SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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
  33 #include <asm/processor.h>
  34 #include <asm/bios_ebda.h>
  35 #include <asm/system.h>
  36 #include <asm/uaccess.h>
  37 #include <asm/pgtable.h>
  38 #include <asm/pgalloc.h>
  39 #include <asm/dma.h>
  40 #include <asm/fixmap.h>
  41 #include <asm/e820.h>
  42 #include <asm/apic.h>
  43 #include <asm/tlb.h>
  44 #include <asm/mmu_context.h>
  45 #include <asm/proto.h>
  46 #include <asm/smp.h>
  47 #include <asm/sections.h>
  48 #include <asm/kdebug.h>
  49 #include <asm/numa.h>
  50 #include <asm/cacheflush.h>
  51 #include <asm/init.h>
  52
  53 /*
  54  * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
  55  * The direct mapping extends to max_pfn_mapped, so that we can directly access
  56  * apertures, ACPI and other tables without having to play with fixmaps.
  57  */
  58 unsigned long max_low_pfn_mapped;
  59 unsigned long max_pfn_mapped;
  60
  61 static unsigned long dma_reserve __initdata;
  62
  63 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  64
  65 static int __init parse_direct_gbpages_off(char *arg)
  66 {
  67         direct_gbpages = 0;
  68         return 0;
  69 }
  70 early_param("nogbpages", parse_direct_gbpages_off);
  71
  72 static int __init parse_direct_gbpages_on(char *arg)
  73 {
  74         direct_gbpages = 1;
  75         return 0;
  76 }
  77 early_param("gbpages", parse_direct_gbpages_on);
  78
  79 /*
  80  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
  81  * physical space so we can cache the place of the first one and move
  82  * around without checking the pgd every time.
  83  */
  84
  85 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
  86 EXPORT_SYMBOL_GPL(__supported_pte_mask);
  87
  88 static int do_not_nx __cpuinitdata;
  89
  90 /*
  91  * noexec=on|off
  92  * Control non-executable mappings for 64-bit processes.
  93  *
  94  * on   Enable (default)
  95  * off  Disable
  96  */
  97 static int __init nonx_setup(char *str)
  98 {
  99         if (!str)
 100                 return -EINVAL;
 101         if (!strncmp(str, "on", 2)) {
 102                 __supported_pte_mask |= _PAGE_NX;
 103                 do_not_nx = 0;
 104         } else if (!strncmp(str, "off", 3)) {
 105                 do_not_nx = 1;
 106                 __supported_pte_mask &= ~_PAGE_NX;
 107         }
 108         return 0;
 109 }
 110 early_param("noexec", nonx_setup);
 111
 112 void __cpuinit check_efer(void)
 113 {
 114         unsigned long efer;
 115
 116         rdmsrl(MSR_EFER, efer);
 117         if (!(efer & EFER_NX) || do_not_nx)
 118                 __supported_pte_mask &= ~_PAGE_NX;
 119 }
 120
 121 int force_personality32;
 122
 123 /*
 124  * noexec32=on|off
 125  * Control non executable heap for 32bit processes.
 126  * To control the stack too use noexec=off
 127  *
 128  * on   PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
 129  * off  PROT_READ implies PROT_EXEC
 130  */
 131 static int __init nonx32_setup(char *str)
 132 {
 133         if (!strcmp(str, "on"))
 134                 force_personality32 &= ~READ_IMPLIES_EXEC;
 135         else if (!strcmp(str, "off"))
 136                 force_personality32 |= READ_IMPLIES_EXEC;
 137         return 1;
 138 }
 139 __setup("noexec32=", nonx32_setup);
 140
 141 /*
 142  * NOTE: This function is marked __ref because it calls __init function
 143  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
 144  */
 145 static __ref void *spp_getpage(void)
 146 {
 147         void *ptr;
 148
 149         if (after_bootmem)
 150                 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
 151         else
 152                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 153
 154         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 155                 panic("set_pte_phys: cannot allocate page data %s\n",
 156                         after_bootmem ? "after bootmem" : "");
 157         }
 158
 159         pr_debug("spp_getpage %p\n", ptr);
 160
 161         return ptr;
 162 }
 163
 164 void
 165 set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
 166 {
 167         pud_t *pud;
 168         pmd_t *pmd;
 169         pte_t *pte;
 170
 171         pud = pud_page + pud_index(vaddr);
 172         if (pud_none(*pud)) {
 173                 pmd = (pmd_t *) spp_getpage();
 174                 pud_populate(&init_mm, pud, pmd);
 175                 if (pmd != pmd_offset(pud, 0)) {
 176                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
 177                                 pmd, pmd_offset(pud, 0));
 178                         return;
 179                 }
 180         }
 181         pmd = pmd_offset(pud, vaddr);
 182         if (pmd_none(*pmd)) {
 183                 pte = (pte_t *) spp_getpage();
 184                 pmd_populate_kernel(&init_mm, pmd, pte);
 185                 if (pte != pte_offset_kernel(pmd, 0)) {
 186                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
 187                         return;
 188                 }
 189         }
 190
 191         pte = pte_offset_kernel(pmd, vaddr);
 192         set_pte(pte, new_pte);
 193
 194         /*
 195          * It's enough to flush this one mapping.
 196          * (PGE mappings get flushed as well)
 197          */
 198         __flush_tlb_one(vaddr);
 199 }
 200
 201 void
 202 set_pte_vaddr(unsigned long vaddr, pte_t pteval)
 203 {
 204         pgd_t *pgd;
 205         pud_t *pud_page;
 206
 207         pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
 208
 209         pgd = pgd_offset_k(vaddr);
 210         if (pgd_none(*pgd)) {
 211                 printk(KERN_ERR
 212                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
 213                 return;
 214         }
 215         pud_page = (pud_t*)pgd_page_vaddr(*pgd);
 216         set_pte_vaddr_pud(pud_page, vaddr, pteval);
 217 }
 218
 219 /*
 220  * Create large page table mappings for a range of physical addresses.
 221  */
 222 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
 223                                                 pgprot_t prot)
 224 {
 225         pgd_t *pgd;
 226         pud_t *pud;
 227         pmd_t *pmd;
 228
 229         BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
 230         for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
 231                 pgd = pgd_offset_k((unsigned long)__va(phys));
 232                 if (pgd_none(*pgd)) {
 233                         pud = (pud_t *) spp_getpage();
 234                         set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
 235                                                 _PAGE_USER));
 236                 }
 237                 pud = pud_offset(pgd, (unsigned long)__va(phys));
 238                 if (pud_none(*pud)) {
 239                         pmd = (pmd_t *) spp_getpage();
 240                         set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
 241                                                 _PAGE_USER));
 242                 }
 243                 pmd = pmd_offset(pud, phys);
 244                 BUG_ON(!pmd_none(*pmd));
 245                 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
 246         }
 247 }
 248
 249 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
 250 {
 251         __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
 252 }
 253
 254 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
 255 {
 256         __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
 257 }
 258
 259 /*
 260  * The head.S code sets up the kernel high mapping:
 261  *
 262  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
 263  *
 264  * phys_addr holds the negative offset to the kernel, which is added
 265  * to the compile time generated pmds. This results in invalid pmds up
 266  * to the point where we hit the physaddr 0 mapping.
 267  *
 268  * We limit the mappings to the region from _text to _end.  _end is
 269  * rounded up to the 2MB boundary. This catches the invalid pmds as
 270  * well, as they are located before _text:
 271  */
 272 void __init cleanup_highmap(void)
 273 {
 274         unsigned long vaddr = __START_KERNEL_map;
 275         unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1;
 276         pmd_t *pmd = level2_kernel_pgt;
 277         pmd_t *last_pmd = pmd + PTRS_PER_PMD;
 278
 279         for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
 280                 if (pmd_none(*pmd))
 281                         continue;
 282                 if (vaddr < (unsigned long) _text || vaddr > end)
 283                         set_pmd(pmd, __pmd(0));
 284         }
 285 }
 286
 287 static __ref void *alloc_low_page(unsigned long *phys)
 288 {
 289         unsigned long pfn = e820_table_end++;
 290         void *adr;
 291
 292         if (after_bootmem) {
 293                 adr = (void *)get_zeroed_page(GFP_ATOMIC);
 294                 *phys = __pa(adr);
 295
 296                 return adr;
 297         }
 298
 299         if (pfn >= e820_table_top)
 300                 panic("alloc_low_page: ran out of memory");
 301
 302         adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
 303         memset(adr, 0, PAGE_SIZE);
 304         *phys  = pfn * PAGE_SIZE;
 305         return adr;
 306 }
 307
 308 static __ref void unmap_low_page(void *adr)
 309 {
 310         if (after_bootmem)
 311                 return;
 312
 313         early_iounmap(adr, PAGE_SIZE);
 314 }
 315
 316 static unsigned long __meminit
 317 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
 318               pgprot_t prot)
 319 {
 320         unsigned pages = 0;
 321         unsigned long last_map_addr = end;
 322         int i;
 323
 324         pte_t *pte = pte_page + pte_index(addr);
 325
 326         for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
 327
 328                 if (addr >= end) {
 329                         if (!after_bootmem) {
 330                                 for(; i < PTRS_PER_PTE; i++, pte++)
 331                                         set_pte(pte, __pte(0));
 332                         }
 333                         break;
 334                 }
 335
 336                 /*
 337                  * We will re-use the existing mapping.
 338                  * Xen for example has some special requirements, like mapping
 339                  * pagetable pages as RO. So assume someone who pre-setup
 340                  * these mappings are more intelligent.
 341                  */
 342                 if (pte_val(*pte)) {
 343                         pages++;
 344                         continue;
 345                 }
 346
 347                 if (0)
 348                         printk("   pte=%p addr=%lx pte=%016lx\n",
 349                                pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
 350                 pages++;
 351                 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
 352                 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
 353         }
 354
 355         update_page_count(PG_LEVEL_4K, pages);
 356
 357         return last_map_addr;
 358 }
 359
 360 static unsigned long __meminit
 361 phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end,
 362                 pgprot_t prot)
 363 {
 364         pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd);
 365
 366         return phys_pte_init(pte, address, end, prot);
 367 }
 368
 369 static unsigned long __meminit
 370 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
 371               unsigned long page_size_mask, pgprot_t prot)
 372 {
 373         unsigned long pages = 0;
 374         unsigned long last_map_addr = end;
 375
 376         int i = pmd_index(address);
 377
 378         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
 379                 unsigned long pte_phys;
 380                 pmd_t *pmd = pmd_page + pmd_index(address);
 381                 pte_t *pte;
 382                 pgprot_t new_prot = prot;
 383
 384                 if (address >= end) {
 385                         if (!after_bootmem) {
 386                                 for (; i < PTRS_PER_PMD; i++, pmd++)
 387                                         set_pmd(pmd, __pmd(0));
 388                         }
 389                         break;
 390                 }
 391
 392                 if (pmd_val(*pmd)) {
 393                         if (!pmd_large(*pmd)) {
 394                                 spin_lock(&init_mm.page_table_lock);
 395                                 last_map_addr = phys_pte_update(pmd, address,
 396                                                                 end, prot);
 397                                 spin_unlock(&init_mm.page_table_lock);
 398                                 continue;
 399                         }
 400                         /*
 401                          * If we are ok with PG_LEVEL_2M mapping, then we will
 402                          * use the existing mapping,
 403                          *
 404                          * Otherwise, we will split the large page mapping but
 405                          * use the same existing protection bits except for
 406                          * large page, so that we don't violate Intel's TLB
 407                          * Application note (317080) which says, while changing
 408                          * the page sizes, new and old translations should
 409                          * not differ with respect to page frame and
 410                          * attributes.
 411                          */
 412                         if (page_size_mask & (1 << PG_LEVEL_2M)) {
 413                                 pages++;
 414                                 continue;
 415                         }
 416                         new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
 417                 }
 418
 419                 if (page_size_mask & (1<<PG_LEVEL_2M)) {
 420                         pages++;
 421                         spin_lock(&init_mm.page_table_lock);
 422                         set_pte((pte_t *)pmd,
 423                                 pfn_pte(address >> PAGE_SHIFT,
 424                                         __pgprot(pgprot_val(prot) | _PAGE_PSE)));
 425                         spin_unlock(&init_mm.page_table_lock);
 426                         last_map_addr = (address & PMD_MASK) + PMD_SIZE;
 427                         continue;
 428                 }
 429
 430                 pte = alloc_low_page(&pte_phys);
 431                 last_map_addr = phys_pte_init(pte, address, end, new_prot);
 432                 unmap_low_page(pte);
 433
 434                 spin_lock(&init_mm.page_table_lock);
 435                 pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
 436                 spin_unlock(&init_mm.page_table_lock);
 437         }
 438         update_page_count(PG_LEVEL_2M, pages);
 439         return last_map_addr;
 440 }
 441
 442 static unsigned long __meminit
 443 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end,
 444                 unsigned long page_size_mask, pgprot_t prot)
 445 {
 446         pmd_t *pmd = pmd_offset(pud, 0);
 447         unsigned long last_map_addr;
 448
 449         last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask, prot);
 450         __flush_tlb_all();
 451         return last_map_addr;
 452 }
 453
 454 static unsigned long __meminit
 455 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
 456                          unsigned long page_size_mask)
 457 {
 458         unsigned long pages = 0;
 459         unsigned long last_map_addr = end;
 460         int i = pud_index(addr);
 461
 462         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
 463                 unsigned long pmd_phys;
 464                 pud_t *pud = pud_page + pud_index(addr);
 465                 pmd_t *pmd;
 466                 pgprot_t prot = PAGE_KERNEL;
 467
 468                 if (addr >= end)
 469                         break;
 470
 471                 if (!after_bootmem &&
 472                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
 473                         set_pud(pud, __pud(0));
 474                         continue;
 475                 }
 476
 477                 if (pud_val(*pud)) {
 478                         if (!pud_large(*pud)) {
 479                                 last_map_addr = phys_pmd_update(pud, addr, end,
 480                                                          page_size_mask, prot);
 481                                 continue;
 482                         }
 483                         /*
 484                          * If we are ok with PG_LEVEL_1G mapping, then we will
 485                          * use the existing mapping.
 486                          *
 487                          * Otherwise, we will split the gbpage mapping but use
 488                          * the same existing protection  bits except for large
 489                          * page, so that we don't violate Intel's TLB
 490                          * Application note (317080) which says, while changing
 491                          * the page sizes, new and old translations should
 492                          * not differ with respect to page frame and
 493                          * attributes.
 494                          */
 495                         if (page_size_mask & (1 << PG_LEVEL_1G)) {
 496                                 pages++;
 497                                 continue;
 498                         }
 499                         prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
 500                 }
 501
 502                 if (page_size_mask & (1<<PG_LEVEL_1G)) {
 503                         pages++;
 504                         spin_lock(&init_mm.page_table_lock);
 505                         set_pte((pte_t *)pud,
 506                                 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 507                         spin_unlock(&init_mm.page_table_lock);
 508                         last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
 509                         continue;
 510                 }
 511
 512                 pmd = alloc_low_page(&pmd_phys);
 513                 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
 514                                               prot);
 515                 unmap_low_page(pmd);
 516
 517                 spin_lock(&init_mm.page_table_lock);
 518                 pud_populate(&init_mm, pud, __va(pmd_phys));
 519                 spin_unlock(&init_mm.page_table_lock);
 520         }
 521         __flush_tlb_all();
 522
 523         update_page_count(PG_LEVEL_1G, pages);
 524
 525         return last_map_addr;
 526 }
 527
 528 static unsigned long __meminit
 529 phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end,
 530                  unsigned long page_size_mask)
 531 {
 532         pud_t *pud;
 533
 534         pud = (pud_t *)pgd_page_vaddr(*pgd);
 535
 536         return phys_pud_init(pud, addr, end, page_size_mask);
 537 }
 538
 539 unsigned long __init
 540 kernel_physical_mapping_init(unsigned long start,
 541                              unsigned long end,
 542                              unsigned long page_size_mask)
 543 {
 544
 545         unsigned long next, last_map_addr = end;
 546
 547         start = (unsigned long)__va(start);
 548         end = (unsigned long)__va(end);
 549
 550         for (; start < end; start = next) {
 551                 pgd_t *pgd = pgd_offset_k(start);
 552                 unsigned long pud_phys;
 553                 pud_t *pud;
 554
 555                 next = (start + PGDIR_SIZE) & PGDIR_MASK;
 556                 if (next > end)
 557                         next = end;
 558
 559                 if (pgd_val(*pgd)) {
 560                         last_map_addr = phys_pud_update(pgd, __pa(start),
 561                                                  __pa(end), page_size_mask);
 562                         continue;
 563                 }
 564
 565                 pud = alloc_low_page(&pud_phys);
 566                 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
 567                                                  page_size_mask);
 568                 unmap_low_page(pud);
 569
 570                 spin_lock(&init_mm.page_table_lock);
 571                 pgd_populate(&init_mm, pgd, __va(pud_phys));
 572                 spin_unlock(&init_mm.page_table_lock);
 573         }
 574         __flush_tlb_all();
 575
 576         return last_map_addr;
 577 }
 578
 579 #ifndef CONFIG_NUMA
 580 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn)
 581 {
 582         unsigned long bootmap_size, bootmap;
 583
 584         bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
 585         bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
 586                                  PAGE_SIZE);
 587         if (bootmap == -1L)
 588                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 589         /* don't touch min_low_pfn */
 590         bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
 591                                          0, end_pfn);
 592         e820_register_active_regions(0, start_pfn, end_pfn);
 593         free_bootmem_with_active_regions(0, end_pfn);
 594         early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
 595         reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
 596 }
 597
 598 void __init paging_init(void)
 599 {
 600         unsigned long max_zone_pfns[MAX_NR_ZONES];
 601
 602         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 603         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 604         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 605         max_zone_pfns[ZONE_NORMAL] = max_pfn;
 606
 607         memory_present(0, 0, max_pfn);
 608         sparse_init();
 609         free_area_init_nodes(max_zone_pfns);
 610 }
 611 #endif
 612
 613 /*
 614  * Memory hotplug specific functions
 615  */
 616 #ifdef CONFIG_MEMORY_HOTPLUG
 617 /*
 618  * Memory is added always to NORMAL zone. This means you will never get
 619  * additional DMA/DMA32 memory.
 620  */
 621 int arch_add_memory(int nid, u64 start, u64 size)
 622 {
 623         struct pglist_data *pgdat = NODE_DATA(nid);
 624         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 625         unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
 626         unsigned long nr_pages = size >> PAGE_SHIFT;
 627         int ret;
 628
 629         last_mapped_pfn = init_memory_mapping(start, start + size);
 630         if (last_mapped_pfn > max_pfn_mapped)
 631                 max_pfn_mapped = last_mapped_pfn;
 632
 633         ret = __add_pages(nid, zone, start_pfn, nr_pages);
 634         WARN_ON_ONCE(ret);
 635
 636         return ret;
 637 }
 638 EXPORT_SYMBOL_GPL(arch_add_memory);
 639
 640 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
 641 int memory_add_physaddr_to_nid(u64 start)
 642 {
 643         return 0;
 644 }
 645 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 646 #endif
 647
 648 #endif /* CONFIG_MEMORY_HOTPLUG */
 649
 650 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
 651                          kcore_modules, kcore_vsyscall;
 652
 653 void __init mem_init(void)
 654 {
 655         long codesize, reservedpages, datasize, initsize;
 656         unsigned long absent_pages;
 657
 658         pci_iommu_alloc();
 659
 660         /* clear_bss() already clear the empty_zero_page */
 661
 662         reservedpages = 0;
 663
 664         /* this will put all low memory onto the freelists */
 665 #ifdef CONFIG_NUMA
 666         totalram_pages = numa_free_all_bootmem();
 667 #else
 668         totalram_pages = free_all_bootmem();
 669 #endif
 670
 671         absent_pages = absent_pages_in_range(0, max_pfn);
 672         reservedpages = max_pfn - totalram_pages - absent_pages;
 673         after_bootmem = 1;
 674
 675         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 676         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 677         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 678
 679         /* Register memory areas for /proc/kcore */
 680         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 681         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 682                    VMALLOC_END-VMALLOC_START);
 683         kclist_add(&kcore_kernel, &_stext, _end - _stext);
 684         kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
 685         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 686                                  VSYSCALL_END - VSYSCALL_START);
 687
 688         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 689                          "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
 690                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 691                 max_pfn << (PAGE_SHIFT-10),
 692                 codesize >> 10,
 693                 absent_pages << (PAGE_SHIFT-10),
 694                 reservedpages << (PAGE_SHIFT-10),
 695                 datasize >> 10,
 696                 initsize >> 10);
 697 }
 698
 699 #ifdef CONFIG_DEBUG_RODATA
 700 const int rodata_test_data = 0xC3;
 701 EXPORT_SYMBOL_GPL(rodata_test_data);
 702
 703 void mark_rodata_ro(void)
 704 {
 705         unsigned long start = PFN_ALIGN(_stext), end = PFN_ALIGN(__end_rodata);
 706         unsigned long rodata_start =
 707                 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
 708
 709 #ifdef CONFIG_DYNAMIC_FTRACE
 710         /* Dynamic tracing modifies the kernel text section */
 711         start = rodata_start;
 712 #endif
 713
 714         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 715                (end - start) >> 10);
 716         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 717
 718         /*
 719          * The rodata section (but not the kernel text!) should also be
 720          * not-executable.
 721          */
 722         set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
 723
 724         rodata_test();
 725
 726 #ifdef CONFIG_CPA_DEBUG
 727         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
 728         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 729
 730         printk(KERN_INFO "Testing CPA: again\n");
 731         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 732 #endif
 733 }
 734
 735 #endif
 736
 737 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
 738                                    int flags)
 739 {
 740 #ifdef CONFIG_NUMA
 741         int nid, next_nid;
 742         int ret;
 743 #endif
 744         unsigned long pfn = phys >> PAGE_SHIFT;
 745
 746         if (pfn >= max_pfn) {
 747                 /*
 748                  * This can happen with kdump kernels when accessing
 749                  * firmware tables:
 750                  */
 751                 if (pfn < max_pfn_mapped)
 752                         return -EFAULT;
 753
 754                 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n",
 755                                 phys, len);
 756                 return -EFAULT;
 757         }
 758
 759         /* Should check here against the e820 map to avoid double free */
 760 #ifdef CONFIG_NUMA
 761         nid = phys_to_nid(phys);
 762         next_nid = phys_to_nid(phys + len - 1);
 763         if (nid == next_nid)
 764                 ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags);
 765         else
 766                 ret = reserve_bootmem(phys, len, flags);
 767
 768         if (ret != 0)
 769                 return ret;
 770
 771 #else
 772         reserve_bootmem(phys, len, BOOTMEM_DEFAULT);
 773 #endif
 774
 775         if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
 776                 dma_reserve += len / PAGE_SIZE;
 777                 set_dma_reserve(dma_reserve);
 778         }
 779
 780         return 0;
 781 }
 782
 783 int kern_addr_valid(unsigned long addr)
 784 {
 785         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 786         pgd_t *pgd;
 787         pud_t *pud;
 788         pmd_t *pmd;
 789         pte_t *pte;
 790
 791         if (above != 0 && above != -1UL)
 792                 return 0;
 793
 794         pgd = pgd_offset_k(addr);
 795         if (pgd_none(*pgd))
 796                 return 0;
 797
 798         pud = pud_offset(pgd, addr);
 799         if (pud_none(*pud))
 800                 return 0;
 801
 802         pmd = pmd_offset(pud, addr);
 803         if (pmd_none(*pmd))
 804                 return 0;
 805
 806         if (pmd_large(*pmd))
 807                 return pfn_valid(pmd_pfn(*pmd));
 808
 809         pte = pte_offset_kernel(pmd, addr);
 810         if (pte_none(*pte))
 811                 return 0;
 812
 813         return pfn_valid(pte_pfn(*pte));
 814 }
 815
 816 /*
 817  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 818  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 819  * not need special handling anymore:
 820  */
 821 static struct vm_area_struct gate_vma = {
 822         .vm_start       = VSYSCALL_START,
 823         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 824         .vm_page_prot   = PAGE_READONLY_EXEC,
 825         .vm_flags       = VM_READ | VM_EXEC
 826 };
 827
 828 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 829 {
 830 #ifdef CONFIG_IA32_EMULATION
 831         if (test_tsk_thread_flag(tsk, TIF_IA32))
 832                 return NULL;
 833 #endif
 834         return &gate_vma;
 835 }
 836
 837 int in_gate_area(struct task_struct *task, unsigned long addr)
 838 {
 839         struct vm_area_struct *vma = get_gate_vma(task);
 840
 841         if (!vma)
 842                 return 0;
 843
 844         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 845 }
 846
 847 /*
 848  * Use this when you have no reliable task/vma, typically from interrupt
 849  * context. It is less reliable than using the task's vma and may give
 850  * false positives:
 851  */
 852 int in_gate_area_no_task(unsigned long addr)
 853 {
 854         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 855 }
 856
 857 const char *arch_vma_name(struct vm_area_struct *vma)
 858 {
 859         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 860                 return "[vdso]";
 861         if (vma == &gate_vma)
 862                 return "[vsyscall]";
 863         return NULL;
 864 }
 865
 866 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 867 /*
 868  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 869  */
 870 static long __meminitdata addr_start, addr_end;
 871 static void __meminitdata *p_start, *p_end;
 872 static int __meminitdata node_start;
 873
 874 int __meminit
 875 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 876 {
 877         unsigned long addr = (unsigned long)start_page;
 878         unsigned long end = (unsigned long)(start_page + size);
 879         unsigned long next;
 880         pgd_t *pgd;
 881         pud_t *pud;
 882         pmd_t *pmd;
 883
 884         for (; addr < end; addr = next) {
 885                 void *p = NULL;
 886
 887                 pgd = vmemmap_pgd_populate(addr, node);
 888                 if (!pgd)
 889                         return -ENOMEM;
 890
 891                 pud = vmemmap_pud_populate(pgd, addr, node);
 892                 if (!pud)
 893                         return -ENOMEM;
 894
 895                 if (!cpu_has_pse) {
 896                         next = (addr + PAGE_SIZE) & PAGE_MASK;
 897                         pmd = vmemmap_pmd_populate(pud, addr, node);
 898
 899                         if (!pmd)
 900                                 return -ENOMEM;
 901
 902                         p = vmemmap_pte_populate(pmd, addr, node);
 903
 904                         if (!p)
 905                                 return -ENOMEM;
 906
 907                         addr_end = addr + PAGE_SIZE;
 908                         p_end = p + PAGE_SIZE;
 909                 } else {
 910                         next = pmd_addr_end(addr, end);
 911
 912                         pmd = pmd_offset(pud, addr);
 913                         if (pmd_none(*pmd)) {
 914                                 pte_t entry;
 915
 916                                 p = vmemmap_alloc_block(PMD_SIZE, node);
 917                                 if (!p)
 918                                         return -ENOMEM;
 919
 920                                 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
 921                                                 PAGE_KERNEL_LARGE);
 922                                 set_pmd(pmd, __pmd(pte_val(entry)));
 923
 924                                 /* check to see if we have contiguous blocks */
 925                                 if (p_end != p || node_start != node) {
 926                                         if (p_start)
 927                                                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
 928                                                        addr_start, addr_end-1, p_start, p_end-1, node_start);
 929                                         addr_start = addr;
 930                                         node_start = node;
 931                                         p_start = p;
 932                                 }
 933
 934                                 addr_end = addr + PMD_SIZE;
 935                                 p_end = p + PMD_SIZE;
 936                         } else
 937                                 vmemmap_verify((pte_t *)pmd, node, addr, next);
 938                 }
 939
 940         }
 941         return 0;
 942 }
 943
 944 void __meminit vmemmap_populate_print_last(void)
 945 {
 946         if (p_start) {
 947                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
 948                         addr_start, addr_end-1, p_start, p_end-1, node_start);
 949                 p_start = NULL;
 950                 p_end = NULL;
 951                 node_start = 0;
 952         }
 953 }
 954 #endif