xref: /openbmc/linux/mm/memblock.c (revision fce96cf0443083e37455eff8f78fd240c621dae3)
1  // SPDX-License-Identifier: GPL-2.0-or-later
2  /*
3   * Procedures for maintaining information about logical memory blocks.
4   *
5   * Peter Bergner, IBM Corp.	June 2001.
6   * Copyright (C) 2001 Peter Bergner.
7   */
8  
9  #include <linux/kernel.h>
10  #include <linux/slab.h>
11  #include <linux/init.h>
12  #include <linux/bitops.h>
13  #include <linux/poison.h>
14  #include <linux/pfn.h>
15  #include <linux/debugfs.h>
16  #include <linux/kmemleak.h>
17  #include <linux/seq_file.h>
18  #include <linux/memblock.h>
19  
20  #include <asm/sections.h>
21  #include <linux/io.h>
22  
23  #include "internal.h"
24  
25  #define INIT_MEMBLOCK_REGIONS			128
26  #define INIT_PHYSMEM_REGIONS			4
27  
28  #ifndef INIT_MEMBLOCK_RESERVED_REGIONS
29  # define INIT_MEMBLOCK_RESERVED_REGIONS		INIT_MEMBLOCK_REGIONS
30  #endif
31  
32  /**
33   * DOC: memblock overview
34   *
35   * Memblock is a method of managing memory regions during the early
36   * boot period when the usual kernel memory allocators are not up and
37   * running.
38   *
39   * Memblock views the system memory as collections of contiguous
40   * regions. There are several types of these collections:
41   *
42   * * ``memory`` - describes the physical memory available to the
43   *   kernel; this may differ from the actual physical memory installed
44   *   in the system, for instance when the memory is restricted with
45   *   ``mem=`` command line parameter
46   * * ``reserved`` - describes the regions that were allocated
47   * * ``physmem`` - describes the actual physical memory available during
48   *   boot regardless of the possible restrictions and memory hot(un)plug;
49   *   the ``physmem`` type is only available on some architectures.
50   *
51   * Each region is represented by struct memblock_region that
52   * defines the region extents, its attributes and NUMA node id on NUMA
53   * systems. Every memory type is described by the struct memblock_type
54   * which contains an array of memory regions along with
55   * the allocator metadata. The "memory" and "reserved" types are nicely
56   * wrapped with struct memblock. This structure is statically
57   * initialized at build time. The region arrays are initially sized to
58   * %INIT_MEMBLOCK_REGIONS for "memory" and %INIT_MEMBLOCK_RESERVED_REGIONS
59   * for "reserved". The region array for "physmem" is initially sized to
60   * %INIT_PHYSMEM_REGIONS.
61   * The memblock_allow_resize() enables automatic resizing of the region
62   * arrays during addition of new regions. This feature should be used
63   * with care so that memory allocated for the region array will not
64   * overlap with areas that should be reserved, for example initrd.
65   *
66   * The early architecture setup should tell memblock what the physical
67   * memory layout is by using memblock_add() or memblock_add_node()
68   * functions. The first function does not assign the region to a NUMA
69   * node and it is appropriate for UMA systems. Yet, it is possible to
70   * use it on NUMA systems as well and assign the region to a NUMA node
71   * later in the setup process using memblock_set_node(). The
72   * memblock_add_node() performs such an assignment directly.
73   *
74   * Once memblock is setup the memory can be allocated using one of the
75   * API variants:
76   *
77   * * memblock_phys_alloc*() - these functions return the **physical**
78   *   address of the allocated memory
79   * * memblock_alloc*() - these functions return the **virtual** address
80   *   of the allocated memory.
81   *
82   * Note, that both API variants use implicit assumptions about allowed
83   * memory ranges and the fallback methods. Consult the documentation
84   * of memblock_alloc_internal() and memblock_alloc_range_nid()
85   * functions for more elaborate description.
86   *
87   * As the system boot progresses, the architecture specific mem_init()
88   * function frees all the memory to the buddy page allocator.
89   *
90   * Unless an architecture enables %CONFIG_ARCH_KEEP_MEMBLOCK, the
91   * memblock data structures (except "physmem") will be discarded after the
92   * system initialization completes.
93   */
94  
95  #ifndef CONFIG_NUMA
96  struct pglist_data __refdata contig_page_data;
97  EXPORT_SYMBOL(contig_page_data);
98  #endif
99  
100  unsigned long max_low_pfn;
101  unsigned long min_low_pfn;
102  unsigned long max_pfn;
103  unsigned long long max_possible_pfn;
104  
105  static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
106  static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_RESERVED_REGIONS] __initdata_memblock;
107  #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
108  static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS];
109  #endif
110  
111  struct memblock memblock __initdata_memblock = {
112  	.memory.regions		= memblock_memory_init_regions,
113  	.memory.cnt		= 1,	/* empty dummy entry */
114  	.memory.max		= INIT_MEMBLOCK_REGIONS,
115  	.memory.name		= "memory",
116  
117  	.reserved.regions	= memblock_reserved_init_regions,
118  	.reserved.cnt		= 1,	/* empty dummy entry */
119  	.reserved.max		= INIT_MEMBLOCK_RESERVED_REGIONS,
120  	.reserved.name		= "reserved",
121  
122  	.bottom_up		= false,
123  	.current_limit		= MEMBLOCK_ALLOC_ANYWHERE,
124  };
125  
126  #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
127  struct memblock_type physmem = {
128  	.regions		= memblock_physmem_init_regions,
129  	.cnt			= 1,	/* empty dummy entry */
130  	.max			= INIT_PHYSMEM_REGIONS,
131  	.name			= "physmem",
132  };
133  #endif
134  
135  /*
136   * keep a pointer to &memblock.memory in the text section to use it in
137   * __next_mem_range() and its helpers.
138   *  For architectures that do not keep memblock data after init, this
139   * pointer will be reset to NULL at memblock_discard()
140   */
141  static __refdata struct memblock_type *memblock_memory = &memblock.memory;
142  
143  #define for_each_memblock_type(i, memblock_type, rgn)			\
144  	for (i = 0, rgn = &memblock_type->regions[0];			\
145  	     i < memblock_type->cnt;					\
146  	     i++, rgn = &memblock_type->regions[i])
147  
148  #define memblock_dbg(fmt, ...)						\
149  	do {								\
150  		if (memblock_debug)					\
151  			pr_info(fmt, ##__VA_ARGS__);			\
152  	} while (0)
153  
154  static int memblock_debug __initdata_memblock;
155  static bool system_has_some_mirror __initdata_memblock = false;
156  static int memblock_can_resize __initdata_memblock;
157  static int memblock_memory_in_slab __initdata_memblock = 0;
158  static int memblock_reserved_in_slab __initdata_memblock = 0;
159  
160  static enum memblock_flags __init_memblock choose_memblock_flags(void)
161  {
162  	return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
163  }
164  
165  /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
166  static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
167  {
168  	return *size = min(*size, PHYS_ADDR_MAX - base);
169  }
170  
171  /*
172   * Address comparison utilities
173   */
174  static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
175  				       phys_addr_t base2, phys_addr_t size2)
176  {
177  	return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
178  }
179  
180  bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
181  					phys_addr_t base, phys_addr_t size)
182  {
183  	unsigned long i;
184  
185  	memblock_cap_size(base, &size);
186  
187  	for (i = 0; i < type->cnt; i++)
188  		if (memblock_addrs_overlap(base, size, type->regions[i].base,
189  					   type->regions[i].size))
190  			break;
191  	return i < type->cnt;
192  }
193  
194  /**
195   * __memblock_find_range_bottom_up - find free area utility in bottom-up
196   * @start: start of candidate range
197   * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
198   *       %MEMBLOCK_ALLOC_ACCESSIBLE
199   * @size: size of free area to find
200   * @align: alignment of free area to find
201   * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
202   * @flags: pick from blocks based on memory attributes
203   *
204   * Utility called from memblock_find_in_range_node(), find free area bottom-up.
205   *
206   * Return:
207   * Found address on success, 0 on failure.
208   */
209  static phys_addr_t __init_memblock
210  __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
211  				phys_addr_t size, phys_addr_t align, int nid,
212  				enum memblock_flags flags)
213  {
214  	phys_addr_t this_start, this_end, cand;
215  	u64 i;
216  
217  	for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
218  		this_start = clamp(this_start, start, end);
219  		this_end = clamp(this_end, start, end);
220  
221  		cand = round_up(this_start, align);
222  		if (cand < this_end && this_end - cand >= size)
223  			return cand;
224  	}
225  
226  	return 0;
227  }
228  
229  /**
230   * __memblock_find_range_top_down - find free area utility, in top-down
231   * @start: start of candidate range
232   * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
233   *       %MEMBLOCK_ALLOC_ACCESSIBLE
234   * @size: size of free area to find
235   * @align: alignment of free area to find
236   * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
237   * @flags: pick from blocks based on memory attributes
238   *
239   * Utility called from memblock_find_in_range_node(), find free area top-down.
240   *
241   * Return:
242   * Found address on success, 0 on failure.
243   */
244  static phys_addr_t __init_memblock
245  __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
246  			       phys_addr_t size, phys_addr_t align, int nid,
247  			       enum memblock_flags flags)
248  {
249  	phys_addr_t this_start, this_end, cand;
250  	u64 i;
251  
252  	for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
253  					NULL) {
254  		this_start = clamp(this_start, start, end);
255  		this_end = clamp(this_end, start, end);
256  
257  		if (this_end < size)
258  			continue;
259  
260  		cand = round_down(this_end - size, align);
261  		if (cand >= this_start)
262  			return cand;
263  	}
264  
265  	return 0;
266  }
267  
268  /**
269   * memblock_find_in_range_node - find free area in given range and node
270   * @size: size of free area to find
271   * @align: alignment of free area to find
272   * @start: start of candidate range
273   * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
274   *       %MEMBLOCK_ALLOC_ACCESSIBLE
275   * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
276   * @flags: pick from blocks based on memory attributes
277   *
278   * Find @size free area aligned to @align in the specified range and node.
279   *
280   * Return:
281   * Found address on success, 0 on failure.
282   */
283  static phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
284  					phys_addr_t align, phys_addr_t start,
285  					phys_addr_t end, int nid,
286  					enum memblock_flags flags)
287  {
288  	/* pump up @end */
289  	if (end == MEMBLOCK_ALLOC_ACCESSIBLE ||
290  	    end == MEMBLOCK_ALLOC_NOLEAKTRACE)
291  		end = memblock.current_limit;
292  
293  	/* avoid allocating the first page */
294  	start = max_t(phys_addr_t, start, PAGE_SIZE);
295  	end = max(start, end);
296  
297  	if (memblock_bottom_up())
298  		return __memblock_find_range_bottom_up(start, end, size, align,
299  						       nid, flags);
300  	else
301  		return __memblock_find_range_top_down(start, end, size, align,
302  						      nid, flags);
303  }
304  
305  /**
306   * memblock_find_in_range - find free area in given range
307   * @start: start of candidate range
308   * @end: end of candidate range, can be %MEMBLOCK_ALLOC_ANYWHERE or
309   *       %MEMBLOCK_ALLOC_ACCESSIBLE
310   * @size: size of free area to find
311   * @align: alignment of free area to find
312   *
313   * Find @size free area aligned to @align in the specified range.
314   *
315   * Return:
316   * Found address on success, 0 on failure.
317   */
318  static phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
319  					phys_addr_t end, phys_addr_t size,
320  					phys_addr_t align)
321  {
322  	phys_addr_t ret;
323  	enum memblock_flags flags = choose_memblock_flags();
324  
325  again:
326  	ret = memblock_find_in_range_node(size, align, start, end,
327  					    NUMA_NO_NODE, flags);
328  
329  	if (!ret && (flags & MEMBLOCK_MIRROR)) {
330  		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
331  			&size);
332  		flags &= ~MEMBLOCK_MIRROR;
333  		goto again;
334  	}
335  
336  	return ret;
337  }
338  
339  static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
340  {
341  	type->total_size -= type->regions[r].size;
342  	memmove(&type->regions[r], &type->regions[r + 1],
343  		(type->cnt - (r + 1)) * sizeof(type->regions[r]));
344  	type->cnt--;
345  
346  	/* Special case for empty arrays */
347  	if (type->cnt == 0) {
348  		WARN_ON(type->total_size != 0);
349  		type->cnt = 1;
350  		type->regions[0].base = 0;
351  		type->regions[0].size = 0;
352  		type->regions[0].flags = 0;
353  		memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
354  	}
355  }
356  
357  #ifndef CONFIG_ARCH_KEEP_MEMBLOCK
358  /**
359   * memblock_discard - discard memory and reserved arrays if they were allocated
360   */
361  void __init memblock_discard(void)
362  {
363  	phys_addr_t addr, size;
364  
365  	if (memblock.reserved.regions != memblock_reserved_init_regions) {
366  		addr = __pa(memblock.reserved.regions);
367  		size = PAGE_ALIGN(sizeof(struct memblock_region) *
368  				  memblock.reserved.max);
369  		if (memblock_reserved_in_slab)
370  			kfree(memblock.reserved.regions);
371  		else
372  			memblock_free_late(addr, size);
373  	}
374  
375  	if (memblock.memory.regions != memblock_memory_init_regions) {
376  		addr = __pa(memblock.memory.regions);
377  		size = PAGE_ALIGN(sizeof(struct memblock_region) *
378  				  memblock.memory.max);
379  		if (memblock_memory_in_slab)
380  			kfree(memblock.memory.regions);
381  		else
382  			memblock_free_late(addr, size);
383  	}
384  
385  	memblock_memory = NULL;
386  }
387  #endif
388  
389  /**
390   * memblock_double_array - double the size of the memblock regions array
391   * @type: memblock type of the regions array being doubled
392   * @new_area_start: starting address of memory range to avoid overlap with
393   * @new_area_size: size of memory range to avoid overlap with
394   *
395   * Double the size of the @type regions array. If memblock is being used to
396   * allocate memory for a new reserved regions array and there is a previously
397   * allocated memory range [@new_area_start, @new_area_start + @new_area_size]
398   * waiting to be reserved, ensure the memory used by the new array does
399   * not overlap.
400   *
401   * Return:
402   * 0 on success, -1 on failure.
403   */
404  static int __init_memblock memblock_double_array(struct memblock_type *type,
405  						phys_addr_t new_area_start,
406  						phys_addr_t new_area_size)
407  {
408  	struct memblock_region *new_array, *old_array;
409  	phys_addr_t old_alloc_size, new_alloc_size;
410  	phys_addr_t old_size, new_size, addr, new_end;
411  	int use_slab = slab_is_available();
412  	int *in_slab;
413  
414  	/* We don't allow resizing until we know about the reserved regions
415  	 * of memory that aren't suitable for allocation
416  	 */
417  	if (!memblock_can_resize)
418  		return -1;
419  
420  	/* Calculate new doubled size */
421  	old_size = type->max * sizeof(struct memblock_region);
422  	new_size = old_size << 1;
423  	/*
424  	 * We need to allocated new one align to PAGE_SIZE,
425  	 *   so we can free them completely later.
426  	 */
427  	old_alloc_size = PAGE_ALIGN(old_size);
428  	new_alloc_size = PAGE_ALIGN(new_size);
429  
430  	/* Retrieve the slab flag */
431  	if (type == &memblock.memory)
432  		in_slab = &memblock_memory_in_slab;
433  	else
434  		in_slab = &memblock_reserved_in_slab;
435  
436  	/* Try to find some space for it */
437  	if (use_slab) {
438  		new_array = kmalloc(new_size, GFP_KERNEL);
439  		addr = new_array ? __pa(new_array) : 0;
440  	} else {
441  		/* only exclude range when trying to double reserved.regions */
442  		if (type != &memblock.reserved)
443  			new_area_start = new_area_size = 0;
444  
445  		addr = memblock_find_in_range(new_area_start + new_area_size,
446  						memblock.current_limit,
447  						new_alloc_size, PAGE_SIZE);
448  		if (!addr && new_area_size)
449  			addr = memblock_find_in_range(0,
450  				min(new_area_start, memblock.current_limit),
451  				new_alloc_size, PAGE_SIZE);
452  
453  		new_array = addr ? __va(addr) : NULL;
454  	}
455  	if (!addr) {
456  		pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
457  		       type->name, type->max, type->max * 2);
458  		return -1;
459  	}
460  
461  	new_end = addr + new_size - 1;
462  	memblock_dbg("memblock: %s is doubled to %ld at [%pa-%pa]",
463  			type->name, type->max * 2, &addr, &new_end);
464  
465  	/*
466  	 * Found space, we now need to move the array over before we add the
467  	 * reserved region since it may be our reserved array itself that is
468  	 * full.
469  	 */
470  	memcpy(new_array, type->regions, old_size);
471  	memset(new_array + type->max, 0, old_size);
472  	old_array = type->regions;
473  	type->regions = new_array;
474  	type->max <<= 1;
475  
476  	/* Free old array. We needn't free it if the array is the static one */
477  	if (*in_slab)
478  		kfree(old_array);
479  	else if (old_array != memblock_memory_init_regions &&
480  		 old_array != memblock_reserved_init_regions)
481  		memblock_free(old_array, old_alloc_size);
482  
483  	/*
484  	 * Reserve the new array if that comes from the memblock.  Otherwise, we
485  	 * needn't do it
486  	 */
487  	if (!use_slab)
488  		BUG_ON(memblock_reserve(addr, new_alloc_size));
489  
490  	/* Update slab flag */
491  	*in_slab = use_slab;
492  
493  	return 0;
494  }
495  
496  /**
497   * memblock_merge_regions - merge neighboring compatible regions
498   * @type: memblock type to scan
499   *
500   * Scan @type and merge neighboring compatible regions.
501   */
502  static void __init_memblock memblock_merge_regions(struct memblock_type *type)
503  {
504  	int i = 0;
505  
506  	/* cnt never goes below 1 */
507  	while (i < type->cnt - 1) {
508  		struct memblock_region *this = &type->regions[i];
509  		struct memblock_region *next = &type->regions[i + 1];
510  
511  		if (this->base + this->size != next->base ||
512  		    memblock_get_region_node(this) !=
513  		    memblock_get_region_node(next) ||
514  		    this->flags != next->flags) {
515  			BUG_ON(this->base + this->size > next->base);
516  			i++;
517  			continue;
518  		}
519  
520  		this->size += next->size;
521  		/* move forward from next + 1, index of which is i + 2 */
522  		memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
523  		type->cnt--;
524  	}
525  }
526  
527  /**
528   * memblock_insert_region - insert new memblock region
529   * @type:	memblock type to insert into
530   * @idx:	index for the insertion point
531   * @base:	base address of the new region
532   * @size:	size of the new region
533   * @nid:	node id of the new region
534   * @flags:	flags of the new region
535   *
536   * Insert new memblock region [@base, @base + @size) into @type at @idx.
537   * @type must already have extra room to accommodate the new region.
538   */
539  static void __init_memblock memblock_insert_region(struct memblock_type *type,
540  						   int idx, phys_addr_t base,
541  						   phys_addr_t size,
542  						   int nid,
543  						   enum memblock_flags flags)
544  {
545  	struct memblock_region *rgn = &type->regions[idx];
546  
547  	BUG_ON(type->cnt >= type->max);
548  	memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
549  	rgn->base = base;
550  	rgn->size = size;
551  	rgn->flags = flags;
552  	memblock_set_region_node(rgn, nid);
553  	type->cnt++;
554  	type->total_size += size;
555  }
556  
557  /**
558   * memblock_add_range - add new memblock region
559   * @type: memblock type to add new region into
560   * @base: base address of the new region
561   * @size: size of the new region
562   * @nid: nid of the new region
563   * @flags: flags of the new region
564   *
565   * Add new memblock region [@base, @base + @size) into @type.  The new region
566   * is allowed to overlap with existing ones - overlaps don't affect already
567   * existing regions.  @type is guaranteed to be minimal (all neighbouring
568   * compatible regions are merged) after the addition.
569   *
570   * Return:
571   * 0 on success, -errno on failure.
572   */
573  static int __init_memblock memblock_add_range(struct memblock_type *type,
574  				phys_addr_t base, phys_addr_t size,
575  				int nid, enum memblock_flags flags)
576  {
577  	bool insert = false;
578  	phys_addr_t obase = base;
579  	phys_addr_t end = base + memblock_cap_size(base, &size);
580  	int idx, nr_new;
581  	struct memblock_region *rgn;
582  
583  	if (!size)
584  		return 0;
585  
586  	/* special case for empty array */
587  	if (type->regions[0].size == 0) {
588  		WARN_ON(type->cnt != 1 || type->total_size);
589  		type->regions[0].base = base;
590  		type->regions[0].size = size;
591  		type->regions[0].flags = flags;
592  		memblock_set_region_node(&type->regions[0], nid);
593  		type->total_size = size;
594  		return 0;
595  	}
596  repeat:
597  	/*
598  	 * The following is executed twice.  Once with %false @insert and
599  	 * then with %true.  The first counts the number of regions needed
600  	 * to accommodate the new area.  The second actually inserts them.
601  	 */
602  	base = obase;
603  	nr_new = 0;
604  
605  	for_each_memblock_type(idx, type, rgn) {
606  		phys_addr_t rbase = rgn->base;
607  		phys_addr_t rend = rbase + rgn->size;
608  
609  		if (rbase >= end)
610  			break;
611  		if (rend <= base)
612  			continue;
613  		/*
614  		 * @rgn overlaps.  If it separates the lower part of new
615  		 * area, insert that portion.
616  		 */
617  		if (rbase > base) {
618  #ifdef CONFIG_NUMA
619  			WARN_ON(nid != memblock_get_region_node(rgn));
620  #endif
621  			WARN_ON(flags != rgn->flags);
622  			nr_new++;
623  			if (insert)
624  				memblock_insert_region(type, idx++, base,
625  						       rbase - base, nid,
626  						       flags);
627  		}
628  		/* area below @rend is dealt with, forget about it */
629  		base = min(rend, end);
630  	}
631  
632  	/* insert the remaining portion */
633  	if (base < end) {
634  		nr_new++;
635  		if (insert)
636  			memblock_insert_region(type, idx, base, end - base,
637  					       nid, flags);
638  	}
639  
640  	if (!nr_new)
641  		return 0;
642  
643  	/*
644  	 * If this was the first round, resize array and repeat for actual
645  	 * insertions; otherwise, merge and return.
646  	 */
647  	if (!insert) {
648  		while (type->cnt + nr_new > type->max)
649  			if (memblock_double_array(type, obase, size) < 0)
650  				return -ENOMEM;
651  		insert = true;
652  		goto repeat;
653  	} else {
654  		memblock_merge_regions(type);
655  		return 0;
656  	}
657  }
658  
659  /**
660   * memblock_add_node - add new memblock region within a NUMA node
661   * @base: base address of the new region
662   * @size: size of the new region
663   * @nid: nid of the new region
664   * @flags: flags of the new region
665   *
666   * Add new memblock region [@base, @base + @size) to the "memory"
667   * type. See memblock_add_range() description for mode details
668   *
669   * Return:
670   * 0 on success, -errno on failure.
671   */
672  int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
673  				      int nid, enum memblock_flags flags)
674  {
675  	phys_addr_t end = base + size - 1;
676  
677  	memblock_dbg("%s: [%pa-%pa] nid=%d flags=%x %pS\n", __func__,
678  		     &base, &end, nid, flags, (void *)_RET_IP_);
679  
680  	return memblock_add_range(&memblock.memory, base, size, nid, flags);
681  }
682  
683  /**
684   * memblock_add - add new memblock region
685   * @base: base address of the new region
686   * @size: size of the new region
687   *
688   * Add new memblock region [@base, @base + @size) to the "memory"
689   * type. See memblock_add_range() description for mode details
690   *
691   * Return:
692   * 0 on success, -errno on failure.
693   */
694  int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
695  {
696  	phys_addr_t end = base + size - 1;
697  
698  	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
699  		     &base, &end, (void *)_RET_IP_);
700  
701  	return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
702  }
703  
704  /**
705   * memblock_isolate_range - isolate given range into disjoint memblocks
706   * @type: memblock type to isolate range for
707   * @base: base of range to isolate
708   * @size: size of range to isolate
709   * @start_rgn: out parameter for the start of isolated region
710   * @end_rgn: out parameter for the end of isolated region
711   *
712   * Walk @type and ensure that regions don't cross the boundaries defined by
713   * [@base, @base + @size).  Crossing regions are split at the boundaries,
714   * which may create at most two more regions.  The index of the first
715   * region inside the range is returned in *@start_rgn and end in *@end_rgn.
716   *
717   * Return:
718   * 0 on success, -errno on failure.
719   */
720  static int __init_memblock memblock_isolate_range(struct memblock_type *type,
721  					phys_addr_t base, phys_addr_t size,
722  					int *start_rgn, int *end_rgn)
723  {
724  	phys_addr_t end = base + memblock_cap_size(base, &size);
725  	int idx;
726  	struct memblock_region *rgn;
727  
728  	*start_rgn = *end_rgn = 0;
729  
730  	if (!size)
731  		return 0;
732  
733  	/* we'll create at most two more regions */
734  	while (type->cnt + 2 > type->max)
735  		if (memblock_double_array(type, base, size) < 0)
736  			return -ENOMEM;
737  
738  	for_each_memblock_type(idx, type, rgn) {
739  		phys_addr_t rbase = rgn->base;
740  		phys_addr_t rend = rbase + rgn->size;
741  
742  		if (rbase >= end)
743  			break;
744  		if (rend <= base)
745  			continue;
746  
747  		if (rbase < base) {
748  			/*
749  			 * @rgn intersects from below.  Split and continue
750  			 * to process the next region - the new top half.
751  			 */
752  			rgn->base = base;
753  			rgn->size -= base - rbase;
754  			type->total_size -= base - rbase;
755  			memblock_insert_region(type, idx, rbase, base - rbase,
756  					       memblock_get_region_node(rgn),
757  					       rgn->flags);
758  		} else if (rend > end) {
759  			/*
760  			 * @rgn intersects from above.  Split and redo the
761  			 * current region - the new bottom half.
762  			 */
763  			rgn->base = end;
764  			rgn->size -= end - rbase;
765  			type->total_size -= end - rbase;
766  			memblock_insert_region(type, idx--, rbase, end - rbase,
767  					       memblock_get_region_node(rgn),
768  					       rgn->flags);
769  		} else {
770  			/* @rgn is fully contained, record it */
771  			if (!*end_rgn)
772  				*start_rgn = idx;
773  			*end_rgn = idx + 1;
774  		}
775  	}
776  
777  	return 0;
778  }
779  
780  static int __init_memblock memblock_remove_range(struct memblock_type *type,
781  					  phys_addr_t base, phys_addr_t size)
782  {
783  	int start_rgn, end_rgn;
784  	int i, ret;
785  
786  	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
787  	if (ret)
788  		return ret;
789  
790  	for (i = end_rgn - 1; i >= start_rgn; i--)
791  		memblock_remove_region(type, i);
792  	return 0;
793  }
794  
795  int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
796  {
797  	phys_addr_t end = base + size - 1;
798  
799  	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
800  		     &base, &end, (void *)_RET_IP_);
801  
802  	return memblock_remove_range(&memblock.memory, base, size);
803  }
804  
805  /**
806   * memblock_free - free boot memory allocation
807   * @ptr: starting address of the  boot memory allocation
808   * @size: size of the boot memory block in bytes
809   *
810   * Free boot memory block previously allocated by memblock_alloc_xx() API.
811   * The freeing memory will not be released to the buddy allocator.
812   */
813  void __init_memblock memblock_free(void *ptr, size_t size)
814  {
815  	if (ptr)
816  		memblock_phys_free(__pa(ptr), size);
817  }
818  
819  /**
820   * memblock_phys_free - free boot memory block
821   * @base: phys starting address of the  boot memory block
822   * @size: size of the boot memory block in bytes
823   *
824   * Free boot memory block previously allocated by memblock_alloc_xx() API.
825   * The freeing memory will not be released to the buddy allocator.
826   */
827  int __init_memblock memblock_phys_free(phys_addr_t base, phys_addr_t size)
828  {
829  	phys_addr_t end = base + size - 1;
830  
831  	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
832  		     &base, &end, (void *)_RET_IP_);
833  
834  	kmemleak_free_part_phys(base, size);
835  	return memblock_remove_range(&memblock.reserved, base, size);
836  }
837  
838  int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
839  {
840  	phys_addr_t end = base + size - 1;
841  
842  	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
843  		     &base, &end, (void *)_RET_IP_);
844  
845  	return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
846  }
847  
848  #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
849  int __init_memblock memblock_physmem_add(phys_addr_t base, phys_addr_t size)
850  {
851  	phys_addr_t end = base + size - 1;
852  
853  	memblock_dbg("%s: [%pa-%pa] %pS\n", __func__,
854  		     &base, &end, (void *)_RET_IP_);
855  
856  	return memblock_add_range(&physmem, base, size, MAX_NUMNODES, 0);
857  }
858  #endif
859  
860  /**
861   * memblock_setclr_flag - set or clear flag for a memory region
862   * @base: base address of the region
863   * @size: size of the region
864   * @set: set or clear the flag
865   * @flag: the flag to update
866   *
867   * This function isolates region [@base, @base + @size), and sets/clears flag
868   *
869   * Return: 0 on success, -errno on failure.
870   */
871  static int __init_memblock memblock_setclr_flag(phys_addr_t base,
872  				phys_addr_t size, int set, int flag)
873  {
874  	struct memblock_type *type = &memblock.memory;
875  	int i, ret, start_rgn, end_rgn;
876  
877  	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
878  	if (ret)
879  		return ret;
880  
881  	for (i = start_rgn; i < end_rgn; i++) {
882  		struct memblock_region *r = &type->regions[i];
883  
884  		if (set)
885  			r->flags |= flag;
886  		else
887  			r->flags &= ~flag;
888  	}
889  
890  	memblock_merge_regions(type);
891  	return 0;
892  }
893  
894  /**
895   * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
896   * @base: the base phys addr of the region
897   * @size: the size of the region
898   *
899   * Return: 0 on success, -errno on failure.
900   */
901  int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
902  {
903  	return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
904  }
905  
906  /**
907   * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
908   * @base: the base phys addr of the region
909   * @size: the size of the region
910   *
911   * Return: 0 on success, -errno on failure.
912   */
913  int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
914  {
915  	return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
916  }
917  
918  /**
919   * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
920   * @base: the base phys addr of the region
921   * @size: the size of the region
922   *
923   * Return: 0 on success, -errno on failure.
924   */
925  int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
926  {
927  	system_has_some_mirror = true;
928  
929  	return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
930  }
931  
932  /**
933   * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
934   * @base: the base phys addr of the region
935   * @size: the size of the region
936   *
937   * The memory regions marked with %MEMBLOCK_NOMAP will not be added to the
938   * direct mapping of the physical memory. These regions will still be
939   * covered by the memory map. The struct page representing NOMAP memory
940   * frames in the memory map will be PageReserved()
941   *
942   * Note: if the memory being marked %MEMBLOCK_NOMAP was allocated from
943   * memblock, the caller must inform kmemleak to ignore that memory
944   *
945   * Return: 0 on success, -errno on failure.
946   */
947  int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
948  {
949  	return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
950  }
951  
952  /**
953   * memblock_clear_nomap - Clear flag MEMBLOCK_NOMAP for a specified region.
954   * @base: the base phys addr of the region
955   * @size: the size of the region
956   *
957   * Return: 0 on success, -errno on failure.
958   */
959  int __init_memblock memblock_clear_nomap(phys_addr_t base, phys_addr_t size)
960  {
961  	return memblock_setclr_flag(base, size, 0, MEMBLOCK_NOMAP);
962  }
963  
964  static bool should_skip_region(struct memblock_type *type,
965  			       struct memblock_region *m,
966  			       int nid, int flags)
967  {
968  	int m_nid = memblock_get_region_node(m);
969  
970  	/* we never skip regions when iterating memblock.reserved or physmem */
971  	if (type != memblock_memory)
972  		return false;
973  
974  	/* only memory regions are associated with nodes, check it */
975  	if (nid != NUMA_NO_NODE && nid != m_nid)
976  		return true;
977  
978  	/* skip hotpluggable memory regions if needed */
979  	if (movable_node_is_enabled() && memblock_is_hotpluggable(m) &&
980  	    !(flags & MEMBLOCK_HOTPLUG))
981  		return true;
982  
983  	/* if we want mirror memory skip non-mirror memory regions */
984  	if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
985  		return true;
986  
987  	/* skip nomap memory unless we were asked for it explicitly */
988  	if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
989  		return true;
990  
991  	/* skip driver-managed memory unless we were asked for it explicitly */
992  	if (!(flags & MEMBLOCK_DRIVER_MANAGED) && memblock_is_driver_managed(m))
993  		return true;
994  
995  	return false;
996  }
997  
998  /**
999   * __next_mem_range - next function for for_each_free_mem_range() etc.
1000   * @idx: pointer to u64 loop variable
1001   * @nid: node selector, %NUMA_NO_NODE for all nodes
1002   * @flags: pick from blocks based on memory attributes
1003   * @type_a: pointer to memblock_type from where the range is taken
1004   * @type_b: pointer to memblock_type which excludes memory from being taken
1005   * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1006   * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1007   * @out_nid: ptr to int for nid of the range, can be %NULL
1008   *
1009   * Find the first area from *@idx which matches @nid, fill the out
1010   * parameters, and update *@idx for the next iteration.  The lower 32bit of
1011   * *@idx contains index into type_a and the upper 32bit indexes the
1012   * areas before each region in type_b.	For example, if type_b regions
1013   * look like the following,
1014   *
1015   *	0:[0-16), 1:[32-48), 2:[128-130)
1016   *
1017   * The upper 32bit indexes the following regions.
1018   *
1019   *	0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
1020   *
1021   * As both region arrays are sorted, the function advances the two indices
1022   * in lockstep and returns each intersection.
1023   */
1024  void __next_mem_range(u64 *idx, int nid, enum memblock_flags flags,
1025  		      struct memblock_type *type_a,
1026  		      struct memblock_type *type_b, phys_addr_t *out_start,
1027  		      phys_addr_t *out_end, int *out_nid)
1028  {
1029  	int idx_a = *idx & 0xffffffff;
1030  	int idx_b = *idx >> 32;
1031  
1032  	if (WARN_ONCE(nid == MAX_NUMNODES,
1033  	"Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1034  		nid = NUMA_NO_NODE;
1035  
1036  	for (; idx_a < type_a->cnt; idx_a++) {
1037  		struct memblock_region *m = &type_a->regions[idx_a];
1038  
1039  		phys_addr_t m_start = m->base;
1040  		phys_addr_t m_end = m->base + m->size;
1041  		int	    m_nid = memblock_get_region_node(m);
1042  
1043  		if (should_skip_region(type_a, m, nid, flags))
1044  			continue;
1045  
1046  		if (!type_b) {
1047  			if (out_start)
1048  				*out_start = m_start;
1049  			if (out_end)
1050  				*out_end = m_end;
1051  			if (out_nid)
1052  				*out_nid = m_nid;
1053  			idx_a++;
1054  			*idx = (u32)idx_a | (u64)idx_b << 32;
1055  			return;
1056  		}
1057  
1058  		/* scan areas before each reservation */
1059  		for (; idx_b < type_b->cnt + 1; idx_b++) {
1060  			struct memblock_region *r;
1061  			phys_addr_t r_start;
1062  			phys_addr_t r_end;
1063  
1064  			r = &type_b->regions[idx_b];
1065  			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1066  			r_end = idx_b < type_b->cnt ?
1067  				r->base : PHYS_ADDR_MAX;
1068  
1069  			/*
1070  			 * if idx_b advanced past idx_a,
1071  			 * break out to advance idx_a
1072  			 */
1073  			if (r_start >= m_end)
1074  				break;
1075  			/* if the two regions intersect, we're done */
1076  			if (m_start < r_end) {
1077  				if (out_start)
1078  					*out_start =
1079  						max(m_start, r_start);
1080  				if (out_end)
1081  					*out_end = min(m_end, r_end);
1082  				if (out_nid)
1083  					*out_nid = m_nid;
1084  				/*
1085  				 * The region which ends first is
1086  				 * advanced for the next iteration.
1087  				 */
1088  				if (m_end <= r_end)
1089  					idx_a++;
1090  				else
1091  					idx_b++;
1092  				*idx = (u32)idx_a | (u64)idx_b << 32;
1093  				return;
1094  			}
1095  		}
1096  	}
1097  
1098  	/* signal end of iteration */
1099  	*idx = ULLONG_MAX;
1100  }
1101  
1102  /**
1103   * __next_mem_range_rev - generic next function for for_each_*_range_rev()
1104   *
1105   * @idx: pointer to u64 loop variable
1106   * @nid: node selector, %NUMA_NO_NODE for all nodes
1107   * @flags: pick from blocks based on memory attributes
1108   * @type_a: pointer to memblock_type from where the range is taken
1109   * @type_b: pointer to memblock_type which excludes memory from being taken
1110   * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
1111   * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
1112   * @out_nid: ptr to int for nid of the range, can be %NULL
1113   *
1114   * Finds the next range from type_a which is not marked as unsuitable
1115   * in type_b.
1116   *
1117   * Reverse of __next_mem_range().
1118   */
1119  void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
1120  					  enum memblock_flags flags,
1121  					  struct memblock_type *type_a,
1122  					  struct memblock_type *type_b,
1123  					  phys_addr_t *out_start,
1124  					  phys_addr_t *out_end, int *out_nid)
1125  {
1126  	int idx_a = *idx & 0xffffffff;
1127  	int idx_b = *idx >> 32;
1128  
1129  	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1130  		nid = NUMA_NO_NODE;
1131  
1132  	if (*idx == (u64)ULLONG_MAX) {
1133  		idx_a = type_a->cnt - 1;
1134  		if (type_b != NULL)
1135  			idx_b = type_b->cnt;
1136  		else
1137  			idx_b = 0;
1138  	}
1139  
1140  	for (; idx_a >= 0; idx_a--) {
1141  		struct memblock_region *m = &type_a->regions[idx_a];
1142  
1143  		phys_addr_t m_start = m->base;
1144  		phys_addr_t m_end = m->base + m->size;
1145  		int m_nid = memblock_get_region_node(m);
1146  
1147  		if (should_skip_region(type_a, m, nid, flags))
1148  			continue;
1149  
1150  		if (!type_b) {
1151  			if (out_start)
1152  				*out_start = m_start;
1153  			if (out_end)
1154  				*out_end = m_end;
1155  			if (out_nid)
1156  				*out_nid = m_nid;
1157  			idx_a--;
1158  			*idx = (u32)idx_a | (u64)idx_b << 32;
1159  			return;
1160  		}
1161  
1162  		/* scan areas before each reservation */
1163  		for (; idx_b >= 0; idx_b--) {
1164  			struct memblock_region *r;
1165  			phys_addr_t r_start;
1166  			phys_addr_t r_end;
1167  
1168  			r = &type_b->regions[idx_b];
1169  			r_start = idx_b ? r[-1].base + r[-1].size : 0;
1170  			r_end = idx_b < type_b->cnt ?
1171  				r->base : PHYS_ADDR_MAX;
1172  			/*
1173  			 * if idx_b advanced past idx_a,
1174  			 * break out to advance idx_a
1175  			 */
1176  
1177  			if (r_end <= m_start)
1178  				break;
1179  			/* if the two regions intersect, we're done */
1180  			if (m_end > r_start) {
1181  				if (out_start)
1182  					*out_start = max(m_start, r_start);
1183  				if (out_end)
1184  					*out_end = min(m_end, r_end);
1185  				if (out_nid)
1186  					*out_nid = m_nid;
1187  				if (m_start >= r_start)
1188  					idx_a--;
1189  				else
1190  					idx_b--;
1191  				*idx = (u32)idx_a | (u64)idx_b << 32;
1192  				return;
1193  			}
1194  		}
1195  	}
1196  	/* signal end of iteration */
1197  	*idx = ULLONG_MAX;
1198  }
1199  
1200  /*
1201   * Common iterator interface used to define for_each_mem_pfn_range().
1202   */
1203  void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1204  				unsigned long *out_start_pfn,
1205  				unsigned long *out_end_pfn, int *out_nid)
1206  {
1207  	struct memblock_type *type = &memblock.memory;
1208  	struct memblock_region *r;
1209  	int r_nid;
1210  
1211  	while (++*idx < type->cnt) {
1212  		r = &type->regions[*idx];
1213  		r_nid = memblock_get_region_node(r);
1214  
1215  		if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1216  			continue;
1217  		if (nid == MAX_NUMNODES || nid == r_nid)
1218  			break;
1219  	}
1220  	if (*idx >= type->cnt) {
1221  		*idx = -1;
1222  		return;
1223  	}
1224  
1225  	if (out_start_pfn)
1226  		*out_start_pfn = PFN_UP(r->base);
1227  	if (out_end_pfn)
1228  		*out_end_pfn = PFN_DOWN(r->base + r->size);
1229  	if (out_nid)
1230  		*out_nid = r_nid;
1231  }
1232  
1233  /**
1234   * memblock_set_node - set node ID on memblock regions
1235   * @base: base of area to set node ID for
1236   * @size: size of area to set node ID for
1237   * @type: memblock type to set node ID for
1238   * @nid: node ID to set
1239   *
1240   * Set the nid of memblock @type regions in [@base, @base + @size) to @nid.
1241   * Regions which cross the area boundaries are split as necessary.
1242   *
1243   * Return:
1244   * 0 on success, -errno on failure.
1245   */
1246  int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1247  				      struct memblock_type *type, int nid)
1248  {
1249  #ifdef CONFIG_NUMA
1250  	int start_rgn, end_rgn;
1251  	int i, ret;
1252  
1253  	ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1254  	if (ret)
1255  		return ret;
1256  
1257  	for (i = start_rgn; i < end_rgn; i++)
1258  		memblock_set_region_node(&type->regions[i], nid);
1259  
1260  	memblock_merge_regions(type);
1261  #endif
1262  	return 0;
1263  }
1264  
1265  #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1266  /**
1267   * __next_mem_pfn_range_in_zone - iterator for for_each_*_range_in_zone()
1268   *
1269   * @idx: pointer to u64 loop variable
1270   * @zone: zone in which all of the memory blocks reside
1271   * @out_spfn: ptr to ulong for start pfn of the range, can be %NULL
1272   * @out_epfn: ptr to ulong for end pfn of the range, can be %NULL
1273   *
1274   * This function is meant to be a zone/pfn specific wrapper for the
1275   * for_each_mem_range type iterators. Specifically they are used in the
1276   * deferred memory init routines and as such we were duplicating much of
1277   * this logic throughout the code. So instead of having it in multiple
1278   * locations it seemed like it would make more sense to centralize this to
1279   * one new iterator that does everything they need.
1280   */
1281  void __init_memblock
1282  __next_mem_pfn_range_in_zone(u64 *idx, struct zone *zone,
1283  			     unsigned long *out_spfn, unsigned long *out_epfn)
1284  {
1285  	int zone_nid = zone_to_nid(zone);
1286  	phys_addr_t spa, epa;
1287  
1288  	__next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1289  			 &memblock.memory, &memblock.reserved,
1290  			 &spa, &epa, NULL);
1291  
1292  	while (*idx != U64_MAX) {
1293  		unsigned long epfn = PFN_DOWN(epa);
1294  		unsigned long spfn = PFN_UP(spa);
1295  
1296  		/*
1297  		 * Verify the end is at least past the start of the zone and
1298  		 * that we have at least one PFN to initialize.
1299  		 */
1300  		if (zone->zone_start_pfn < epfn && spfn < epfn) {
1301  			/* if we went too far just stop searching */
1302  			if (zone_end_pfn(zone) <= spfn) {
1303  				*idx = U64_MAX;
1304  				break;
1305  			}
1306  
1307  			if (out_spfn)
1308  				*out_spfn = max(zone->zone_start_pfn, spfn);
1309  			if (out_epfn)
1310  				*out_epfn = min(zone_end_pfn(zone), epfn);
1311  
1312  			return;
1313  		}
1314  
1315  		__next_mem_range(idx, zone_nid, MEMBLOCK_NONE,
1316  				 &memblock.memory, &memblock.reserved,
1317  				 &spa, &epa, NULL);
1318  	}
1319  
1320  	/* signal end of iteration */
1321  	if (out_spfn)
1322  		*out_spfn = ULONG_MAX;
1323  	if (out_epfn)
1324  		*out_epfn = 0;
1325  }
1326  
1327  #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1328  
1329  /**
1330   * memblock_alloc_range_nid - allocate boot memory block
1331   * @size: size of memory block to be allocated in bytes
1332   * @align: alignment of the region and block's size
1333   * @start: the lower bound of the memory region to allocate (phys address)
1334   * @end: the upper bound of the memory region to allocate (phys address)
1335   * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1336   * @exact_nid: control the allocation fall back to other nodes
1337   *
1338   * The allocation is performed from memory region limited by
1339   * memblock.current_limit if @end == %MEMBLOCK_ALLOC_ACCESSIBLE.
1340   *
1341   * If the specified node can not hold the requested memory and @exact_nid
1342   * is false, the allocation falls back to any node in the system.
1343   *
1344   * For systems with memory mirroring, the allocation is attempted first
1345   * from the regions with mirroring enabled and then retried from any
1346   * memory region.
1347   *
1348   * In addition, function sets the min_count to 0 using kmemleak_alloc_phys for
1349   * allocated boot memory block, so that it is never reported as leaks.
1350   *
1351   * Return:
1352   * Physical address of allocated memory block on success, %0 on failure.
1353   */
1354  phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1355  					phys_addr_t align, phys_addr_t start,
1356  					phys_addr_t end, int nid,
1357  					bool exact_nid)
1358  {
1359  	enum memblock_flags flags = choose_memblock_flags();
1360  	phys_addr_t found;
1361  
1362  	if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1363  		nid = NUMA_NO_NODE;
1364  
1365  	if (!align) {
1366  		/* Can't use WARNs this early in boot on powerpc */
1367  		dump_stack();
1368  		align = SMP_CACHE_BYTES;
1369  	}
1370  
1371  again:
1372  	found = memblock_find_in_range_node(size, align, start, end, nid,
1373  					    flags);
1374  	if (found && !memblock_reserve(found, size))
1375  		goto done;
1376  
1377  	if (nid != NUMA_NO_NODE && !exact_nid) {
1378  		found = memblock_find_in_range_node(size, align, start,
1379  						    end, NUMA_NO_NODE,
1380  						    flags);
1381  		if (found && !memblock_reserve(found, size))
1382  			goto done;
1383  	}
1384  
1385  	if (flags & MEMBLOCK_MIRROR) {
1386  		flags &= ~MEMBLOCK_MIRROR;
1387  		pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1388  			&size);
1389  		goto again;
1390  	}
1391  
1392  	return 0;
1393  
1394  done:
1395  	/*
1396  	 * Skip kmemleak for those places like kasan_init() and
1397  	 * early_pgtable_alloc() due to high volume.
1398  	 */
1399  	if (end != MEMBLOCK_ALLOC_NOLEAKTRACE)
1400  		/*
1401  		 * The min_count is set to 0 so that memblock allocated
1402  		 * blocks are never reported as leaks. This is because many
1403  		 * of these blocks are only referred via the physical
1404  		 * address which is not looked up by kmemleak.
1405  		 */
1406  		kmemleak_alloc_phys(found, size, 0, 0);
1407  
1408  	return found;
1409  }
1410  
1411  /**
1412   * memblock_phys_alloc_range - allocate a memory block inside specified range
1413   * @size: size of memory block to be allocated in bytes
1414   * @align: alignment of the region and block's size
1415   * @start: the lower bound of the memory region to allocate (physical address)
1416   * @end: the upper bound of the memory region to allocate (physical address)
1417   *
1418   * Allocate @size bytes in the between @start and @end.
1419   *
1420   * Return: physical address of the allocated memory block on success,
1421   * %0 on failure.
1422   */
1423  phys_addr_t __init memblock_phys_alloc_range(phys_addr_t size,
1424  					     phys_addr_t align,
1425  					     phys_addr_t start,
1426  					     phys_addr_t end)
1427  {
1428  	memblock_dbg("%s: %llu bytes align=0x%llx from=%pa max_addr=%pa %pS\n",
1429  		     __func__, (u64)size, (u64)align, &start, &end,
1430  		     (void *)_RET_IP_);
1431  	return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1432  					false);
1433  }
1434  
1435  /**
1436   * memblock_phys_alloc_try_nid - allocate a memory block from specified NUMA node
1437   * @size: size of memory block to be allocated in bytes
1438   * @align: alignment of the region and block's size
1439   * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1440   *
1441   * Allocates memory block from the specified NUMA node. If the node
1442   * has no available memory, attempts to allocated from any node in the
1443   * system.
1444   *
1445   * Return: physical address of the allocated memory block on success,
1446   * %0 on failure.
1447   */
1448  phys_addr_t __init memblock_phys_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1449  {
1450  	return memblock_alloc_range_nid(size, align, 0,
1451  					MEMBLOCK_ALLOC_ACCESSIBLE, nid, false);
1452  }
1453  
1454  /**
1455   * memblock_alloc_internal - allocate boot memory block
1456   * @size: size of memory block to be allocated in bytes
1457   * @align: alignment of the region and block's size
1458   * @min_addr: the lower bound of the memory region to allocate (phys address)
1459   * @max_addr: the upper bound of the memory region to allocate (phys address)
1460   * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1461   * @exact_nid: control the allocation fall back to other nodes
1462   *
1463   * Allocates memory block using memblock_alloc_range_nid() and
1464   * converts the returned physical address to virtual.
1465   *
1466   * The @min_addr limit is dropped if it can not be satisfied and the allocation
1467   * will fall back to memory below @min_addr. Other constraints, such
1468   * as node and mirrored memory will be handled again in
1469   * memblock_alloc_range_nid().
1470   *
1471   * Return:
1472   * Virtual address of allocated memory block on success, NULL on failure.
1473   */
1474  static void * __init memblock_alloc_internal(
1475  				phys_addr_t size, phys_addr_t align,
1476  				phys_addr_t min_addr, phys_addr_t max_addr,
1477  				int nid, bool exact_nid)
1478  {
1479  	phys_addr_t alloc;
1480  
1481  	/*
1482  	 * Detect any accidental use of these APIs after slab is ready, as at
1483  	 * this moment memblock may be deinitialized already and its
1484  	 * internal data may be destroyed (after execution of memblock_free_all)
1485  	 */
1486  	if (WARN_ON_ONCE(slab_is_available()))
1487  		return kzalloc_node(size, GFP_NOWAIT, nid);
1488  
1489  	if (max_addr > memblock.current_limit)
1490  		max_addr = memblock.current_limit;
1491  
1492  	alloc = memblock_alloc_range_nid(size, align, min_addr, max_addr, nid,
1493  					exact_nid);
1494  
1495  	/* retry allocation without lower limit */
1496  	if (!alloc && min_addr)
1497  		alloc = memblock_alloc_range_nid(size, align, 0, max_addr, nid,
1498  						exact_nid);
1499  
1500  	if (!alloc)
1501  		return NULL;
1502  
1503  	return phys_to_virt(alloc);
1504  }
1505  
1506  /**
1507   * memblock_alloc_exact_nid_raw - allocate boot memory block on the exact node
1508   * without zeroing memory
1509   * @size: size of memory block to be allocated in bytes
1510   * @align: alignment of the region and block's size
1511   * @min_addr: the lower bound of the memory region from where the allocation
1512   *	  is preferred (phys address)
1513   * @max_addr: the upper bound of the memory region from where the allocation
1514   *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1515   *	      allocate only from memory limited by memblock.current_limit value
1516   * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1517   *
1518   * Public function, provides additional debug information (including caller
1519   * info), if enabled. Does not zero allocated memory.
1520   *
1521   * Return:
1522   * Virtual address of allocated memory block on success, NULL on failure.
1523   */
1524  void * __init memblock_alloc_exact_nid_raw(
1525  			phys_addr_t size, phys_addr_t align,
1526  			phys_addr_t min_addr, phys_addr_t max_addr,
1527  			int nid)
1528  {
1529  	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1530  		     __func__, (u64)size, (u64)align, nid, &min_addr,
1531  		     &max_addr, (void *)_RET_IP_);
1532  
1533  	return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1534  				       true);
1535  }
1536  
1537  /**
1538   * memblock_alloc_try_nid_raw - allocate boot memory block without zeroing
1539   * memory and without panicking
1540   * @size: size of memory block to be allocated in bytes
1541   * @align: alignment of the region and block's size
1542   * @min_addr: the lower bound of the memory region from where the allocation
1543   *	  is preferred (phys address)
1544   * @max_addr: the upper bound of the memory region from where the allocation
1545   *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1546   *	      allocate only from memory limited by memblock.current_limit value
1547   * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1548   *
1549   * Public function, provides additional debug information (including caller
1550   * info), if enabled. Does not zero allocated memory, does not panic if request
1551   * cannot be satisfied.
1552   *
1553   * Return:
1554   * Virtual address of allocated memory block on success, NULL on failure.
1555   */
1556  void * __init memblock_alloc_try_nid_raw(
1557  			phys_addr_t size, phys_addr_t align,
1558  			phys_addr_t min_addr, phys_addr_t max_addr,
1559  			int nid)
1560  {
1561  	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1562  		     __func__, (u64)size, (u64)align, nid, &min_addr,
1563  		     &max_addr, (void *)_RET_IP_);
1564  
1565  	return memblock_alloc_internal(size, align, min_addr, max_addr, nid,
1566  				       false);
1567  }
1568  
1569  /**
1570   * memblock_alloc_try_nid - allocate boot memory block
1571   * @size: size of memory block to be allocated in bytes
1572   * @align: alignment of the region and block's size
1573   * @min_addr: the lower bound of the memory region from where the allocation
1574   *	  is preferred (phys address)
1575   * @max_addr: the upper bound of the memory region from where the allocation
1576   *	      is preferred (phys address), or %MEMBLOCK_ALLOC_ACCESSIBLE to
1577   *	      allocate only from memory limited by memblock.current_limit value
1578   * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1579   *
1580   * Public function, provides additional debug information (including caller
1581   * info), if enabled. This function zeroes the allocated memory.
1582   *
1583   * Return:
1584   * Virtual address of allocated memory block on success, NULL on failure.
1585   */
1586  void * __init memblock_alloc_try_nid(
1587  			phys_addr_t size, phys_addr_t align,
1588  			phys_addr_t min_addr, phys_addr_t max_addr,
1589  			int nid)
1590  {
1591  	void *ptr;
1592  
1593  	memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=%pa max_addr=%pa %pS\n",
1594  		     __func__, (u64)size, (u64)align, nid, &min_addr,
1595  		     &max_addr, (void *)_RET_IP_);
1596  	ptr = memblock_alloc_internal(size, align,
1597  					   min_addr, max_addr, nid, false);
1598  	if (ptr)
1599  		memset(ptr, 0, size);
1600  
1601  	return ptr;
1602  }
1603  
1604  /**
1605   * memblock_free_late - free pages directly to buddy allocator
1606   * @base: phys starting address of the  boot memory block
1607   * @size: size of the boot memory block in bytes
1608   *
1609   * This is only useful when the memblock allocator has already been torn
1610   * down, but we are still initializing the system.  Pages are released directly
1611   * to the buddy allocator.
1612   */
1613  void __init memblock_free_late(phys_addr_t base, phys_addr_t size)
1614  {
1615  	phys_addr_t cursor, end;
1616  
1617  	end = base + size - 1;
1618  	memblock_dbg("%s: [%pa-%pa] %pS\n",
1619  		     __func__, &base, &end, (void *)_RET_IP_);
1620  	kmemleak_free_part_phys(base, size);
1621  	cursor = PFN_UP(base);
1622  	end = PFN_DOWN(base + size);
1623  
1624  	for (; cursor < end; cursor++) {
1625  		memblock_free_pages(pfn_to_page(cursor), cursor, 0);
1626  		totalram_pages_inc();
1627  	}
1628  }
1629  
1630  /*
1631   * Remaining API functions
1632   */
1633  
1634  phys_addr_t __init_memblock memblock_phys_mem_size(void)
1635  {
1636  	return memblock.memory.total_size;
1637  }
1638  
1639  phys_addr_t __init_memblock memblock_reserved_size(void)
1640  {
1641  	return memblock.reserved.total_size;
1642  }
1643  
1644  /* lowest address */
1645  phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1646  {
1647  	return memblock.memory.regions[0].base;
1648  }
1649  
1650  phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1651  {
1652  	int idx = memblock.memory.cnt - 1;
1653  
1654  	return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1655  }
1656  
1657  static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1658  {
1659  	phys_addr_t max_addr = PHYS_ADDR_MAX;
1660  	struct memblock_region *r;
1661  
1662  	/*
1663  	 * translate the memory @limit size into the max address within one of
1664  	 * the memory memblock regions, if the @limit exceeds the total size
1665  	 * of those regions, max_addr will keep original value PHYS_ADDR_MAX
1666  	 */
1667  	for_each_mem_region(r) {
1668  		if (limit <= r->size) {
1669  			max_addr = r->base + limit;
1670  			break;
1671  		}
1672  		limit -= r->size;
1673  	}
1674  
1675  	return max_addr;
1676  }
1677  
1678  void __init memblock_enforce_memory_limit(phys_addr_t limit)
1679  {
1680  	phys_addr_t max_addr;
1681  
1682  	if (!limit)
1683  		return;
1684  
1685  	max_addr = __find_max_addr(limit);
1686  
1687  	/* @limit exceeds the total size of the memory, do nothing */
1688  	if (max_addr == PHYS_ADDR_MAX)
1689  		return;
1690  
1691  	/* truncate both memory and reserved regions */
1692  	memblock_remove_range(&memblock.memory, max_addr,
1693  			      PHYS_ADDR_MAX);
1694  	memblock_remove_range(&memblock.reserved, max_addr,
1695  			      PHYS_ADDR_MAX);
1696  }
1697  
1698  void __init memblock_cap_memory_range(phys_addr_t base, phys_addr_t size)
1699  {
1700  	int start_rgn, end_rgn;
1701  	int i, ret;
1702  
1703  	if (!size)
1704  		return;
1705  
1706  	if (!memblock_memory->total_size) {
1707  		pr_warn("%s: No memory registered yet\n", __func__);
1708  		return;
1709  	}
1710  
1711  	ret = memblock_isolate_range(&memblock.memory, base, size,
1712  						&start_rgn, &end_rgn);
1713  	if (ret)
1714  		return;
1715  
1716  	/* remove all the MAP regions */
1717  	for (i = memblock.memory.cnt - 1; i >= end_rgn; i--)
1718  		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1719  			memblock_remove_region(&memblock.memory, i);
1720  
1721  	for (i = start_rgn - 1; i >= 0; i--)
1722  		if (!memblock_is_nomap(&memblock.memory.regions[i]))
1723  			memblock_remove_region(&memblock.memory, i);
1724  
1725  	/* truncate the reserved regions */
1726  	memblock_remove_range(&memblock.reserved, 0, base);
1727  	memblock_remove_range(&memblock.reserved,
1728  			base + size, PHYS_ADDR_MAX);
1729  }
1730  
1731  void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1732  {
1733  	phys_addr_t max_addr;
1734  
1735  	if (!limit)
1736  		return;
1737  
1738  	max_addr = __find_max_addr(limit);
1739  
1740  	/* @limit exceeds the total size of the memory, do nothing */
1741  	if (max_addr == PHYS_ADDR_MAX)
1742  		return;
1743  
1744  	memblock_cap_memory_range(0, max_addr);
1745  }
1746  
1747  static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1748  {
1749  	unsigned int left = 0, right = type->cnt;
1750  
1751  	do {
1752  		unsigned int mid = (right + left) / 2;
1753  
1754  		if (addr < type->regions[mid].base)
1755  			right = mid;
1756  		else if (addr >= (type->regions[mid].base +
1757  				  type->regions[mid].size))
1758  			left = mid + 1;
1759  		else
1760  			return mid;
1761  	} while (left < right);
1762  	return -1;
1763  }
1764  
1765  bool __init_memblock memblock_is_reserved(phys_addr_t addr)
1766  {
1767  	return memblock_search(&memblock.reserved, addr) != -1;
1768  }
1769  
1770  bool __init_memblock memblock_is_memory(phys_addr_t addr)
1771  {
1772  	return memblock_search(&memblock.memory, addr) != -1;
1773  }
1774  
1775  bool __init_memblock memblock_is_map_memory(phys_addr_t addr)
1776  {
1777  	int i = memblock_search(&memblock.memory, addr);
1778  
1779  	if (i == -1)
1780  		return false;
1781  	return !memblock_is_nomap(&memblock.memory.regions[i]);
1782  }
1783  
1784  int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1785  			 unsigned long *start_pfn, unsigned long *end_pfn)
1786  {
1787  	struct memblock_type *type = &memblock.memory;
1788  	int mid = memblock_search(type, PFN_PHYS(pfn));
1789  
1790  	if (mid == -1)
1791  		return -1;
1792  
1793  	*start_pfn = PFN_DOWN(type->regions[mid].base);
1794  	*end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1795  
1796  	return memblock_get_region_node(&type->regions[mid]);
1797  }
1798  
1799  /**
1800   * memblock_is_region_memory - check if a region is a subset of memory
1801   * @base: base of region to check
1802   * @size: size of region to check
1803   *
1804   * Check if the region [@base, @base + @size) is a subset of a memory block.
1805   *
1806   * Return:
1807   * 0 if false, non-zero if true
1808   */
1809  bool __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1810  {
1811  	int idx = memblock_search(&memblock.memory, base);
1812  	phys_addr_t end = base + memblock_cap_size(base, &size);
1813  
1814  	if (idx == -1)
1815  		return false;
1816  	return (memblock.memory.regions[idx].base +
1817  		 memblock.memory.regions[idx].size) >= end;
1818  }
1819  
1820  /**
1821   * memblock_is_region_reserved - check if a region intersects reserved memory
1822   * @base: base of region to check
1823   * @size: size of region to check
1824   *
1825   * Check if the region [@base, @base + @size) intersects a reserved
1826   * memory block.
1827   *
1828   * Return:
1829   * True if they intersect, false if not.
1830   */
1831  bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1832  {
1833  	return memblock_overlaps_region(&memblock.reserved, base, size);
1834  }
1835  
1836  void __init_memblock memblock_trim_memory(phys_addr_t align)
1837  {
1838  	phys_addr_t start, end, orig_start, orig_end;
1839  	struct memblock_region *r;
1840  
1841  	for_each_mem_region(r) {
1842  		orig_start = r->base;
1843  		orig_end = r->base + r->size;
1844  		start = round_up(orig_start, align);
1845  		end = round_down(orig_end, align);
1846  
1847  		if (start == orig_start && end == orig_end)
1848  			continue;
1849  
1850  		if (start < end) {
1851  			r->base = start;
1852  			r->size = end - start;
1853  		} else {
1854  			memblock_remove_region(&memblock.memory,
1855  					       r - memblock.memory.regions);
1856  			r--;
1857  		}
1858  	}
1859  }
1860  
1861  void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1862  {
1863  	memblock.current_limit = limit;
1864  }
1865  
1866  phys_addr_t __init_memblock memblock_get_current_limit(void)
1867  {
1868  	return memblock.current_limit;
1869  }
1870  
1871  static void __init_memblock memblock_dump(struct memblock_type *type)
1872  {
1873  	phys_addr_t base, end, size;
1874  	enum memblock_flags flags;
1875  	int idx;
1876  	struct memblock_region *rgn;
1877  
1878  	pr_info(" %s.cnt  = 0x%lx\n", type->name, type->cnt);
1879  
1880  	for_each_memblock_type(idx, type, rgn) {
1881  		char nid_buf[32] = "";
1882  
1883  		base = rgn->base;
1884  		size = rgn->size;
1885  		end = base + size - 1;
1886  		flags = rgn->flags;
1887  #ifdef CONFIG_NUMA
1888  		if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1889  			snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1890  				 memblock_get_region_node(rgn));
1891  #endif
1892  		pr_info(" %s[%#x]\t[%pa-%pa], %pa bytes%s flags: %#x\n",
1893  			type->name, idx, &base, &end, &size, nid_buf, flags);
1894  	}
1895  }
1896  
1897  static void __init_memblock __memblock_dump_all(void)
1898  {
1899  	pr_info("MEMBLOCK configuration:\n");
1900  	pr_info(" memory size = %pa reserved size = %pa\n",
1901  		&memblock.memory.total_size,
1902  		&memblock.reserved.total_size);
1903  
1904  	memblock_dump(&memblock.memory);
1905  	memblock_dump(&memblock.reserved);
1906  #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1907  	memblock_dump(&physmem);
1908  #endif
1909  }
1910  
1911  void __init_memblock memblock_dump_all(void)
1912  {
1913  	if (memblock_debug)
1914  		__memblock_dump_all();
1915  }
1916  
1917  void __init memblock_allow_resize(void)
1918  {
1919  	memblock_can_resize = 1;
1920  }
1921  
1922  static int __init early_memblock(char *p)
1923  {
1924  	if (p && strstr(p, "debug"))
1925  		memblock_debug = 1;
1926  	return 0;
1927  }
1928  early_param("memblock", early_memblock);
1929  
1930  static void __init free_memmap(unsigned long start_pfn, unsigned long end_pfn)
1931  {
1932  	struct page *start_pg, *end_pg;
1933  	phys_addr_t pg, pgend;
1934  
1935  	/*
1936  	 * Convert start_pfn/end_pfn to a struct page pointer.
1937  	 */
1938  	start_pg = pfn_to_page(start_pfn - 1) + 1;
1939  	end_pg = pfn_to_page(end_pfn - 1) + 1;
1940  
1941  	/*
1942  	 * Convert to physical addresses, and round start upwards and end
1943  	 * downwards.
1944  	 */
1945  	pg = PAGE_ALIGN(__pa(start_pg));
1946  	pgend = __pa(end_pg) & PAGE_MASK;
1947  
1948  	/*
1949  	 * If there are free pages between these, free the section of the
1950  	 * memmap array.
1951  	 */
1952  	if (pg < pgend)
1953  		memblock_phys_free(pg, pgend - pg);
1954  }
1955  
1956  /*
1957   * The mem_map array can get very big.  Free the unused area of the memory map.
1958   */
1959  static void __init free_unused_memmap(void)
1960  {
1961  	unsigned long start, end, prev_end = 0;
1962  	int i;
1963  
1964  	if (!IS_ENABLED(CONFIG_HAVE_ARCH_PFN_VALID) ||
1965  	    IS_ENABLED(CONFIG_SPARSEMEM_VMEMMAP))
1966  		return;
1967  
1968  	/*
1969  	 * This relies on each bank being in address order.
1970  	 * The banks are sorted previously in bootmem_init().
1971  	 */
1972  	for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, NULL) {
1973  #ifdef CONFIG_SPARSEMEM
1974  		/*
1975  		 * Take care not to free memmap entries that don't exist
1976  		 * due to SPARSEMEM sections which aren't present.
1977  		 */
1978  		start = min(start, ALIGN(prev_end, PAGES_PER_SECTION));
1979  #endif
1980  		/*
1981  		 * Align down here since many operations in VM subsystem
1982  		 * presume that there are no holes in the memory map inside
1983  		 * a pageblock
1984  		 */
1985  		start = round_down(start, pageblock_nr_pages);
1986  
1987  		/*
1988  		 * If we had a previous bank, and there is a space
1989  		 * between the current bank and the previous, free it.
1990  		 */
1991  		if (prev_end && prev_end < start)
1992  			free_memmap(prev_end, start);
1993  
1994  		/*
1995  		 * Align up here since many operations in VM subsystem
1996  		 * presume that there are no holes in the memory map inside
1997  		 * a pageblock
1998  		 */
1999  		prev_end = ALIGN(end, pageblock_nr_pages);
2000  	}
2001  
2002  #ifdef CONFIG_SPARSEMEM
2003  	if (!IS_ALIGNED(prev_end, PAGES_PER_SECTION)) {
2004  		prev_end = ALIGN(end, pageblock_nr_pages);
2005  		free_memmap(prev_end, ALIGN(prev_end, PAGES_PER_SECTION));
2006  	}
2007  #endif
2008  }
2009  
2010  static void __init __free_pages_memory(unsigned long start, unsigned long end)
2011  {
2012  	int order;
2013  
2014  	while (start < end) {
2015  		order = min(MAX_ORDER - 1UL, __ffs(start));
2016  
2017  		while (start + (1UL << order) > end)
2018  			order--;
2019  
2020  		memblock_free_pages(pfn_to_page(start), start, order);
2021  
2022  		start += (1UL << order);
2023  	}
2024  }
2025  
2026  static unsigned long __init __free_memory_core(phys_addr_t start,
2027  				 phys_addr_t end)
2028  {
2029  	unsigned long start_pfn = PFN_UP(start);
2030  	unsigned long end_pfn = min_t(unsigned long,
2031  				      PFN_DOWN(end), max_low_pfn);
2032  
2033  	if (start_pfn >= end_pfn)
2034  		return 0;
2035  
2036  	__free_pages_memory(start_pfn, end_pfn);
2037  
2038  	return end_pfn - start_pfn;
2039  }
2040  
2041  static void __init memmap_init_reserved_pages(void)
2042  {
2043  	struct memblock_region *region;
2044  	phys_addr_t start, end;
2045  	u64 i;
2046  
2047  	/* initialize struct pages for the reserved regions */
2048  	for_each_reserved_mem_range(i, &start, &end)
2049  		reserve_bootmem_region(start, end);
2050  
2051  	/* and also treat struct pages for the NOMAP regions as PageReserved */
2052  	for_each_mem_region(region) {
2053  		if (memblock_is_nomap(region)) {
2054  			start = region->base;
2055  			end = start + region->size;
2056  			reserve_bootmem_region(start, end);
2057  		}
2058  	}
2059  }
2060  
2061  static unsigned long __init free_low_memory_core_early(void)
2062  {
2063  	unsigned long count = 0;
2064  	phys_addr_t start, end;
2065  	u64 i;
2066  
2067  	memblock_clear_hotplug(0, -1);
2068  
2069  	memmap_init_reserved_pages();
2070  
2071  	/*
2072  	 * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
2073  	 *  because in some case like Node0 doesn't have RAM installed
2074  	 *  low ram will be on Node1
2075  	 */
2076  	for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
2077  				NULL)
2078  		count += __free_memory_core(start, end);
2079  
2080  	return count;
2081  }
2082  
2083  static int reset_managed_pages_done __initdata;
2084  
2085  void reset_node_managed_pages(pg_data_t *pgdat)
2086  {
2087  	struct zone *z;
2088  
2089  	for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
2090  		atomic_long_set(&z->managed_pages, 0);
2091  }
2092  
2093  void __init reset_all_zones_managed_pages(void)
2094  {
2095  	struct pglist_data *pgdat;
2096  
2097  	if (reset_managed_pages_done)
2098  		return;
2099  
2100  	for_each_online_pgdat(pgdat)
2101  		reset_node_managed_pages(pgdat);
2102  
2103  	reset_managed_pages_done = 1;
2104  }
2105  
2106  /**
2107   * memblock_free_all - release free pages to the buddy allocator
2108   */
2109  void __init memblock_free_all(void)
2110  {
2111  	unsigned long pages;
2112  
2113  	free_unused_memmap();
2114  	reset_all_zones_managed_pages();
2115  
2116  	pages = free_low_memory_core_early();
2117  	totalram_pages_add(pages);
2118  }
2119  
2120  #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_ARCH_KEEP_MEMBLOCK)
2121  
2122  static int memblock_debug_show(struct seq_file *m, void *private)
2123  {
2124  	struct memblock_type *type = m->private;
2125  	struct memblock_region *reg;
2126  	int i;
2127  	phys_addr_t end;
2128  
2129  	for (i = 0; i < type->cnt; i++) {
2130  		reg = &type->regions[i];
2131  		end = reg->base + reg->size - 1;
2132  
2133  		seq_printf(m, "%4d: ", i);
2134  		seq_printf(m, "%pa..%pa\n", &reg->base, &end);
2135  	}
2136  	return 0;
2137  }
2138  DEFINE_SHOW_ATTRIBUTE(memblock_debug);
2139  
2140  static int __init memblock_init_debugfs(void)
2141  {
2142  	struct dentry *root = debugfs_create_dir("memblock", NULL);
2143  
2144  	debugfs_create_file("memory", 0444, root,
2145  			    &memblock.memory, &memblock_debug_fops);
2146  	debugfs_create_file("reserved", 0444, root,
2147  			    &memblock.reserved, &memblock_debug_fops);
2148  #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
2149  	debugfs_create_file("physmem", 0444, root, &physmem,
2150  			    &memblock_debug_fops);
2151  #endif
2152  
2153  	return 0;
2154  }
2155  __initcall(memblock_init_debugfs);
2156  
2157  #endif /* CONFIG_DEBUG_FS */
2158