arch/x86/mm/init_64.c

   1 /*
   2  *  linux/arch/x86_64/mm/init.c
   3  *
   4  *  Copyright (C) 1995  Linus Torvalds
   5  *  Copyright (C) 2000  Pavel Machek <pavel@ucw.cz>
   6  *  Copyright (C) 2002,2003 Andi Kleen <ak@suse.de>
   7  */
   8
   9 #include <linux/signal.h>
  10 #include <linux/sched.h>
  11 #include <linux/kernel.h>
  12 #include <linux/errno.h>
  13 #include <linux/string.h>
  14 #include <linux/types.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/mman.h>
  17 #include <linux/mm.h>
  18 #include <linux/swap.h>
  19 #include <linux/smp.h>
  20 #include <linux/init.h>
  21 #include <linux/initrd.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/bootmem.h>
  24 #include <linux/memblock.h>
  25 #include <linux/proc_fs.h>
  26 #include <linux/pci.h>
  27 #include <linux/pfn.h>
  28 #include <linux/poison.h>
  29 #include <linux/dma-mapping.h>
  30 #include <linux/module.h>
  31 #include <linux/memory.h>
  32 #include <linux/memory_hotplug.h>
  33 #include <linux/memremap.h>
  34 #include <linux/nmi.h>
  35 #include <linux/gfp.h>
  36 #include <linux/kcore.h>
  37
  38 #include <asm/processor.h>
  39 #include <asm/bios_ebda.h>
  40 #include <asm/uaccess.h>
  41 #include <asm/pgtable.h>
  42 #include <asm/pgalloc.h>
  43 #include <asm/dma.h>
  44 #include <asm/fixmap.h>
  45 #include <asm/e820.h>
  46 #include <asm/apic.h>
  47 #include <asm/tlb.h>
  48 #include <asm/mmu_context.h>
  49 #include <asm/proto.h>
  50 #include <asm/smp.h>
  51 #include <asm/sections.h>
  52 #include <asm/kdebug.h>
  53 #include <asm/numa.h>
  54 #include <asm/cacheflush.h>
  55 #include <asm/init.h>
  56 #include <asm/setup.h>
  57
  58 #include "mm_internal.h"
  59
  60 static void ident_pmd_init(unsigned long pmd_flag, pmd_t *pmd_page,
  61                            unsigned long addr, unsigned long end)
  62 {
  63         addr &= PMD_MASK;
  64         for (; addr < end; addr += PMD_SIZE) {
  65                 pmd_t *pmd = pmd_page + pmd_index(addr);
  66
  67                 if (!pmd_present(*pmd))
  68                         set_pmd(pmd, __pmd(addr | pmd_flag));
  69         }
  70 }
  71 static int ident_pud_init(struct x86_mapping_info *info, pud_t *pud_page,
  72                           unsigned long addr, unsigned long end)
  73 {
  74         unsigned long next;
  75
  76         for (; addr < end; addr = next) {
  77                 pud_t *pud = pud_page + pud_index(addr);
  78                 pmd_t *pmd;
  79
  80                 next = (addr & PUD_MASK) + PUD_SIZE;
  81                 if (next > end)
  82                         next = end;
  83
  84                 if (pud_present(*pud)) {
  85                         pmd = pmd_offset(pud, 0);
  86                         ident_pmd_init(info->pmd_flag, pmd, addr, next);
  87                         continue;
  88                 }
  89                 pmd = (pmd_t *)info->alloc_pgt_page(info->context);
  90                 if (!pmd)
  91                         return -ENOMEM;
  92                 ident_pmd_init(info->pmd_flag, pmd, addr, next);
  93                 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
  94         }
  95
  96         return 0;
  97 }
  98
  99 int kernel_ident_mapping_init(struct x86_mapping_info *info, pgd_t *pgd_page,
 100                               unsigned long addr, unsigned long end)
 101 {
 102         unsigned long next;
 103         int result;
 104         int off = info->kernel_mapping ? pgd_index(__PAGE_OFFSET) : 0;
 105
 106         for (; addr < end; addr = next) {
 107                 pgd_t *pgd = pgd_page + pgd_index(addr) + off;
 108                 pud_t *pud;
 109
 110                 next = (addr & PGDIR_MASK) + PGDIR_SIZE;
 111                 if (next > end)
 112                         next = end;
 113
 114                 if (pgd_present(*pgd)) {
 115                         pud = pud_offset(pgd, 0);
 116                         result = ident_pud_init(info, pud, addr, next);
 117                         if (result)
 118                                 return result;
 119                         continue;
 120                 }
 121
 122                 pud = (pud_t *)info->alloc_pgt_page(info->context);
 123                 if (!pud)
 124                         return -ENOMEM;
 125                 result = ident_pud_init(info, pud, addr, next);
 126                 if (result)
 127                         return result;
 128                 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
 129         }
 130
 131         return 0;
 132 }
 133
 134 /*
 135  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
 136  * physical space so we can cache the place of the first one and move
 137  * around without checking the pgd every time.
 138  */
 139
 140 pteval_t __supported_pte_mask __read_mostly = ~0;
 141 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 142
 143 int force_personality32;
 144
 145 /*
 146  * noexec32=on|off
 147  * Control non executable heap for 32bit processes.
 148  * To control the stack too use noexec=off
 149  *
 150  * on   PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
 151  * off  PROT_READ implies PROT_EXEC
 152  */
 153 static int __init nonx32_setup(char *str)
 154 {
 155         if (!strcmp(str, "on"))
 156                 force_personality32 &= ~READ_IMPLIES_EXEC;
 157         else if (!strcmp(str, "off"))
 158                 force_personality32 |= READ_IMPLIES_EXEC;
 159         return 1;
 160 }
 161 __setup("noexec32=", nonx32_setup);
 162
 163 /*
 164  * When memory was added/removed make sure all the processes MM have
 165  * suitable PGD entries in the local PGD level page.
 166  */
 167 void sync_global_pgds(unsigned long start, unsigned long end, int removed)
 168 {
 169         unsigned long address;
 170
 171         for (address = start; address <= end; address += PGDIR_SIZE) {
 172                 const pgd_t *pgd_ref = pgd_offset_k(address);
 173                 struct page *page;
 174
 175                 /*
 176                  * When it is called after memory hot remove, pgd_none()
 177                  * returns true. In this case (removed == 1), we must clear
 178                  * the PGD entries in the local PGD level page.
 179                  */
 180                 if (pgd_none(*pgd_ref) && !removed)
 181                         continue;
 182
 183                 spin_lock(&pgd_lock);
 184                 list_for_each_entry(page, &pgd_list, lru) {
 185                         pgd_t *pgd;
 186                         spinlock_t *pgt_lock;
 187
 188                         pgd = (pgd_t *)page_address(page) + pgd_index(address);
 189                         /* the pgt_lock only for Xen */
 190                         pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
 191                         spin_lock(pgt_lock);
 192
 193                         if (!pgd_none(*pgd_ref) && !pgd_none(*pgd))
 194                                 BUG_ON(pgd_page_vaddr(*pgd)
 195                                        != pgd_page_vaddr(*pgd_ref));
 196
 197                         if (removed) {
 198                                 if (pgd_none(*pgd_ref) && !pgd_none(*pgd))
 199                                         pgd_clear(pgd);
 200                         } else {
 201                                 if (pgd_none(*pgd))
 202                                         set_pgd(pgd, *pgd_ref);
 203                         }
 204
 205                         spin_unlock(pgt_lock);
 206                 }
 207                 spin_unlock(&pgd_lock);
 208         }
 209 }
 210
 211 /*
 212  * NOTE: This function is marked __ref because it calls __init function
 213  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
 214  */
 215 static __ref void *spp_getpage(void)
 216 {
 217         void *ptr;
 218
 219         if (after_bootmem)
 220                 ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
 221         else
 222                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 223
 224         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 225                 panic("set_pte_phys: cannot allocate page data %s\n",
 226                         after_bootmem ? "after bootmem" : "");
 227         }
 228
 229         pr_debug("spp_getpage %p\n", ptr);
 230
 231         return ptr;
 232 }
 233
 234 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
 235 {
 236         if (pgd_none(*pgd)) {
 237                 pud_t *pud = (pud_t *)spp_getpage();
 238                 pgd_populate(&init_mm, pgd, pud);
 239                 if (pud != pud_offset(pgd, 0))
 240                         printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
 241                                pud, pud_offset(pgd, 0));
 242         }
 243         return pud_offset(pgd, vaddr);
 244 }
 245
 246 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
 247 {
 248         if (pud_none(*pud)) {
 249                 pmd_t *pmd = (pmd_t *) spp_getpage();
 250                 pud_populate(&init_mm, pud, pmd);
 251                 if (pmd != pmd_offset(pud, 0))
 252                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
 253                                pmd, pmd_offset(pud, 0));
 254         }
 255         return pmd_offset(pud, vaddr);
 256 }
 257
 258 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
 259 {
 260         if (pmd_none(*pmd)) {
 261                 pte_t *pte = (pte_t *) spp_getpage();
 262                 pmd_populate_kernel(&init_mm, pmd, pte);
 263                 if (pte != pte_offset_kernel(pmd, 0))
 264                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
 265         }
 266         return pte_offset_kernel(pmd, vaddr);
 267 }
 268
 269 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
 270 {
 271         pud_t *pud;
 272         pmd_t *pmd;
 273         pte_t *pte;
 274
 275         pud = pud_page + pud_index(vaddr);
 276         pmd = fill_pmd(pud, vaddr);
 277         pte = fill_pte(pmd, vaddr);
 278
 279         set_pte(pte, new_pte);
 280
 281         /*
 282          * It's enough to flush this one mapping.
 283          * (PGE mappings get flushed as well)
 284          */
 285         __flush_tlb_one(vaddr);
 286 }
 287
 288 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
 289 {
 290         pgd_t *pgd;
 291         pud_t *pud_page;
 292
 293         pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
 294
 295         pgd = pgd_offset_k(vaddr);
 296         if (pgd_none(*pgd)) {
 297                 printk(KERN_ERR
 298                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
 299                 return;
 300         }
 301         pud_page = (pud_t*)pgd_page_vaddr(*pgd);
 302         set_pte_vaddr_pud(pud_page, vaddr, pteval);
 303 }
 304
 305 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
 306 {
 307         pgd_t *pgd;
 308         pud_t *pud;
 309
 310         pgd = pgd_offset_k(vaddr);
 311         pud = fill_pud(pgd, vaddr);
 312         return fill_pmd(pud, vaddr);
 313 }
 314
 315 pte_t * __init populate_extra_pte(unsigned long vaddr)
 316 {
 317         pmd_t *pmd;
 318
 319         pmd = populate_extra_pmd(vaddr);
 320         return fill_pte(pmd, vaddr);
 321 }
 322
 323 /*
 324  * Create large page table mappings for a range of physical addresses.
 325  */
 326 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
 327                                         enum page_cache_mode cache)
 328 {
 329         pgd_t *pgd;
 330         pud_t *pud;
 331         pmd_t *pmd;
 332         pgprot_t prot;
 333
 334         pgprot_val(prot) = pgprot_val(PAGE_KERNEL_LARGE) |
 335                 pgprot_val(pgprot_4k_2_large(cachemode2pgprot(cache)));
 336         BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
 337         for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
 338                 pgd = pgd_offset_k((unsigned long)__va(phys));
 339                 if (pgd_none(*pgd)) {
 340                         pud = (pud_t *) spp_getpage();
 341                         set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
 342                                                 _PAGE_USER));
 343                 }
 344                 pud = pud_offset(pgd, (unsigned long)__va(phys));
 345                 if (pud_none(*pud)) {
 346                         pmd = (pmd_t *) spp_getpage();
 347                         set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
 348                                                 _PAGE_USER));
 349                 }
 350                 pmd = pmd_offset(pud, phys);
 351                 BUG_ON(!pmd_none(*pmd));
 352                 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
 353         }
 354 }
 355
 356 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
 357 {
 358         __init_extra_mapping(phys, size, _PAGE_CACHE_MODE_WB);
 359 }
 360
 361 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
 362 {
 363         __init_extra_mapping(phys, size, _PAGE_CACHE_MODE_UC);
 364 }
 365
 366 /*
 367  * The head.S code sets up the kernel high mapping:
 368  *
 369  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
 370  *
 371  * phys_base holds the negative offset to the kernel, which is added
 372  * to the compile time generated pmds. This results in invalid pmds up
 373  * to the point where we hit the physaddr 0 mapping.
 374  *
 375  * We limit the mappings to the region from _text to _brk_end.  _brk_end
 376  * is rounded up to the 2MB boundary. This catches the invalid pmds as
 377  * well, as they are located before _text:
 378  */
 379 void __init cleanup_highmap(void)
 380 {
 381         unsigned long vaddr = __START_KERNEL_map;
 382         unsigned long vaddr_end = __START_KERNEL_map + KERNEL_IMAGE_SIZE;
 383         unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
 384         pmd_t *pmd = level2_kernel_pgt;
 385
 386         /*
 387          * Native path, max_pfn_mapped is not set yet.
 388          * Xen has valid max_pfn_mapped set in
 389          *      arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
 390          */
 391         if (max_pfn_mapped)
 392                 vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
 393
 394         for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
 395                 if (pmd_none(*pmd))
 396                         continue;
 397                 if (vaddr < (unsigned long) _text || vaddr > end)
 398                         set_pmd(pmd, __pmd(0));
 399         }
 400 }
 401
 402 static unsigned long __meminit
 403 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
 404               pgprot_t prot)
 405 {
 406         unsigned long pages = 0, next;
 407         unsigned long last_map_addr = end;
 408         int i;
 409
 410         pte_t *pte = pte_page + pte_index(addr);
 411
 412         for (i = pte_index(addr); i < PTRS_PER_PTE; i++, addr = next, pte++) {
 413                 next = (addr & PAGE_MASK) + PAGE_SIZE;
 414                 if (addr >= end) {
 415                         if (!after_bootmem &&
 416                             !e820_any_mapped(addr & PAGE_MASK, next, E820_RAM) &&
 417                             !e820_any_mapped(addr & PAGE_MASK, next, E820_RESERVED_KERN))
 418                                 set_pte(pte, __pte(0));
 419                         continue;
 420                 }
 421
 422                 /*
 423                  * We will re-use the existing mapping.
 424                  * Xen for example has some special requirements, like mapping
 425                  * pagetable pages as RO. So assume someone who pre-setup
 426                  * these mappings are more intelligent.
 427                  */
 428                 if (pte_val(*pte)) {
 429                         if (!after_bootmem)
 430                                 pages++;
 431                         continue;
 432                 }
 433
 434                 if (0)
 435                         printk("   pte=%p addr=%lx pte=%016lx\n",
 436                                pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
 437                 pages++;
 438                 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
 439                 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
 440         }
 441
 442         update_page_count(PG_LEVEL_4K, pages);
 443
 444         return last_map_addr;
 445 }
 446
 447 static unsigned long __meminit
 448 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
 449               unsigned long page_size_mask, pgprot_t prot)
 450 {
 451         unsigned long pages = 0, next;
 452         unsigned long last_map_addr = end;
 453
 454         int i = pmd_index(address);
 455
 456         for (; i < PTRS_PER_PMD; i++, address = next) {
 457                 pmd_t *pmd = pmd_page + pmd_index(address);
 458                 pte_t *pte;
 459                 pgprot_t new_prot = prot;
 460
 461                 next = (address & PMD_MASK) + PMD_SIZE;
 462                 if (address >= end) {
 463                         if (!after_bootmem &&
 464                             !e820_any_mapped(address & PMD_MASK, next, E820_RAM) &&
 465                             !e820_any_mapped(address & PMD_MASK, next, E820_RESERVED_KERN))
 466                                 set_pmd(pmd, __pmd(0));
 467                         continue;
 468                 }
 469
 470                 if (pmd_val(*pmd)) {
 471                         if (!pmd_large(*pmd)) {
 472                                 spin_lock(&init_mm.page_table_lock);
 473                                 pte = (pte_t *)pmd_page_vaddr(*pmd);
 474                                 last_map_addr = phys_pte_init(pte, address,
 475                                                                 end, prot);
 476                                 spin_unlock(&init_mm.page_table_lock);
 477                                 continue;
 478                         }
 479                         /*
 480                          * If we are ok with PG_LEVEL_2M mapping, then we will
 481                          * use the existing mapping,
 482                          *
 483                          * Otherwise, we will split the large page mapping but
 484                          * use the same existing protection bits except for
 485                          * large page, so that we don't violate Intel's TLB
 486                          * Application note (317080) which says, while changing
 487                          * the page sizes, new and old translations should
 488                          * not differ with respect to page frame and
 489                          * attributes.
 490                          */
 491                         if (page_size_mask & (1 << PG_LEVEL_2M)) {
 492                                 if (!after_bootmem)
 493                                         pages++;
 494                                 last_map_addr = next;
 495                                 continue;
 496                         }
 497                         new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
 498                 }
 499
 500                 if (page_size_mask & (1<<PG_LEVEL_2M)) {
 501                         pages++;
 502                         spin_lock(&init_mm.page_table_lock);
 503                         set_pte((pte_t *)pmd,
 504                                 pfn_pte((address & PMD_MASK) >> PAGE_SHIFT,
 505                                         __pgprot(pgprot_val(prot) | _PAGE_PSE)));
 506                         spin_unlock(&init_mm.page_table_lock);
 507                         last_map_addr = next;
 508                         continue;
 509                 }
 510
 511                 pte = alloc_low_page();
 512                 last_map_addr = phys_pte_init(pte, address, end, new_prot);
 513
 514                 spin_lock(&init_mm.page_table_lock);
 515                 pmd_populate_kernel(&init_mm, pmd, pte);
 516                 spin_unlock(&init_mm.page_table_lock);
 517         }
 518         update_page_count(PG_LEVEL_2M, pages);
 519         return last_map_addr;
 520 }
 521
 522 static unsigned long __meminit
 523 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
 524                          unsigned long page_size_mask)
 525 {
 526         unsigned long pages = 0, next;
 527         unsigned long last_map_addr = end;
 528         int i = pud_index(addr);
 529
 530         for (; i < PTRS_PER_PUD; i++, addr = next) {
 531                 pud_t *pud = pud_page + pud_index(addr);
 532                 pmd_t *pmd;
 533                 pgprot_t prot = PAGE_KERNEL;
 534
 535                 next = (addr & PUD_MASK) + PUD_SIZE;
 536                 if (addr >= end) {
 537                         if (!after_bootmem &&
 538                             !e820_any_mapped(addr & PUD_MASK, next, E820_RAM) &&
 539                             !e820_any_mapped(addr & PUD_MASK, next, E820_RESERVED_KERN))
 540                                 set_pud(pud, __pud(0));
 541                         continue;
 542                 }
 543
 544                 if (pud_val(*pud)) {
 545                         if (!pud_large(*pud)) {
 546                                 pmd = pmd_offset(pud, 0);
 547                                 last_map_addr = phys_pmd_init(pmd, addr, end,
 548                                                          page_size_mask, prot);
 549                                 __flush_tlb_all();
 550                                 continue;
 551                         }
 552                         /*
 553                          * If we are ok with PG_LEVEL_1G mapping, then we will
 554                          * use the existing mapping.
 555                          *
 556                          * Otherwise, we will split the gbpage mapping but use
 557                          * the same existing protection  bits except for large
 558                          * page, so that we don't violate Intel's TLB
 559                          * Application note (317080) which says, while changing
 560                          * the page sizes, new and old translations should
 561                          * not differ with respect to page frame and
 562                          * attributes.
 563                          */
 564                         if (page_size_mask & (1 << PG_LEVEL_1G)) {
 565                                 if (!after_bootmem)
 566                                         pages++;
 567                                 last_map_addr = next;
 568                                 continue;
 569                         }
 570                         prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
 571                 }
 572
 573                 if (page_size_mask & (1<<PG_LEVEL_1G)) {
 574                         pages++;
 575                         spin_lock(&init_mm.page_table_lock);
 576                         set_pte((pte_t *)pud,
 577                                 pfn_pte((addr & PUD_MASK) >> PAGE_SHIFT,
 578                                         PAGE_KERNEL_LARGE));
 579                         spin_unlock(&init_mm.page_table_lock);
 580                         last_map_addr = next;
 581                         continue;
 582                 }
 583
 584                 pmd = alloc_low_page();
 585                 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
 586                                               prot);
 587
 588                 spin_lock(&init_mm.page_table_lock);
 589                 pud_populate(&init_mm, pud, pmd);
 590                 spin_unlock(&init_mm.page_table_lock);
 591         }
 592         __flush_tlb_all();
 593
 594         update_page_count(PG_LEVEL_1G, pages);
 595
 596         return last_map_addr;
 597 }
 598
 599 unsigned long __meminit
 600 kernel_physical_mapping_init(unsigned long start,
 601                              unsigned long end,
 602                              unsigned long page_size_mask)
 603 {
 604         bool pgd_changed = false;
 605         unsigned long next, last_map_addr = end;
 606         unsigned long addr;
 607
 608         start = (unsigned long)__va(start);
 609         end = (unsigned long)__va(end);
 610         addr = start;
 611
 612         for (; start < end; start = next) {
 613                 pgd_t *pgd = pgd_offset_k(start);
 614                 pud_t *pud;
 615
 616                 next = (start & PGDIR_MASK) + PGDIR_SIZE;
 617
 618                 if (pgd_val(*pgd)) {
 619                         pud = (pud_t *)pgd_page_vaddr(*pgd);
 620                         last_map_addr = phys_pud_init(pud, __pa(start),
 621                                                  __pa(end), page_size_mask);
 622                         continue;
 623                 }
 624
 625                 pud = alloc_low_page();
 626                 last_map_addr = phys_pud_init(pud, __pa(start), __pa(end),
 627                                                  page_size_mask);
 628
 629                 spin_lock(&init_mm.page_table_lock);
 630                 pgd_populate(&init_mm, pgd, pud);
 631                 spin_unlock(&init_mm.page_table_lock);
 632                 pgd_changed = true;
 633         }
 634
 635         if (pgd_changed)
 636                 sync_global_pgds(addr, end - 1, 0);
 637
 638         __flush_tlb_all();
 639
 640         return last_map_addr;
 641 }
 642
 643 #ifndef CONFIG_NUMA
 644 void __init initmem_init(void)
 645 {
 646         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
 647 }
 648 #endif
 649
 650 void __init paging_init(void)
 651 {
 652         sparse_memory_present_with_active_regions(MAX_NUMNODES);
 653         sparse_init();
 654
 655         /*
 656          * clear the default setting with node 0
 657          * note: don't use nodes_clear here, that is really clearing when
 658          *       numa support is not compiled in, and later node_set_state
 659          *       will not set it back.
 660          */
 661         node_clear_state(0, N_MEMORY);
 662         if (N_MEMORY != N_NORMAL_MEMORY)
 663                 node_clear_state(0, N_NORMAL_MEMORY);
 664
 665         zone_sizes_init();
 666 }
 667
 668 /*
 669  * Memory hotplug specific functions
 670  */
 671 #ifdef CONFIG_MEMORY_HOTPLUG
 672 /*
 673  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
 674  * updating.
 675  */
 676 static void  update_end_of_memory_vars(u64 start, u64 size)
 677 {
 678         unsigned long end_pfn = PFN_UP(start + size);
 679
 680         if (end_pfn > max_pfn) {
 681                 max_pfn = end_pfn;
 682                 max_low_pfn = end_pfn;
 683                 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 684         }
 685 }
 686
 687 /*
 688  * Memory is added always to NORMAL zone. This means you will never get
 689  * additional DMA/DMA32 memory.
 690  */
 691 int arch_add_memory(int nid, u64 start, u64 size, bool for_device)
 692 {
 693         struct pglist_data *pgdat = NODE_DATA(nid);
 694         struct zone *zone = pgdat->node_zones +
 695                 zone_for_memory(nid, start, size, ZONE_NORMAL, for_device);
 696         unsigned long start_pfn = start >> PAGE_SHIFT;
 697         unsigned long nr_pages = size >> PAGE_SHIFT;
 698         int ret;
 699
 700         init_memory_mapping(start, start + size);
 701
 702         ret = __add_pages(nid, zone, start_pfn, nr_pages);
 703         WARN_ON_ONCE(ret);
 704
 705         /* update max_pfn, max_low_pfn and high_memory */
 706         update_end_of_memory_vars(start, size);
 707
 708         return ret;
 709 }
 710 EXPORT_SYMBOL_GPL(arch_add_memory);
 711
 712 #define PAGE_INUSE 0xFD
 713
 714 static void __meminit free_pagetable(struct page *page, int order)
 715 {
 716         unsigned long magic;
 717         unsigned int nr_pages = 1 << order;
 718         struct vmem_altmap *altmap = to_vmem_altmap((unsigned long) page);
 719
 720         if (altmap) {
 721                 vmem_altmap_free(altmap, nr_pages);
 722                 return;
 723         }
 724
 725         /* bootmem page has reserved flag */
 726         if (PageReserved(page)) {
 727                 __ClearPageReserved(page);
 728
 729                 magic = (unsigned long)page->lru.next;
 730                 if (magic == SECTION_INFO || magic == MIX_SECTION_INFO) {
 731                         while (nr_pages--)
 732                                 put_page_bootmem(page++);
 733                 } else
 734                         while (nr_pages--)
 735                                 free_reserved_page(page++);
 736         } else
 737                 free_pages((unsigned long)page_address(page), order);
 738 }
 739
 740 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
 741 {
 742         pte_t *pte;
 743         int i;
 744
 745         for (i = 0; i < PTRS_PER_PTE; i++) {
 746                 pte = pte_start + i;
 747                 if (pte_val(*pte))
 748                         return;
 749         }
 750
 751         /* free a pte talbe */
 752         free_pagetable(pmd_page(*pmd), 0);
 753         spin_lock(&init_mm.page_table_lock);
 754         pmd_clear(pmd);
 755         spin_unlock(&init_mm.page_table_lock);
 756 }
 757
 758 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud)
 759 {
 760         pmd_t *pmd;
 761         int i;
 762
 763         for (i = 0; i < PTRS_PER_PMD; i++) {
 764                 pmd = pmd_start + i;
 765                 if (pmd_val(*pmd))
 766                         return;
 767         }
 768
 769         /* free a pmd talbe */
 770         free_pagetable(pud_page(*pud), 0);
 771         spin_lock(&init_mm.page_table_lock);
 772         pud_clear(pud);
 773         spin_unlock(&init_mm.page_table_lock);
 774 }
 775
 776 /* Return true if pgd is changed, otherwise return false. */
 777 static bool __meminit free_pud_table(pud_t *pud_start, pgd_t *pgd)
 778 {
 779         pud_t *pud;
 780         int i;
 781
 782         for (i = 0; i < PTRS_PER_PUD; i++) {
 783                 pud = pud_start + i;
 784                 if (pud_val(*pud))
 785                         return false;
 786         }
 787
 788         /* free a pud table */
 789         free_pagetable(pgd_page(*pgd), 0);
 790         spin_lock(&init_mm.page_table_lock);
 791         pgd_clear(pgd);
 792         spin_unlock(&init_mm.page_table_lock);
 793
 794         return true;
 795 }
 796
 797 static void __meminit
 798 remove_pte_table(pte_t *pte_start, unsigned long addr, unsigned long end,
 799                  bool direct)
 800 {
 801         unsigned long next, pages = 0;
 802         pte_t *pte;
 803         void *page_addr;
 804         phys_addr_t phys_addr;
 805
 806         pte = pte_start + pte_index(addr);
 807         for (; addr < end; addr = next, pte++) {
 808                 next = (addr + PAGE_SIZE) & PAGE_MASK;
 809                 if (next > end)
 810                         next = end;
 811
 812                 if (!pte_present(*pte))
 813                         continue;
 814
 815                 /*
 816                  * We mapped [0,1G) memory as identity mapping when
 817                  * initializing, in arch/x86/kernel/head_64.S. These
 818                  * pagetables cannot be removed.
 819                  */
 820                 phys_addr = pte_val(*pte) + (addr & PAGE_MASK);
 821                 if (phys_addr < (phys_addr_t)0x40000000)
 822                         return;
 823
 824                 if (PAGE_ALIGNED(addr) && PAGE_ALIGNED(next)) {
 825                         /*
 826                          * Do not free direct mapping pages since they were
 827                          * freed when offlining, or simplely not in use.
 828                          */
 829                         if (!direct)
 830                                 free_pagetable(pte_page(*pte), 0);
 831
 832                         spin_lock(&init_mm.page_table_lock);
 833                         pte_clear(&init_mm, addr, pte);
 834                         spin_unlock(&init_mm.page_table_lock);
 835
 836                         /* For non-direct mapping, pages means nothing. */
 837                         pages++;
 838                 } else {
 839                         /*
 840                          * If we are here, we are freeing vmemmap pages since
 841                          * direct mapped memory ranges to be freed are aligned.
 842                          *
 843                          * If we are not removing the whole page, it means
 844                          * other page structs in this page are being used and
 845                          * we canot remove them. So fill the unused page_structs
 846                          * with 0xFD, and remove the page when it is wholly
 847                          * filled with 0xFD.
 848                          */
 849                         memset((void *)addr, PAGE_INUSE, next - addr);
 850
 851                         page_addr = page_address(pte_page(*pte));
 852                         if (!memchr_inv(page_addr, PAGE_INUSE, PAGE_SIZE)) {
 853                                 free_pagetable(pte_page(*pte), 0);
 854
 855                                 spin_lock(&init_mm.page_table_lock);
 856                                 pte_clear(&init_mm, addr, pte);
 857                                 spin_unlock(&init_mm.page_table_lock);
 858                         }
 859                 }
 860         }
 861
 862         /* Call free_pte_table() in remove_pmd_table(). */
 863         flush_tlb_all();
 864         if (direct)
 865                 update_page_count(PG_LEVEL_4K, -pages);
 866 }
 867
 868 static void __meminit
 869 remove_pmd_table(pmd_t *pmd_start, unsigned long addr, unsigned long end,
 870                  bool direct)
 871 {
 872         unsigned long next, pages = 0;
 873         pte_t *pte_base;
 874         pmd_t *pmd;
 875         void *page_addr;
 876
 877         pmd = pmd_start + pmd_index(addr);
 878         for (; addr < end; addr = next, pmd++) {
 879                 next = pmd_addr_end(addr, end);
 880
 881                 if (!pmd_present(*pmd))
 882                         continue;
 883
 884                 if (pmd_large(*pmd)) {
 885                         if (IS_ALIGNED(addr, PMD_SIZE) &&
 886                             IS_ALIGNED(next, PMD_SIZE)) {
 887                                 if (!direct)
 888                                         free_pagetable(pmd_page(*pmd),
 889                                                        get_order(PMD_SIZE));
 890
 891                                 spin_lock(&init_mm.page_table_lock);
 892                                 pmd_clear(pmd);
 893                                 spin_unlock(&init_mm.page_table_lock);
 894                                 pages++;
 895                         } else {
 896                                 /* If here, we are freeing vmemmap pages. */
 897                                 memset((void *)addr, PAGE_INUSE, next - addr);
 898
 899                                 page_addr = page_address(pmd_page(*pmd));
 900                                 if (!memchr_inv(page_addr, PAGE_INUSE,
 901                                                 PMD_SIZE)) {
 902                                         free_pagetable(pmd_page(*pmd),
 903                                                        get_order(PMD_SIZE));
 904
 905                                         spin_lock(&init_mm.page_table_lock);
 906                                         pmd_clear(pmd);
 907                                         spin_unlock(&init_mm.page_table_lock);
 908                                 }
 909                         }
 910
 911                         continue;
 912                 }
 913
 914                 pte_base = (pte_t *)pmd_page_vaddr(*pmd);
 915                 remove_pte_table(pte_base, addr, next, direct);
 916                 free_pte_table(pte_base, pmd);
 917         }
 918
 919         /* Call free_pmd_table() in remove_pud_table(). */
 920         if (direct)
 921                 update_page_count(PG_LEVEL_2M, -pages);
 922 }
 923
 924 static void __meminit
 925 remove_pud_table(pud_t *pud_start, unsigned long addr, unsigned long end,
 926                  bool direct)
 927 {
 928         unsigned long next, pages = 0;
 929         pmd_t *pmd_base;
 930         pud_t *pud;
 931         void *page_addr;
 932
 933         pud = pud_start + pud_index(addr);
 934         for (; addr < end; addr = next, pud++) {
 935                 next = pud_addr_end(addr, end);
 936
 937                 if (!pud_present(*pud))
 938                         continue;
 939
 940                 if (pud_large(*pud)) {
 941                         if (IS_ALIGNED(addr, PUD_SIZE) &&
 942                             IS_ALIGNED(next, PUD_SIZE)) {
 943                                 if (!direct)
 944                                         free_pagetable(pud_page(*pud),
 945                                                        get_order(PUD_SIZE));
 946
 947                                 spin_lock(&init_mm.page_table_lock);
 948                                 pud_clear(pud);
 949                                 spin_unlock(&init_mm.page_table_lock);
 950                                 pages++;
 951                         } else {
 952                                 /* If here, we are freeing vmemmap pages. */
 953                                 memset((void *)addr, PAGE_INUSE, next - addr);
 954
 955                                 page_addr = page_address(pud_page(*pud));
 956                                 if (!memchr_inv(page_addr, PAGE_INUSE,
 957                                                 PUD_SIZE)) {
 958                                         free_pagetable(pud_page(*pud),
 959                                                        get_order(PUD_SIZE));
 960
 961                                         spin_lock(&init_mm.page_table_lock);
 962                                         pud_clear(pud);
 963                                         spin_unlock(&init_mm.page_table_lock);
 964                                 }
 965                         }
 966
 967                         continue;
 968                 }
 969
 970                 pmd_base = (pmd_t *)pud_page_vaddr(*pud);
 971                 remove_pmd_table(pmd_base, addr, next, direct);
 972                 free_pmd_table(pmd_base, pud);
 973         }
 974
 975         if (direct)
 976                 update_page_count(PG_LEVEL_1G, -pages);
 977 }
 978
 979 /* start and end are both virtual address. */
 980 static void __meminit
 981 remove_pagetable(unsigned long start, unsigned long end, bool direct)
 982 {
 983         unsigned long next;
 984         unsigned long addr;
 985         pgd_t *pgd;
 986         pud_t *pud;
 987         bool pgd_changed = false;
 988
 989         for (addr = start; addr < end; addr = next) {
 990                 next = pgd_addr_end(addr, end);
 991
 992                 pgd = pgd_offset_k(addr);
 993                 if (!pgd_present(*pgd))
 994                         continue;
 995
 996                 pud = (pud_t *)pgd_page_vaddr(*pgd);
 997                 remove_pud_table(pud, addr, next, direct);
 998                 if (free_pud_table(pud, pgd))
 999                         pgd_changed = true;
1000         }
1001
1002         if (pgd_changed)
1003                 sync_global_pgds(start, end - 1, 1);
1004
1005         flush_tlb_all();
1006 }
1007
1008 void __ref vmemmap_free(unsigned long start, unsigned long end)
1009 {
1010         remove_pagetable(start, end, false);
1011 }
1012
1013 #ifdef CONFIG_MEMORY_HOTREMOVE
1014 static void __meminit
1015 kernel_physical_mapping_remove(unsigned long start, unsigned long end)
1016 {
1017         start = (unsigned long)__va(start);
1018         end = (unsigned long)__va(end);
1019
1020         remove_pagetable(start, end, true);
1021 }
1022
1023 int __ref arch_remove_memory(u64 start, u64 size)
1024 {
1025         unsigned long start_pfn = start >> PAGE_SHIFT;
1026         unsigned long nr_pages = size >> PAGE_SHIFT;
1027         struct page *page = pfn_to_page(start_pfn);
1028         struct vmem_altmap *altmap;
1029         struct zone *zone;
1030         int ret;
1031
1032         /* With altmap the first mapped page is offset from @start */
1033         altmap = to_vmem_altmap((unsigned long) page);
1034         if (altmap)
1035                 page += vmem_altmap_offset(altmap);
1036         zone = page_zone(page);
1037         ret = __remove_pages(zone, start_pfn, nr_pages);
1038         WARN_ON_ONCE(ret);
1039         kernel_physical_mapping_remove(start, start + size);
1040
1041         return ret;
1042 }
1043 #endif
1044 #endif /* CONFIG_MEMORY_HOTPLUG */
1045
1046 static struct kcore_list kcore_vsyscall;
1047
1048 static void __init register_page_bootmem_info(void)
1049 {
1050 #ifdef CONFIG_NUMA
1051         int i;
1052
1053         for_each_online_node(i)
1054                 register_page_bootmem_info_node(NODE_DATA(i));
1055 #endif
1056 }
1057
1058 void __init mem_init(void)
1059 {
1060         pci_iommu_alloc();
1061
1062         /* clear_bss() already clear the empty_zero_page */
1063
1064         register_page_bootmem_info();
1065
1066         /* this will put all memory onto the freelists */
1067         free_all_bootmem();
1068         after_bootmem = 1;
1069
1070         /* Register memory areas for /proc/kcore */
1071         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_ADDR,
1072                          PAGE_SIZE, KCORE_OTHER);
1073
1074         mem_init_print_info(NULL);
1075 }
1076
1077 #ifdef CONFIG_DEBUG_RODATA
1078 const int rodata_test_data = 0xC3;
1079 EXPORT_SYMBOL_GPL(rodata_test_data);
1080
1081 int kernel_set_to_readonly;
1082
1083 void set_kernel_text_rw(void)
1084 {
1085         unsigned long start = PFN_ALIGN(_text);
1086         unsigned long end = PFN_ALIGN(__stop___ex_table);
1087
1088         if (!kernel_set_to_readonly)
1089                 return;
1090
1091         pr_debug("Set kernel text: %lx - %lx for read write\n",
1092                  start, end);
1093
1094         /*
1095          * Make the kernel identity mapping for text RW. Kernel text
1096          * mapping will always be RO. Refer to the comment in
1097          * static_protections() in pageattr.c
1098          */
1099         set_memory_rw(start, (end - start) >> PAGE_SHIFT);
1100 }
1101
1102 void set_kernel_text_ro(void)
1103 {
1104         unsigned long start = PFN_ALIGN(_text);
1105         unsigned long end = PFN_ALIGN(__stop___ex_table);
1106
1107         if (!kernel_set_to_readonly)
1108                 return;
1109
1110         pr_debug("Set kernel text: %lx - %lx for read only\n",
1111                  start, end);
1112
1113         /*
1114          * Set the kernel identity mapping for text RO.
1115          */
1116         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1117 }
1118
1119 void mark_rodata_ro(void)
1120 {
1121         unsigned long start = PFN_ALIGN(_text);
1122         unsigned long rodata_start = PFN_ALIGN(__start_rodata);
1123         unsigned long end = (unsigned long) &__end_rodata_hpage_align;
1124         unsigned long text_end = PFN_ALIGN(&__stop___ex_table);
1125         unsigned long rodata_end = PFN_ALIGN(&__end_rodata);
1126         unsigned long all_end;
1127
1128         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1129                (end - start) >> 10);
1130         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1131
1132         kernel_set_to_readonly = 1;
1133
1134         /*
1135          * The rodata/data/bss/brk section (but not the kernel text!)
1136          * should also be not-executable.
1137          *
1138          * We align all_end to PMD_SIZE because the existing mapping
1139          * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1140          * split the PMD and the reminder between _brk_end and the end
1141          * of the PMD will remain mapped executable.
1142          *
1143          * Any PMD which was setup after the one which covers _brk_end
1144          * has been zapped already via cleanup_highmem().
1145          */
1146         all_end = roundup((unsigned long)_brk_end, PMD_SIZE);
1147         set_memory_nx(text_end, (all_end - text_end) >> PAGE_SHIFT);
1148
1149         rodata_test();
1150
1151 #ifdef CONFIG_CPA_DEBUG
1152         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
1153         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
1154
1155         printk(KERN_INFO "Testing CPA: again\n");
1156         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
1157 #endif
1158
1159         free_init_pages("unused kernel",
1160                         (unsigned long) __va(__pa_symbol(text_end)),
1161                         (unsigned long) __va(__pa_symbol(rodata_start)));
1162         free_init_pages("unused kernel",
1163                         (unsigned long) __va(__pa_symbol(rodata_end)),
1164                         (unsigned long) __va(__pa_symbol(_sdata)));
1165
1166         debug_checkwx();
1167 }
1168
1169 #endif
1170
1171 int kern_addr_valid(unsigned long addr)
1172 {
1173         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
1174         pgd_t *pgd;
1175         pud_t *pud;
1176         pmd_t *pmd;
1177         pte_t *pte;
1178
1179         if (above != 0 && above != -1UL)
1180                 return 0;
1181
1182         pgd = pgd_offset_k(addr);
1183         if (pgd_none(*pgd))
1184                 return 0;
1185
1186         pud = pud_offset(pgd, addr);
1187         if (pud_none(*pud))
1188                 return 0;
1189
1190         if (pud_large(*pud))
1191                 return pfn_valid(pud_pfn(*pud));
1192
1193         pmd = pmd_offset(pud, addr);
1194         if (pmd_none(*pmd))
1195                 return 0;
1196
1197         if (pmd_large(*pmd))
1198                 return pfn_valid(pmd_pfn(*pmd));
1199
1200         pte = pte_offset_kernel(pmd, addr);
1201         if (pte_none(*pte))
1202                 return 0;
1203
1204         return pfn_valid(pte_pfn(*pte));
1205 }
1206
1207 static unsigned long probe_memory_block_size(void)
1208 {
1209         /* start from 2g */
1210         unsigned long bz = 1UL<<31;
1211
1212         if (totalram_pages >= (64ULL << (30 - PAGE_SHIFT))) {
1213                 pr_info("Using 2GB memory block size for large-memory system\n");
1214                 return 2UL * 1024 * 1024 * 1024;
1215         }
1216
1217         /* less than 64g installed */
1218         if ((max_pfn << PAGE_SHIFT) < (16UL << 32))
1219                 return MIN_MEMORY_BLOCK_SIZE;
1220
1221         /* get the tail size */
1222         while (bz > MIN_MEMORY_BLOCK_SIZE) {
1223                 if (!((max_pfn << PAGE_SHIFT) & (bz - 1)))
1224                         break;
1225                 bz >>= 1;
1226         }
1227
1228         printk(KERN_DEBUG "memory block size : %ldMB\n", bz >> 20);
1229
1230         return bz;
1231 }
1232
1233 static unsigned long memory_block_size_probed;
1234 unsigned long memory_block_size_bytes(void)
1235 {
1236         if (!memory_block_size_probed)
1237                 memory_block_size_probed = probe_memory_block_size();
1238
1239         return memory_block_size_probed;
1240 }
1241
1242 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1243 /*
1244  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1245  */
1246 static long __meminitdata addr_start, addr_end;
1247 static void __meminitdata *p_start, *p_end;
1248 static int __meminitdata node_start;
1249
1250 static int __meminit vmemmap_populate_hugepages(unsigned long start,
1251                 unsigned long end, int node, struct vmem_altmap *altmap)
1252 {
1253         unsigned long addr;
1254         unsigned long next;
1255         pgd_t *pgd;
1256         pud_t *pud;
1257         pmd_t *pmd;
1258
1259         for (addr = start; addr < end; addr = next) {
1260                 next = pmd_addr_end(addr, end);
1261
1262                 pgd = vmemmap_pgd_populate(addr, node);
1263                 if (!pgd)
1264                         return -ENOMEM;
1265
1266                 pud = vmemmap_pud_populate(pgd, addr, node);
1267                 if (!pud)
1268                         return -ENOMEM;
1269
1270                 pmd = pmd_offset(pud, addr);
1271                 if (pmd_none(*pmd)) {
1272                         void *p;
1273
1274                         p = __vmemmap_alloc_block_buf(PMD_SIZE, node, altmap);
1275                         if (p) {
1276                                 pte_t entry;
1277
1278                                 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1279                                                 PAGE_KERNEL_LARGE);
1280                                 set_pmd(pmd, __pmd(pte_val(entry)));
1281
1282                                 /* check to see if we have contiguous blocks */
1283                                 if (p_end != p || node_start != node) {
1284                                         if (p_start)
1285                                                 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1286                                                        addr_start, addr_end-1, p_start, p_end-1, node_start);
1287                                         addr_start = addr;
1288                                         node_start = node;
1289                                         p_start = p;
1290                                 }
1291
1292                                 addr_end = addr + PMD_SIZE;
1293                                 p_end = p + PMD_SIZE;
1294                                 continue;
1295                         } else if (altmap)
1296                                 return -ENOMEM; /* no fallback */
1297                 } else if (pmd_large(*pmd)) {
1298                         vmemmap_verify((pte_t *)pmd, node, addr, next);
1299                         continue;
1300                 }
1301                 pr_warn_once("vmemmap: falling back to regular page backing\n");
1302                 if (vmemmap_populate_basepages(addr, next, node))
1303                         return -ENOMEM;
1304         }
1305         return 0;
1306 }
1307
1308 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
1309 {
1310         struct vmem_altmap *altmap = to_vmem_altmap(start);
1311         int err;
1312
1313         if (cpu_has_pse)
1314                 err = vmemmap_populate_hugepages(start, end, node, altmap);
1315         else if (altmap) {
1316                 pr_err_once("%s: no cpu support for altmap allocations\n",
1317                                 __func__);
1318                 err = -ENOMEM;
1319         } else
1320                 err = vmemmap_populate_basepages(start, end, node);
1321         if (!err)
1322                 sync_global_pgds(start, end - 1, 0);
1323         return err;
1324 }
1325
1326 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1327 void register_page_bootmem_memmap(unsigned long section_nr,
1328                                   struct page *start_page, unsigned long size)
1329 {
1330         unsigned long addr = (unsigned long)start_page;
1331         unsigned long end = (unsigned long)(start_page + size);
1332         unsigned long next;
1333         pgd_t *pgd;
1334         pud_t *pud;
1335         pmd_t *pmd;
1336         unsigned int nr_pages;
1337         struct page *page;
1338
1339         for (; addr < end; addr = next) {
1340                 pte_t *pte = NULL;
1341
1342                 pgd = pgd_offset_k(addr);
1343                 if (pgd_none(*pgd)) {
1344                         next = (addr + PAGE_SIZE) & PAGE_MASK;
1345                         continue;
1346                 }
1347                 get_page_bootmem(section_nr, pgd_page(*pgd), MIX_SECTION_INFO);
1348
1349                 pud = pud_offset(pgd, addr);
1350                 if (pud_none(*pud)) {
1351                         next = (addr + PAGE_SIZE) & PAGE_MASK;
1352                         continue;
1353                 }
1354                 get_page_bootmem(section_nr, pud_page(*pud), MIX_SECTION_INFO);
1355
1356                 if (!cpu_has_pse) {
1357                         next = (addr + PAGE_SIZE) & PAGE_MASK;
1358                         pmd = pmd_offset(pud, addr);
1359                         if (pmd_none(*pmd))
1360                                 continue;
1361                         get_page_bootmem(section_nr, pmd_page(*pmd),
1362                                          MIX_SECTION_INFO);
1363
1364                         pte = pte_offset_kernel(pmd, addr);
1365                         if (pte_none(*pte))
1366                                 continue;
1367                         get_page_bootmem(section_nr, pte_page(*pte),
1368                                          SECTION_INFO);
1369                 } else {
1370                         next = pmd_addr_end(addr, end);
1371
1372                         pmd = pmd_offset(pud, addr);
1373                         if (pmd_none(*pmd))
1374                                 continue;
1375
1376                         nr_pages = 1 << (get_order(PMD_SIZE));
1377                         page = pmd_page(*pmd);
1378                         while (nr_pages--)
1379                                 get_page_bootmem(section_nr, page++,
1380                                                  SECTION_INFO);
1381                 }
1382         }
1383 }
1384 #endif
1385
1386 void __meminit vmemmap_populate_print_last(void)
1387 {
1388         if (p_start) {
1389                 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1390                         addr_start, addr_end-1, p_start, p_end-1, node_start);
1391                 p_start = NULL;
1392                 p_end = NULL;
1393                 node_start = 0;
1394         }
1395 }
1396 #endif