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@suse.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/proc_fs.h>
  25 #include <linux/pci.h>
  26 #include <linux/pfn.h>
  27 #include <linux/poison.h>
  28 #include <linux/dma-mapping.h>
  29 #include <linux/module.h>
  30 #include <linux/memory_hotplug.h>
  31 #include <linux/nmi.h>
  32 #include <linux/gfp.h>
  33
  34 #include <asm/processor.h>
  35 #include <asm/bios_ebda.h>
  36 #include <asm/system.h>
  37 #include <asm/uaccess.h>
  38 #include <asm/pgtable.h>
  39 #include <asm/pgalloc.h>
  40 #include <asm/dma.h>
  41 #include <asm/fixmap.h>
  42 #include <asm/e820.h>
  43 #include <asm/apic.h>
  44 #include <asm/tlb.h>
  45 #include <asm/mmu_context.h>
  46 #include <asm/proto.h>
  47 #include <asm/smp.h>
  48 #include <asm/sections.h>
  49 #include <asm/kdebug.h>
  50 #include <asm/numa.h>
  51 #include <asm/cacheflush.h>
  52 #include <asm/init.h>
  53 #include <linux/bootmem.h>
  54
  55 static unsigned long dma_reserve __initdata;
  56
  57 static int __init parse_direct_gbpages_off(char *arg)
  58 {
  59         direct_gbpages = 0;
  60         return 0;
  61 }
  62 early_param("nogbpages", parse_direct_gbpages_off);
  63
  64 static int __init parse_direct_gbpages_on(char *arg)
  65 {
  66         direct_gbpages = 1;
  67         return 0;
  68 }
  69 early_param("gbpages", parse_direct_gbpages_on);
  70
  71 /*
  72  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
  73  * physical space so we can cache the place of the first one and move
  74  * around without checking the pgd every time.
  75  */
  76
  77 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
  78 EXPORT_SYMBOL_GPL(__supported_pte_mask);
  79
  80 int force_personality32;
  81
  82 /*
  83  * noexec32=on|off
  84  * Control non executable heap for 32bit processes.
  85  * To control the stack too use noexec=off
  86  *
  87  * on   PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
  88  * off  PROT_READ implies PROT_EXEC
  89  */
  90 static int __init nonx32_setup(char *str)
  91 {
  92         if (!strcmp(str, "on"))
  93                 force_personality32 &= ~READ_IMPLIES_EXEC;
  94         else if (!strcmp(str, "off"))
  95                 force_personality32 |= READ_IMPLIES_EXEC;
  96         return 1;
  97 }
  98 __setup("noexec32=", nonx32_setup);
  99
 100 /*
 101  * NOTE: This function is marked __ref because it calls __init function
 102  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
 103  */
 104 static __ref void *spp_getpage(void)
 105 {
 106         void *ptr;
 107
 108         if (after_bootmem)
 109                 ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
 110         else
 111                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 112
 113         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 114                 panic("set_pte_phys: cannot allocate page data %s\n",
 115                         after_bootmem ? "after bootmem" : "");
 116         }
 117
 118         pr_debug("spp_getpage %p\n", ptr);
 119
 120         return ptr;
 121 }
 122
 123 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
 124 {
 125         if (pgd_none(*pgd)) {
 126                 pud_t *pud = (pud_t *)spp_getpage();
 127                 pgd_populate(&init_mm, pgd, pud);
 128                 if (pud != pud_offset(pgd, 0))
 129                         printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
 130                                pud, pud_offset(pgd, 0));
 131         }
 132         return pud_offset(pgd, vaddr);
 133 }
 134
 135 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
 136 {
 137         if (pud_none(*pud)) {
 138                 pmd_t *pmd = (pmd_t *) spp_getpage();
 139                 pud_populate(&init_mm, pud, pmd);
 140                 if (pmd != pmd_offset(pud, 0))
 141                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
 142                                pmd, pmd_offset(pud, 0));
 143         }
 144         return pmd_offset(pud, vaddr);
 145 }
 146
 147 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
 148 {
 149         if (pmd_none(*pmd)) {
 150                 pte_t *pte = (pte_t *) spp_getpage();
 151                 pmd_populate_kernel(&init_mm, pmd, pte);
 152                 if (pte != pte_offset_kernel(pmd, 0))
 153                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
 154         }
 155         return pte_offset_kernel(pmd, vaddr);
 156 }
 157
 158 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
 159 {
 160         pud_t *pud;
 161         pmd_t *pmd;
 162         pte_t *pte;
 163
 164         pud = pud_page + pud_index(vaddr);
 165         pmd = fill_pmd(pud, vaddr);
 166         pte = fill_pte(pmd, vaddr);
 167
 168         set_pte(pte, new_pte);
 169
 170         /*
 171          * It's enough to flush this one mapping.
 172          * (PGE mappings get flushed as well)
 173          */
 174         __flush_tlb_one(vaddr);
 175 }
 176
 177 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
 178 {
 179         pgd_t *pgd;
 180         pud_t *pud_page;
 181
 182         pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
 183
 184         pgd = pgd_offset_k(vaddr);
 185         if (pgd_none(*pgd)) {
 186                 printk(KERN_ERR
 187                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
 188                 return;
 189         }
 190         pud_page = (pud_t*)pgd_page_vaddr(*pgd);
 191         set_pte_vaddr_pud(pud_page, vaddr, pteval);
 192 }
 193
 194 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
 195 {
 196         pgd_t *pgd;
 197         pud_t *pud;
 198
 199         pgd = pgd_offset_k(vaddr);
 200         pud = fill_pud(pgd, vaddr);
 201         return fill_pmd(pud, vaddr);
 202 }
 203
 204 pte_t * __init populate_extra_pte(unsigned long vaddr)
 205 {
 206         pmd_t *pmd;
 207
 208         pmd = populate_extra_pmd(vaddr);
 209         return fill_pte(pmd, vaddr);
 210 }
 211
 212 /*
 213  * Create large page table mappings for a range of physical addresses.
 214  */
 215 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
 216                                                 pgprot_t prot)
 217 {
 218         pgd_t *pgd;
 219         pud_t *pud;
 220         pmd_t *pmd;
 221
 222         BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
 223         for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
 224                 pgd = pgd_offset_k((unsigned long)__va(phys));
 225                 if (pgd_none(*pgd)) {
 226                         pud = (pud_t *) spp_getpage();
 227                         set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
 228                                                 _PAGE_USER));
 229                 }
 230                 pud = pud_offset(pgd, (unsigned long)__va(phys));
 231                 if (pud_none(*pud)) {
 232                         pmd = (pmd_t *) spp_getpage();
 233                         set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
 234                                                 _PAGE_USER));
 235                 }
 236                 pmd = pmd_offset(pud, phys);
 237                 BUG_ON(!pmd_none(*pmd));
 238                 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
 239         }
 240 }
 241
 242 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
 243 {
 244         __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
 245 }
 246
 247 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
 248 {
 249         __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
 250 }
 251
 252 /*
 253  * The head.S code sets up the kernel high mapping:
 254  *
 255  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
 256  *
 257  * phys_addr holds the negative offset to the kernel, which is added
 258  * to the compile time generated pmds. This results in invalid pmds up
 259  * to the point where we hit the physaddr 0 mapping.
 260  *
 261  * We limit the mappings to the region from _text to _end.  _end is
 262  * rounded up to the 2MB boundary. This catches the invalid pmds as
 263  * well, as they are located before _text:
 264  */
 265 void __init cleanup_highmap(void)
 266 {
 267         unsigned long vaddr = __START_KERNEL_map;
 268         unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1;
 269         pmd_t *pmd = level2_kernel_pgt;
 270         pmd_t *last_pmd = pmd + PTRS_PER_PMD;
 271
 272         for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
 273                 if (pmd_none(*pmd))
 274                         continue;
 275                 if (vaddr < (unsigned long) _text || vaddr > end)
 276                         set_pmd(pmd, __pmd(0));
 277         }
 278 }
 279
 280 static __ref void *alloc_low_page(unsigned long *phys)
 281 {
 282         unsigned long pfn = e820_table_end++;
 283         void *adr;
 284
 285         if (after_bootmem) {
 286                 adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
 287                 *phys = __pa(adr);
 288
 289                 return adr;
 290         }
 291
 292         if (pfn >= e820_table_top)
 293                 panic("alloc_low_page: ran out of memory");
 294
 295         adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
 296         memset(adr, 0, PAGE_SIZE);
 297         *phys  = pfn * PAGE_SIZE;
 298         return adr;
 299 }
 300
 301 static __ref void unmap_low_page(void *adr)
 302 {
 303         if (after_bootmem)
 304                 return;
 305
 306         early_iounmap(adr, PAGE_SIZE);
 307 }
 308
 309 static unsigned long __meminit
 310 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
 311               pgprot_t prot)
 312 {
 313         unsigned pages = 0;
 314         unsigned long last_map_addr = end;
 315         int i;
 316
 317         pte_t *pte = pte_page + pte_index(addr);
 318
 319         for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
 320
 321                 if (addr >= end) {
 322                         if (!after_bootmem) {
 323                                 for(; i < PTRS_PER_PTE; i++, pte++)
 324                                         set_pte(pte, __pte(0));
 325                         }
 326                         break;
 327                 }
 328
 329                 /*
 330                  * We will re-use the existing mapping.
 331                  * Xen for example has some special requirements, like mapping
 332                  * pagetable pages as RO. So assume someone who pre-setup
 333                  * these mappings are more intelligent.
 334                  */
 335                 if (pte_val(*pte)) {
 336                         pages++;
 337                         continue;
 338                 }
 339
 340                 if (0)
 341                         printk("   pte=%p addr=%lx pte=%016lx\n",
 342                                pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
 343                 pages++;
 344                 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
 345                 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
 346         }
 347
 348         update_page_count(PG_LEVEL_4K, pages);
 349
 350         return last_map_addr;
 351 }
 352
 353 static unsigned long __meminit
 354 phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end,
 355                 pgprot_t prot)
 356 {
 357         pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd);
 358
 359         return phys_pte_init(pte, address, end, prot);
 360 }
 361
 362 static unsigned long __meminit
 363 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
 364               unsigned long page_size_mask, pgprot_t prot)
 365 {
 366         unsigned long pages = 0;
 367         unsigned long last_map_addr = end;
 368
 369         int i = pmd_index(address);
 370
 371         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
 372                 unsigned long pte_phys;
 373                 pmd_t *pmd = pmd_page + pmd_index(address);
 374                 pte_t *pte;
 375                 pgprot_t new_prot = prot;
 376
 377                 if (address >= end) {
 378                         if (!after_bootmem) {
 379                                 for (; i < PTRS_PER_PMD; i++, pmd++)
 380                                         set_pmd(pmd, __pmd(0));
 381                         }
 382                         break;
 383                 }
 384
 385                 if (pmd_val(*pmd)) {
 386                         if (!pmd_large(*pmd)) {
 387                                 spin_lock(&init_mm.page_table_lock);
 388                                 last_map_addr = phys_pte_update(pmd, address,
 389                                                                 end, prot);
 390                                 spin_unlock(&init_mm.page_table_lock);
 391                                 continue;
 392                         }
 393                         /*
 394                          * If we are ok with PG_LEVEL_2M mapping, then we will
 395                          * use the existing mapping,
 396                          *
 397                          * Otherwise, we will split the large page mapping but
 398                          * use the same existing protection bits except for
 399                          * large page, so that we don't violate Intel's TLB
 400                          * Application note (317080) which says, while changing
 401                          * the page sizes, new and old translations should
 402                          * not differ with respect to page frame and
 403                          * attributes.
 404                          */
 405                         if (page_size_mask & (1 << PG_LEVEL_2M)) {
 406                                 pages++;
 407                                 continue;
 408                         }
 409                         new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
 410                 }
 411
 412                 if (page_size_mask & (1<<PG_LEVEL_2M)) {
 413                         pages++;
 414                         spin_lock(&init_mm.page_table_lock);
 415                         set_pte((pte_t *)pmd,
 416                                 pfn_pte(address >> PAGE_SHIFT,
 417                                         __pgprot(pgprot_val(prot) | _PAGE_PSE)));
 418                         spin_unlock(&init_mm.page_table_lock);
 419                         last_map_addr = (address & PMD_MASK) + PMD_SIZE;
 420                         continue;
 421                 }
 422
 423                 pte = alloc_low_page(&pte_phys);
 424                 last_map_addr = phys_pte_init(pte, address, end, new_prot);
 425                 unmap_low_page(pte);
 426
 427                 spin_lock(&init_mm.page_table_lock);
 428                 pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
 429                 spin_unlock(&init_mm.page_table_lock);
 430         }
 431         update_page_count(PG_LEVEL_2M, pages);
 432         return last_map_addr;
 433 }
 434
 435 static unsigned long __meminit
 436 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end,
 437                 unsigned long page_size_mask, pgprot_t prot)
 438 {
 439         pmd_t *pmd = pmd_offset(pud, 0);
 440         unsigned long last_map_addr;
 441
 442         last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask, prot);
 443         __flush_tlb_all();
 444         return last_map_addr;
 445 }
 446
 447 static unsigned long __meminit
 448 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
 449                          unsigned long page_size_mask)
 450 {
 451         unsigned long pages = 0;
 452         unsigned long last_map_addr = end;
 453         int i = pud_index(addr);
 454
 455         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
 456                 unsigned long pmd_phys;
 457                 pud_t *pud = pud_page + pud_index(addr);
 458                 pmd_t *pmd;
 459                 pgprot_t prot = PAGE_KERNEL;
 460
 461                 if (addr >= end)
 462                         break;
 463
 464                 if (!after_bootmem &&
 465                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
 466                         set_pud(pud, __pud(0));
 467                         continue;
 468                 }
 469
 470                 if (pud_val(*pud)) {
 471                         if (!pud_large(*pud)) {
 472                                 last_map_addr = phys_pmd_update(pud, addr, end,
 473                                                          page_size_mask, prot);
 474                                 continue;
 475                         }
 476                         /*
 477                          * If we are ok with PG_LEVEL_1G mapping, then we will
 478                          * use the existing mapping.
 479                          *
 480                          * Otherwise, we will split the gbpage mapping but use
 481                          * the same existing protection  bits except for large
 482                          * page, so that we don't violate Intel's TLB
 483                          * Application note (317080) which says, while changing
 484                          * the page sizes, new and old translations should
 485                          * not differ with respect to page frame and
 486                          * attributes.
 487                          */
 488                         if (page_size_mask & (1 << PG_LEVEL_1G)) {
 489                                 pages++;
 490                                 continue;
 491                         }
 492                         prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
 493                 }
 494
 495                 if (page_size_mask & (1<<PG_LEVEL_1G)) {
 496                         pages++;
 497                         spin_lock(&init_mm.page_table_lock);
 498                         set_pte((pte_t *)pud,
 499                                 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 500                         spin_unlock(&init_mm.page_table_lock);
 501                         last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
 502                         continue;
 503                 }
 504
 505                 pmd = alloc_low_page(&pmd_phys);
 506                 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
 507                                               prot);
 508                 unmap_low_page(pmd);
 509
 510                 spin_lock(&init_mm.page_table_lock);
 511                 pud_populate(&init_mm, pud, __va(pmd_phys));
 512                 spin_unlock(&init_mm.page_table_lock);
 513         }
 514         __flush_tlb_all();
 515
 516         update_page_count(PG_LEVEL_1G, pages);
 517
 518         return last_map_addr;
 519 }
 520
 521 static unsigned long __meminit
 522 phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end,
 523                  unsigned long page_size_mask)
 524 {
 525         pud_t *pud;
 526
 527         pud = (pud_t *)pgd_page_vaddr(*pgd);
 528
 529         return phys_pud_init(pud, addr, end, page_size_mask);
 530 }
 531
 532 unsigned long __meminit
 533 kernel_physical_mapping_init(unsigned long start,
 534                              unsigned long end,
 535                              unsigned long page_size_mask)
 536 {
 537
 538         unsigned long next, last_map_addr = end;
 539
 540         start = (unsigned long)__va(start);
 541         end = (unsigned long)__va(end);
 542
 543         for (; start < end; start = next) {
 544                 pgd_t *pgd = pgd_offset_k(start);
 545                 unsigned long pud_phys;
 546                 pud_t *pud;
 547
 548                 next = (start + PGDIR_SIZE) & PGDIR_MASK;
 549                 if (next > end)
 550                         next = end;
 551
 552                 if (pgd_val(*pgd)) {
 553                         last_map_addr = phys_pud_update(pgd, __pa(start),
 554                                                  __pa(end), page_size_mask);
 555                         continue;
 556                 }
 557
 558                 pud = alloc_low_page(&pud_phys);
 559                 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
 560                                                  page_size_mask);
 561                 unmap_low_page(pud);
 562
 563                 spin_lock(&init_mm.page_table_lock);
 564                 pgd_populate(&init_mm, pgd, __va(pud_phys));
 565                 spin_unlock(&init_mm.page_table_lock);
 566         }
 567         __flush_tlb_all();
 568
 569         return last_map_addr;
 570 }
 571
 572 #ifndef CONFIG_NUMA
 573 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn,
 574                                 int acpi, int k8)
 575 {
 576 #ifndef CONFIG_NO_BOOTMEM
 577         unsigned long bootmap_size, bootmap;
 578
 579         bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
 580         bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
 581                                  PAGE_SIZE);
 582         if (bootmap == -1L)
 583                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 584         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
 585         /* don't touch min_low_pfn */
 586         bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
 587                                          0, end_pfn);
 588         e820_register_active_regions(0, start_pfn, end_pfn);
 589         free_bootmem_with_active_regions(0, end_pfn);
 590 #else
 591         e820_register_active_regions(0, start_pfn, end_pfn);
 592 #endif
 593 }
 594 #endif
 595
 596 void __init paging_init(void)
 597 {
 598         unsigned long max_zone_pfns[MAX_NR_ZONES];
 599
 600         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 601         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 602         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 603         max_zone_pfns[ZONE_NORMAL] = max_pfn;
 604
 605         sparse_memory_present_with_active_regions(MAX_NUMNODES);
 606         sparse_init();
 607
 608         /*
 609          * clear the default setting with node 0
 610          * note: don't use nodes_clear here, that is really clearing when
 611          *       numa support is not compiled in, and later node_set_state
 612          *       will not set it back.
 613          */
 614         node_clear_state(0, N_NORMAL_MEMORY);
 615
 616         free_area_init_nodes(max_zone_pfns);
 617 }
 618
 619 /*
 620  * Memory hotplug specific functions
 621  */
 622 #ifdef CONFIG_MEMORY_HOTPLUG
 623 /*
 624  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
 625  * updating.
 626  */
 627 static void  update_end_of_memory_vars(u64 start, u64 size)
 628 {
 629         unsigned long end_pfn = PFN_UP(start + size);
 630
 631         if (end_pfn > max_pfn) {
 632                 max_pfn = end_pfn;
 633                 max_low_pfn = end_pfn;
 634                 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 635         }
 636 }
 637
 638 /*
 639  * Memory is added always to NORMAL zone. This means you will never get
 640  * additional DMA/DMA32 memory.
 641  */
 642 int arch_add_memory(int nid, u64 start, u64 size)
 643 {
 644         struct pglist_data *pgdat = NODE_DATA(nid);
 645         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 646         unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
 647         unsigned long nr_pages = size >> PAGE_SHIFT;
 648         int ret;
 649
 650         last_mapped_pfn = init_memory_mapping(start, start + size);
 651         if (last_mapped_pfn > max_pfn_mapped)
 652                 max_pfn_mapped = last_mapped_pfn;
 653
 654         ret = __add_pages(nid, zone, start_pfn, nr_pages);
 655         WARN_ON_ONCE(ret);
 656
 657         /* update max_pfn, max_low_pfn and high_memory */
 658         update_end_of_memory_vars(start, size);
 659
 660         return ret;
 661 }
 662 EXPORT_SYMBOL_GPL(arch_add_memory);
 663
 664 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
 665 int memory_add_physaddr_to_nid(u64 start)
 666 {
 667         return 0;
 668 }
 669 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 670 #endif
 671
 672 #endif /* CONFIG_MEMORY_HOTPLUG */
 673
 674 static struct kcore_list kcore_vsyscall;
 675
 676 void __init mem_init(void)
 677 {
 678         long codesize, reservedpages, datasize, initsize;
 679         unsigned long absent_pages;
 680
 681         pci_iommu_alloc();
 682
 683         /* clear_bss() already clear the empty_zero_page */
 684
 685         reservedpages = 0;
 686
 687         /* this will put all low memory onto the freelists */
 688 #ifdef CONFIG_NUMA
 689         totalram_pages = numa_free_all_bootmem();
 690 #else
 691         totalram_pages = free_all_bootmem();
 692 #endif
 693
 694         absent_pages = absent_pages_in_range(0, max_pfn);
 695         reservedpages = max_pfn - totalram_pages - absent_pages;
 696         after_bootmem = 1;
 697
 698         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 699         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 700         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 701
 702         /* Register memory areas for /proc/kcore */
 703         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 704                          VSYSCALL_END - VSYSCALL_START, KCORE_OTHER);
 705
 706         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 707                          "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
 708                 nr_free_pages() << (PAGE_SHIFT-10),
 709                 max_pfn << (PAGE_SHIFT-10),
 710                 codesize >> 10,
 711                 absent_pages << (PAGE_SHIFT-10),
 712                 reservedpages << (PAGE_SHIFT-10),
 713                 datasize >> 10,
 714                 initsize >> 10);
 715 }
 716
 717 #ifdef CONFIG_DEBUG_RODATA
 718 const int rodata_test_data = 0xC3;
 719 EXPORT_SYMBOL_GPL(rodata_test_data);
 720
 721 int kernel_set_to_readonly;
 722
 723 void set_kernel_text_rw(void)
 724 {
 725         unsigned long start = PFN_ALIGN(_text);
 726         unsigned long end = PFN_ALIGN(__stop___ex_table);
 727
 728         if (!kernel_set_to_readonly)
 729                 return;
 730
 731         pr_debug("Set kernel text: %lx - %lx for read write\n",
 732                  start, end);
 733
 734         /*
 735          * Make the kernel identity mapping for text RW. Kernel text
 736          * mapping will always be RO. Refer to the comment in
 737          * static_protections() in pageattr.c
 738          */
 739         set_memory_rw(start, (end - start) >> PAGE_SHIFT);
 740 }
 741
 742 void set_kernel_text_ro(void)
 743 {
 744         unsigned long start = PFN_ALIGN(_text);
 745         unsigned long end = PFN_ALIGN(__stop___ex_table);
 746
 747         if (!kernel_set_to_readonly)
 748                 return;
 749
 750         pr_debug("Set kernel text: %lx - %lx for read only\n",
 751                  start, end);
 752
 753         /*
 754          * Set the kernel identity mapping for text RO.
 755          */
 756         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 757 }
 758
 759 void mark_rodata_ro(void)
 760 {
 761         unsigned long start = PFN_ALIGN(_text);
 762         unsigned long rodata_start =
 763                 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
 764         unsigned long end = (unsigned long) &__end_rodata_hpage_align;
 765         unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table);
 766         unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata);
 767         unsigned long data_start = (unsigned long) &_sdata;
 768
 769         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 770                (end - start) >> 10);
 771         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 772
 773         kernel_set_to_readonly = 1;
 774
 775         /*
 776          * The rodata section (but not the kernel text!) should also be
 777          * not-executable.
 778          */
 779         set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
 780
 781         rodata_test();
 782
 783 #ifdef CONFIG_CPA_DEBUG
 784         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
 785         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 786
 787         printk(KERN_INFO "Testing CPA: again\n");
 788         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 789 #endif
 790
 791         free_init_pages("unused kernel memory",
 792                         (unsigned long) page_address(virt_to_page(text_end)),
 793                         (unsigned long)
 794                                  page_address(virt_to_page(rodata_start)));
 795         free_init_pages("unused kernel memory",
 796                         (unsigned long) page_address(virt_to_page(rodata_end)),
 797                         (unsigned long) page_address(virt_to_page(data_start)));
 798 }
 799
 800 #endif
 801
 802 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
 803                                    int flags)
 804 {
 805 #ifdef CONFIG_NUMA
 806         int nid, next_nid;
 807         int ret;
 808 #endif
 809         unsigned long pfn = phys >> PAGE_SHIFT;
 810
 811         if (pfn >= max_pfn) {
 812                 /*
 813                  * This can happen with kdump kernels when accessing
 814                  * firmware tables:
 815                  */
 816                 if (pfn < max_pfn_mapped)
 817                         return -EFAULT;
 818
 819                 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n",
 820                                 phys, len);
 821                 return -EFAULT;
 822         }
 823
 824         /* Should check here against the e820 map to avoid double free */
 825 #ifdef CONFIG_NUMA
 826         nid = phys_to_nid(phys);
 827         next_nid = phys_to_nid(phys + len - 1);
 828         if (nid == next_nid)
 829                 ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags);
 830         else
 831                 ret = reserve_bootmem(phys, len, flags);
 832
 833         if (ret != 0)
 834                 return ret;
 835
 836 #else
 837         reserve_bootmem(phys, len, flags);
 838 #endif
 839
 840         if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
 841                 dma_reserve += len / PAGE_SIZE;
 842                 set_dma_reserve(dma_reserve);
 843         }
 844
 845         return 0;
 846 }
 847
 848 int kern_addr_valid(unsigned long addr)
 849 {
 850         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 851         pgd_t *pgd;
 852         pud_t *pud;
 853         pmd_t *pmd;
 854         pte_t *pte;
 855
 856         if (above != 0 && above != -1UL)
 857                 return 0;
 858
 859         pgd = pgd_offset_k(addr);
 860         if (pgd_none(*pgd))
 861                 return 0;
 862
 863         pud = pud_offset(pgd, addr);
 864         if (pud_none(*pud))
 865                 return 0;
 866
 867         pmd = pmd_offset(pud, addr);
 868         if (pmd_none(*pmd))
 869                 return 0;
 870
 871         if (pmd_large(*pmd))
 872                 return pfn_valid(pmd_pfn(*pmd));
 873
 874         pte = pte_offset_kernel(pmd, addr);
 875         if (pte_none(*pte))
 876                 return 0;
 877
 878         return pfn_valid(pte_pfn(*pte));
 879 }
 880
 881 /*
 882  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 883  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 884  * not need special handling anymore:
 885  */
 886 static struct vm_area_struct gate_vma = {
 887         .vm_start       = VSYSCALL_START,
 888         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 889         .vm_page_prot   = PAGE_READONLY_EXEC,
 890         .vm_flags       = VM_READ | VM_EXEC
 891 };
 892
 893 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 894 {
 895 #ifdef CONFIG_IA32_EMULATION
 896         if (test_tsk_thread_flag(tsk, TIF_IA32))
 897                 return NULL;
 898 #endif
 899         return &gate_vma;
 900 }
 901
 902 int in_gate_area(struct task_struct *task, unsigned long addr)
 903 {
 904         struct vm_area_struct *vma = get_gate_vma(task);
 905
 906         if (!vma)
 907                 return 0;
 908
 909         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 910 }
 911
 912 /*
 913  * Use this when you have no reliable task/vma, typically from interrupt
 914  * context. It is less reliable than using the task's vma and may give
 915  * false positives:
 916  */
 917 int in_gate_area_no_task(unsigned long addr)
 918 {
 919         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 920 }
 921
 922 const char *arch_vma_name(struct vm_area_struct *vma)
 923 {
 924         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 925                 return "[vdso]";
 926         if (vma == &gate_vma)
 927                 return "[vsyscall]";
 928         return NULL;
 929 }
 930
 931 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 932 /*
 933  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 934  */
 935 static long __meminitdata addr_start, addr_end;
 936 static void __meminitdata *p_start, *p_end;
 937 static int __meminitdata node_start;
 938
 939 int __meminit
 940 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 941 {
 942         unsigned long addr = (unsigned long)start_page;
 943         unsigned long end = (unsigned long)(start_page + size);
 944         unsigned long next;
 945         pgd_t *pgd;
 946         pud_t *pud;
 947         pmd_t *pmd;
 948
 949         for (; addr < end; addr = next) {
 950                 void *p = NULL;
 951
 952                 pgd = vmemmap_pgd_populate(addr, node);
 953                 if (!pgd)
 954                         return -ENOMEM;
 955
 956                 pud = vmemmap_pud_populate(pgd, addr, node);
 957                 if (!pud)
 958                         return -ENOMEM;
 959
 960                 if (!cpu_has_pse) {
 961                         next = (addr + PAGE_SIZE) & PAGE_MASK;
 962                         pmd = vmemmap_pmd_populate(pud, addr, node);
 963
 964                         if (!pmd)
 965                                 return -ENOMEM;
 966
 967                         p = vmemmap_pte_populate(pmd, addr, node);
 968
 969                         if (!p)
 970                                 return -ENOMEM;
 971
 972                         addr_end = addr + PAGE_SIZE;
 973                         p_end = p + PAGE_SIZE;
 974                 } else {
 975                         next = pmd_addr_end(addr, end);
 976
 977                         pmd = pmd_offset(pud, addr);
 978                         if (pmd_none(*pmd)) {
 979                                 pte_t entry;
 980
 981                                 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
 982                                 if (!p)
 983                                         return -ENOMEM;
 984
 985                                 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
 986                                                 PAGE_KERNEL_LARGE);
 987                                 set_pmd(pmd, __pmd(pte_val(entry)));
 988
 989                                 /* check to see if we have contiguous blocks */
 990                                 if (p_end != p || node_start != node) {
 991                                         if (p_start)
 992                                                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
 993                                                        addr_start, addr_end-1, p_start, p_end-1, node_start);
 994                                         addr_start = addr;
 995                                         node_start = node;
 996                                         p_start = p;
 997                                 }
 998
 999                                 addr_end = addr + PMD_SIZE;
1000                                 p_end = p + PMD_SIZE;
1001                         } else
1002                                 vmemmap_verify((pte_t *)pmd, node, addr, next);
1003                 }
1004
1005         }
1006         return 0;
1007 }
1008
1009 void __meminit vmemmap_populate_print_last(void)
1010 {
1011         if (p_start) {
1012                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1013                         addr_start, addr_end-1, p_start, p_end-1, node_start);
1014                 p_start = NULL;
1015                 p_end = NULL;
1016                 node_start = 0;
1017         }
1018 }
1019 #endif