only exclude range when trying to double reserved.regions */
394 		if (type != &memblock.reserved)
395 			new_area_start = new_area_size = 0;
396 
397 		addr = memblock_find_in_range(new_area_start + new_area_size,
398 						memblock.current_limit,
399 						new_alloc_size, PAGE_SIZE);
400 		if (!addr && new_area_size)
401 			addr = memblock_find_in_range(0,
402 				min(new_area_start, memblock.current_limit),
403 				new_alloc_size, PAGE_SIZE);
404 
405 		new_array = addr ? __va(addr) : NULL;
406 	}
407 	if (!addr) {
408 		pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
409 		       memblock_type_name(type), type->max, type->max * 2);
410 		return -1;
411 	}
412 
413 	memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
414 			memblock_type_name(type), type->max * 2, (u64)addr,
415 			(u64)addr + new_size - 1);
416 
417 	/*
418 	 * Found space, we now need to move the array over before we add the
419 	 * reserved region since it may be our reserved array itself that is
420 	 * full.
421 	 */
422 	memcpy(new_array, type->regions, old_size);
423 	memset(new_array + type->max, 0, old_size);
424 	old_array = type->regions;
425 	type->regions = new_array;
426 	type->max <<= 1;
427 
428 	/* Free old array. We needn't free it if the array is the static one */
429 	if (*in_slab)
430 		kfree(old_array);
431 	else if (old_array != memblock_memory_init_regions &&
432 		 old_array != memblock_reserved_init_regions)
433 		memblock_free(__pa(old_array), old_alloc_size);
434 
435 	/*
436 	 * Reserve the new array if that comes from the memblock.  Otherwise, we
437 	 * needn't do it
438 	 */
439 	if (!use_slab)
440 		BUG_ON(memblock_reserve(addr, new_alloc_size));
441 
442 	/* Update slab flag */
443 	*in_slab = use_slab;
444 
445 	return 0;
446 }
447 
448 /**
449  * memblock_merge_regions - merge neighboring compatible regions
450  * @type: memblock type to scan
451  *
452  * Scan @type and merge neighboring compatible regions.
453  */
454 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
455 {
456 	int i = 0;
457 
458 	/* cnt never goes below 1 */
459 	while (i < type->cnt - 1) {
460 		struct memblock_region *this = &type->regions[i];
461 		struct memblock_region *next = &type->regions[i + 1];
462 
463 		if (this->base + this->size != next->base ||
464 		    memblock_get_region_node(this) !=
465 		    memblock_get_region_node(next) ||
466 		    this->flags != next->flags) {
467 			BUG_ON(this->base + this->size > next->base);
468 			i++;
469 			continue;
470 		}
471 
472 		this->size += next->size;
473 		/* move forward from next + 1, index of which is i + 2 */
474 		memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
475 		type->cnt--;
476 	}
477 }
478 
479 /**
480  * memblock_insert_region - insert new memblock region
481  * @type:	memblock type to insert into
482  * @idx:	index for the insertion point
483  * @base:	base address of the new region
484  * @size:	size of the new region
485  * @nid:	node id of the new region
486  * @flags:	flags of the new region
487  *
488  * Insert new memblock region [@base,@base+@size) into @type at @idx.
489  * @type must already have extra room to accomodate the new region.
490  */
491 static void __init_memblock memblock_insert_region(struct memblock_type *type,
492 						   int idx, phys_addr_t base,
493 						   phys_addr_t size,
494 						   int nid, unsigned long flags)
495 {
496 	struct memblock_region *rgn = &type->regions[idx];
497 
498 	BUG_ON(type->cnt >= type->max);
499 	memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
500 	rgn->base = base;
501 	rgn->size = size;
502 	rgn->flags = flags;
503 	memblock_set_region_node(rgn, nid);
504 	type->cnt++;
505 	type->total_size += size;
506 }
507 
508 /**
509  * memblock_add_range - add new memblock region
510  * @type: memblock type to add new region into
511  * @base: base address of the new region
512  * @size: size of the new region
513  * @nid: nid of the new region
514  * @flags: flags of the new region
515  *
516  * Add new memblock region [@base,@base+@size) into @type.  The new region
517  * is allowed to overlap with existing ones - overlaps don't affect already
518  * existing regions.  @type is guaranteed to be minimal (all neighbouring
519  * compatible regions are merged) after the addition.
520  *
521  * RETURNS:
522  * 0 on success, -errno on failure.
523  */
524 int __init_memblock memblock_add_range(struct memblock_type *type,
525 				phys_addr_t base, phys_addr_t size,
526 				int nid, unsigned long flags)
527 {
528 	bool insert = false;
529 	phys_addr_t obase = base;
530 	phys_addr_t end = base + memblock_cap_size(base, &size);
531 	int i, nr_new;
532 
533 	if (!size)
534 		return 0;
535 
536 	/* special case for empty array */
537 	if (type->regions[0].size == 0) {
538 		WARN_ON(type->cnt != 1 || type->total_size);
539 		type->regions[0].base = base;
540 		type->regions[0].size = size;
541 		type->regions[0].flags = flags;
542 		memblock_set_region_node(&type->regions[0], nid);
543 		type->total_size = size;
544 		return 0;
545 	}
546 repeat:
547 	/*
548 	 * The following is executed twice.  Once with %false @insert and
549 	 * then with %true.  The first counts the number of regions needed
550 	 * to accomodate the new area.  The second actually inserts them.
551 	 */
552 	base = obase;
553 	nr_new = 0;
554 
555 	for (i = 0; i < type->cnt; i++) {
556 		struct memblock_region *rgn = &type->regions[i];
557 		phys_addr_t rbase = rgn->base;
558 		phys_addr_t rend = rbase + rgn->size;
559 
560 		if (rbase >= end)
561 			break;
562 		if (rend <= base)
563 			continue;
564 		/*
565 		 * @rgn overlaps.  If it separates the lower part of new
566 		 * area, insert that portion.
567 		 */
568 		if (rbase > base) {
569 			nr_new++;
570 			if (insert)
571 				memblock_insert_region(type, i++, 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, i, base, end - base,
584 					       nid, flags);
585 	}
586 
587 	/*
588 	 * If this was the first round, resize array and repeat for actual
589 	 * insertions; otherwise, merge and return.
590 	 */
591 	if (!insert) {
592 		while (type->cnt + nr_new > type->max)
593 			if (memblock_double_array(type, obase, size) < 0)
594 				return -ENOMEM;
595 		insert = true;
596 		goto repeat;
597 	} else {
598 		memblock_merge_regions(type);
599 		return 0;
600 	}
601 }
602 
603 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
604 				       int nid)
605 {
606 	return memblock_add_range(&memblock.memory, base, size, nid, 0);
607 }
608 
609 static int __init_memblock memblock_add_region(phys_addr_t base,
610 						phys_addr_t size,
611 						int nid,
612 						unsigned long flags)
613 {
614 	struct memblock_type *_rgn = &memblock.memory;
615 
616 	memblock_dbg("memblock_add: [%#016llx-%#016llx] flags %#02lx %pF\n",
617 		     (unsigned long long)base,
618 		     (unsigned long long)base + size - 1,
619 		     flags, (void *)_RET_IP_);
620 
621 	return memblock_add_range(_rgn, base, size, nid, flags);
622 }
623 
624 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
625 {
626 	return memblock_add_region(base, size, MAX_NUMNODES, 0);
627 }
628 
629 /**
630  * memblock_isolate_range - isolate given range into disjoint memblocks
631  * @type: memblock type to isolate range for
632  * @base: base of range to isolate
633  * @size: size of range to isolate
634  * @start_rgn: out parameter for the start of isolated region
635  * @end_rgn: out parameter for the end of isolated region
636  *
637  * Walk @type and ensure that regions don't cross the boundaries defined by
638  * [@base,@base+@size).  Crossing regions are split at the boundaries,
639  * which may create at most two more regions.  The index of the first
640  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
641  *
642  * RETURNS:
643  * 0 on success, -errno on failure.
644  */
645 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
646 					phys_addr_t base, phys_addr_t size,
647 					int *start_rgn, int *end_rgn)
648 {
649 	phys_addr_t end = base + memblock_cap_size(base, &size);
650 	int i;
651 
652 	*start_rgn = *end_rgn = 0;
653 
654 	if (!size)
655 		return 0;
656 
657 	/* we'll create at most two more regions */
658 	while (type->cnt + 2 > type->max)
659 		if (memblock_double_array(type, base, size) < 0)
660 			return -ENOMEM;
661 
662 	for (i = 0; i < type->cnt; i++) {
663 		struct memblock_region *rgn = &type->regions[i];
664 		phys_addr_t rbase = rgn->base;
665 		phys_addr_t rend = rbase + rgn->size;
666 
667 		if (rbase >= end)
668 			break;
669 		if (rend <= base)
670 			continue;
671 
672 		if (rbase < base) {
673 			/*
674 			 * @rgn intersects from below.  Split and continue
675 			 * to process the next region - the new top half.
676 			 */
677 			rgn->base = base;
678 			rgn->size -= base - rbase;
679 			type->total_size -= base - rbase;
680 			memblock_insert_region(type, i, rbase, base - rbase,
681 					       memblock_get_region_node(rgn),
682 					       rgn->flags);
683 		} else if (rend > end) {
684 			/*
685 			 * @rgn intersects from above.  Split and redo the
686 			 * current region - the new bottom half.
687 			 */
688 			rgn->base = end;
689 			rgn->size -= end - rbase;
690 			type->total_size -= end - rbase;
691 			memblock_insert_region(type, i--, rbase, end - rbase,
692 					       memblock_get_region_node(rgn),
693 					       rgn->flags);
694 		} else {
695 			/* @rgn is fully contained, record it */
696 			if (!*end_rgn)
697 				*start_rgn = i;
698 			*end_rgn = i + 1;
699 		}
700 	}
701 
702 	return 0;
703 }
704 
705 int __init_memblock memblock_remove_range(struct memblock_type *type,
706 					  phys_addr_t base, phys_addr_t size)
707 {
708 	int start_rgn, end_rgn;
709 	int i, ret;
710 
711 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
712 	if (ret)
713 		return ret;
714 
715 	for (i = end_rgn - 1; i >= start_rgn; i--)
716 		memblock_remove_region(type, i);
717 	return 0;
718 }
719 
720 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
721 {
722 	return memblock_remove_range(&memblock.memory, base, size);
723 }
724 
725 
726 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
727 {
728 	memblock_dbg("   memblock_free: [%#016llx-%#016llx] %pF\n",
729 		     (unsigned long long)base,
730 		     (unsigned long long)base + size - 1,
731 		     (void *)_RET_IP_);
732 
733 	kmemleak_free_part(__va(base), size);
734 	return memblock_remove_range(&memblock.reserved, base, size);
735 }
736 
737 static int __init_memblock memblock_reserve_region(phys_addr_t base,
738 						   phys_addr_t size,
739 						   int nid,
740 						   unsigned long flags)
741 {
742 	struct memblock_type *type = &memblock.reserved;
743 
744 	memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
745 		     (unsigned long long)base,
746 		     (unsigned long long)base + size - 1,
747 		     flags, (void *)_RET_IP_);
748 
749 	return memblock_add_range(type, base, size, nid, flags);
750 }
751 
752 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
753 {
754 	return memblock_reserve_region(base, size, MAX_NUMNODES, 0);
755 }
756 
757 /**
758  *
759  * This function isolates region [@base, @base + @size), and sets/clears flag
760  *
761  * Return 0 on succees, -errno on failure.
762  */
763 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
764 				phys_addr_t size, int set, int flag)
765 {
766 	struct memblock_type *type = &memblock.memory;
767 	int i, ret, start_rgn, end_rgn;
768 
769 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
770 	if (ret)
771 		return ret;
772 
773 	for (i = start_rgn; i < end_rgn; i++)
774 		if (set)
775 			memblock_set_region_flags(&type->regions[i], flag);
776 		else
777 			memblock_clear_region_flags(&type->regions[i], flag);
778 
779 	memblock_merge_regions(type);
780 	return 0;
781 }
782 
783 /**
784  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
785  * @base: the base phys addr of the region
786  * @size: the size of the region
787  *
788  * Return 0 on succees, -errno on failure.
789  */
790 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
791 {
792 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
793 }
794 
795 /**
796  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
797  * @base: the base phys addr of the region
798  * @size: the size of the region
799  *
800  * Return 0 on succees, -errno on failure.
801  */
802 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
803 {
804 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
805 }
806 
807 /**
808  * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
809  * @base: the base phys addr of the region
810  * @size: the size of the region
811  *
812  * Return 0 on succees, -errno on failure.
813  */
814 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
815 {
816 	system_has_some_mirror = true;
817 
818 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
819 }
820 
821 
822 /**
823  * __next_reserved_mem_region - next function for for_each_reserved_region()
824  * @idx: pointer to u64 loop variable
825  * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
826  * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
827  *
828  * Iterate over all reserved memory regions.
829  */
830 void __init_memblock __next_reserved_mem_region(u64 *idx,
831 					   phys_addr_t *out_start,
832 					   phys_addr_t *out_end)
833 {
834 	struct memblock_type *rsv = &memblock.reserved;
835 
836 	if (*idx >= 0 && *idx < rsv->cnt) {
837 		struct memblock_region *r = &rsv->regions[*idx];
838 		phys_addr_t base = r->base;
839 		phys_addr_t size = r->size;
840 
841 		if (out_start)
842 			*out_start = base;
843 		if (out_end)
844 			*out_end = base + size - 1;
845 
846 		*idx += 1;
847 		return;
848 	}
849 
850 	/* signal end of iteration */
851 	*idx = ULLONG_MAX;
852 }
853 
854 /**
855  * __next__mem_range - next function for for_each_free_mem_range() etc.
856  * @idx: pointer to u64 loop variable
857  * @nid: node selector, %NUMA_NO_NODE for all nodes
858  * @flags: pick from blocks based on memory attributes
859  * @type_a: pointer to memblock_type from where the range is taken
860  * @type_b: pointer to memblock_type which excludes memory from being taken
861  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
862  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
863  * @out_nid: ptr to int for nid of the range, can be %NULL
864  *
865  * Find the first area from *@idx which matches @nid, fill the out
866  * parameters, and update *@idx for the next iteration.  The lower 32bit of
867  * *@idx contains index into type_a and the upper 32bit indexes the
868  * areas before each region in type_b.	For example, if type_b regions
869  * look like the following,
870  *
871  *	0:[0-16), 1:[32-48), 2:[128-130)
872  *
873  * The upper 32bit indexes the following regions.
874  *
875  *	0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
876  *
877  * As both region arrays are sorted, the function advances the two indices
878  * in lockstep and returns each intersection.
879  */
880 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
881 				      struct memblock_type *type_a,
882 				      struct memblock_type *type_b,
883 				      phys_addr_t *out_start,
884 				      phys_addr_t *out_end, int *out_nid)
885 {
886 	int idx_a = *idx & 0xffffffff;
887 	int idx_b = *idx >> 32;
888 
889 	if (WARN_ONCE(nid == MAX_NUMNODES,
890 	"Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
891 		nid = NUMA_NO_NODE;
892 
893 	for (; idx_a < type_a->cnt; idx_a++) {
894 		struct memblock_region *m = &type_a->regions[idx_a];
895 
896 		phys_addr_t m_start = m->base;
897 		phys_addr_t m_end = m->base + m->size;
898 		int	    m_nid = memblock_get_region_node(m);
899 
900 		/* only memory regions are associated with nodes, check it */
901 		if (nid != NUMA_NO_NODE && nid != m_nid)
902 			continue;
903 
904 		/* skip hotpluggable memory regions if needed */
905 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
906 			continue;
907 
908 		/* if we want mirror memory skip non-mirror memory regions */
909 		if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
910 			continue;
911 
912 		if (!type_b) {
913 			if (out_start)
914 				*out_start = m_start;
915 			if (out_end)
916 				*out_end = m_end;
917 			if (out_nid)
918 				*out_nid = m_nid;
919 			idx_a++;
920 			*idx = (u32)idx_a | (u64)idx_b << 32;
921 			return;
922 		}
923 
924 		/* scan areas before each reservation */
925 		for (; idx_b < type_b->cnt + 1; idx_b++) {
926 			struct memblock_region *r;
927 			phys_addr_t r_start;
928 			phys_addr_t r_end;
929 
930 			r = &type_b->regions[idx_b];
931 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
932 			r_end = idx_b < type_b->cnt ?
933 				r->base : ULLONG_MAX;
934 
935 			/*
936 			 * if idx_b advanced past idx_a,
937 			 * break out to advance idx_a
938 			 */
939 			if (r_start >= m_end)
940 				break;
941 			/* if the two regions intersect, we're done */
942 			if (m_start < r_end) {
943 				if (out_start)
944 					*out_start =
945 						max(m_start, r_start);
946 				if (out_end)
947 					*out_end = min(m_end, r_end);
948 				if (out_nid)
949 					*out_nid = m_nid;
950 				/*
951 				 * The region which ends first is
952 				 * advanced for the next iteration.
953 				 */
954 				if (m_end <= r_end)
955 					idx_a++;
956 				else
957 					idx_b++;
958 				*idx = (u32)idx_a | (u64)idx_b << 32;
959 				return;
960 			}
961 		}
962 	}
963 
964 	/* signal end of iteration */
965 	*idx = ULLONG_MAX;
966 }
967 
968 /**
969  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
970  *
971  * Finds the next range from type_a which is not marked as unsuitable
972  * in type_b.
973  *
974  * @idx: pointer to u64 loop variable
975  * @nid: nid: node selector, %NUMA_NO_NODE for all nodes
976  * @flags: pick from blocks based on memory attributes
977  * @type_a: pointer to memblock_type from where the range is taken
978  * @type_b: pointer to memblock_type which excludes memory from being taken
979  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
980  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
981  * @out_nid: ptr to int for nid of the range, can be %NULL
982  *
983  * Reverse of __next_mem_range().
984  */
985 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
986 					  struct memblock_type *type_a,
987 					  struct memblock_type *type_b,
988 					  phys_addr_t *out_start,
989 					  phys_addr_t *out_end, int *out_nid)
990 {
991 	int idx_a = *idx & 0xffffffff;
992 	int idx_b = *idx >> 32;
993 
994 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
995 		nid = NUMA_NO_NODE;
996 
997 	if (*idx == (u64)ULLONG_MAX) {
998 		idx_a = type_a->cnt - 1;
999 		idx_b = type_b->cnt;
1000 	}
1001 
1002 	for (; idx_a >= 0; idx_a--) {
1003 		struct memblock_region *m = &type_a->regions[idx_a];
1004 
1005 		phys_addr_t m_start = m->base;
1006 		phys_addr_t m_end = m->base + m->size;
1007 		int m_nid = memblock_get_region_node(m);
1008 
1009 		/* only memory regions are associated with nodes, check it */
1010 		if (nid != NUMA_NO_NODE && nid != m_nid)
1011 			continue;
1012 
1013 		/* skip hotpluggable memory regions if needed */
1014 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1015 			continue;
1016 
1017 		/* if we want mirror memory skip non-mirror memory regions */
1018 		if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1019 			continue;
1020 
1021 		if (!type_b) {
1022 			if (out_start)
1023 				*out_start = m_start;
1024 			if (out_end)
1025 				*out_end = m_end;
1026 			if (out_nid)
1027 				*out_nid = m_nid;
1028 			idx_a++;
1029 			*idx = (u32)idx_a | (u64)idx_b << 32;
1030 			return;
1031 		}
1032 
1033 		/* scan areas before each reservation */
1034 		for (; idx_b >= 0; idx_b--) {
1035 			struct memblock_region *r;
1036 			phys_addr_t r_start;
1037 			phys_addr_t r_end;
1038 
1039 			r = &type_b->regions[idx_b];
1040 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1041 			r_end = idx_b < type_b->cnt ?
1042 				r->base : ULLONG_MAX;
1043 			/*
1044 			 * if idx_b advanced past idx_a,
1045 			 * break out to advance idx_a
1046 			 */
1047 
1048 			if (r_end <= m_start)
1049 				break;
1050 			/* if the two regions intersect, we're done */
1051 			if (m_end > r_start) {
1052 				if (out_start)
1053 					*out_start = max(m_start, r_start);
1054 				if (out_end)
1055 					*out_end = min(m_end, r_end);
1056 				if (out_nid)
1057 					*out_nid = m_nid;
1058 				if (m_start >= r_start)
1059 					idx_a--;
1060 				else
1061 					idx_b--;
1062 				*idx = (u32)idx_a | (u64)idx_b << 32;
1063 				return;
1064 			}
1065 		}
1066 	}
1067 	/* signal end of iteration */
1068 	*idx = ULLONG_MAX;
1069 }
1070 
1071 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1072 /*
1073  * Common iterator interface used to define for_each_mem_range().
1074  */
1075 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1076 				unsigned long *out_start_pfn,
1077 				unsigned long *out_end_pfn, int *out_nid)
1078 {
1079 	struct memblock_type *type = &memblock.memory;
1080 	struct memblock_region *r;
1081 
1082 	while (++*idx < type->cnt) {
1083 		r = &type->regions[*idx];
1084 
1085 		if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1086 			continue;
1087 		if (nid == MAX_NUMNODES || nid == r->nid)
1088 			break;
1089 	}
1090 	if (*idx >= type->cnt) {
1091 		*idx = -1;
1092 		return;
1093 	}
1094 
1095 	if (out_start_pfn)
1096 		*out_start_pfn = PFN_UP(r->base);
1097 	if (out_end_pfn)
1098 		*out_end_pfn = PFN_DOWN(r->base + r->size);
1099 	if (out_nid)
1100 		*out_nid = r->nid;
1101 }
1102 
1103 /**
1104  * memblock_set_node - set node ID on memblock regions
1105  * @base: base of area to set node ID for
1106  * @size: size of area to set node ID for
1107  * @type: memblock type to set node ID for
1108  * @nid: node ID to set
1109  *
1110  * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1111  * Regions which cross the area boundaries are split as necessary.
1112  *
1113  * RETURNS:
1114  * 0 on success, -errno on failure.
1115  */
1116 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1117 				      struct memblock_type *type, int nid)
1118 {
1119 	int start_rgn, end_rgn;
1120 	int i, ret;
1121 
1122 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1123 	if (ret)
1124 		return ret;
1125 
1126 	for (i = start_rgn; i < end_rgn; i++)
1127 		memblock_set_region_node(&type->regions[i], nid);
1128 
1129 	memblock_merge_regions(type);
1130 	return 0;
1131 }
1132 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1133 
1134 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1135 					phys_addr_t align, phys_addr_t start,
1136 					phys_addr_t end, int nid, ulong flags)
1137 {
1138 	phys_addr_t found;
1139 
1140 	if (!align)
1141 		align = SMP_CACHE_BYTES;
1142 
1143 	found = memblock_find_in_range_node(size, align, start, end, nid,
1144 					    flags);
1145 	if (found && !memblock_reserve(found, size)) {
1146 		/*
1147 		 * The min_count is set to 0 so that memblock allocations are
1148 		 * never reported as leaks.
1149 		 */
1150 		kmemleak_alloc(__va(found), size, 0, 0);
1151 		return found;
1152 	}
1153 	return 0;
1154 }
1155 
1156 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1157 					phys_addr_t start, phys_addr_t end,
1158 					ulong flags)
1159 {
1160 	return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1161 					flags);
1162 }
1163 
1164 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1165 					phys_addr_t align, phys_addr_t max_addr,
1166 					int nid, ulong flags)
1167 {
1168 	return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1169 }
1170 
1171 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1172 {
1173 	ulong flags = choose_memblock_flags();
1174 	phys_addr_t ret;
1175 
1176 again:
1177 	ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1178 				      nid, flags);
1179 
1180 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
1181 		flags &= ~MEMBLOCK_MIRROR;
1182 		goto again;
1183 	}
1184 	return ret;
1185 }
1186 
1187 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1188 {
1189 	return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1190 				       MEMBLOCK_NONE);
1191 }
1192 
1193 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1194 {
1195 	phys_addr_t alloc;
1196 
1197 	alloc = __memblock_alloc_base(size, align, max_addr);
1198 
1199 	if (alloc == 0)
1200 		panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1201 		      (unsigned long long) size, (unsigned long long) max_addr);
1202 
1203 	return alloc;
1204 }
1205 
1206 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1207 {
1208 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1209 }
1210 
1211 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1212 {
1213 	phys_addr_t res = memblock_alloc_nid(size, align, nid);
1214 
1215 	if (res)
1216 		return res;
1217 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1218 }
1219 
1220 /**
1221  * memblock_virt_alloc_internal - allocate boot memory block
1222  * @size: size of memory block to be allocated in bytes
1223  * @align: alignment of the region and block's size
1224  * @min_addr: the lower bound of the memory region to allocate (phys address)
1225  * @max_addr: the upper bound of the memory region to allocate (phys address)
1226  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1227  *
1228  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1229  * will fall back to memory below @min_addr. Also, allocation may fall back
1230  * to any node in the system if the specified node can not
1231  * hold the requested memory.
1232  *
1233  * The allocation is performed from memory region limited by
1234  * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1235  *
1236  * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1237  *
1238  * The phys address of allocated boot memory block is converted to virtual and
1239  * allocated memory is reset to 0.
1240  *
1241  * In addition, function sets the min_count to 0 using kmemleak_alloc for
1242  * allocated boot memory block, so that it is never reported as leaks.
1243  *
1244  * RETURNS:
1245  * Virtual address of allocated memory block on success, NULL on failure.
1246  */
1247 static void * __init memblock_virt_alloc_internal(
1248 				phys_addr_t size, phys_addr_t align,
1249 				phys_addr_t min_addr, phys_addr_t max_addr,
1250 				int nid)
1251 {
1252 	phys_addr_t alloc;
1253 	void *ptr;
1254 	ulong flags = choose_memblock_flags();
1255 
1256 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1257 		nid = NUMA_NO_NODE;
1258 
1259 	/*
1260 	 * Detect any accidental use of these APIs after slab is ready, as at
1261 	 * this moment memblock may be deinitialized already and its
1262 	 * internal data may be destroyed (after execution of free_all_bootmem)
1263 	 */
1264 	if (WARN_ON_ONCE(slab_is_available()))
1265 		return kzalloc_node(size, GFP_NOWAIT, nid);
1266 
1267 	if (!align)
1268 		align = SMP_CACHE_BYTES;
1269 
1270 	if (max_addr > memblock.current_limit)
1271 		max_addr = memblock.current_limit;
1272 
1273 again:
1274 	alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1275 					    nid, flags);
1276 	if (alloc)
1277 		goto done;
1278 
1279 	if (nid != NUMA_NO_NODE) {
1280 		alloc = memblock_find_in_range_node(size, align, min_addr,
1281 						    max_addr, NUMA_NO_NODE,
1282 						    flags);
1283 		if (alloc)
1284 			goto done;
1285 	}
1286 
1287 	if (min_addr) {
1288 		min_addr = 0;
1289 		goto again;
1290 	}
1291 
1292 	if (flags & MEMBLOCK_MIRROR) {
1293 		flags &= ~MEMBLOCK_MIRROR;
1294 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1295 			&size);
1296 		goto again;
1297 	}
1298 
1299 	return NULL;
1300 done:
1301 	memblock_reserve(alloc, size);
1302 	ptr = phys_to_virt(alloc);
1303 	memset(ptr, 0, size);
1304 
1305 	/*
1306 	 * The min_count is set to 0 so that bootmem allocated blocks
1307 	 * are never reported as leaks. This is because many of these blocks
1308 	 * are only referred via the physical address which is not
1309 	 * looked up by kmemleak.
1310 	 */
1311 	kmemleak_alloc(ptr, size, 0, 0);
1312 
1313 	return ptr;
1314 }
1315 
1316 /**
1317  * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1318  * @size: size of memory block to be allocated in bytes
1319  * @align: alignment of the region and block's size
1320  * @min_addr: the lower bound of the memory region from where the allocation
1321  *	  is preferred (phys address)
1322  * @max_addr: the upper bound of the memory region from where the allocation
1323  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1324  *	      allocate only from memory limited by memblock.current_limit value
1325  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1326  *
1327  * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1328  * additional debug information (including caller info), if enabled.
1329  *
1330  * RETURNS:
1331  * Virtual address of allocated memory block on success, NULL on failure.
1332  */
1333 void * __init memblock_virt_alloc_try_nid_nopanic(
1334 				phys_addr_t size, phys_addr_t align,
1335 				phys_addr_t min_addr, phys_addr_t max_addr,
1336 				int nid)
1337 {
1338 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1339 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1340 		     (u64)max_addr, (void *)_RET_IP_);
1341 	return memblock_virt_alloc_internal(size, align, min_addr,
1342 					     max_addr, nid);
1343 }
1344 
1345 /**
1346  * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1347  * @size: size of memory block to be allocated in bytes
1348  * @align: alignment of the region and block's size
1349  * @min_addr: the lower bound of the memory region from where the allocation
1350  *	  is preferred (phys address)
1351  * @max_addr: the upper bound of the memory region from where the allocation
1352  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1353  *	      allocate only from memory limited by memblock.current_limit value
1354  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1355  *
1356  * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1357  * which provides debug information (including caller info), if enabled,
1358  * and panics if the request can not be satisfied.
1359  *
1360  * RETURNS:
1361  * Virtual address of allocated memory block on success, NULL on failure.
1362  */
1363 void * __init memblock_virt_alloc_try_nid(
1364 			phys_addr_t size, phys_addr_t align,
1365 			phys_addr_t min_addr, phys_addr_t max_addr,
1366 			int nid)
1367 {
1368 	void *ptr;
1369 
1370 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1371 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1372 		     (u64)max_addr, (void *)_RET_IP_);
1373 	ptr = memblock_virt_alloc_internal(size, align,
1374 					   min_addr, max_addr, nid);
1375 	if (ptr)
1376 		return ptr;
1377 
1378 	panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1379 	      __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1380 	      (u64)max_addr);
1381 	return NULL;
1382 }
1383 
1384 /**
1385  * __memblock_free_early - free boot memory block
1386  * @base: phys starting address of the  boot memory block
1387  * @size: size of the boot memory block in bytes
1388  *
1389  * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1390  * The freeing memory will not be released to the buddy allocator.
1391  */
1392 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1393 {
1394 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1395 		     __func__, (u64)base, (u64)base + size - 1,
1396 		     (void *)_RET_IP_);
1397 	kmemleak_free_part(__va(base), size);
1398 	memblock_remove_range(&memblock.reserved, base, size);
1399 }
1400 
1401 /*
1402  * __memblock_free_late - free bootmem block pages directly to buddy allocator
1403  * @addr: phys starting address of the  boot memory block
1404  * @size: size of the boot memory block in bytes
1405  *
1406  * This is only useful when the bootmem allocator has already been torn
1407  * down, but we are still initializing the system.  Pages are released directly
1408  * to the buddy allocator, no bootmem metadata is updated because it is gone.
1409  */
1410 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1411 {
1412 	u64 cursor, end;
1413 
1414 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1415 		     __func__, (u64)base, (u64)base + size - 1,
1416 		     (void *)_RET_IP_);
1417 	kmemleak_free_part(__va(base), size);
1418 	cursor = PFN_UP(base);
1419 	end = PFN_DOWN(base + size);
1420 
1421 	for (; cursor < end; cursor++) {
1422 		__free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1423 		totalram_pages++;
1424 	}
1425 }
1426 
1427 /*
1428  * Remaining API functions
1429  */
1430 
1431 phys_addr_t __init memblock_phys_mem_size(void)
1432 {
1433 	return memblock.memory.total_size;
1434 }
1435 
1436 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1437 {
1438 	unsigned long pages = 0;
1439 	struct memblock_region *r;
1440 	unsigned long start_pfn, end_pfn;
1441 
1442 	for_each_memblock(memory, r) {
1443 		start_pfn = memblock_region_memory_base_pfn(r);
1444 		end_pfn = memblock_region_memory_end_pfn(r);
1445 		start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1446 		end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1447 		pages += end_pfn - start_pfn;
1448 	}
1449 
1450 	return PFN_PHYS(pages);
1451 }
1452 
1453 /* lowest address */
1454 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1455 {
1456 	return memblock.memory.regions[0].base;
1457 }
1458 
1459 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1460 {
1461 	int idx = memblock.memory.cnt - 1;
1462 
1463 	return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1464 }
1465 
1466 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1467 {
1468 	phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1469 	struct memblock_region *r;
1470 
1471 	if (!limit)
1472 		return;
1473 
1474 	/* find out max address */
1475 	for_each_memblock(memory, r) {
1476 		if (limit <= r->size) {
1477 			max_addr = r->base + limit;
1478 			break;
1479 		}
1480 		limit -= r->size;
1481 	}
1482 
1483 	/* truncate both memory and reserved regions */
1484 	memblock_remove_range(&memblock.memory, max_addr,
1485 			      (phys_addr_t)ULLONG_MAX);
1486 	memblock_remove_range(&memblock.reserved, max_addr,
1487 			      (phys_addr_t)ULLONG_MAX);
1488 }
1489 
1490 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1491 {
1492 	unsigned int left = 0, right = type->cnt;
1493 
1494 	do {
1495 		unsigned int mid = (right + left) / 2;
1496 
1497 		if (addr < type->regions[mid].base)
1498 			right = mid;
1499 		else if (addr >= (type->regions[mid].base +
1500 				  type->regions[mid].size))
1501 			left = mid + 1;
1502 		else
1503 			return mid;
1504 	} while (left < right);
1505 	return -1;
1506 }
1507 
1508 int __init memblock_is_reserved(phys_addr_t addr)
1509 {
1510 	return memblock_search(&memblock.reserved, addr) != -1;
1511 }
1512 
1513 int __init_memblock memblock_is_memory(phys_addr_t addr)
1514 {
1515 	return memblock_search(&memblock.memory, addr) != -1;
1516 }
1517 
1518 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1519 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1520 			 unsigned long *start_pfn, unsigned long *end_pfn)
1521 {
1522 	struct memblock_type *type = &memblock.memory;
1523 	int mid = memblock_search(type, PFN_PHYS(pfn));
1524 
1525 	if (mid == -1)
1526 		return -1;
1527 
1528 	*start_pfn = PFN_DOWN(type->regions[mid].base);
1529 	*end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1530 
1531 	return type->regions[mid].nid;
1532 }
1533 #endif
1534 
1535 /**
1536  * memblock_is_region_memory - check if a region is a subset of memory
1537  * @base: base of region to check
1538  * @size: size of region to check
1539  *
1540  * Check if the region [@base, @base+@size) is a subset of a memory block.
1541  *
1542  * RETURNS:
1543  * 0 if false, non-zero if true
1544  */
1545 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1546 {
1547 	int idx = memblock_search(&memblock.memory, base);
1548 	phys_addr_t end = base + memblock_cap_size(base, &size);
1549 
1550 	if (idx == -1)
1551 		return 0;
1552 	return memblock.memory.regions[idx].base <= base &&
1553 		(memblock.memory.regions[idx].base +
1554 		 memblock.memory.regions[idx].size) >= end;
1555 }
1556 
1557 /**
1558  * memblock_is_region_reserved - check if a region intersects reserved memory
1559  * @base: base of region to check
1560  * @size: size of region to check
1561  *
1562  * Check if the region [@base, @base+@size) intersects a reserved memory block.
1563  *
1564  * RETURNS:
1565  * 0 if false, non-zero if true
1566  */
1567 int __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1568 {
1569 	memblock_cap_size(base, &size);
1570 	return memblock_overlaps_region(&memblock.reserved, base, size) >= 0;
1571 }
1572 
1573 void __init_memblock memblock_trim_memory(phys_addr_t align)
1574 {
1575 	phys_addr_t start, end, orig_start, orig_end;
1576 	struct memblock_region *r;
1577 
1578 	for_each_memblock(memory, r) {
1579 		orig_start = r->base;
1580 		orig_end = r->base + r->size;
1581 		start = round_up(orig_start, align);
1582 		end = round_down(orig_end, align);
1583 
1584 		if (start == orig_start && end == orig_end)
1585 			continue;
1586 
1587 		if (start < end) {
1588 			r->base = start;
1589 			r->size = end - start;
1590 		} else {
1591 			memblock_remove_region(&memblock.memory,
1592 					       r - memblock.memory.regions);
1593 			r--;
1594 		}
1595 	}
1596 }
1597 
1598 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1599 {
1600 	memblock.current_limit = limit;
1601 }
1602 
1603 phys_addr_t __init_memblock memblock_get_current_limit(void)
1604 {
1605 	return memblock.current_limit;
1606 }
1607 
1608 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1609 {
1610 	unsigned long long base, size;
1611 	unsigned long flags;
1612 	int i;
1613 
1614 	pr_info(" %s.cnt  = 0x%lx\n", name, type->cnt);
1615 
1616 	for (i = 0; i < type->cnt; i++) {
1617 		struct memblock_region *rgn = &type->regions[i];
1618 		char nid_buf[32] = "";
1619 
1620 		base = rgn->base;
1621 		size = rgn->size;
1622 		flags = rgn->flags;
1623 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1624 		if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1625 			snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1626 				 memblock_get_region_node(rgn));
1627 #endif
1628 		pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1629 			name, i, base, base + size - 1, size, nid_buf, flags);
1630 	}
1631 }
1632 
1633 void __init_memblock __memblock_dump_all(void)
1634 {
1635 	pr_info("MEMBLOCK configuration:\n");
1636 	pr_info(" memory size = %#llx reserved size = %#llx\n",
1637 		(unsigned long long)memblock.memory.total_size,
1638 		(unsigned long long)memblock.reserved.total_size);
1639 
1640 	memblock_dump(&memblock.memory, "memory");
1641 	memblock_dump(&memblock.reserved, "reserved");
1642 }
1643 
1644 void __init memblock_allow_resize(void)
1645 {
1646 	memblock_can_resize = 1;
1647 }
1648 
1649 static int __init early_memblock(char *p)
1650 {
1651 	if (p && strstr(p, "debug"))
1652 		memblock_debug = 1;
1653 	return 0;
1654 }
1655 early_param("memblock", early_memblock);
1656 
1657 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1658 
1659 static int memblock_debug_show(struct seq_file *m, void *private)
1660 {
1661 	struct memblock_type *type = m->private;
1662 	struct memblock_region *reg;
1663 	int i;
1664 
1665 	for (i = 0; i < type->cnt; i++) {
1666 		reg = &type->regions[i];
1667 		seq_printf(m, "%4d: ", i);
1668 		if (sizeof(phys_addr_t) == 4)
1669 			seq_printf(m, "0x%08lx..0x%08lx\n",
1670 				   (unsigned long)reg->base,
1671 				   (unsigned long)(reg->base + reg->size - 1));
1672 		else
1673 			seq_printf(m, "0x%016llx..0x%016llx\n",
1674 				   (unsigned long long)reg->base,
1675 				   (unsigned long long)(reg->base + reg->size - 1));
1676 
1677 	}
1678 	return 0;
1679 }
1680 
1681 static int memblock_debug_open(struct inode *inode, struct file *file)
1682 {
1683 	return single_open(file, memblock_debug_show, inode->i_private);
1684 }
1685 
1686 static const struct file_operations memblock_debug_fops = {
1687 	.open = memblock_debug_open,
1688 	.read = seq_read,
1689 	.llseek = seq_lseek,
1690 	.release = single_release,
1691 };
1692 
1693 static int __init memblock_init_debugfs(void)
1694 {
1695 	struct dentry *root = debugfs_create_dir("memblock", NULL);
1696 	if (!root)
1697 		return -ENXIO;
1698 	debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1699 	debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1700 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1701 	debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1702 #endif
1703 
1704 	return 0;
1705 }
1706 __initcall(memblock_init_debugfs);
1707 
1708 #endif /* CONFIG_DEBUG_FS */
1709