mm/nobootmem.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  bootmem - A boot-time physical memory allocator and configurator
   4  *
   5  *  Copyright (C) 1999 Ingo Molnar
   6  *                1999 Kanoj Sarcar, SGI
   7  *                2008 Johannes Weiner
   8  *
   9  * Access to this subsystem has to be serialized externally (which is true
  10  * for the boot process anyway).
  11  */
  12 #include <linux/init.h>
  13 #include <linux/pfn.h>
  14 #include <linux/slab.h>
  15 #include <linux/export.h>
  16 #include <linux/kmemleak.h>
  17 #include <linux/range.h>
  18 #include <linux/memblock.h>
  19 #include <linux/bootmem.h>
  20
  21 #include <asm/bug.h>
  22 #include <asm/io.h>
  23
  24 #include "internal.h"
  25
  26 #ifndef CONFIG_HAVE_MEMBLOCK
  27 #error CONFIG_HAVE_MEMBLOCK not defined
  28 #endif
  29
  30 #ifndef CONFIG_NEED_MULTIPLE_NODES
  31 struct pglist_data __refdata contig_page_data;
  32 EXPORT_SYMBOL(contig_page_data);
  33 #endif
  34
  35 unsigned long max_low_pfn;
  36 unsigned long min_low_pfn;
  37 unsigned long max_pfn;
  38 unsigned long long max_possible_pfn;
  39
  40 static void * __init __alloc_memory_core_early(int nid, u64 size, u64 align,
  41                                         u64 goal, u64 limit)
  42 {
  43         void *ptr;
  44         u64 addr;
  45         enum memblock_flags flags = choose_memblock_flags();
  46
  47         if (limit > memblock.current_limit)
  48                 limit = memblock.current_limit;
  49
  50 again:
  51         addr = memblock_find_in_range_node(size, align, goal, limit, nid,
  52                                            flags);
  53         if (!addr && (flags & MEMBLOCK_MIRROR)) {
  54                 flags &= ~MEMBLOCK_MIRROR;
  55                 pr_warn("Could not allocate %pap bytes of mirrored memory\n",
  56                         &size);
  57                 goto again;
  58         }
  59         if (!addr)
  60                 return NULL;
  61
  62         if (memblock_reserve(addr, size))
  63                 return NULL;
  64
  65         ptr = phys_to_virt(addr);
  66         memset(ptr, 0, size);
  67         /*
  68          * The min_count is set to 0 so that bootmem allocated blocks
  69          * are never reported as leaks.
  70          */
  71         kmemleak_alloc(ptr, size, 0, 0);
  72         return ptr;
  73 }
  74
  75 /**
  76  * free_bootmem_late - free bootmem pages directly to page allocator
  77  * @addr: starting address of the range
  78  * @size: size of the range in bytes
  79  *
  80  * This is only useful when the bootmem allocator has already been torn
  81  * down, but we are still initializing the system.  Pages are given directly
  82  * to the page allocator, no bootmem metadata is updated because it is gone.
  83  */
  84 void __init free_bootmem_late(unsigned long addr, unsigned long size)
  85 {
  86         unsigned long cursor, end;
  87
  88         kmemleak_free_part_phys(addr, size);
  89
  90         cursor = PFN_UP(addr);
  91         end = PFN_DOWN(addr + size);
  92
  93         for (; cursor < end; cursor++) {
  94                 __free_pages_bootmem(pfn_to_page(cursor), cursor, 0);
  95                 totalram_pages++;
  96         }
  97 }
  98
  99 static void __init __free_pages_memory(unsigned long start, unsigned long end)
 100 {
 101         int order;
 102
 103         while (start < end) {
 104                 order = min(MAX_ORDER - 1UL, __ffs(start));
 105
 106                 while (start + (1UL << order) > end)
 107                         order--;
 108
 109                 __free_pages_bootmem(pfn_to_page(start), start, order);
 110
 111                 start += (1UL << order);
 112         }
 113 }
 114
 115 static unsigned long __init __free_memory_core(phys_addr_t start,
 116                                  phys_addr_t end)
 117 {
 118         unsigned long start_pfn = PFN_UP(start);
 119         unsigned long end_pfn = min_t(unsigned long,
 120                                       PFN_DOWN(end), max_low_pfn);
 121
 122         if (start_pfn >= end_pfn)
 123                 return 0;
 124
 125         __free_pages_memory(start_pfn, end_pfn);
 126
 127         return end_pfn - start_pfn;
 128 }
 129
 130 static unsigned long __init free_low_memory_core_early(void)
 131 {
 132         unsigned long count = 0;
 133         phys_addr_t start, end;
 134         u64 i;
 135
 136         memblock_clear_hotplug(0, -1);
 137
 138         for_each_reserved_mem_region(i, &start, &end)
 139                 reserve_bootmem_region(start, end);
 140
 141         /*
 142          * We need to use NUMA_NO_NODE instead of NODE_DATA(0)->node_id
 143          *  because in some case like Node0 doesn't have RAM installed
 144          *  low ram will be on Node1
 145          */
 146         for_each_free_mem_range(i, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
 147                                 NULL)
 148                 count += __free_memory_core(start, end);
 149
 150         return count;
 151 }
 152
 153 static int reset_managed_pages_done __initdata;
 154
 155 void reset_node_managed_pages(pg_data_t *pgdat)
 156 {
 157         struct zone *z;
 158
 159         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
 160                 z->managed_pages = 0;
 161 }
 162
 163 void __init reset_all_zones_managed_pages(void)
 164 {
 165         struct pglist_data *pgdat;
 166
 167         if (reset_managed_pages_done)
 168                 return;
 169
 170         for_each_online_pgdat(pgdat)
 171                 reset_node_managed_pages(pgdat);
 172
 173         reset_managed_pages_done = 1;
 174 }
 175
 176 /**
 177  * free_all_bootmem - release free pages to the buddy allocator
 178  *
 179  * Return: the number of pages actually released.
 180  */
 181 unsigned long __init free_all_bootmem(void)
 182 {
 183         unsigned long pages;
 184
 185         reset_all_zones_managed_pages();
 186
 187         pages = free_low_memory_core_early();
 188         totalram_pages += pages;
 189
 190         return pages;
 191 }
 192
 193 /**
 194  * free_bootmem_node - mark a page range as usable
 195  * @pgdat: node the range resides on
 196  * @physaddr: starting physical address of the range
 197  * @size: size of the range in bytes
 198  *
 199  * Partial pages will be considered reserved and left as they are.
 200  *
 201  * The range must reside completely on the specified node.
 202  */
 203 void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
 204                               unsigned long size)
 205 {
 206         memblock_free(physaddr, size);
 207 }
 208
 209 /**
 210  * free_bootmem - mark a page range as usable
 211  * @addr: starting physical address of the range
 212  * @size: size of the range in bytes
 213  *
 214  * Partial pages will be considered reserved and left as they are.
 215  *
 216  * The range must be contiguous but may span node boundaries.
 217  */
 218 void __init free_bootmem(unsigned long addr, unsigned long size)
 219 {
 220         memblock_free(addr, size);
 221 }
 222
 223 static void * __init ___alloc_bootmem_nopanic(unsigned long size,
 224                                         unsigned long align,
 225                                         unsigned long goal,
 226                                         unsigned long limit)
 227 {
 228         void *ptr;
 229
 230         if (WARN_ON_ONCE(slab_is_available()))
 231                 return kzalloc(size, GFP_NOWAIT);
 232
 233 restart:
 234
 235         ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align, goal, limit);
 236
 237         if (ptr)
 238                 return ptr;
 239
 240         if (goal != 0) {
 241                 goal = 0;
 242                 goto restart;
 243         }
 244
 245         return NULL;
 246 }
 247
 248 /**
 249  * __alloc_bootmem_nopanic - allocate boot memory without panicking
 250  * @size: size of the request in bytes
 251  * @align: alignment of the region
 252  * @goal: preferred starting address of the region
 253  *
 254  * The goal is dropped if it can not be satisfied and the allocation will
 255  * fall back to memory below @goal.
 256  *
 257  * Allocation may happen on any node in the system.
 258  *
 259  * Return: address of the allocated region or %NULL on failure.
 260  */
 261 void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
 262                                         unsigned long goal)
 263 {
 264         unsigned long limit = -1UL;
 265
 266         return ___alloc_bootmem_nopanic(size, align, goal, limit);
 267 }
 268
 269 static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
 270                                         unsigned long goal, unsigned long limit)
 271 {
 272         void *mem = ___alloc_bootmem_nopanic(size, align, goal, limit);
 273
 274         if (mem)
 275                 return mem;
 276         /*
 277          * Whoops, we cannot satisfy the allocation request.
 278          */
 279         pr_alert("bootmem alloc of %lu bytes failed!\n", size);
 280         panic("Out of memory");
 281         return NULL;
 282 }
 283
 284 /**
 285  * __alloc_bootmem - allocate boot memory
 286  * @size: size of the request in bytes
 287  * @align: alignment of the region
 288  * @goal: preferred starting address of the region
 289  *
 290  * The goal is dropped if it can not be satisfied and the allocation will
 291  * fall back to memory below @goal.
 292  *
 293  * Allocation may happen on any node in the system.
 294  *
 295  * The function panics if the request can not be satisfied.
 296  *
 297  * Return: address of the allocated region.
 298  */
 299 void * __init __alloc_bootmem(unsigned long size, unsigned long align,
 300                               unsigned long goal)
 301 {
 302         unsigned long limit = -1UL;
 303
 304         return ___alloc_bootmem(size, align, goal, limit);
 305 }
 306
 307 void * __init ___alloc_bootmem_node_nopanic(pg_data_t *pgdat,
 308                                                    unsigned long size,
 309                                                    unsigned long align,
 310                                                    unsigned long goal,
 311                                                    unsigned long limit)
 312 {
 313         void *ptr;
 314
 315 again:
 316         ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
 317                                         goal, limit);
 318         if (ptr)
 319                 return ptr;
 320
 321         ptr = __alloc_memory_core_early(NUMA_NO_NODE, size, align,
 322                                         goal, limit);
 323         if (ptr)
 324                 return ptr;
 325
 326         if (goal) {
 327                 goal = 0;
 328                 goto again;
 329         }
 330
 331         return NULL;
 332 }
 333
 334 void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
 335                                    unsigned long align, unsigned long goal)
 336 {
 337         if (WARN_ON_ONCE(slab_is_available()))
 338                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 339
 340         return ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, 0);
 341 }
 342
 343 static void * __init ___alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 344                                     unsigned long align, unsigned long goal,
 345                                     unsigned long limit)
 346 {
 347         void *ptr;
 348
 349         ptr = ___alloc_bootmem_node_nopanic(pgdat, size, align, goal, limit);
 350         if (ptr)
 351                 return ptr;
 352
 353         pr_alert("bootmem alloc of %lu bytes failed!\n", size);
 354         panic("Out of memory");
 355         return NULL;
 356 }
 357
 358 /**
 359  * __alloc_bootmem_node - allocate boot memory from a specific node
 360  * @pgdat: node to allocate from
 361  * @size: size of the request in bytes
 362  * @align: alignment of the region
 363  * @goal: preferred starting address of the region
 364  *
 365  * The goal is dropped if it can not be satisfied and the allocation will
 366  * fall back to memory below @goal.
 367  *
 368  * Allocation may fall back to any node in the system if the specified node
 369  * can not hold the requested memory.
 370  *
 371  * The function panics if the request can not be satisfied.
 372  *
 373  * Return: address of the allocated region.
 374  */
 375 void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
 376                                    unsigned long align, unsigned long goal)
 377 {
 378         if (WARN_ON_ONCE(slab_is_available()))
 379                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 380
 381         return ___alloc_bootmem_node(pgdat, size, align, goal, 0);
 382 }
 383
 384 void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
 385                                    unsigned long align, unsigned long goal)
 386 {
 387         return __alloc_bootmem_node(pgdat, size, align, goal);
 388 }
 389
 390
 391 /**
 392  * __alloc_bootmem_low - allocate low boot memory
 393  * @size: size of the request in bytes
 394  * @align: alignment of the region
 395  * @goal: preferred starting address of the region
 396  *
 397  * The goal is dropped if it can not be satisfied and the allocation will
 398  * fall back to memory below @goal.
 399  *
 400  * Allocation may happen on any node in the system.
 401  *
 402  * The function panics if the request can not be satisfied.
 403  *
 404  * Return: address of the allocated region.
 405  */
 406 void * __init __alloc_bootmem_low(unsigned long size, unsigned long align,
 407                                   unsigned long goal)
 408 {
 409         return ___alloc_bootmem(size, align, goal, ARCH_LOW_ADDRESS_LIMIT);
 410 }
 411
 412 void * __init __alloc_bootmem_low_nopanic(unsigned long size,
 413                                           unsigned long align,
 414                                           unsigned long goal)
 415 {
 416         return ___alloc_bootmem_nopanic(size, align, goal,
 417                                         ARCH_LOW_ADDRESS_LIMIT);
 418 }
 419
 420 /**
 421  * __alloc_bootmem_low_node - allocate low boot memory from a specific node
 422  * @pgdat: node to allocate from
 423  * @size: size of the request in bytes
 424  * @align: alignment of the region
 425  * @goal: preferred starting address of the region
 426  *
 427  * The goal is dropped if it can not be satisfied and the allocation will
 428  * fall back to memory below @goal.
 429  *
 430  * Allocation may fall back to any node in the system if the specified node
 431  * can not hold the requested memory.
 432  *
 433  * The function panics if the request can not be satisfied.
 434  *
 435  * Return: address of the allocated region.
 436  */
 437 void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
 438                                        unsigned long align, unsigned long goal)
 439 {
 440         if (WARN_ON_ONCE(slab_is_available()))
 441                 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
 442
 443         return ___alloc_bootmem_node(pgdat, size, align, goal,
 444                                      ARCH_LOW_ADDRESS_LIMIT);
 445 }