xref: /openbmc/linux/mm/memblock.c (revision f59f1caf)
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/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 	.memory.name		= "memory",
39 
40 	.reserved.regions	= memblock_reserved_init_regions,
41 	.reserved.cnt		= 1,	/* empty dummy entry */
42 	.reserved.max		= INIT_MEMBLOCK_REGIONS,
43 	.reserved.name		= "reserved",
44 
45 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
46 	.physmem.regions	= memblock_physmem_init_regions,
47 	.physmem.cnt		= 1,	/* empty dummy entry */
48 	.physmem.max		= INIT_PHYSMEM_REGIONS,
49 	.physmem.name		= "physmem",
50 #endif
51 
52 	.bottom_up		= false,
53 	.current_limit		= MEMBLOCK_ALLOC_ANYWHERE,
54 };
55 
56 int memblock_debug __initdata_memblock;
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 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
68 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
69 {
70 	return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
71 }
72 
73 /*
74  * Address comparison utilities
75  */
76 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
77 				       phys_addr_t base2, phys_addr_t size2)
78 {
79 	return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
80 }
81 
82 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
83 					phys_addr_t base, phys_addr_t size)
84 {
85 	unsigned long i;
86 
87 	for (i = 0; i < type->cnt; i++)
88 		if (memblock_addrs_overlap(base, size, type->regions[i].base,
89 					   type->regions[i].size))
90 			break;
91 	return i < type->cnt;
92 }
93 
94 /*
95  * __memblock_find_range_bottom_up - find free area utility in bottom-up
96  * @start: start of candidate range
97  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
98  * @size: size of free area to find
99  * @align: alignment of free area to find
100  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
101  * @flags: pick from blocks based on memory attributes
102  *
103  * Utility called from memblock_find_in_range_node(), find free area bottom-up.
104  *
105  * RETURNS:
106  * Found address on success, 0 on failure.
107  */
108 static phys_addr_t __init_memblock
109 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
110 				phys_addr_t size, phys_addr_t align, int nid,
111 				ulong flags)
112 {
113 	phys_addr_t this_start, this_end, cand;
114 	u64 i;
115 
116 	for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
117 		this_start = clamp(this_start, start, end);
118 		this_end = clamp(this_end, start, end);
119 
120 		cand = round_up(this_start, align);
121 		if (cand < this_end && this_end - cand >= size)
122 			return cand;
123 	}
124 
125 	return 0;
126 }
127 
128 /**
129  * __memblock_find_range_top_down - find free area utility, in top-down
130  * @start: start of candidate range
131  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
132  * @size: size of free area to find
133  * @align: alignment of free area to find
134  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
135  * @flags: pick from blocks based on memory attributes
136  *
137  * Utility called from memblock_find_in_range_node(), find free area top-down.
138  *
139  * RETURNS:
140  * Found address on success, 0 on failure.
141  */
142 static phys_addr_t __init_memblock
143 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
144 			       phys_addr_t size, phys_addr_t align, int nid,
145 			       ulong flags)
146 {
147 	phys_addr_t this_start, this_end, cand;
148 	u64 i;
149 
150 	for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
151 					NULL) {
152 		this_start = clamp(this_start, start, end);
153 		this_end = clamp(this_end, start, end);
154 
155 		if (this_end < size)
156 			continue;
157 
158 		cand = round_down(this_end - size, align);
159 		if (cand >= this_start)
160 			return cand;
161 	}
162 
163 	return 0;
164 }
165 
166 /**
167  * memblock_find_in_range_node - find free area in given range and node
168  * @size: size of free area to find
169  * @align: alignment of free area to find
170  * @start: start of candidate range
171  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
172  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
173  * @flags: pick from blocks based on memory attributes
174  *
175  * Find @size free area aligned to @align in the specified range and node.
176  *
177  * When allocation direction is bottom-up, the @start should be greater
178  * than the end of the kernel image. Otherwise, it will be trimmed. The
179  * reason is that we want the bottom-up allocation just near the kernel
180  * image so it is highly likely that the allocated memory and the kernel
181  * will reside in the same node.
182  *
183  * If bottom-up allocation failed, will try to allocate memory top-down.
184  *
185  * RETURNS:
186  * Found address on success, 0 on failure.
187  */
188 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
189 					phys_addr_t align, phys_addr_t start,
190 					phys_addr_t end, int nid, ulong flags)
191 {
192 	phys_addr_t kernel_end, ret;
193 
194 	/* pump up @end */
195 	if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
196 		end = memblock.current_limit;
197 
198 	/* avoid allocating the first page */
199 	start = max_t(phys_addr_t, start, PAGE_SIZE);
200 	end = max(start, end);
201 	kernel_end = __pa_symbol(_end);
202 
203 	/*
204 	 * try bottom-up allocation only when bottom-up mode
205 	 * is set and @end is above the kernel image.
206 	 */
207 	if (memblock_bottom_up() && end > kernel_end) {
208 		phys_addr_t bottom_up_start;
209 
210 		/* make sure we will allocate above the kernel */
211 		bottom_up_start = max(start, kernel_end);
212 
213 		/* ok, try bottom-up allocation first */
214 		ret = __memblock_find_range_bottom_up(bottom_up_start, end,
215 						      size, align, nid, flags);
216 		if (ret)
217 			return ret;
218 
219 		/*
220 		 * we always limit bottom-up allocation above the kernel,
221 		 * but top-down allocation doesn't have the limit, so
222 		 * retrying top-down allocation may succeed when bottom-up
223 		 * allocation failed.
224 		 *
225 		 * bottom-up allocation is expected to be fail very rarely,
226 		 * so we use WARN_ONCE() here to see the stack trace if
227 		 * fail happens.
228 		 */
229 		WARN_ONCE(1, "memblock: bottom-up allocation failed, memory hotunplug may be affected\n");
230 	}
231 
232 	return __memblock_find_range_top_down(start, end, size, align, nid,
233 					      flags);
234 }
235 
236 /**
237  * memblock_find_in_range - find free area in given range
238  * @start: start of candidate range
239  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
240  * @size: size of free area to find
241  * @align: alignment of free area to find
242  *
243  * Find @size free area aligned to @align in the specified range.
244  *
245  * RETURNS:
246  * Found address on success, 0 on failure.
247  */
248 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
249 					phys_addr_t end, phys_addr_t size,
250 					phys_addr_t align)
251 {
252 	phys_addr_t ret;
253 	ulong flags = choose_memblock_flags();
254 
255 again:
256 	ret = memblock_find_in_range_node(size, align, start, end,
257 					    NUMA_NO_NODE, flags);
258 
259 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
260 		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
261 			&size);
262 		flags &= ~MEMBLOCK_MIRROR;
263 		goto again;
264 	}
265 
266 	return ret;
267 }
268 
269 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
270 {
271 	type->total_size -= type->regions[r].size;
272 	memmove(&type->regions[r], &type->regions[r + 1],
273 		(type->cnt - (r + 1)) * sizeof(type->regions[r]));
274 	type->cnt--;
275 
276 	/* Special case for empty arrays */
277 	if (type->cnt == 0) {
278 		WARN_ON(type->total_size != 0);
279 		type->cnt = 1;
280 		type->regions[0].base = 0;
281 		type->regions[0].size = 0;
282 		type->regions[0].flags = 0;
283 		memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
284 	}
285 }
286 
287 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
288 /**
289  * Discard memory and reserved arrays if they were allocated
290  */
291 void __init memblock_discard(void)
292 {
293 	phys_addr_t addr, size;
294 
295 	if (memblock.reserved.regions != memblock_reserved_init_regions) {
296 		addr = __pa(memblock.reserved.regions);
297 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
298 				  memblock.reserved.max);
299 		__memblock_free_late(addr, size);
300 	}
301 
302 	if (memblock.memory.regions != memblock_memory_init_regions) {
303 		addr = __pa(memblock.memory.regions);
304 		size = PAGE_ALIGN(sizeof(struct memblock_region) *
305 				  memblock.memory.max);
306 		__memblock_free_late(addr, size);
307 	}
308 }
309 #endif
310 
311 /**
312  * memblock_double_array - double the size of the memblock regions array
313  * @type: memblock type of the regions array being doubled
314  * @new_area_start: starting address of memory range to avoid overlap with
315  * @new_area_size: size of memory range to avoid overlap with
316  *
317  * Double the size of the @type regions array. If memblock is being used to
318  * allocate memory for a new reserved regions array and there is a previously
319  * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
320  * waiting to be reserved, ensure the memory used by the new array does
321  * not overlap.
322  *
323  * RETURNS:
324  * 0 on success, -1 on failure.
325  */
326 static int __init_memblock memblock_double_array(struct memblock_type *type,
327 						phys_addr_t new_area_start,
328 						phys_addr_t new_area_size)
329 {
330 	struct memblock_region *new_array, *old_array;
331 	phys_addr_t old_alloc_size, new_alloc_size;
332 	phys_addr_t old_size, new_size, addr;
333 	int use_slab = slab_is_available();
334 	int *in_slab;
335 
336 	/* We don't allow resizing until we know about the reserved regions
337 	 * of memory that aren't suitable for allocation
338 	 */
339 	if (!memblock_can_resize)
340 		return -1;
341 
342 	/* Calculate new doubled size */
343 	old_size = type->max * sizeof(struct memblock_region);
344 	new_size = old_size << 1;
345 	/*
346 	 * We need to allocated new one align to PAGE_SIZE,
347 	 *   so we can free them completely later.
348 	 */
349 	old_alloc_size = PAGE_ALIGN(old_size);
350 	new_alloc_size = PAGE_ALIGN(new_size);
351 
352 	/* Retrieve the slab flag */
353 	if (type == &memblock.memory)
354 		in_slab = &memblock_memory_in_slab;
355 	else
356 		in_slab = &memblock_reserved_in_slab;
357 
358 	/* Try to find some space for it.
359 	 *
360 	 * WARNING: We assume that either slab_is_available() and we use it or
361 	 * we use MEMBLOCK for allocations. That means that this is unsafe to
362 	 * use when bootmem is currently active (unless bootmem itself is
363 	 * implemented on top of MEMBLOCK which isn't the case yet)
364 	 *
365 	 * This should however not be an issue for now, as we currently only
366 	 * call into MEMBLOCK while it's still active, or much later when slab
367 	 * is active for memory hotplug operations
368 	 */
369 	if (use_slab) {
370 		new_array = kmalloc(new_size, GFP_KERNEL);
371 		addr = new_array ? __pa(new_array) : 0;
372 	} else {
373 		/* only exclude range when trying to double reserved.regions */
374 		if (type != &memblock.reserved)
375 			new_area_start = new_area_size = 0;
376 
377 		addr = memblock_find_in_range(new_area_start + new_area_size,
378 						memblock.current_limit,
379 						new_alloc_size, PAGE_SIZE);
380 		if (!addr && new_area_size)
381 			addr = memblock_find_in_range(0,
382 				min(new_area_start, memblock.current_limit),
383 				new_alloc_size, PAGE_SIZE);
384 
385 		new_array = addr ? __va(addr) : NULL;
386 	}
387 	if (!addr) {
388 		pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
389 		       type->name, type->max, type->max * 2);
390 		return -1;
391 	}
392 
393 	memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
394 			type->name, type->max * 2, (u64)addr,
395 			(u64)addr + new_size - 1);
396 
397 	/*
398 	 * Found space, we now need to move the array over before we add the
399 	 * reserved region since it may be our reserved array itself that is
400 	 * full.
401 	 */
402 	memcpy(new_array, type->regions, old_size);
403 	memset(new_array + type->max, 0, old_size);
404 	old_array = type->regions;
405 	type->regions = new_array;
406 	type->max <<= 1;
407 
408 	/* Free old array. We needn't free it if the array is the static one */
409 	if (*in_slab)
410 		kfree(old_array);
411 	else if (old_array != memblock_memory_init_regions &&
412 		 old_array != memblock_reserved_init_regions)
413 		memblock_free(__pa(old_array), old_alloc_size);
414 
415 	/*
416 	 * Reserve the new array if that comes from the memblock.  Otherwise, we
417 	 * needn't do it
418 	 */
419 	if (!use_slab)
420 		BUG_ON(memblock_reserve(addr, new_alloc_size));
421 
422 	/* Update slab flag */
423 	*in_slab = use_slab;
424 
425 	return 0;
426 }
427 
428 /**
429  * memblock_merge_regions - merge neighboring compatible regions
430  * @type: memblock type to scan
431  *
432  * Scan @type and merge neighboring compatible regions.
433  */
434 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
435 {
436 	int i = 0;
437 
438 	/* cnt never goes below 1 */
439 	while (i < type->cnt - 1) {
440 		struct memblock_region *this = &type->regions[i];
441 		struct memblock_region *next = &type->regions[i + 1];
442 
443 		if (this->base + this->size != next->base ||
444 		    memblock_get_region_node(this) !=
445 		    memblock_get_region_node(next) ||
446 		    this->flags != next->flags) {
447 			BUG_ON(this->base + this->size > next->base);
448 			i++;
449 			continue;
450 		}
451 
452 		this->size += next->size;
453 		/* move forward from next + 1, index of which is i + 2 */
454 		memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
455 		type->cnt--;
456 	}
457 }
458 
459 /**
460  * memblock_insert_region - insert new memblock region
461  * @type:	memblock type to insert into
462  * @idx:	index for the insertion point
463  * @base:	base address of the new region
464  * @size:	size of the new region
465  * @nid:	node id of the new region
466  * @flags:	flags of the new region
467  *
468  * Insert new memblock region [@base,@base+@size) into @type at @idx.
469  * @type must already have extra room to accommodate the new region.
470  */
471 static void __init_memblock memblock_insert_region(struct memblock_type *type,
472 						   int idx, phys_addr_t base,
473 						   phys_addr_t size,
474 						   int nid, unsigned long flags)
475 {
476 	struct memblock_region *rgn = &type->regions[idx];
477 
478 	BUG_ON(type->cnt >= type->max);
479 	memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
480 	rgn->base = base;
481 	rgn->size = size;
482 	rgn->flags = flags;
483 	memblock_set_region_node(rgn, nid);
484 	type->cnt++;
485 	type->total_size += size;
486 }
487 
488 /**
489  * memblock_add_range - add new memblock region
490  * @type: memblock type to add new region into
491  * @base: base address of the new region
492  * @size: size of the new region
493  * @nid: nid of the new region
494  * @flags: flags of the new region
495  *
496  * Add new memblock region [@base,@base+@size) into @type.  The new region
497  * is allowed to overlap with existing ones - overlaps don't affect already
498  * existing regions.  @type is guaranteed to be minimal (all neighbouring
499  * compatible regions are merged) after the addition.
500  *
501  * RETURNS:
502  * 0 on success, -errno on failure.
503  */
504 int __init_memblock memblock_add_range(struct memblock_type *type,
505 				phys_addr_t base, phys_addr_t size,
506 				int nid, unsigned long flags)
507 {
508 	bool insert = false;
509 	phys_addr_t obase = base;
510 	phys_addr_t end = base + memblock_cap_size(base, &size);
511 	int idx, nr_new;
512 	struct memblock_region *rgn;
513 
514 	if (!size)
515 		return 0;
516 
517 	/* special case for empty array */
518 	if (type->regions[0].size == 0) {
519 		WARN_ON(type->cnt != 1 || type->total_size);
520 		type->regions[0].base = base;
521 		type->regions[0].size = size;
522 		type->regions[0].flags = flags;
523 		memblock_set_region_node(&type->regions[0], nid);
524 		type->total_size = size;
525 		return 0;
526 	}
527 repeat:
528 	/*
529 	 * The following is executed twice.  Once with %false @insert and
530 	 * then with %true.  The first counts the number of regions needed
531 	 * to accommodate the new area.  The second actually inserts them.
532 	 */
533 	base = obase;
534 	nr_new = 0;
535 
536 	for_each_memblock_type(idx, type, rgn) {
537 		phys_addr_t rbase = rgn->base;
538 		phys_addr_t rend = rbase + rgn->size;
539 
540 		if (rbase >= end)
541 			break;
542 		if (rend <= base)
543 			continue;
544 		/*
545 		 * @rgn overlaps.  If it separates the lower part of new
546 		 * area, insert that portion.
547 		 */
548 		if (rbase > base) {
549 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
550 			WARN_ON(nid != memblock_get_region_node(rgn));
551 #endif
552 			WARN_ON(flags != rgn->flags);
553 			nr_new++;
554 			if (insert)
555 				memblock_insert_region(type, idx++, base,
556 						       rbase - base, nid,
557 						       flags);
558 		}
559 		/* area below @rend is dealt with, forget about it */
560 		base = min(rend, end);
561 	}
562 
563 	/* insert the remaining portion */
564 	if (base < end) {
565 		nr_new++;
566 		if (insert)
567 			memblock_insert_region(type, idx, base, end - base,
568 					       nid, flags);
569 	}
570 
571 	if (!nr_new)
572 		return 0;
573 
574 	/*
575 	 * If this was the first round, resize array and repeat for actual
576 	 * insertions; otherwise, merge and return.
577 	 */
578 	if (!insert) {
579 		while (type->cnt + nr_new > type->max)
580 			if (memblock_double_array(type, obase, size) < 0)
581 				return -ENOMEM;
582 		insert = true;
583 		goto repeat;
584 	} else {
585 		memblock_merge_regions(type);
586 		return 0;
587 	}
588 }
589 
590 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
591 				       int nid)
592 {
593 	return memblock_add_range(&memblock.memory, base, size, nid, 0);
594 }
595 
596 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
597 {
598 	phys_addr_t end = base + size - 1;
599 
600 	memblock_dbg("memblock_add: [%pa-%pa] %pF\n",
601 		     &base, &end, (void *)_RET_IP_);
602 
603 	return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
604 }
605 
606 /**
607  * memblock_isolate_range - isolate given range into disjoint memblocks
608  * @type: memblock type to isolate range for
609  * @base: base of range to isolate
610  * @size: size of range to isolate
611  * @start_rgn: out parameter for the start of isolated region
612  * @end_rgn: out parameter for the end of isolated region
613  *
614  * Walk @type and ensure that regions don't cross the boundaries defined by
615  * [@base,@base+@size).  Crossing regions are split at the boundaries,
616  * which may create at most two more regions.  The index of the first
617  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
618  *
619  * RETURNS:
620  * 0 on success, -errno on failure.
621  */
622 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
623 					phys_addr_t base, phys_addr_t size,
624 					int *start_rgn, int *end_rgn)
625 {
626 	phys_addr_t end = base + memblock_cap_size(base, &size);
627 	int idx;
628 	struct memblock_region *rgn;
629 
630 	*start_rgn = *end_rgn = 0;
631 
632 	if (!size)
633 		return 0;
634 
635 	/* we'll create at most two more regions */
636 	while (type->cnt + 2 > type->max)
637 		if (memblock_double_array(type, base, size) < 0)
638 			return -ENOMEM;
639 
640 	for_each_memblock_type(idx, type, rgn) {
641 		phys_addr_t rbase = rgn->base;
642 		phys_addr_t rend = rbase + rgn->size;
643 
644 		if (rbase >= end)
645 			break;
646 		if (rend <= base)
647 			continue;
648 
649 		if (rbase < base) {
650 			/*
651 			 * @rgn intersects from below.  Split and continue
652 			 * to process the next region - the new top half.
653 			 */
654 			rgn->base = base;
655 			rgn->size -= base - rbase;
656 			type->total_size -= base - rbase;
657 			memblock_insert_region(type, idx, rbase, base - rbase,
658 					       memblock_get_region_node(rgn),
659 					       rgn->flags);
660 		} else if (rend > end) {
661 			/*
662 			 * @rgn intersects from above.  Split and redo the
663 			 * current region - the new bottom half.
664 			 */
665 			rgn->base = end;
666 			rgn->size -= end - rbase;
667 			type->total_size -= end - rbase;
668 			memblock_insert_region(type, idx--, rbase, end - rbase,
669 					       memblock_get_region_node(rgn),
670 					       rgn->flags);
671 		} else {
672 			/* @rgn is fully contained, record it */
673 			if (!*end_rgn)
674 				*start_rgn = idx;
675 			*end_rgn = idx + 1;
676 		}
677 	}
678 
679 	return 0;
680 }
681 
682 static int __init_memblock memblock_remove_range(struct memblock_type *type,
683 					  phys_addr_t base, phys_addr_t size)
684 {
685 	int start_rgn, end_rgn;
686 	int i, ret;
687 
688 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
689 	if (ret)
690 		return ret;
691 
692 	for (i = end_rgn - 1; i >= start_rgn; i--)
693 		memblock_remove_region(type, i);
694 	return 0;
695 }
696 
697 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
698 {
699 	return memblock_remove_range(&memblock.memory, base, size);
700 }
701 
702 
703 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
704 {
705 	phys_addr_t end = base + size - 1;
706 
707 	memblock_dbg("   memblock_free: [%pa-%pa] %pF\n",
708 		     &base, &end, (void *)_RET_IP_);
709 
710 	kmemleak_free_part_phys(base, size);
711 	return memblock_remove_range(&memblock.reserved, base, size);
712 }
713 
714 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
715 {
716 	phys_addr_t end = base + size - 1;
717 
718 	memblock_dbg("memblock_reserve: [%pa-%pa] %pF\n",
719 		     &base, &end, (void *)_RET_IP_);
720 
721 	return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
722 }
723 
724 /**
725  *
726  * This function isolates region [@base, @base + @size), and sets/clears flag
727  *
728  * Return 0 on success, -errno on failure.
729  */
730 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
731 				phys_addr_t size, int set, int flag)
732 {
733 	struct memblock_type *type = &memblock.memory;
734 	int i, ret, start_rgn, end_rgn;
735 
736 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
737 	if (ret)
738 		return ret;
739 
740 	for (i = start_rgn; i < end_rgn; i++)
741 		if (set)
742 			memblock_set_region_flags(&type->regions[i], flag);
743 		else
744 			memblock_clear_region_flags(&type->regions[i], flag);
745 
746 	memblock_merge_regions(type);
747 	return 0;
748 }
749 
750 /**
751  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
752  * @base: the base phys addr of the region
753  * @size: the size of the region
754  *
755  * Return 0 on success, -errno on failure.
756  */
757 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
758 {
759 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
760 }
761 
762 /**
763  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
764  * @base: the base phys addr of the region
765  * @size: the size of the region
766  *
767  * Return 0 on success, -errno on failure.
768  */
769 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
770 {
771 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
772 }
773 
774 /**
775  * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
776  * @base: the base phys addr of the region
777  * @size: the size of the region
778  *
779  * Return 0 on success, -errno on failure.
780  */
781 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
782 {
783 	system_has_some_mirror = true;
784 
785 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
786 }
787 
788 /**
789  * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
790  * @base: the base phys addr of the region
791  * @size: the size of the region
792  *
793  * Return 0 on success, -errno on failure.
794  */
795 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
796 {
797 	return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
798 }
799 
800 /**
801  * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
802  * @base: the base phys addr of the region
803  * @size: the size of the region
804  *
805  * Return 0 on success, -errno on failure.
806  */
807 int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
808 {
809 	return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
810 }
811 
812 /**
813  * __next_reserved_mem_region - next function for for_each_reserved_region()
814  * @idx: pointer to u64 loop variable
815  * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
816  * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
817  *
818  * Iterate over all reserved memory regions.
819  */
820 void __init_memblock __next_reserved_mem_region(u64 *idx,
821 					   phys_addr_t *out_start,
822 					   phys_addr_t *out_end)
823 {
824 	struct memblock_type *type = &memblock.reserved;
825 
826 	if (*idx < type->cnt) {
827 		struct memblock_region *r = &type->regions[*idx];
828 		phys_addr_t base = r->base;
829 		phys_addr_t size = r->size;
830 
831 		if (out_start)
832 			*out_start = base;
833 		if (out_end)
834 			*out_end = base + size - 1;
835 
836 		*idx += 1;
837 		return;
838 	}
839 
840 	/* signal end of iteration */
841 	*idx = ULLONG_MAX;
842 }
843 
844 /**
845  * __next__mem_range - next function for for_each_free_mem_range() etc.
846  * @idx: pointer to u64 loop variable
847  * @nid: node selector, %NUMA_NO_NODE for all nodes
848  * @flags: pick from blocks based on memory attributes
849  * @type_a: pointer to memblock_type from where the range is taken
850  * @type_b: pointer to memblock_type which excludes memory from being taken
851  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
852  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
853  * @out_nid: ptr to int for nid of the range, can be %NULL
854  *
855  * Find the first area from *@idx which matches @nid, fill the out
856  * parameters, and update *@idx for the next iteration.  The lower 32bit of
857  * *@idx contains index into type_a and the upper 32bit indexes the
858  * areas before each region in type_b.	For example, if type_b regions
859  * look like the following,
860  *
861  *	0:[0-16), 1:[32-48), 2:[128-130)
862  *
863  * The upper 32bit indexes the following regions.
864  *
865  *	0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
866  *
867  * As both region arrays are sorted, the function advances the two indices
868  * in lockstep and returns each intersection.
869  */
870 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
871 				      struct memblock_type *type_a,
872 				      struct memblock_type *type_b,
873 				      phys_addr_t *out_start,
874 				      phys_addr_t *out_end, int *out_nid)
875 {
876 	int idx_a = *idx & 0xffffffff;
877 	int idx_b = *idx >> 32;
878 
879 	if (WARN_ONCE(nid == MAX_NUMNODES,
880 	"Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
881 		nid = NUMA_NO_NODE;
882 
883 	for (; idx_a < type_a->cnt; idx_a++) {
884 		struct memblock_region *m = &type_a->regions[idx_a];
885 
886 		phys_addr_t m_start = m->base;
887 		phys_addr_t m_end = m->base + m->size;
888 		int	    m_nid = memblock_get_region_node(m);
889 
890 		/* only memory regions are associated with nodes, check it */
891 		if (nid != NUMA_NO_NODE && nid != m_nid)
892 			continue;
893 
894 		/* skip hotpluggable memory regions if needed */
895 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
896 			continue;
897 
898 		/* if we want mirror memory skip non-mirror memory regions */
899 		if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
900 			continue;
901 
902 		/* skip nomap memory unless we were asked for it explicitly */
903 		if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
904 			continue;
905 
906 		if (!type_b) {
907 			if (out_start)
908 				*out_start = m_start;
909 			if (out_end)
910 				*out_end = m_end;
911 			if (out_nid)
912 				*out_nid = m_nid;
913 			idx_a++;
914 			*idx = (u32)idx_a | (u64)idx_b << 32;
915 			return;
916 		}
917 
918 		/* scan areas before each reservation */
919 		for (; idx_b < type_b->cnt + 1; idx_b++) {
920 			struct memblock_region *r;
921 			phys_addr_t r_start;
922 			phys_addr_t r_end;
923 
924 			r = &type_b->regions[idx_b];
925 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
926 			r_end = idx_b < type_b->cnt ?
927 				r->base : ULLONG_MAX;
928 
929 			/*
930 			 * if idx_b advanced past idx_a,
931 			 * break out to advance idx_a
932 			 */
933 			if (r_start >= m_end)
934 				break;
935 			/* if the two regions intersect, we're done */
936 			if (m_start < r_end) {
937 				if (out_start)
938 					*out_start =
939 						max(m_start, r_start);
940 				if (out_end)
941 					*out_end = min(m_end, r_end);
942 				if (out_nid)
943 					*out_nid = m_nid;
944 				/*
945 				 * The region which ends first is
946 				 * advanced for the next iteration.
947 				 */
948 				if (m_end <= r_end)
949 					idx_a++;
950 				else
951 					idx_b++;
952 				*idx = (u32)idx_a | (u64)idx_b << 32;
953 				return;
954 			}
955 		}
956 	}
957 
958 	/* signal end of iteration */
959 	*idx = ULLONG_MAX;
960 }
961 
962 /**
963  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
964  *
965  * Finds the next range from type_a which is not marked as unsuitable
966  * in type_b.
967  *
968  * @idx: pointer to u64 loop variable
969  * @nid: node selector, %NUMA_NO_NODE for all nodes
970  * @flags: pick from blocks based on memory attributes
971  * @type_a: pointer to memblock_type from where the range is taken
972  * @type_b: pointer to memblock_type which excludes memory from being taken
973  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
974  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
975  * @out_nid: ptr to int for nid of the range, can be %NULL
976  *
977  * Reverse of __next_mem_range().
978  */
979 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
980 					  struct memblock_type *type_a,
981 					  struct memblock_type *type_b,
982 					  phys_addr_t *out_start,
983 					  phys_addr_t *out_end, int *out_nid)
984 {
985 	int idx_a = *idx & 0xffffffff;
986 	int idx_b = *idx >> 32;
987 
988 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
989 		nid = NUMA_NO_NODE;
990 
991 	if (*idx == (u64)ULLONG_MAX) {
992 		idx_a = type_a->cnt - 1;
993 		if (type_b != NULL)
994 			idx_b = type_b->cnt;
995 		else
996 			idx_b = 0;
997 	}
998 
999 	for (; idx_a >= 0; idx_a--) {
1000 		struct memblock_region *m = &type_a->regions[idx_a];
1001 
1002 		phys_addr_t m_start = m->base;
1003 		phys_addr_t m_end = m->base + m->size;
1004 		int m_nid = memblock_get_region_node(m);
1005 
1006 		/* only memory regions are associated with nodes, check it */
1007 		if (nid != NUMA_NO_NODE && nid != m_nid)
1008 			continue;
1009 
1010 		/* skip hotpluggable memory regions if needed */
1011 		if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1012 			continue;
1013 
1014 		/* if we want mirror memory skip non-mirror memory regions */
1015 		if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1016 			continue;
1017 
1018 		/* skip nomap memory unless we were asked for it explicitly */
1019 		if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1020 			continue;
1021 
1022 		if (!type_b) {
1023 			if (out_start)
1024 				*out_start = m_start;
1025 			if (out_end)
1026 				*out_end = m_end;
1027 			if (out_nid)
1028 				*out_nid = m_nid;
1029 			idx_a--;
1030 			*idx = (u32)idx_a | (u64)idx_b << 32;
1031 			return;
1032 		}
1033 
1034 		/* scan areas before each reservation */
1035 		for (; idx_b >= 0; idx_b--) {
1036 			struct memblock_region *r;
1037 			phys_addr_t r_start;
1038 			phys_addr_t r_end;
1039 
1040 			r = &type_b->regions[idx_b];
1041 			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1042 			r_end = idx_b < type_b->cnt ?
1043 				r->base : ULLONG_MAX;
1044 			/*
1045 			 * if idx_b advanced past idx_a,
1046 			 * break out to advance idx_a
1047 			 */
1048 
1049 			if (r_end <= m_start)
1050 				break;
1051 			/* if the two regions intersect, we're done */
1052 			if (m_end > r_start) {
1053 				if (out_start)
1054 					*out_start = max(m_start, r_start);
1055 				if (out_end)
1056 					*out_end = min(m_end, r_end);
1057 				if (out_nid)
1058 					*out_nid = m_nid;
1059 				if (m_start >= r_start)
1060 					idx_a--;
1061 				else
1062 					idx_b--;
1063 				*idx = (u32)idx_a | (u64)idx_b << 32;
1064 				return;
1065 			}
1066 		}
1067 	}
1068 	/* signal end of iteration */
1069 	*idx = ULLONG_MAX;
1070 }
1071 
1072 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1073 /*
1074  * Common iterator interface used to define for_each_mem_range().
1075  */
1076 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1077 				unsigned long *out_start_pfn,
1078 				unsigned long *out_end_pfn, int *out_nid)
1079 {
1080 	struct memblock_type *type = &memblock.memory;
1081 	struct memblock_region *r;
1082 
1083 	while (++*idx < type->cnt) {
1084 		r = &type->regions[*idx];
1085 
1086 		if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1087 			continue;
1088 		if (nid == MAX_NUMNODES || nid == r->nid)
1089 			break;
1090 	}
1091 	if (*idx >= type->cnt) {
1092 		*idx = -1;
1093 		return;
1094 	}
1095 
1096 	if (out_start_pfn)
1097 		*out_start_pfn = PFN_UP(r->base);
1098 	if (out_end_pfn)
1099 		*out_end_pfn = PFN_DOWN(r->base + r->size);
1100 	if (out_nid)
1101 		*out_nid = r->nid;
1102 }
1103 
1104 /**
1105  * memblock_set_node - set node ID on memblock regions
1106  * @base: base of area to set node ID for
1107  * @size: size of area to set node ID for
1108  * @type: memblock type to set node ID for
1109  * @nid: node ID to set
1110  *
1111  * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1112  * Regions which cross the area boundaries are split as necessary.
1113  *
1114  * RETURNS:
1115  * 0 on success, -errno on failure.
1116  */
1117 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1118 				      struct memblock_type *type, int nid)
1119 {
1120 	int start_rgn, end_rgn;
1121 	int i, ret;
1122 
1123 	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1124 	if (ret)
1125 		return ret;
1126 
1127 	for (i = start_rgn; i < end_rgn; i++)
1128 		memblock_set_region_node(&type->regions[i], nid);
1129 
1130 	memblock_merge_regions(type);
1131 	return 0;
1132 }
1133 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1134 
1135 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1136 					phys_addr_t align, phys_addr_t start,
1137 					phys_addr_t end, int nid, ulong flags)
1138 {
1139 	phys_addr_t found;
1140 
1141 	if (!align)
1142 		align = SMP_CACHE_BYTES;
1143 
1144 	found = memblock_find_in_range_node(size, align, start, end, nid,
1145 					    flags);
1146 	if (found && !memblock_reserve(found, size)) {
1147 		/*
1148 		 * The min_count is set to 0 so that memblock allocations are
1149 		 * never reported as leaks.
1150 		 */
1151 		kmemleak_alloc_phys(found, size, 0, 0);
1152 		return found;
1153 	}
1154 	return 0;
1155 }
1156 
1157 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1158 					phys_addr_t start, phys_addr_t end,
1159 					ulong flags)
1160 {
1161 	return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1162 					flags);
1163 }
1164 
1165 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1166 					phys_addr_t align, phys_addr_t max_addr,
1167 					int nid, ulong flags)
1168 {
1169 	return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1170 }
1171 
1172 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1173 {
1174 	ulong flags = choose_memblock_flags();
1175 	phys_addr_t ret;
1176 
1177 again:
1178 	ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1179 				      nid, flags);
1180 
1181 	if (!ret && (flags & MEMBLOCK_MIRROR)) {
1182 		flags &= ~MEMBLOCK_MIRROR;
1183 		goto again;
1184 	}
1185 	return ret;
1186 }
1187 
1188 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1189 {
1190 	return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1191 				       MEMBLOCK_NONE);
1192 }
1193 
1194 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1195 {
1196 	phys_addr_t alloc;
1197 
1198 	alloc = __memblock_alloc_base(size, align, max_addr);
1199 
1200 	if (alloc == 0)
1201 		panic("ERROR: Failed to allocate %pa bytes below %pa.\n",
1202 		      &size, &max_addr);
1203 
1204 	return alloc;
1205 }
1206 
1207 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1208 {
1209 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1210 }
1211 
1212 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1213 {
1214 	phys_addr_t res = memblock_alloc_nid(size, align, nid);
1215 
1216 	if (res)
1217 		return res;
1218 	return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1219 }
1220 
1221 /**
1222  * memblock_virt_alloc_internal - allocate boot memory block
1223  * @size: size of memory block to be allocated in bytes
1224  * @align: alignment of the region and block's size
1225  * @min_addr: the lower bound of the memory region to allocate (phys address)
1226  * @max_addr: the upper bound of the memory region to allocate (phys address)
1227  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1228  *
1229  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1230  * will fall back to memory below @min_addr. Also, allocation may fall back
1231  * to any node in the system if the specified node can not
1232  * hold the requested memory.
1233  *
1234  * The allocation is performed from memory region limited by
1235  * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1236  *
1237  * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1238  *
1239  * The phys address of allocated boot memory block is converted to virtual and
1240  * allocated memory is reset to 0.
1241  *
1242  * In addition, function sets the min_count to 0 using kmemleak_alloc for
1243  * allocated boot memory block, so that it is never reported as leaks.
1244  *
1245  * RETURNS:
1246  * Virtual address of allocated memory block on success, NULL on failure.
1247  */
1248 static void * __init memblock_virt_alloc_internal(
1249 				phys_addr_t size, phys_addr_t align,
1250 				phys_addr_t min_addr, phys_addr_t max_addr,
1251 				int nid)
1252 {
1253 	phys_addr_t alloc;
1254 	void *ptr;
1255 	ulong flags = choose_memblock_flags();
1256 
1257 	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1258 		nid = NUMA_NO_NODE;
1259 
1260 	/*
1261 	 * Detect any accidental use of these APIs after slab is ready, as at
1262 	 * this moment memblock may be deinitialized already and its
1263 	 * internal data may be destroyed (after execution of free_all_bootmem)
1264 	 */
1265 	if (WARN_ON_ONCE(slab_is_available()))
1266 		return kzalloc_node(size, GFP_NOWAIT, nid);
1267 
1268 	if (!align)
1269 		align = SMP_CACHE_BYTES;
1270 
1271 	if (max_addr > memblock.current_limit)
1272 		max_addr = memblock.current_limit;
1273 again:
1274 	alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1275 					    nid, flags);
1276 	if (alloc && !memblock_reserve(alloc, size))
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 && !memblock_reserve(alloc, size))
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 	ptr = phys_to_virt(alloc);
1302 
1303 	/*
1304 	 * The min_count is set to 0 so that bootmem allocated blocks
1305 	 * are never reported as leaks. This is because many of these blocks
1306 	 * are only referred via the physical address which is not
1307 	 * looked up by kmemleak.
1308 	 */
1309 	kmemleak_alloc(ptr, size, 0, 0);
1310 
1311 	return ptr;
1312 }
1313 
1314 /**
1315  * memblock_virt_alloc_try_nid_raw - allocate boot memory block without zeroing
1316  * memory and without panicking
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 function, provides additional debug information (including caller
1327  * info), if enabled. Does not zero allocated memory, does not panic if request
1328  * cannot be satisfied.
1329  *
1330  * RETURNS:
1331  * Virtual address of allocated memory block on success, NULL on failure.
1332  */
1333 void * __init memblock_virt_alloc_try_nid_raw(
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 	void *ptr;
1339 
1340 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1341 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1342 		     (u64)max_addr, (void *)_RET_IP_);
1343 
1344 	ptr = memblock_virt_alloc_internal(size, align,
1345 					   min_addr, max_addr, nid);
1346 #ifdef CONFIG_DEBUG_VM
1347 	if (ptr && size > 0)
1348 		memset(ptr, 0xff, size);
1349 #endif
1350 	return ptr;
1351 }
1352 
1353 /**
1354  * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1355  * @size: size of memory block to be allocated in bytes
1356  * @align: alignment of the region and block's size
1357  * @min_addr: the lower bound of the memory region from where the allocation
1358  *	  is preferred (phys address)
1359  * @max_addr: the upper bound of the memory region from where the allocation
1360  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1361  *	      allocate only from memory limited by memblock.current_limit value
1362  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1363  *
1364  * Public function, provides additional debug information (including caller
1365  * info), if enabled. This function zeroes the allocated memory.
1366  *
1367  * RETURNS:
1368  * Virtual address of allocated memory block on success, NULL on failure.
1369  */
1370 void * __init memblock_virt_alloc_try_nid_nopanic(
1371 				phys_addr_t size, phys_addr_t align,
1372 				phys_addr_t min_addr, phys_addr_t max_addr,
1373 				int nid)
1374 {
1375 	void *ptr;
1376 
1377 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1378 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1379 		     (u64)max_addr, (void *)_RET_IP_);
1380 
1381 	ptr = memblock_virt_alloc_internal(size, align,
1382 					   min_addr, max_addr, nid);
1383 	if (ptr)
1384 		memset(ptr, 0, size);
1385 	return ptr;
1386 }
1387 
1388 /**
1389  * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1390  * @size: size of memory block to be allocated in bytes
1391  * @align: alignment of the region and block's size
1392  * @min_addr: the lower bound of the memory region from where the allocation
1393  *	  is preferred (phys address)
1394  * @max_addr: the upper bound of the memory region from where the allocation
1395  *	      is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1396  *	      allocate only from memory limited by memblock.current_limit value
1397  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1398  *
1399  * Public panicking version of memblock_virt_alloc_try_nid_nopanic()
1400  * which provides debug information (including caller info), if enabled,
1401  * and panics if the request can not be satisfied.
1402  *
1403  * RETURNS:
1404  * Virtual address of allocated memory block on success, NULL on failure.
1405  */
1406 void * __init memblock_virt_alloc_try_nid(
1407 			phys_addr_t size, phys_addr_t align,
1408 			phys_addr_t min_addr, phys_addr_t max_addr,
1409 			int nid)
1410 {
1411 	void *ptr;
1412 
1413 	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1414 		     __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1415 		     (u64)max_addr, (void *)_RET_IP_);
1416 	ptr = memblock_virt_alloc_internal(size, align,
1417 					   min_addr, max_addr, nid);
1418 	if (ptr) {
1419 		memset(ptr, 0, size);
1420 		return ptr;
1421 	}
1422 
1423 	panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1424 	      __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1425 	      (u64)max_addr);
1426 	return NULL;
1427 }
1428 
1429 /**
1430  * __memblock_free_early - free boot memory block
1431  * @base: phys starting address of the  boot memory block
1432  * @size: size of the boot memory block in bytes
1433  *
1434  * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1435  * The freeing memory will not be released to the buddy allocator.
1436  */
1437 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1438 {
1439 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1440 		     __func__, (u64)base, (u64)base + size - 1,
1441 		     (void *)_RET_IP_);
1442 	kmemleak_free_part_phys(base, size);
1443 	memblock_remove_range(&memblock.reserved, base, size);
1444 }
1445 
1446 /*
1447  * __memblock_free_late - free bootmem block pages directly to buddy allocator
1448  * @addr: phys starting address of the  boot memory block
1449  * @size: size of the boot memory block in bytes
1450  *
1451  * This is only useful when the bootmem allocator has already been torn
1452  * down, but we are still initializing the system.  Pages are released directly
1453  * to the buddy allocator, no bootmem metadata is updated because it is gone.
1454  */
1455 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1456 {
1457 	u64 cursor, end;
1458 
1459 	memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1460 		     __func__, (u64)base, (u64)base + size - 1,
1461 		     (void *)_RET_IP_);
1462 	kmemleak_free_part_phys(base, size);
1463 	cursor = PFN_UP(base);
1464 	end = PFN_DOWN(base + size);
1465 
1466 	for (; cursor < end; cursor++) {
1467 		__free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1468 		totalram_pages++;
1469 	}
1470 }
1471 
1472 /*
1473  * Remaining API functions
1474  */
1475 
1476 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1477 {
1478 	return memblock.memory.total_size;
1479 }
1480 
1481 phys_addr_t __init_memblock memblock_reserved_size(void)
1482 {
1483 	return memblock.reserved.total_size;
1484 }
1485 
1486 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1487 {
1488 	unsigned long pages = 0;
1489 	struct memblock_region *r;
1490 	unsigned long start_pfn, end_pfn;
1491 
1492 	for_each_memblock(memory, r) {
1493 		start_pfn = memblock_region_memory_base_pfn(r);
1494 		end_pfn = memblock_region_memory_end_pfn(r);
1495 		start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1496 		end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1497 		pages += end_pfn - start_pfn;
1498 	}
1499 
1500 	return PFN_PHYS(pages);
1501 }
1502 
1503 /* lowest address */
1504 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1505 {
1506 	return memblock.memory.regions[0].base;
1507 }
1508 
1509 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1510 {
1511 	int idx = memblock.memory.cnt - 1;
1512 
1513 	return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1514 }
1515 
1516 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1517 {
1518 	phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1519 	struct memblock_region *r;
1520 
1521 	/*
1522 	 * translate the memory @limit size into the max address within one of
1523 	 * the memory memblock regions, if the @limit exceeds the total size
1524 	 * of those regions, max_addr will keep original value ULLONG_MAX
1525 	 */
1526 	for_each_memblock(memory, r) {
1527 		if (limit <= r->size) {
1528 			max_addr = r->base + limit;
1529 			break;
1530 		}
1531 		limit -= r->size;
1532 	}
1533 
1534 	return max_addr;
1535 }
1536 
1537 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1538 {
1539 	phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1540 
1541 	if (!limit)
1542 		return;
1543 
1544 	max_addr = __find_max_addr(limit);
1545 
1546 	/* @limit exceeds the total size of the memory, do nothing */
1547 	if (max_addr == (phys_addr_t)ULLONG_MAX)
1548 		return;
1549 
1550 	/* truncate both memory and reserved regions */
1551 	memblock_remove_range(&memblock.memory, max_addr,
1552 			      (phys_addr_t)ULLONG_MAX);
1553 	memblock_remove_range(&memblock.reserved, max_addr,
1554 			      (phys_addr_t)ULLONG_MAX);
1555 }
1556 
1557 void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1558 {
1559 	int start_rgn, end_rgn;
1560 	int i, ret;
1561 
1562 	if (!size)
1563 		return;
1564 
1565 	ret = memblock_isolate_range(&memblock.memory, base, size,
1566 						&start_rgn, &end_rgn);
1567 	if (ret)
1568 		return;
1569 
1570 	/* remove all the MAP regions */
1571 	for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1572 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1573 			memblock_remove_region(&memblock.memory, i);
1574 
1575 	for (i = start_rgn - 1; i >= 0; i--)
1576 		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1577 			memblock_remove_region(&memblock.memory, i);
1578 
1579 	/* truncate the reserved regions */
1580 	memblock_remove_range(&memblock.reserved, 0, base);
1581 	memblock_remove_range(&memblock.reserved,
1582 			base + size, (phys_addr_t)ULLONG_MAX);
1583 }
1584 
1585 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1586 {
1587 	phys_addr_t max_addr;
1588 
1589 	if (!limit)
1590 		return;
1591 
1592 	max_addr = __find_max_addr(limit);
1593 
1594 	/* @limit exceeds the total size of the memory, do nothing */
1595 	if (max_addr == (phys_addr_t)ULLONG_MAX)
1596 		return;
1597 
1598 	memblock_cap_memory_range(0, max_addr);
1599 }
1600 
1601 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1602 {
1603 	unsigned int left = 0, right = type->cnt;
1604 
1605 	do {
1606 		unsigned int mid = (right + left) / 2;
1607 
1608 		if (addr < type->regions[mid].base)
1609 			right = mid;
1610 		else if (addr >= (type->regions[mid].base +
1611 				  type->regions[mid].size))
1612 			left = mid + 1;
1613 		else
1614 			return mid;
1615 	} while (left < right);
1616 	return -1;
1617 }
1618 
1619 bool __init memblock_is_reserved(phys_addr_t addr)
1620 {
1621 	return memblock_search(&memblock.reserved, addr) != -1;
1622 }
1623 
1624 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1625 {
1626 	return memblock_search(&memblock.memory, addr) != -1;
1627 }
1628 
1629 bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1630 {
1631 	int i = memblock_search(&memblock.memory, addr);
1632 
1633 	if (i == -1)
1634 		return false;
1635 	return !memblock_is_nomap(&memblock.memory.regions[i]);
1636 }
1637 
1638 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1639 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1640 			 unsigned long *start_pfn, unsigned long *end_pfn)
1641 {
1642 	struct memblock_type *type = &memblock.memory;
1643 	int mid = memblock_search(type, PFN_PHYS(pfn));
1644 
1645 	if (mid == -1)
1646 		return -1;
1647 
1648 	*start_pfn = PFN_DOWN(type->regions[mid].base);
1649 	*end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1650 
1651 	return type->regions[mid].nid;
1652 }
1653 #endif
1654 
1655 /**
1656  * memblock_is_region_memory - check if a region is a subset of memory
1657  * @base: base of region to check
1658  * @size: size of region to check
1659  *
1660  * Check if the region [@base, @base+@size) is a subset of a memory block.
1661  *
1662  * RETURNS:
1663  * 0 if false, non-zero if true
1664  */
1665 bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1666 {
1667 	int idx = memblock_search(&memblock.memory, base);
1668 	phys_addr_t end = base + memblock_cap_size(base, &size);
1669 
1670 	if (idx == -1)
1671 		return false;
1672 	return (memblock.memory.regions[idx].base +
1673 		 memblock.memory.regions[idx].size) >= end;
1674 }
1675 
1676 /**
1677  * memblock_is_region_reserved - check if a region intersects reserved memory
1678  * @base: base of region to check
1679  * @size: size of region to check
1680  *
1681  * Check if the region [@base, @base+@size) intersects a reserved memory block.
1682  *
1683  * RETURNS:
1684  * True if they intersect, false if not.
1685  */
1686 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1687 {
1688 	memblock_cap_size(base, &size);
1689 	return memblock_overlaps_region(&memblock.reserved, base, size);
1690 }
1691 
1692 void __init_memblock memblock_trim_memory(phys_addr_t align)
1693 {
1694 	phys_addr_t start, end, orig_start, orig_end;
1695 	struct memblock_region *r;
1696 
1697 	for_each_memblock(memory, r) {
1698 		orig_start = r->base;
1699 		orig_end = r->base + r->size;
1700 		start = round_up(orig_start, align);
1701 		end = round_down(orig_end, align);
1702 
1703 		if (start == orig_start && end == orig_end)
1704 			continue;
1705 
1706 		if (start < end) {
1707 			r->base = start;
1708 			r->size = end - start;
1709 		} else {
1710 			memblock_remove_region(&memblock.memory,
1711 					       r - memblock.memory.regions);
1712 			r--;
1713 		}
1714 	}
1715 }
1716 
1717 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1718 {
1719 	memblock.current_limit = limit;
1720 }
1721 
1722 phys_addr_t __init_memblock memblock_get_current_limit(void)
1723 {
1724 	return memblock.current_limit;
1725 }
1726 
1727 static void __init_memblock memblock_dump(struct memblock_type *type)
1728 {
1729 	phys_addr_t base, end, size;
1730 	unsigned long flags;
1731 	int idx;
1732 	struct memblock_region *rgn;
1733 
1734 	pr_info(" %s.cnt  = 0x%lx\n", type->name, type->cnt);
1735 
1736 	for_each_memblock_type(idx, type, rgn) {
1737 		char nid_buf[32] = "";
1738 
1739 		base = rgn->base;
1740 		size = rgn->size;
1741 		end = base + size - 1;
1742 		flags = rgn->flags;
1743 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1744 		if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1745 			snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1746 				 memblock_get_region_node(rgn));
1747 #endif
1748 		pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#lx\n",
1749 			type->name, idx, &base, &end, &size, nid_buf, flags);
1750 	}
1751 }
1752 
1753 extern unsigned long __init_memblock
1754 memblock_reserved_memory_within(phys_addr_t start_addr, phys_addr_t end_addr)
1755 {
1756 	struct memblock_region *rgn;
1757 	unsigned long size = 0;
1758 	int idx;
1759 
1760 	for_each_memblock_type(idx, (&memblock.reserved), rgn) {
1761 		phys_addr_t start, end;
1762 
1763 		if (rgn->base + rgn->size < start_addr)
1764 			continue;
1765 		if (rgn->base > end_addr)
1766 			continue;
1767 
1768 		start = rgn->base;
1769 		end = start + rgn->size;
1770 		size += end - start;
1771 	}
1772 
1773 	return size;
1774 }
1775 
1776 void __init_memblock __memblock_dump_all(void)
1777 {
1778 	pr_info("MEMBLOCK configuration:\n");
1779 	pr_info(" memory size = %pa reserved size = %pa\n",
1780 		&memblock.memory.total_size,
1781 		&memblock.reserved.total_size);
1782 
1783 	memblock_dump(&memblock.memory);
1784 	memblock_dump(&memblock.reserved);
1785 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1786 	memblock_dump(&memblock.physmem);
1787 #endif
1788 }
1789 
1790 void __init memblock_allow_resize(void)
1791 {
1792 	memblock_can_resize = 1;
1793 }
1794 
1795 static int __init early_memblock(char *p)
1796 {
1797 	if (p && strstr(p, "debug"))
1798 		memblock_debug = 1;
1799 	return 0;
1800 }
1801 early_param("memblock", early_memblock);
1802 
1803 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1804 
1805 static int memblock_debug_show(struct seq_file *m, void *private)
1806 {
1807 	struct memblock_type *type = m->private;
1808 	struct memblock_region *reg;
1809 	int i;
1810 	phys_addr_t end;
1811 
1812 	for (i = 0; i < type->cnt; i++) {
1813 		reg = &type->regions[i];
1814 		end = reg->base + reg->size - 1;
1815 
1816 		seq_printf(m, "%4d: ", i);
1817 		seq_printf(m, "%pa..%pa\n", &reg->base, &end);
1818 	}
1819 	return 0;
1820 }
1821 
1822 static int memblock_debug_open(struct inode *inode, struct file *file)
1823 {
1824 	return single_open(file, memblock_debug_show, inode->i_private);
1825 }
1826 
1827 static const struct file_operations memblock_debug_fops = {
1828 	.open = memblock_debug_open,
1829 	.read = seq_read,
1830 	.llseek = seq_lseek,
1831 	.release = single_release,
1832 };
1833 
1834 static int __init memblock_init_debugfs(void)
1835 {
1836 	struct dentry *root = debugfs_create_dir("memblock", NULL);
1837 	if (!root)
1838 		return -ENXIO;
1839 	debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1840 	debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1841 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1842 	debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1843 #endif
1844 
1845 	return 0;
1846 }
1847 __initcall(memblock_init_debugfs);
1848 
1849 #endif /* CONFIG_DEBUG_FS */
1850