Back to home page

LXR

 
 

    


0001 /*
0002  * Procedures for maintaining information about logical memory blocks.
0003  *
0004  * Peter Bergner, IBM Corp. June 2001.
0005  * Copyright (C) 2001 Peter Bergner.
0006  *
0007  *      This program is free software; you can redistribute it and/or
0008  *      modify it under the terms of the GNU General Public License
0009  *      as published by the Free Software Foundation; either version
0010  *      2 of the License, or (at your option) any later version.
0011  */
0012 
0013 #include <linux/kernel.h>
0014 #include <linux/slab.h>
0015 #include <linux/init.h>
0016 #include <linux/bitops.h>
0017 #include <linux/poison.h>
0018 #include <linux/pfn.h>
0019 #include <linux/debugfs.h>
0020 #include <linux/seq_file.h>
0021 #include <linux/memblock.h>
0022 
0023 #include <asm/sections.h>
0024 #include <linux/io.h>
0025 
0026 #include "internal.h"
0027 
0028 static struct memblock_region memblock_memory_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
0029 static struct memblock_region memblock_reserved_init_regions[INIT_MEMBLOCK_REGIONS] __initdata_memblock;
0030 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
0031 static struct memblock_region memblock_physmem_init_regions[INIT_PHYSMEM_REGIONS] __initdata_memblock;
0032 #endif
0033 
0034 struct memblock memblock __initdata_memblock = {
0035     .memory.regions     = memblock_memory_init_regions,
0036     .memory.cnt     = 1,    /* empty dummy entry */
0037     .memory.max     = INIT_MEMBLOCK_REGIONS,
0038 
0039     .reserved.regions   = memblock_reserved_init_regions,
0040     .reserved.cnt       = 1,    /* empty dummy entry */
0041     .reserved.max       = INIT_MEMBLOCK_REGIONS,
0042 
0043 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
0044     .physmem.regions    = memblock_physmem_init_regions,
0045     .physmem.cnt        = 1,    /* empty dummy entry */
0046     .physmem.max        = INIT_PHYSMEM_REGIONS,
0047 #endif
0048 
0049     .bottom_up      = false,
0050     .current_limit      = MEMBLOCK_ALLOC_ANYWHERE,
0051 };
0052 
0053 int memblock_debug __initdata_memblock;
0054 #ifdef CONFIG_MOVABLE_NODE
0055 bool movable_node_enabled __initdata_memblock = false;
0056 #endif
0057 static bool system_has_some_mirror __initdata_memblock = false;
0058 static int memblock_can_resize __initdata_memblock;
0059 static int memblock_memory_in_slab __initdata_memblock = 0;
0060 static int memblock_reserved_in_slab __initdata_memblock = 0;
0061 
0062 ulong __init_memblock choose_memblock_flags(void)
0063 {
0064     return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE;
0065 }
0066 
0067 /* inline so we don't get a warning when pr_debug is compiled out */
0068 static __init_memblock const char *
0069 memblock_type_name(struct memblock_type *type)
0070 {
0071     if (type == &memblock.memory)
0072         return "memory";
0073     else if (type == &memblock.reserved)
0074         return "reserved";
0075     else
0076         return "unknown";
0077 }
0078 
0079 /* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
0080 static inline phys_addr_t memblock_cap_size(phys_addr_t base, phys_addr_t *size)
0081 {
0082     return *size = min(*size, (phys_addr_t)ULLONG_MAX - base);
0083 }
0084 
0085 /*
0086  * Address comparison utilities
0087  */
0088 static unsigned long __init_memblock memblock_addrs_overlap(phys_addr_t base1, phys_addr_t size1,
0089                        phys_addr_t base2, phys_addr_t size2)
0090 {
0091     return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
0092 }
0093 
0094 bool __init_memblock memblock_overlaps_region(struct memblock_type *type,
0095                     phys_addr_t base, phys_addr_t size)
0096 {
0097     unsigned long i;
0098 
0099     for (i = 0; i < type->cnt; i++)
0100         if (memblock_addrs_overlap(base, size, type->regions[i].base,
0101                        type->regions[i].size))
0102             break;
0103     return i < type->cnt;
0104 }
0105 
0106 /*
0107  * __memblock_find_range_bottom_up - find free area utility in bottom-up
0108  * @start: start of candidate range
0109  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
0110  * @size: size of free area to find
0111  * @align: alignment of free area to find
0112  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
0113  * @flags: pick from blocks based on memory attributes
0114  *
0115  * Utility called from memblock_find_in_range_node(), find free area bottom-up.
0116  *
0117  * RETURNS:
0118  * Found address on success, 0 on failure.
0119  */
0120 static phys_addr_t __init_memblock
0121 __memblock_find_range_bottom_up(phys_addr_t start, phys_addr_t end,
0122                 phys_addr_t size, phys_addr_t align, int nid,
0123                 ulong flags)
0124 {
0125     phys_addr_t this_start, this_end, cand;
0126     u64 i;
0127 
0128     for_each_free_mem_range(i, nid, flags, &this_start, &this_end, NULL) {
0129         this_start = clamp(this_start, start, end);
0130         this_end = clamp(this_end, start, end);
0131 
0132         cand = round_up(this_start, align);
0133         if (cand < this_end && this_end - cand >= size)
0134             return cand;
0135     }
0136 
0137     return 0;
0138 }
0139 
0140 /**
0141  * __memblock_find_range_top_down - find free area utility, in top-down
0142  * @start: start of candidate range
0143  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
0144  * @size: size of free area to find
0145  * @align: alignment of free area to find
0146  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
0147  * @flags: pick from blocks based on memory attributes
0148  *
0149  * Utility called from memblock_find_in_range_node(), find free area top-down.
0150  *
0151  * RETURNS:
0152  * Found address on success, 0 on failure.
0153  */
0154 static phys_addr_t __init_memblock
0155 __memblock_find_range_top_down(phys_addr_t start, phys_addr_t end,
0156                    phys_addr_t size, phys_addr_t align, int nid,
0157                    ulong flags)
0158 {
0159     phys_addr_t this_start, this_end, cand;
0160     u64 i;
0161 
0162     for_each_free_mem_range_reverse(i, nid, flags, &this_start, &this_end,
0163                     NULL) {
0164         this_start = clamp(this_start, start, end);
0165         this_end = clamp(this_end, start, end);
0166 
0167         if (this_end < size)
0168             continue;
0169 
0170         cand = round_down(this_end - size, align);
0171         if (cand >= this_start)
0172             return cand;
0173     }
0174 
0175     return 0;
0176 }
0177 
0178 /**
0179  * memblock_find_in_range_node - find free area in given range and node
0180  * @size: size of free area to find
0181  * @align: alignment of free area to find
0182  * @start: start of candidate range
0183  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
0184  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
0185  * @flags: pick from blocks based on memory attributes
0186  *
0187  * Find @size free area aligned to @align in the specified range and node.
0188  *
0189  * When allocation direction is bottom-up, the @start should be greater
0190  * than the end of the kernel image. Otherwise, it will be trimmed. The
0191  * reason is that we want the bottom-up allocation just near the kernel
0192  * image so it is highly likely that the allocated memory and the kernel
0193  * will reside in the same node.
0194  *
0195  * If bottom-up allocation failed, will try to allocate memory top-down.
0196  *
0197  * RETURNS:
0198  * Found address on success, 0 on failure.
0199  */
0200 phys_addr_t __init_memblock memblock_find_in_range_node(phys_addr_t size,
0201                     phys_addr_t align, phys_addr_t start,
0202                     phys_addr_t end, int nid, ulong flags)
0203 {
0204     phys_addr_t kernel_end, ret;
0205 
0206     /* pump up @end */
0207     if (end == MEMBLOCK_ALLOC_ACCESSIBLE)
0208         end = memblock.current_limit;
0209 
0210     /* avoid allocating the first page */
0211     start = max_t(phys_addr_t, start, PAGE_SIZE);
0212     end = max(start, end);
0213     kernel_end = __pa_symbol(_end);
0214 
0215     /*
0216      * try bottom-up allocation only when bottom-up mode
0217      * is set and @end is above the kernel image.
0218      */
0219     if (memblock_bottom_up() && end > kernel_end) {
0220         phys_addr_t bottom_up_start;
0221 
0222         /* make sure we will allocate above the kernel */
0223         bottom_up_start = max(start, kernel_end);
0224 
0225         /* ok, try bottom-up allocation first */
0226         ret = __memblock_find_range_bottom_up(bottom_up_start, end,
0227                               size, align, nid, flags);
0228         if (ret)
0229             return ret;
0230 
0231         /*
0232          * we always limit bottom-up allocation above the kernel,
0233          * but top-down allocation doesn't have the limit, so
0234          * retrying top-down allocation may succeed when bottom-up
0235          * allocation failed.
0236          *
0237          * bottom-up allocation is expected to be fail very rarely,
0238          * so we use WARN_ONCE() here to see the stack trace if
0239          * fail happens.
0240          */
0241         WARN_ONCE(1, "memblock: bottom-up allocation failed, memory hotunplug may be affected\n");
0242     }
0243 
0244     return __memblock_find_range_top_down(start, end, size, align, nid,
0245                           flags);
0246 }
0247 
0248 /**
0249  * memblock_find_in_range - find free area in given range
0250  * @start: start of candidate range
0251  * @end: end of candidate range, can be %MEMBLOCK_ALLOC_{ANYWHERE|ACCESSIBLE}
0252  * @size: size of free area to find
0253  * @align: alignment of free area to find
0254  *
0255  * Find @size free area aligned to @align in the specified range.
0256  *
0257  * RETURNS:
0258  * Found address on success, 0 on failure.
0259  */
0260 phys_addr_t __init_memblock memblock_find_in_range(phys_addr_t start,
0261                     phys_addr_t end, phys_addr_t size,
0262                     phys_addr_t align)
0263 {
0264     phys_addr_t ret;
0265     ulong flags = choose_memblock_flags();
0266 
0267 again:
0268     ret = memblock_find_in_range_node(size, align, start, end,
0269                         NUMA_NO_NODE, flags);
0270 
0271     if (!ret && (flags & MEMBLOCK_MIRROR)) {
0272         pr_warn("Could not allocate %pap bytes of mirrored memory\n",
0273             &size);
0274         flags &= ~MEMBLOCK_MIRROR;
0275         goto again;
0276     }
0277 
0278     return ret;
0279 }
0280 
0281 static void __init_memblock memblock_remove_region(struct memblock_type *type, unsigned long r)
0282 {
0283     type->total_size -= type->regions[r].size;
0284     memmove(&type->regions[r], &type->regions[r + 1],
0285         (type->cnt - (r + 1)) * sizeof(type->regions[r]));
0286     type->cnt--;
0287 
0288     /* Special case for empty arrays */
0289     if (type->cnt == 0) {
0290         WARN_ON(type->total_size != 0);
0291         type->cnt = 1;
0292         type->regions[0].base = 0;
0293         type->regions[0].size = 0;
0294         type->regions[0].flags = 0;
0295         memblock_set_region_node(&type->regions[0], MAX_NUMNODES);
0296     }
0297 }
0298 
0299 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
0300 
0301 phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info(
0302                     phys_addr_t *addr)
0303 {
0304     if (memblock.reserved.regions == memblock_reserved_init_regions)
0305         return 0;
0306 
0307     *addr = __pa(memblock.reserved.regions);
0308 
0309     return PAGE_ALIGN(sizeof(struct memblock_region) *
0310               memblock.reserved.max);
0311 }
0312 
0313 phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info(
0314                     phys_addr_t *addr)
0315 {
0316     if (memblock.memory.regions == memblock_memory_init_regions)
0317         return 0;
0318 
0319     *addr = __pa(memblock.memory.regions);
0320 
0321     return PAGE_ALIGN(sizeof(struct memblock_region) *
0322               memblock.memory.max);
0323 }
0324 
0325 #endif
0326 
0327 /**
0328  * memblock_double_array - double the size of the memblock regions array
0329  * @type: memblock type of the regions array being doubled
0330  * @new_area_start: starting address of memory range to avoid overlap with
0331  * @new_area_size: size of memory range to avoid overlap with
0332  *
0333  * Double the size of the @type regions array. If memblock is being used to
0334  * allocate memory for a new reserved regions array and there is a previously
0335  * allocated memory range [@new_area_start,@new_area_start+@new_area_size]
0336  * waiting to be reserved, ensure the memory used by the new array does
0337  * not overlap.
0338  *
0339  * RETURNS:
0340  * 0 on success, -1 on failure.
0341  */
0342 static int __init_memblock memblock_double_array(struct memblock_type *type,
0343                         phys_addr_t new_area_start,
0344                         phys_addr_t new_area_size)
0345 {
0346     struct memblock_region *new_array, *old_array;
0347     phys_addr_t old_alloc_size, new_alloc_size;
0348     phys_addr_t old_size, new_size, addr;
0349     int use_slab = slab_is_available();
0350     int *in_slab;
0351 
0352     /* We don't allow resizing until we know about the reserved regions
0353      * of memory that aren't suitable for allocation
0354      */
0355     if (!memblock_can_resize)
0356         return -1;
0357 
0358     /* Calculate new doubled size */
0359     old_size = type->max * sizeof(struct memblock_region);
0360     new_size = old_size << 1;
0361     /*
0362      * We need to allocated new one align to PAGE_SIZE,
0363      *   so we can free them completely later.
0364      */
0365     old_alloc_size = PAGE_ALIGN(old_size);
0366     new_alloc_size = PAGE_ALIGN(new_size);
0367 
0368     /* Retrieve the slab flag */
0369     if (type == &memblock.memory)
0370         in_slab = &memblock_memory_in_slab;
0371     else
0372         in_slab = &memblock_reserved_in_slab;
0373 
0374     /* Try to find some space for it.
0375      *
0376      * WARNING: We assume that either slab_is_available() and we use it or
0377      * we use MEMBLOCK for allocations. That means that this is unsafe to
0378      * use when bootmem is currently active (unless bootmem itself is
0379      * implemented on top of MEMBLOCK which isn't the case yet)
0380      *
0381      * This should however not be an issue for now, as we currently only
0382      * call into MEMBLOCK while it's still active, or much later when slab
0383      * is active for memory hotplug operations
0384      */
0385     if (use_slab) {
0386         new_array = kmalloc(new_size, GFP_KERNEL);
0387         addr = new_array ? __pa(new_array) : 0;
0388     } else {
0389         /* only exclude range when trying to double reserved.regions */
0390         if (type != &memblock.reserved)
0391             new_area_start = new_area_size = 0;
0392 
0393         addr = memblock_find_in_range(new_area_start + new_area_size,
0394                         memblock.current_limit,
0395                         new_alloc_size, PAGE_SIZE);
0396         if (!addr && new_area_size)
0397             addr = memblock_find_in_range(0,
0398                 min(new_area_start, memblock.current_limit),
0399                 new_alloc_size, PAGE_SIZE);
0400 
0401         new_array = addr ? __va(addr) : NULL;
0402     }
0403     if (!addr) {
0404         pr_err("memblock: Failed to double %s array from %ld to %ld entries !\n",
0405                memblock_type_name(type), type->max, type->max * 2);
0406         return -1;
0407     }
0408 
0409     memblock_dbg("memblock: %s is doubled to %ld at [%#010llx-%#010llx]",
0410             memblock_type_name(type), type->max * 2, (u64)addr,
0411             (u64)addr + new_size - 1);
0412 
0413     /*
0414      * Found space, we now need to move the array over before we add the
0415      * reserved region since it may be our reserved array itself that is
0416      * full.
0417      */
0418     memcpy(new_array, type->regions, old_size);
0419     memset(new_array + type->max, 0, old_size);
0420     old_array = type->regions;
0421     type->regions = new_array;
0422     type->max <<= 1;
0423 
0424     /* Free old array. We needn't free it if the array is the static one */
0425     if (*in_slab)
0426         kfree(old_array);
0427     else if (old_array != memblock_memory_init_regions &&
0428          old_array != memblock_reserved_init_regions)
0429         memblock_free(__pa(old_array), old_alloc_size);
0430 
0431     /*
0432      * Reserve the new array if that comes from the memblock.  Otherwise, we
0433      * needn't do it
0434      */
0435     if (!use_slab)
0436         BUG_ON(memblock_reserve(addr, new_alloc_size));
0437 
0438     /* Update slab flag */
0439     *in_slab = use_slab;
0440 
0441     return 0;
0442 }
0443 
0444 /**
0445  * memblock_merge_regions - merge neighboring compatible regions
0446  * @type: memblock type to scan
0447  *
0448  * Scan @type and merge neighboring compatible regions.
0449  */
0450 static void __init_memblock memblock_merge_regions(struct memblock_type *type)
0451 {
0452     int i = 0;
0453 
0454     /* cnt never goes below 1 */
0455     while (i < type->cnt - 1) {
0456         struct memblock_region *this = &type->regions[i];
0457         struct memblock_region *next = &type->regions[i + 1];
0458 
0459         if (this->base + this->size != next->base ||
0460             memblock_get_region_node(this) !=
0461             memblock_get_region_node(next) ||
0462             this->flags != next->flags) {
0463             BUG_ON(this->base + this->size > next->base);
0464             i++;
0465             continue;
0466         }
0467 
0468         this->size += next->size;
0469         /* move forward from next + 1, index of which is i + 2 */
0470         memmove(next, next + 1, (type->cnt - (i + 2)) * sizeof(*next));
0471         type->cnt--;
0472     }
0473 }
0474 
0475 /**
0476  * memblock_insert_region - insert new memblock region
0477  * @type:   memblock type to insert into
0478  * @idx:    index for the insertion point
0479  * @base:   base address of the new region
0480  * @size:   size of the new region
0481  * @nid:    node id of the new region
0482  * @flags:  flags of the new region
0483  *
0484  * Insert new memblock region [@base,@base+@size) into @type at @idx.
0485  * @type must already have extra room to accommodate the new region.
0486  */
0487 static void __init_memblock memblock_insert_region(struct memblock_type *type,
0488                            int idx, phys_addr_t base,
0489                            phys_addr_t size,
0490                            int nid, unsigned long flags)
0491 {
0492     struct memblock_region *rgn = &type->regions[idx];
0493 
0494     BUG_ON(type->cnt >= type->max);
0495     memmove(rgn + 1, rgn, (type->cnt - idx) * sizeof(*rgn));
0496     rgn->base = base;
0497     rgn->size = size;
0498     rgn->flags = flags;
0499     memblock_set_region_node(rgn, nid);
0500     type->cnt++;
0501     type->total_size += size;
0502 }
0503 
0504 /**
0505  * memblock_add_range - add new memblock region
0506  * @type: memblock type to add new region into
0507  * @base: base address of the new region
0508  * @size: size of the new region
0509  * @nid: nid of the new region
0510  * @flags: flags of the new region
0511  *
0512  * Add new memblock region [@base,@base+@size) into @type.  The new region
0513  * is allowed to overlap with existing ones - overlaps don't affect already
0514  * existing regions.  @type is guaranteed to be minimal (all neighbouring
0515  * compatible regions are merged) after the addition.
0516  *
0517  * RETURNS:
0518  * 0 on success, -errno on failure.
0519  */
0520 int __init_memblock memblock_add_range(struct memblock_type *type,
0521                 phys_addr_t base, phys_addr_t size,
0522                 int nid, unsigned long flags)
0523 {
0524     bool insert = false;
0525     phys_addr_t obase = base;
0526     phys_addr_t end = base + memblock_cap_size(base, &size);
0527     int idx, nr_new;
0528     struct memblock_region *rgn;
0529 
0530     if (!size)
0531         return 0;
0532 
0533     /* special case for empty array */
0534     if (type->regions[0].size == 0) {
0535         WARN_ON(type->cnt != 1 || type->total_size);
0536         type->regions[0].base = base;
0537         type->regions[0].size = size;
0538         type->regions[0].flags = flags;
0539         memblock_set_region_node(&type->regions[0], nid);
0540         type->total_size = size;
0541         return 0;
0542     }
0543 repeat:
0544     /*
0545      * The following is executed twice.  Once with %false @insert and
0546      * then with %true.  The first counts the number of regions needed
0547      * to accommodate the new area.  The second actually inserts them.
0548      */
0549     base = obase;
0550     nr_new = 0;
0551 
0552     for_each_memblock_type(type, rgn) {
0553         phys_addr_t rbase = rgn->base;
0554         phys_addr_t rend = rbase + rgn->size;
0555 
0556         if (rbase >= end)
0557             break;
0558         if (rend <= base)
0559             continue;
0560         /*
0561          * @rgn overlaps.  If it separates the lower part of new
0562          * area, insert that portion.
0563          */
0564         if (rbase > base) {
0565 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
0566             WARN_ON(nid != memblock_get_region_node(rgn));
0567 #endif
0568             WARN_ON(flags != rgn->flags);
0569             nr_new++;
0570             if (insert)
0571                 memblock_insert_region(type, idx++, base,
0572                                rbase - base, nid,
0573                                flags);
0574         }
0575         /* area below @rend is dealt with, forget about it */
0576         base = min(rend, end);
0577     }
0578 
0579     /* insert the remaining portion */
0580     if (base < end) {
0581         nr_new++;
0582         if (insert)
0583             memblock_insert_region(type, idx, base, end - base,
0584                            nid, flags);
0585     }
0586 
0587     if (!nr_new)
0588         return 0;
0589 
0590     /*
0591      * If this was the first round, resize array and repeat for actual
0592      * insertions; otherwise, merge and return.
0593      */
0594     if (!insert) {
0595         while (type->cnt + nr_new > type->max)
0596             if (memblock_double_array(type, obase, size) < 0)
0597                 return -ENOMEM;
0598         insert = true;
0599         goto repeat;
0600     } else {
0601         memblock_merge_regions(type);
0602         return 0;
0603     }
0604 }
0605 
0606 int __init_memblock memblock_add_node(phys_addr_t base, phys_addr_t size,
0607                        int nid)
0608 {
0609     return memblock_add_range(&memblock.memory, base, size, nid, 0);
0610 }
0611 
0612 int __init_memblock memblock_add(phys_addr_t base, phys_addr_t size)
0613 {
0614     memblock_dbg("memblock_add: [%#016llx-%#016llx] flags %#02lx %pF\n",
0615              (unsigned long long)base,
0616              (unsigned long long)base + size - 1,
0617              0UL, (void *)_RET_IP_);
0618 
0619     return memblock_add_range(&memblock.memory, base, size, MAX_NUMNODES, 0);
0620 }
0621 
0622 /**
0623  * memblock_isolate_range - isolate given range into disjoint memblocks
0624  * @type: memblock type to isolate range for
0625  * @base: base of range to isolate
0626  * @size: size of range to isolate
0627  * @start_rgn: out parameter for the start of isolated region
0628  * @end_rgn: out parameter for the end of isolated region
0629  *
0630  * Walk @type and ensure that regions don't cross the boundaries defined by
0631  * [@base,@base+@size).  Crossing regions are split at the boundaries,
0632  * which may create at most two more regions.  The index of the first
0633  * region inside the range is returned in *@start_rgn and end in *@end_rgn.
0634  *
0635  * RETURNS:
0636  * 0 on success, -errno on failure.
0637  */
0638 static int __init_memblock memblock_isolate_range(struct memblock_type *type,
0639                     phys_addr_t base, phys_addr_t size,
0640                     int *start_rgn, int *end_rgn)
0641 {
0642     phys_addr_t end = base + memblock_cap_size(base, &size);
0643     int idx;
0644     struct memblock_region *rgn;
0645 
0646     *start_rgn = *end_rgn = 0;
0647 
0648     if (!size)
0649         return 0;
0650 
0651     /* we'll create at most two more regions */
0652     while (type->cnt + 2 > type->max)
0653         if (memblock_double_array(type, base, size) < 0)
0654             return -ENOMEM;
0655 
0656     for_each_memblock_type(type, rgn) {
0657         phys_addr_t rbase = rgn->base;
0658         phys_addr_t rend = rbase + rgn->size;
0659 
0660         if (rbase >= end)
0661             break;
0662         if (rend <= base)
0663             continue;
0664 
0665         if (rbase < base) {
0666             /*
0667              * @rgn intersects from below.  Split and continue
0668              * to process the next region - the new top half.
0669              */
0670             rgn->base = base;
0671             rgn->size -= base - rbase;
0672             type->total_size -= base - rbase;
0673             memblock_insert_region(type, idx, rbase, base - rbase,
0674                            memblock_get_region_node(rgn),
0675                            rgn->flags);
0676         } else if (rend > end) {
0677             /*
0678              * @rgn intersects from above.  Split and redo the
0679              * current region - the new bottom half.
0680              */
0681             rgn->base = end;
0682             rgn->size -= end - rbase;
0683             type->total_size -= end - rbase;
0684             memblock_insert_region(type, idx--, rbase, end - rbase,
0685                            memblock_get_region_node(rgn),
0686                            rgn->flags);
0687         } else {
0688             /* @rgn is fully contained, record it */
0689             if (!*end_rgn)
0690                 *start_rgn = idx;
0691             *end_rgn = idx + 1;
0692         }
0693     }
0694 
0695     return 0;
0696 }
0697 
0698 static int __init_memblock memblock_remove_range(struct memblock_type *type,
0699                       phys_addr_t base, phys_addr_t size)
0700 {
0701     int start_rgn, end_rgn;
0702     int i, ret;
0703 
0704     ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
0705     if (ret)
0706         return ret;
0707 
0708     for (i = end_rgn - 1; i >= start_rgn; i--)
0709         memblock_remove_region(type, i);
0710     return 0;
0711 }
0712 
0713 int __init_memblock memblock_remove(phys_addr_t base, phys_addr_t size)
0714 {
0715     return memblock_remove_range(&memblock.memory, base, size);
0716 }
0717 
0718 
0719 int __init_memblock memblock_free(phys_addr_t base, phys_addr_t size)
0720 {
0721     memblock_dbg("   memblock_free: [%#016llx-%#016llx] %pF\n",
0722              (unsigned long long)base,
0723              (unsigned long long)base + size - 1,
0724              (void *)_RET_IP_);
0725 
0726     kmemleak_free_part_phys(base, size);
0727     return memblock_remove_range(&memblock.reserved, base, size);
0728 }
0729 
0730 int __init_memblock memblock_reserve(phys_addr_t base, phys_addr_t size)
0731 {
0732     memblock_dbg("memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF\n",
0733              (unsigned long long)base,
0734              (unsigned long long)base + size - 1,
0735              0UL, (void *)_RET_IP_);
0736 
0737     return memblock_add_range(&memblock.reserved, base, size, MAX_NUMNODES, 0);
0738 }
0739 
0740 /**
0741  *
0742  * This function isolates region [@base, @base + @size), and sets/clears flag
0743  *
0744  * Return 0 on success, -errno on failure.
0745  */
0746 static int __init_memblock memblock_setclr_flag(phys_addr_t base,
0747                 phys_addr_t size, int set, int flag)
0748 {
0749     struct memblock_type *type = &memblock.memory;
0750     int i, ret, start_rgn, end_rgn;
0751 
0752     ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
0753     if (ret)
0754         return ret;
0755 
0756     for (i = start_rgn; i < end_rgn; i++)
0757         if (set)
0758             memblock_set_region_flags(&type->regions[i], flag);
0759         else
0760             memblock_clear_region_flags(&type->regions[i], flag);
0761 
0762     memblock_merge_regions(type);
0763     return 0;
0764 }
0765 
0766 /**
0767  * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG.
0768  * @base: the base phys addr of the region
0769  * @size: the size of the region
0770  *
0771  * Return 0 on success, -errno on failure.
0772  */
0773 int __init_memblock memblock_mark_hotplug(phys_addr_t base, phys_addr_t size)
0774 {
0775     return memblock_setclr_flag(base, size, 1, MEMBLOCK_HOTPLUG);
0776 }
0777 
0778 /**
0779  * memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region.
0780  * @base: the base phys addr of the region
0781  * @size: the size of the region
0782  *
0783  * Return 0 on success, -errno on failure.
0784  */
0785 int __init_memblock memblock_clear_hotplug(phys_addr_t base, phys_addr_t size)
0786 {
0787     return memblock_setclr_flag(base, size, 0, MEMBLOCK_HOTPLUG);
0788 }
0789 
0790 /**
0791  * memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR.
0792  * @base: the base phys addr of the region
0793  * @size: the size of the region
0794  *
0795  * Return 0 on success, -errno on failure.
0796  */
0797 int __init_memblock memblock_mark_mirror(phys_addr_t base, phys_addr_t size)
0798 {
0799     system_has_some_mirror = true;
0800 
0801     return memblock_setclr_flag(base, size, 1, MEMBLOCK_MIRROR);
0802 }
0803 
0804 /**
0805  * memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP.
0806  * @base: the base phys addr of the region
0807  * @size: the size of the region
0808  *
0809  * Return 0 on success, -errno on failure.
0810  */
0811 int __init_memblock memblock_mark_nomap(phys_addr_t base, phys_addr_t size)
0812 {
0813     return memblock_setclr_flag(base, size, 1, MEMBLOCK_NOMAP);
0814 }
0815 
0816 /**
0817  * __next_reserved_mem_region - next function for for_each_reserved_region()
0818  * @idx: pointer to u64 loop variable
0819  * @out_start: ptr to phys_addr_t for start address of the region, can be %NULL
0820  * @out_end: ptr to phys_addr_t for end address of the region, can be %NULL
0821  *
0822  * Iterate over all reserved memory regions.
0823  */
0824 void __init_memblock __next_reserved_mem_region(u64 *idx,
0825                        phys_addr_t *out_start,
0826                        phys_addr_t *out_end)
0827 {
0828     struct memblock_type *type = &memblock.reserved;
0829 
0830     if (*idx < type->cnt) {
0831         struct memblock_region *r = &type->regions[*idx];
0832         phys_addr_t base = r->base;
0833         phys_addr_t size = r->size;
0834 
0835         if (out_start)
0836             *out_start = base;
0837         if (out_end)
0838             *out_end = base + size - 1;
0839 
0840         *idx += 1;
0841         return;
0842     }
0843 
0844     /* signal end of iteration */
0845     *idx = ULLONG_MAX;
0846 }
0847 
0848 /**
0849  * __next__mem_range - next function for for_each_free_mem_range() etc.
0850  * @idx: pointer to u64 loop variable
0851  * @nid: node selector, %NUMA_NO_NODE for all nodes
0852  * @flags: pick from blocks based on memory attributes
0853  * @type_a: pointer to memblock_type from where the range is taken
0854  * @type_b: pointer to memblock_type which excludes memory from being taken
0855  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
0856  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
0857  * @out_nid: ptr to int for nid of the range, can be %NULL
0858  *
0859  * Find the first area from *@idx which matches @nid, fill the out
0860  * parameters, and update *@idx for the next iteration.  The lower 32bit of
0861  * *@idx contains index into type_a and the upper 32bit indexes the
0862  * areas before each region in type_b.  For example, if type_b regions
0863  * look like the following,
0864  *
0865  *  0:[0-16), 1:[32-48), 2:[128-130)
0866  *
0867  * The upper 32bit indexes the following regions.
0868  *
0869  *  0:[0-0), 1:[16-32), 2:[48-128), 3:[130-MAX)
0870  *
0871  * As both region arrays are sorted, the function advances the two indices
0872  * in lockstep and returns each intersection.
0873  */
0874 void __init_memblock __next_mem_range(u64 *idx, int nid, ulong flags,
0875                       struct memblock_type *type_a,
0876                       struct memblock_type *type_b,
0877                       phys_addr_t *out_start,
0878                       phys_addr_t *out_end, int *out_nid)
0879 {
0880     int idx_a = *idx & 0xffffffff;
0881     int idx_b = *idx >> 32;
0882 
0883     if (WARN_ONCE(nid == MAX_NUMNODES,
0884     "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
0885         nid = NUMA_NO_NODE;
0886 
0887     for (; idx_a < type_a->cnt; idx_a++) {
0888         struct memblock_region *m = &type_a->regions[idx_a];
0889 
0890         phys_addr_t m_start = m->base;
0891         phys_addr_t m_end = m->base + m->size;
0892         int     m_nid = memblock_get_region_node(m);
0893 
0894         /* only memory regions are associated with nodes, check it */
0895         if (nid != NUMA_NO_NODE && nid != m_nid)
0896             continue;
0897 
0898         /* skip hotpluggable memory regions if needed */
0899         if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
0900             continue;
0901 
0902         /* if we want mirror memory skip non-mirror memory regions */
0903         if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
0904             continue;
0905 
0906         /* skip nomap memory unless we were asked for it explicitly */
0907         if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
0908             continue;
0909 
0910         if (!type_b) {
0911             if (out_start)
0912                 *out_start = m_start;
0913             if (out_end)
0914                 *out_end = m_end;
0915             if (out_nid)
0916                 *out_nid = m_nid;
0917             idx_a++;
0918             *idx = (u32)idx_a | (u64)idx_b << 32;
0919             return;
0920         }
0921 
0922         /* scan areas before each reservation */
0923         for (; idx_b < type_b->cnt + 1; idx_b++) {
0924             struct memblock_region *r;
0925             phys_addr_t r_start;
0926             phys_addr_t r_end;
0927 
0928             r = &type_b->regions[idx_b];
0929             r_start = idx_b ? r[-1].base + r[-1].size : 0;
0930             r_end = idx_b < type_b->cnt ?
0931                 r->base : ULLONG_MAX;
0932 
0933             /*
0934              * if idx_b advanced past idx_a,
0935              * break out to advance idx_a
0936              */
0937             if (r_start >= m_end)
0938                 break;
0939             /* if the two regions intersect, we're done */
0940             if (m_start < r_end) {
0941                 if (out_start)
0942                     *out_start =
0943                         max(m_start, r_start);
0944                 if (out_end)
0945                     *out_end = min(m_end, r_end);
0946                 if (out_nid)
0947                     *out_nid = m_nid;
0948                 /*
0949                  * The region which ends first is
0950                  * advanced for the next iteration.
0951                  */
0952                 if (m_end <= r_end)
0953                     idx_a++;
0954                 else
0955                     idx_b++;
0956                 *idx = (u32)idx_a | (u64)idx_b << 32;
0957                 return;
0958             }
0959         }
0960     }
0961 
0962     /* signal end of iteration */
0963     *idx = ULLONG_MAX;
0964 }
0965 
0966 /**
0967  * __next_mem_range_rev - generic next function for for_each_*_range_rev()
0968  *
0969  * Finds the next range from type_a which is not marked as unsuitable
0970  * in type_b.
0971  *
0972  * @idx: pointer to u64 loop variable
0973  * @nid: node selector, %NUMA_NO_NODE for all nodes
0974  * @flags: pick from blocks based on memory attributes
0975  * @type_a: pointer to memblock_type from where the range is taken
0976  * @type_b: pointer to memblock_type which excludes memory from being taken
0977  * @out_start: ptr to phys_addr_t for start address of the range, can be %NULL
0978  * @out_end: ptr to phys_addr_t for end address of the range, can be %NULL
0979  * @out_nid: ptr to int for nid of the range, can be %NULL
0980  *
0981  * Reverse of __next_mem_range().
0982  */
0983 void __init_memblock __next_mem_range_rev(u64 *idx, int nid, ulong flags,
0984                       struct memblock_type *type_a,
0985                       struct memblock_type *type_b,
0986                       phys_addr_t *out_start,
0987                       phys_addr_t *out_end, int *out_nid)
0988 {
0989     int idx_a = *idx & 0xffffffff;
0990     int idx_b = *idx >> 32;
0991 
0992     if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
0993         nid = NUMA_NO_NODE;
0994 
0995     if (*idx == (u64)ULLONG_MAX) {
0996         idx_a = type_a->cnt - 1;
0997         if (type_b != NULL)
0998             idx_b = type_b->cnt;
0999         else
1000             idx_b = 0;
1001     }
1002 
1003     for (; idx_a >= 0; idx_a--) {
1004         struct memblock_region *m = &type_a->regions[idx_a];
1005 
1006         phys_addr_t m_start = m->base;
1007         phys_addr_t m_end = m->base + m->size;
1008         int m_nid = memblock_get_region_node(m);
1009 
1010         /* only memory regions are associated with nodes, check it */
1011         if (nid != NUMA_NO_NODE && nid != m_nid)
1012             continue;
1013 
1014         /* skip hotpluggable memory regions if needed */
1015         if (movable_node_is_enabled() && memblock_is_hotpluggable(m))
1016             continue;
1017 
1018         /* if we want mirror memory skip non-mirror memory regions */
1019         if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m))
1020             continue;
1021 
1022         /* skip nomap memory unless we were asked for it explicitly */
1023         if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m))
1024             continue;
1025 
1026         if (!type_b) {
1027             if (out_start)
1028                 *out_start = m_start;
1029             if (out_end)
1030                 *out_end = m_end;
1031             if (out_nid)
1032                 *out_nid = m_nid;
1033             idx_a--;
1034             *idx = (u32)idx_a | (u64)idx_b << 32;
1035             return;
1036         }
1037 
1038         /* scan areas before each reservation */
1039         for (; idx_b >= 0; idx_b--) {
1040             struct memblock_region *r;
1041             phys_addr_t r_start;
1042             phys_addr_t r_end;
1043 
1044             r = &type_b->regions[idx_b];
1045             r_start = idx_b ? r[-1].base + r[-1].size : 0;
1046             r_end = idx_b < type_b->cnt ?
1047                 r->base : ULLONG_MAX;
1048             /*
1049              * if idx_b advanced past idx_a,
1050              * break out to advance idx_a
1051              */
1052 
1053             if (r_end <= m_start)
1054                 break;
1055             /* if the two regions intersect, we're done */
1056             if (m_end > r_start) {
1057                 if (out_start)
1058                     *out_start = max(m_start, r_start);
1059                 if (out_end)
1060                     *out_end = min(m_end, r_end);
1061                 if (out_nid)
1062                     *out_nid = m_nid;
1063                 if (m_start >= r_start)
1064                     idx_a--;
1065                 else
1066                     idx_b--;
1067                 *idx = (u32)idx_a | (u64)idx_b << 32;
1068                 return;
1069             }
1070         }
1071     }
1072     /* signal end of iteration */
1073     *idx = ULLONG_MAX;
1074 }
1075 
1076 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1077 /*
1078  * Common iterator interface used to define for_each_mem_range().
1079  */
1080 void __init_memblock __next_mem_pfn_range(int *idx, int nid,
1081                 unsigned long *out_start_pfn,
1082                 unsigned long *out_end_pfn, int *out_nid)
1083 {
1084     struct memblock_type *type = &memblock.memory;
1085     struct memblock_region *r;
1086 
1087     while (++*idx < type->cnt) {
1088         r = &type->regions[*idx];
1089 
1090         if (PFN_UP(r->base) >= PFN_DOWN(r->base + r->size))
1091             continue;
1092         if (nid == MAX_NUMNODES || nid == r->nid)
1093             break;
1094     }
1095     if (*idx >= type->cnt) {
1096         *idx = -1;
1097         return;
1098     }
1099 
1100     if (out_start_pfn)
1101         *out_start_pfn = PFN_UP(r->base);
1102     if (out_end_pfn)
1103         *out_end_pfn = PFN_DOWN(r->base + r->size);
1104     if (out_nid)
1105         *out_nid = r->nid;
1106 }
1107 
1108 /**
1109  * memblock_set_node - set node ID on memblock regions
1110  * @base: base of area to set node ID for
1111  * @size: size of area to set node ID for
1112  * @type: memblock type to set node ID for
1113  * @nid: node ID to set
1114  *
1115  * Set the nid of memblock @type regions in [@base,@base+@size) to @nid.
1116  * Regions which cross the area boundaries are split as necessary.
1117  *
1118  * RETURNS:
1119  * 0 on success, -errno on failure.
1120  */
1121 int __init_memblock memblock_set_node(phys_addr_t base, phys_addr_t size,
1122                       struct memblock_type *type, int nid)
1123 {
1124     int start_rgn, end_rgn;
1125     int i, ret;
1126 
1127     ret = memblock_isolate_range(type, base, size, &start_rgn, &end_rgn);
1128     if (ret)
1129         return ret;
1130 
1131     for (i = start_rgn; i < end_rgn; i++)
1132         memblock_set_region_node(&type->regions[i], nid);
1133 
1134     memblock_merge_regions(type);
1135     return 0;
1136 }
1137 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
1138 
1139 static phys_addr_t __init memblock_alloc_range_nid(phys_addr_t size,
1140                     phys_addr_t align, phys_addr_t start,
1141                     phys_addr_t end, int nid, ulong flags)
1142 {
1143     phys_addr_t found;
1144 
1145     if (!align)
1146         align = SMP_CACHE_BYTES;
1147 
1148     found = memblock_find_in_range_node(size, align, start, end, nid,
1149                         flags);
1150     if (found && !memblock_reserve(found, size)) {
1151         /*
1152          * The min_count is set to 0 so that memblock allocations are
1153          * never reported as leaks.
1154          */
1155         kmemleak_alloc_phys(found, size, 0, 0);
1156         return found;
1157     }
1158     return 0;
1159 }
1160 
1161 phys_addr_t __init memblock_alloc_range(phys_addr_t size, phys_addr_t align,
1162                     phys_addr_t start, phys_addr_t end,
1163                     ulong flags)
1164 {
1165     return memblock_alloc_range_nid(size, align, start, end, NUMA_NO_NODE,
1166                     flags);
1167 }
1168 
1169 static phys_addr_t __init memblock_alloc_base_nid(phys_addr_t size,
1170                     phys_addr_t align, phys_addr_t max_addr,
1171                     int nid, ulong flags)
1172 {
1173     return memblock_alloc_range_nid(size, align, 0, max_addr, nid, flags);
1174 }
1175 
1176 phys_addr_t __init memblock_alloc_nid(phys_addr_t size, phys_addr_t align, int nid)
1177 {
1178     ulong flags = choose_memblock_flags();
1179     phys_addr_t ret;
1180 
1181 again:
1182     ret = memblock_alloc_base_nid(size, align, MEMBLOCK_ALLOC_ACCESSIBLE,
1183                       nid, flags);
1184 
1185     if (!ret && (flags & MEMBLOCK_MIRROR)) {
1186         flags &= ~MEMBLOCK_MIRROR;
1187         goto again;
1188     }
1189     return ret;
1190 }
1191 
1192 phys_addr_t __init __memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1193 {
1194     return memblock_alloc_base_nid(size, align, max_addr, NUMA_NO_NODE,
1195                        MEMBLOCK_NONE);
1196 }
1197 
1198 phys_addr_t __init memblock_alloc_base(phys_addr_t size, phys_addr_t align, phys_addr_t max_addr)
1199 {
1200     phys_addr_t alloc;
1201 
1202     alloc = __memblock_alloc_base(size, align, max_addr);
1203 
1204     if (alloc == 0)
1205         panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
1206               (unsigned long long) size, (unsigned long long) max_addr);
1207 
1208     return alloc;
1209 }
1210 
1211 phys_addr_t __init memblock_alloc(phys_addr_t size, phys_addr_t align)
1212 {
1213     return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1214 }
1215 
1216 phys_addr_t __init memblock_alloc_try_nid(phys_addr_t size, phys_addr_t align, int nid)
1217 {
1218     phys_addr_t res = memblock_alloc_nid(size, align, nid);
1219 
1220     if (res)
1221         return res;
1222     return memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
1223 }
1224 
1225 /**
1226  * memblock_virt_alloc_internal - allocate boot memory block
1227  * @size: size of memory block to be allocated in bytes
1228  * @align: alignment of the region and block's size
1229  * @min_addr: the lower bound of the memory region to allocate (phys address)
1230  * @max_addr: the upper bound of the memory region to allocate (phys address)
1231  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1232  *
1233  * The @min_addr limit is dropped if it can not be satisfied and the allocation
1234  * will fall back to memory below @min_addr. Also, allocation may fall back
1235  * to any node in the system if the specified node can not
1236  * hold the requested memory.
1237  *
1238  * The allocation is performed from memory region limited by
1239  * memblock.current_limit if @max_addr == %BOOTMEM_ALLOC_ACCESSIBLE.
1240  *
1241  * The memory block is aligned on SMP_CACHE_BYTES if @align == 0.
1242  *
1243  * The phys address of allocated boot memory block is converted to virtual and
1244  * allocated memory is reset to 0.
1245  *
1246  * In addition, function sets the min_count to 0 using kmemleak_alloc for
1247  * allocated boot memory block, so that it is never reported as leaks.
1248  *
1249  * RETURNS:
1250  * Virtual address of allocated memory block on success, NULL on failure.
1251  */
1252 static void * __init memblock_virt_alloc_internal(
1253                 phys_addr_t size, phys_addr_t align,
1254                 phys_addr_t min_addr, phys_addr_t max_addr,
1255                 int nid)
1256 {
1257     phys_addr_t alloc;
1258     void *ptr;
1259     ulong flags = choose_memblock_flags();
1260 
1261     if (WARN_ONCE(nid == MAX_NUMNODES, "Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead\n"))
1262         nid = NUMA_NO_NODE;
1263 
1264     /*
1265      * Detect any accidental use of these APIs after slab is ready, as at
1266      * this moment memblock may be deinitialized already and its
1267      * internal data may be destroyed (after execution of free_all_bootmem)
1268      */
1269     if (WARN_ON_ONCE(slab_is_available()))
1270         return kzalloc_node(size, GFP_NOWAIT, nid);
1271 
1272     if (!align)
1273         align = SMP_CACHE_BYTES;
1274 
1275     if (max_addr > memblock.current_limit)
1276         max_addr = memblock.current_limit;
1277 
1278 again:
1279     alloc = memblock_find_in_range_node(size, align, min_addr, max_addr,
1280                         nid, flags);
1281     if (alloc)
1282         goto done;
1283 
1284     if (nid != NUMA_NO_NODE) {
1285         alloc = memblock_find_in_range_node(size, align, min_addr,
1286                             max_addr, NUMA_NO_NODE,
1287                             flags);
1288         if (alloc)
1289             goto done;
1290     }
1291 
1292     if (min_addr) {
1293         min_addr = 0;
1294         goto again;
1295     }
1296 
1297     if (flags & MEMBLOCK_MIRROR) {
1298         flags &= ~MEMBLOCK_MIRROR;
1299         pr_warn("Could not allocate %pap bytes of mirrored memory\n",
1300             &size);
1301         goto again;
1302     }
1303 
1304     return NULL;
1305 done:
1306     memblock_reserve(alloc, size);
1307     ptr = phys_to_virt(alloc);
1308     memset(ptr, 0, size);
1309 
1310     /*
1311      * The min_count is set to 0 so that bootmem allocated blocks
1312      * are never reported as leaks. This is because many of these blocks
1313      * are only referred via the physical address which is not
1314      * looked up by kmemleak.
1315      */
1316     kmemleak_alloc(ptr, size, 0, 0);
1317 
1318     return ptr;
1319 }
1320 
1321 /**
1322  * memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
1323  * @size: size of memory block to be allocated in bytes
1324  * @align: alignment of the region and block's size
1325  * @min_addr: the lower bound of the memory region from where the allocation
1326  *    is preferred (phys address)
1327  * @max_addr: the upper bound of the memory region from where the allocation
1328  *        is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1329  *        allocate only from memory limited by memblock.current_limit value
1330  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1331  *
1332  * Public version of _memblock_virt_alloc_try_nid_nopanic() which provides
1333  * additional debug information (including caller info), if enabled.
1334  *
1335  * RETURNS:
1336  * Virtual address of allocated memory block on success, NULL on failure.
1337  */
1338 void * __init memblock_virt_alloc_try_nid_nopanic(
1339                 phys_addr_t size, phys_addr_t align,
1340                 phys_addr_t min_addr, phys_addr_t max_addr,
1341                 int nid)
1342 {
1343     memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1344              __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1345              (u64)max_addr, (void *)_RET_IP_);
1346     return memblock_virt_alloc_internal(size, align, min_addr,
1347                          max_addr, nid);
1348 }
1349 
1350 /**
1351  * memblock_virt_alloc_try_nid - allocate boot memory block with panicking
1352  * @size: size of memory block to be allocated in bytes
1353  * @align: alignment of the region and block's size
1354  * @min_addr: the lower bound of the memory region from where the allocation
1355  *    is preferred (phys address)
1356  * @max_addr: the upper bound of the memory region from where the allocation
1357  *        is preferred (phys address), or %BOOTMEM_ALLOC_ACCESSIBLE to
1358  *        allocate only from memory limited by memblock.current_limit value
1359  * @nid: nid of the free area to find, %NUMA_NO_NODE for any node
1360  *
1361  * Public panicking version of _memblock_virt_alloc_try_nid_nopanic()
1362  * which provides debug information (including caller info), if enabled,
1363  * and panics if the request can not be satisfied.
1364  *
1365  * RETURNS:
1366  * Virtual address of allocated memory block on success, NULL on failure.
1367  */
1368 void * __init memblock_virt_alloc_try_nid(
1369             phys_addr_t size, phys_addr_t align,
1370             phys_addr_t min_addr, phys_addr_t max_addr,
1371             int nid)
1372 {
1373     void *ptr;
1374 
1375     memblock_dbg("%s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF\n",
1376              __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1377              (u64)max_addr, (void *)_RET_IP_);
1378     ptr = memblock_virt_alloc_internal(size, align,
1379                        min_addr, max_addr, nid);
1380     if (ptr)
1381         return ptr;
1382 
1383     panic("%s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx\n",
1384           __func__, (u64)size, (u64)align, nid, (u64)min_addr,
1385           (u64)max_addr);
1386     return NULL;
1387 }
1388 
1389 /**
1390  * __memblock_free_early - free boot memory block
1391  * @base: phys starting address of the  boot memory block
1392  * @size: size of the boot memory block in bytes
1393  *
1394  * Free boot memory block previously allocated by memblock_virt_alloc_xx() API.
1395  * The freeing memory will not be released to the buddy allocator.
1396  */
1397 void __init __memblock_free_early(phys_addr_t base, phys_addr_t size)
1398 {
1399     memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1400              __func__, (u64)base, (u64)base + size - 1,
1401              (void *)_RET_IP_);
1402     kmemleak_free_part_phys(base, size);
1403     memblock_remove_range(&memblock.reserved, base, size);
1404 }
1405 
1406 /*
1407  * __memblock_free_late - free bootmem block pages directly to buddy allocator
1408  * @addr: phys starting address of the  boot memory block
1409  * @size: size of the boot memory block in bytes
1410  *
1411  * This is only useful when the bootmem allocator has already been torn
1412  * down, but we are still initializing the system.  Pages are released directly
1413  * to the buddy allocator, no bootmem metadata is updated because it is gone.
1414  */
1415 void __init __memblock_free_late(phys_addr_t base, phys_addr_t size)
1416 {
1417     u64 cursor, end;
1418 
1419     memblock_dbg("%s: [%#016llx-%#016llx] %pF\n",
1420              __func__, (u64)base, (u64)base + size - 1,
1421              (void *)_RET_IP_);
1422     kmemleak_free_part_phys(base, size);
1423     cursor = PFN_UP(base);
1424     end = PFN_DOWN(base + size);
1425 
1426     for (; cursor < end; cursor++) {
1427         __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
1428         totalram_pages++;
1429     }
1430 }
1431 
1432 /*
1433  * Remaining API functions
1434  */
1435 
1436 phys_addr_t __init_memblock memblock_phys_mem_size(void)
1437 {
1438     return memblock.memory.total_size;
1439 }
1440 
1441 phys_addr_t __init_memblock memblock_reserved_size(void)
1442 {
1443     return memblock.reserved.total_size;
1444 }
1445 
1446 phys_addr_t __init memblock_mem_size(unsigned long limit_pfn)
1447 {
1448     unsigned long pages = 0;
1449     struct memblock_region *r;
1450     unsigned long start_pfn, end_pfn;
1451 
1452     for_each_memblock(memory, r) {
1453         start_pfn = memblock_region_memory_base_pfn(r);
1454         end_pfn = memblock_region_memory_end_pfn(r);
1455         start_pfn = min_t(unsigned long, start_pfn, limit_pfn);
1456         end_pfn = min_t(unsigned long, end_pfn, limit_pfn);
1457         pages += end_pfn - start_pfn;
1458     }
1459 
1460     return PFN_PHYS(pages);
1461 }
1462 
1463 /* lowest address */
1464 phys_addr_t __init_memblock memblock_start_of_DRAM(void)
1465 {
1466     return memblock.memory.regions[0].base;
1467 }
1468 
1469 phys_addr_t __init_memblock memblock_end_of_DRAM(void)
1470 {
1471     int idx = memblock.memory.cnt - 1;
1472 
1473     return (memblock.memory.regions[idx].base + memblock.memory.regions[idx].size);
1474 }
1475 
1476 static phys_addr_t __init_memblock __find_max_addr(phys_addr_t limit)
1477 {
1478     phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1479     struct memblock_region *r;
1480 
1481     /*
1482      * translate the memory @limit size into the max address within one of
1483      * the memory memblock regions, if the @limit exceeds the total size
1484      * of those regions, max_addr will keep original value ULLONG_MAX
1485      */
1486     for_each_memblock(memory, r) {
1487         if (limit <= r->size) {
1488             max_addr = r->base + limit;
1489             break;
1490         }
1491         limit -= r->size;
1492     }
1493 
1494     return max_addr;
1495 }
1496 
1497 void __init memblock_enforce_memory_limit(phys_addr_t limit)
1498 {
1499     phys_addr_t max_addr = (phys_addr_t)ULLONG_MAX;
1500 
1501     if (!limit)
1502         return;
1503 
1504     max_addr = __find_max_addr(limit);
1505 
1506     /* @limit exceeds the total size of the memory, do nothing */
1507     if (max_addr == (phys_addr_t)ULLONG_MAX)
1508         return;
1509 
1510     /* truncate both memory and reserved regions */
1511     memblock_remove_range(&memblock.memory, max_addr,
1512                   (phys_addr_t)ULLONG_MAX);
1513     memblock_remove_range(&memblock.reserved, max_addr,
1514                   (phys_addr_t)ULLONG_MAX);
1515 }
1516 
1517 void __init memblock_mem_limit_remove_map(phys_addr_t limit)
1518 {
1519     struct memblock_type *type = &memblock.memory;
1520     phys_addr_t max_addr;
1521     int i, ret, start_rgn, end_rgn;
1522 
1523     if (!limit)
1524         return;
1525 
1526     max_addr = __find_max_addr(limit);
1527 
1528     /* @limit exceeds the total size of the memory, do nothing */
1529     if (max_addr == (phys_addr_t)ULLONG_MAX)
1530         return;
1531 
1532     ret = memblock_isolate_range(type, max_addr, (phys_addr_t)ULLONG_MAX,
1533                 &start_rgn, &end_rgn);
1534     if (ret)
1535         return;
1536 
1537     /* remove all the MAP regions above the limit */
1538     for (i = end_rgn - 1; i >= start_rgn; i--) {
1539         if (!memblock_is_nomap(&type->regions[i]))
1540             memblock_remove_region(type, i);
1541     }
1542     /* truncate the reserved regions */
1543     memblock_remove_range(&memblock.reserved, max_addr,
1544                   (phys_addr_t)ULLONG_MAX);
1545 }
1546 
1547 static int __init_memblock memblock_search(struct memblock_type *type, phys_addr_t addr)
1548 {
1549     unsigned int left = 0, right = type->cnt;
1550 
1551     do {
1552         unsigned int mid = (right + left) / 2;
1553 
1554         if (addr < type->regions[mid].base)
1555             right = mid;
1556         else if (addr >= (type->regions[mid].base +
1557                   type->regions[mid].size))
1558             left = mid + 1;
1559         else
1560             return mid;
1561     } while (left < right);
1562     return -1;
1563 }
1564 
1565 bool __init memblock_is_reserved(phys_addr_t addr)
1566 {
1567     return memblock_search(&memblock.reserved, addr) != -1;
1568 }
1569 
1570 bool __init_memblock memblock_is_memory(phys_addr_t addr)
1571 {
1572     return memblock_search(&memblock.memory, addr) != -1;
1573 }
1574 
1575 int __init_memblock memblock_is_map_memory(phys_addr_t addr)
1576 {
1577     int i = memblock_search(&memblock.memory, addr);
1578 
1579     if (i == -1)
1580         return false;
1581     return !memblock_is_nomap(&memblock.memory.regions[i]);
1582 }
1583 
1584 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1585 int __init_memblock memblock_search_pfn_nid(unsigned long pfn,
1586              unsigned long *start_pfn, unsigned long *end_pfn)
1587 {
1588     struct memblock_type *type = &memblock.memory;
1589     int mid = memblock_search(type, PFN_PHYS(pfn));
1590 
1591     if (mid == -1)
1592         return -1;
1593 
1594     *start_pfn = PFN_DOWN(type->regions[mid].base);
1595     *end_pfn = PFN_DOWN(type->regions[mid].base + type->regions[mid].size);
1596 
1597     return type->regions[mid].nid;
1598 }
1599 #endif
1600 
1601 /**
1602  * memblock_is_region_memory - check if a region is a subset of memory
1603  * @base: base of region to check
1604  * @size: size of region to check
1605  *
1606  * Check if the region [@base, @base+@size) is a subset of a memory block.
1607  *
1608  * RETURNS:
1609  * 0 if false, non-zero if true
1610  */
1611 int __init_memblock memblock_is_region_memory(phys_addr_t base, phys_addr_t size)
1612 {
1613     int idx = memblock_search(&memblock.memory, base);
1614     phys_addr_t end = base + memblock_cap_size(base, &size);
1615 
1616     if (idx == -1)
1617         return 0;
1618     return memblock.memory.regions[idx].base <= base &&
1619         (memblock.memory.regions[idx].base +
1620          memblock.memory.regions[idx].size) >= end;
1621 }
1622 
1623 /**
1624  * memblock_is_region_reserved - check if a region intersects reserved memory
1625  * @base: base of region to check
1626  * @size: size of region to check
1627  *
1628  * Check if the region [@base, @base+@size) intersects a reserved memory block.
1629  *
1630  * RETURNS:
1631  * True if they intersect, false if not.
1632  */
1633 bool __init_memblock memblock_is_region_reserved(phys_addr_t base, phys_addr_t size)
1634 {
1635     memblock_cap_size(base, &size);
1636     return memblock_overlaps_region(&memblock.reserved, base, size);
1637 }
1638 
1639 void __init_memblock memblock_trim_memory(phys_addr_t align)
1640 {
1641     phys_addr_t start, end, orig_start, orig_end;
1642     struct memblock_region *r;
1643 
1644     for_each_memblock(memory, r) {
1645         orig_start = r->base;
1646         orig_end = r->base + r->size;
1647         start = round_up(orig_start, align);
1648         end = round_down(orig_end, align);
1649 
1650         if (start == orig_start && end == orig_end)
1651             continue;
1652 
1653         if (start < end) {
1654             r->base = start;
1655             r->size = end - start;
1656         } else {
1657             memblock_remove_region(&memblock.memory,
1658                            r - memblock.memory.regions);
1659             r--;
1660         }
1661     }
1662 }
1663 
1664 void __init_memblock memblock_set_current_limit(phys_addr_t limit)
1665 {
1666     memblock.current_limit = limit;
1667 }
1668 
1669 phys_addr_t __init_memblock memblock_get_current_limit(void)
1670 {
1671     return memblock.current_limit;
1672 }
1673 
1674 static void __init_memblock memblock_dump(struct memblock_type *type, char *name)
1675 {
1676     unsigned long long base, size;
1677     unsigned long flags;
1678     int idx;
1679     struct memblock_region *rgn;
1680 
1681     pr_info(" %s.cnt  = 0x%lx\n", name, type->cnt);
1682 
1683     for_each_memblock_type(type, rgn) {
1684         char nid_buf[32] = "";
1685 
1686         base = rgn->base;
1687         size = rgn->size;
1688         flags = rgn->flags;
1689 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
1690         if (memblock_get_region_node(rgn) != MAX_NUMNODES)
1691             snprintf(nid_buf, sizeof(nid_buf), " on node %d",
1692                  memblock_get_region_node(rgn));
1693 #endif
1694         pr_info(" %s[%#x]\t[%#016llx-%#016llx], %#llx bytes%s flags: %#lx\n",
1695             name, idx, base, base + size - 1, size, nid_buf, flags);
1696     }
1697 }
1698 
1699 void __init_memblock __memblock_dump_all(void)
1700 {
1701     pr_info("MEMBLOCK configuration:\n");
1702     pr_info(" memory size = %#llx reserved size = %#llx\n",
1703         (unsigned long long)memblock.memory.total_size,
1704         (unsigned long long)memblock.reserved.total_size);
1705 
1706     memblock_dump(&memblock.memory, "memory");
1707     memblock_dump(&memblock.reserved, "reserved");
1708 }
1709 
1710 void __init memblock_allow_resize(void)
1711 {
1712     memblock_can_resize = 1;
1713 }
1714 
1715 static int __init early_memblock(char *p)
1716 {
1717     if (p && strstr(p, "debug"))
1718         memblock_debug = 1;
1719     return 0;
1720 }
1721 early_param("memblock", early_memblock);
1722 
1723 #if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
1724 
1725 static int memblock_debug_show(struct seq_file *m, void *private)
1726 {
1727     struct memblock_type *type = m->private;
1728     struct memblock_region *reg;
1729     int i;
1730 
1731     for (i = 0; i < type->cnt; i++) {
1732         reg = &type->regions[i];
1733         seq_printf(m, "%4d: ", i);
1734         if (sizeof(phys_addr_t) == 4)
1735             seq_printf(m, "0x%08lx..0x%08lx\n",
1736                    (unsigned long)reg->base,
1737                    (unsigned long)(reg->base + reg->size - 1));
1738         else
1739             seq_printf(m, "0x%016llx..0x%016llx\n",
1740                    (unsigned long long)reg->base,
1741                    (unsigned long long)(reg->base + reg->size - 1));
1742 
1743     }
1744     return 0;
1745 }
1746 
1747 static int memblock_debug_open(struct inode *inode, struct file *file)
1748 {
1749     return single_open(file, memblock_debug_show, inode->i_private);
1750 }
1751 
1752 static const struct file_operations memblock_debug_fops = {
1753     .open = memblock_debug_open,
1754     .read = seq_read,
1755     .llseek = seq_lseek,
1756     .release = single_release,
1757 };
1758 
1759 static int __init memblock_init_debugfs(void)
1760 {
1761     struct dentry *root = debugfs_create_dir("memblock", NULL);
1762     if (!root)
1763         return -ENXIO;
1764     debugfs_create_file("memory", S_IRUGO, root, &memblock.memory, &memblock_debug_fops);
1765     debugfs_create_file("reserved", S_IRUGO, root, &memblock.reserved, &memblock_debug_fops);
1766 #ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
1767     debugfs_create_file("physmem", S_IRUGO, root, &memblock.physmem, &memblock_debug_fops);
1768 #endif
1769 
1770     return 0;
1771 }
1772 __initcall(memblock_init_debugfs);
1773 
1774 #endif /* CONFIG_DEBUG_FS */