~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/memblock.c

Version: ~ [ linux-5.16-rc3 ] ~ [ linux-5.15.5 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.82 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.162 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.218 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.256 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.291 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.293 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp