err.no Git - 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/system.h>
  35 #include <asm/uaccess.h>
  36 #include <asm/pgtable.h>
  37 #include <asm/pgalloc.h>
  38 #include <asm/dma.h>
  39 #include <asm/fixmap.h>
  40 #include <asm/e820.h>
  41 #include <asm/apic.h>
  42 #include <asm/tlb.h>
  43 #include <asm/mmu_context.h>
  44 #include <asm/proto.h>
  45 #include <asm/smp.h>
  46 #include <asm/sections.h>
  47 #include <asm/kdebug.h>
  48 #include <asm/numa.h>
  49 #include <asm/cacheflush.h>
  50
  51 /*
  52  * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
  53  * The direct mapping extends to max_pfn_mapped, so that we can directly access
  54  * apertures, ACPI and other tables without having to play with fixmaps.
  55  */
  56 unsigned long max_pfn_mapped;
  57
  58 static unsigned long dma_reserve __initdata;
  59
  60 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  61
  62 int direct_gbpages __meminitdata
  63 #ifdef CONFIG_DIRECT_GBPAGES
  64                                 = 1
  65 #endif
  66 ;
  67
  68 static int __init parse_direct_gbpages_off(char *arg)
  69 {
  70         direct_gbpages = 0;
  71         return 0;
  72 }
  73 early_param("nogbpages", parse_direct_gbpages_off);
  74
  75 static int __init parse_direct_gbpages_on(char *arg)
  76 {
  77         direct_gbpages = 1;
  78         return 0;
  79 }
  80 early_param("gbpages", parse_direct_gbpages_on);
  81
  82 /*
  83  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
  84  * physical space so we can cache the place of the first one and move
  85  * around without checking the pgd every time.
  86  */
  87
  88 void show_mem(void)
  89 {
  90         long i, total = 0, reserved = 0;
  91         long shared = 0, cached = 0;
  92         struct page *page;
  93         pg_data_t *pgdat;
  94
  95         printk(KERN_INFO "Mem-info:\n");
  96         show_free_areas();
  97         for_each_online_pgdat(pgdat) {
  98                 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
  99                         /*
 100                          * This loop can take a while with 256 GB and
 101                          * 4k pages so defer the NMI watchdog:
 102                          */
 103                         if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
 104                                 touch_nmi_watchdog();
 105
 106                         if (!pfn_valid(pgdat->node_start_pfn + i))
 107                                 continue;
 108
 109                         page = pfn_to_page(pgdat->node_start_pfn + i);
 110                         total++;
 111                         if (PageReserved(page))
 112                                 reserved++;
 113                         else if (PageSwapCache(page))
 114                                 cached++;
 115                         else if (page_count(page))
 116                                 shared += page_count(page) - 1;
 117                 }
 118         }
 119         printk(KERN_INFO "%lu pages of RAM\n",          total);
 120         printk(KERN_INFO "%lu reserved pages\n",        reserved);
 121         printk(KERN_INFO "%lu pages shared\n",          shared);
 122         printk(KERN_INFO "%lu pages swap cached\n",     cached);
 123 }
 124
 125 int after_bootmem;
 126
 127 static __init void *spp_getpage(void)
 128 {
 129         void *ptr;
 130
 131         if (after_bootmem)
 132                 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
 133         else
 134                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 135
 136         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 137                 panic("set_pte_phys: cannot allocate page data %s\n",
 138                         after_bootmem ? "after bootmem" : "");
 139         }
 140
 141         pr_debug("spp_getpage %p\n", ptr);
 142
 143         return ptr;
 144 }
 145
 146 void
 147 set_pte_vaddr(unsigned long vaddr, pte_t new_pte)
 148 {
 149         pgd_t *pgd;
 150         pud_t *pud;
 151         pmd_t *pmd;
 152         pte_t *pte;
 153
 154         pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(new_pte));
 155
 156         pgd = pgd_offset_k(vaddr);
 157         if (pgd_none(*pgd)) {
 158                 printk(KERN_ERR
 159                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
 160                 return;
 161         }
 162         pud = pud_offset(pgd, vaddr);
 163         if (pud_none(*pud)) {
 164                 pmd = (pmd_t *) spp_getpage();
 165                 pud_populate(&init_mm, pud, pmd);
 166                 if (pmd != pmd_offset(pud, 0)) {
 167                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
 168                                 pmd, pmd_offset(pud, 0));
 169                         return;
 170                 }
 171         }
 172         pmd = pmd_offset(pud, vaddr);
 173         if (pmd_none(*pmd)) {
 174                 pte = (pte_t *) spp_getpage();
 175                 pmd_populate_kernel(&init_mm, pmd, pte);
 176                 if (pte != pte_offset_kernel(pmd, 0)) {
 177                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
 178                         return;
 179                 }
 180         }
 181
 182         pte = pte_offset_kernel(pmd, vaddr);
 183         if (!pte_none(*pte) && pte_val(new_pte) &&
 184             pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask))
 185                 pte_ERROR(*pte);
 186         set_pte(pte, new_pte);
 187
 188         /*
 189          * It's enough to flush this one mapping.
 190          * (PGE mappings get flushed as well)
 191          */
 192         __flush_tlb_one(vaddr);
 193 }
 194
 195 /*
 196  * The head.S code sets up the kernel high mapping:
 197  *
 198  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
 199  *
 200  * phys_addr holds the negative offset to the kernel, which is added
 201  * to the compile time generated pmds. This results in invalid pmds up
 202  * to the point where we hit the physaddr 0 mapping.
 203  *
 204  * We limit the mappings to the region from _text to _end.  _end is
 205  * rounded up to the 2MB boundary. This catches the invalid pmds as
 206  * well, as they are located before _text:
 207  */
 208 void __init cleanup_highmap(void)
 209 {
 210         unsigned long vaddr = __START_KERNEL_map;
 211         unsigned long end = round_up((unsigned long)_end, PMD_SIZE) - 1;
 212         pmd_t *pmd = level2_kernel_pgt;
 213         pmd_t *last_pmd = pmd + PTRS_PER_PMD;
 214
 215         for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
 216                 if (pmd_none(*pmd))
 217                         continue;
 218                 if (vaddr < (unsigned long) _text || vaddr > end)
 219                         set_pmd(pmd, __pmd(0));
 220         }
 221 }
 222
 223 static unsigned long __initdata table_start;
 224 static unsigned long __meminitdata table_end;
 225 static unsigned long __meminitdata table_top;
 226
 227 static __meminit void *alloc_low_page(unsigned long *phys)
 228 {
 229         unsigned long pfn = table_end++;
 230         void *adr;
 231
 232         if (after_bootmem) {
 233                 adr = (void *)get_zeroed_page(GFP_ATOMIC);
 234                 *phys = __pa(adr);
 235
 236                 return adr;
 237         }
 238
 239         if (pfn >= table_top)
 240                 panic("alloc_low_page: ran out of memory");
 241
 242         adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
 243         memset(adr, 0, PAGE_SIZE);
 244         *phys  = pfn * PAGE_SIZE;
 245         return adr;
 246 }
 247
 248 static __meminit void unmap_low_page(void *adr)
 249 {
 250         if (after_bootmem)
 251                 return;
 252
 253         early_iounmap(adr, PAGE_SIZE);
 254 }
 255
 256 static unsigned long __meminit
 257 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
 258 {
 259         unsigned long pages = 0;
 260
 261         int i = pmd_index(address);
 262
 263         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
 264                 pmd_t *pmd = pmd_page + pmd_index(address);
 265
 266                 if (address >= end) {
 267                         if (!after_bootmem) {
 268                                 for (; i < PTRS_PER_PMD; i++, pmd++)
 269                                         set_pmd(pmd, __pmd(0));
 270                         }
 271                         break;
 272                 }
 273
 274                 if (pmd_val(*pmd))
 275                         continue;
 276
 277                 pages++;
 278                 set_pte((pte_t *)pmd,
 279                         pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 280         }
 281         update_page_count(PG_LEVEL_2M, pages);
 282         return address;
 283 }
 284
 285 static unsigned long __meminit
 286 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
 287 {
 288         pmd_t *pmd = pmd_offset(pud, 0);
 289         unsigned long last_map_addr;
 290
 291         spin_lock(&init_mm.page_table_lock);
 292         last_map_addr = phys_pmd_init(pmd, address, end);
 293         spin_unlock(&init_mm.page_table_lock);
 294         __flush_tlb_all();
 295         return last_map_addr;
 296 }
 297
 298 static unsigned long __meminit
 299 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
 300 {
 301         unsigned long pages = 0;
 302         unsigned long last_map_addr = end;
 303         int i = pud_index(addr);
 304
 305         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
 306                 unsigned long pmd_phys;
 307                 pud_t *pud = pud_page + pud_index(addr);
 308                 pmd_t *pmd;
 309
 310                 if (addr >= end)
 311                         break;
 312
 313                 if (!after_bootmem &&
 314                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
 315                         set_pud(pud, __pud(0));
 316                         continue;
 317                 }
 318
 319                 if (pud_val(*pud)) {
 320                         if (!pud_large(*pud))
 321                                 last_map_addr = phys_pmd_update(pud, addr, end);
 322                         continue;
 323                 }
 324
 325                 if (direct_gbpages) {
 326                         pages++;
 327                         set_pte((pte_t *)pud,
 328                                 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 329                         last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
 330                         continue;
 331                 }
 332
 333                 pmd = alloc_low_page(&pmd_phys);
 334
 335                 spin_lock(&init_mm.page_table_lock);
 336                 pud_populate(&init_mm, pud, __va(pmd_phys));
 337                 last_map_addr = phys_pmd_init(pmd, addr, end);
 338                 spin_unlock(&init_mm.page_table_lock);
 339
 340                 unmap_low_page(pmd);
 341         }
 342         __flush_tlb_all();
 343         update_page_count(PG_LEVEL_1G, pages);
 344
 345         return last_map_addr;
 346 }
 347
 348 static void __init find_early_table_space(unsigned long end)
 349 {
 350         unsigned long puds, pmds, tables, start;
 351
 352         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 353         tables = round_up(puds * sizeof(pud_t), PAGE_SIZE);
 354         if (!direct_gbpages) {
 355                 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 356                 tables += round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
 357         }
 358
 359         /*
 360          * RED-PEN putting page tables only on node 0 could
 361          * cause a hotspot and fill up ZONE_DMA. The page tables
 362          * need roughly 0.5KB per GB.
 363          */
 364         start = 0x8000;
 365         table_start = find_e820_area(start, end, tables, PAGE_SIZE);
 366         if (table_start == -1UL)
 367                 panic("Cannot find space for the kernel page tables");
 368
 369         table_start >>= PAGE_SHIFT;
 370         table_end = table_start;
 371         table_top = table_start + (tables >> PAGE_SHIFT);
 372
 373         printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
 374                 end, table_start << PAGE_SHIFT, table_top << PAGE_SHIFT);
 375 }
 376
 377 static void __init init_gbpages(void)
 378 {
 379         if (direct_gbpages && cpu_has_gbpages)
 380                 printk(KERN_INFO "Using GB pages for direct mapping\n");
 381         else
 382                 direct_gbpages = 0;
 383 }
 384
 385 #ifdef CONFIG_MEMTEST
 386
 387 static void __init memtest(unsigned long start_phys, unsigned long size,
 388                                  unsigned pattern)
 389 {
 390         unsigned long i;
 391         unsigned long *start;
 392         unsigned long start_bad;
 393         unsigned long last_bad;
 394         unsigned long val;
 395         unsigned long start_phys_aligned;
 396         unsigned long count;
 397         unsigned long incr;
 398
 399         switch (pattern) {
 400         case 0:
 401                 val = 0UL;
 402                 break;
 403         case 1:
 404                 val = -1UL;
 405                 break;
 406         case 2:
 407                 val = 0x5555555555555555UL;
 408                 break;
 409         case 3:
 410                 val = 0xaaaaaaaaaaaaaaaaUL;
 411                 break;
 412         default:
 413                 return;
 414         }
 415
 416         incr = sizeof(unsigned long);
 417         start_phys_aligned = ALIGN(start_phys, incr);
 418         count = (size - (start_phys_aligned - start_phys))/incr;
 419         start = __va(start_phys_aligned);
 420         start_bad = 0;
 421         last_bad = 0;
 422
 423         for (i = 0; i < count; i++)
 424                 start[i] = val;
 425         for (i = 0; i < count; i++, start++, start_phys_aligned += incr) {
 426                 if (*start != val) {
 427                         if (start_phys_aligned == last_bad + incr) {
 428                                 last_bad += incr;
 429                         } else {
 430                                 if (start_bad) {
 431                                         printk(KERN_CONT "\n  %016lx bad mem addr %016lx - %016lx reserved",
 432                                                 val, start_bad, last_bad + incr);
 433                                         reserve_early(start_bad, last_bad - start_bad, "BAD RAM");
 434                                 }
 435                                 start_bad = last_bad = start_phys_aligned;
 436                         }
 437                 }
 438         }
 439         if (start_bad) {
 440                 printk(KERN_CONT "\n  %016lx bad mem addr %016lx - %016lx reserved",
 441                         val, start_bad, last_bad + incr);
 442                 reserve_early(start_bad, last_bad - start_bad, "BAD RAM");
 443         }
 444
 445 }
 446
 447 /* default is disabled */
 448 static int memtest_pattern __initdata;
 449
 450 static int __init parse_memtest(char *arg)
 451 {
 452         if (arg)
 453                 memtest_pattern = simple_strtoul(arg, NULL, 0);
 454         return 0;
 455 }
 456
 457 early_param("memtest", parse_memtest);
 458
 459 static void __init early_memtest(unsigned long start, unsigned long end)
 460 {
 461         u64 t_start, t_size;
 462         unsigned pattern;
 463
 464         if (!memtest_pattern)
 465                 return;
 466
 467         printk(KERN_INFO "early_memtest: pattern num %d", memtest_pattern);
 468         for (pattern = 0; pattern < memtest_pattern; pattern++) {
 469                 t_start = start;
 470                 t_size = 0;
 471                 while (t_start < end) {
 472                         t_start = find_e820_area_size(t_start, &t_size, 1);
 473
 474                         /* done ? */
 475                         if (t_start >= end)
 476                                 break;
 477                         if (t_start + t_size > end)
 478                                 t_size = end - t_start;
 479
 480                         printk(KERN_CONT "\n  %016llx - %016llx pattern %d",
 481                                 (unsigned long long)t_start,
 482                                 (unsigned long long)t_start + t_size, pattern);
 483
 484                         memtest(t_start, t_size, pattern);
 485
 486                         t_start += t_size;
 487                 }
 488         }
 489         printk(KERN_CONT "\n");
 490 }
 491 #else
 492 static void __init early_memtest(unsigned long start, unsigned long end)
 493 {
 494 }
 495 #endif
 496
 497 /*
 498  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 499  * This runs before bootmem is initialized and gets pages directly from
 500  * the physical memory. To access them they are temporarily mapped.
 501  */
 502 unsigned long __init_refok init_memory_mapping(unsigned long start, unsigned long end)
 503 {
 504         unsigned long next, last_map_addr = end;
 505         unsigned long start_phys = start, end_phys = end;
 506
 507         printk(KERN_INFO "init_memory_mapping\n");
 508
 509         /*
 510          * Find space for the kernel direct mapping tables.
 511          *
 512          * Later we should allocate these tables in the local node of the
 513          * memory mapped. Unfortunately this is done currently before the
 514          * nodes are discovered.
 515          */
 516         if (!after_bootmem) {
 517                 init_gbpages();
 518                 find_early_table_space(end);
 519         }
 520
 521         start = (unsigned long)__va(start);
 522         end = (unsigned long)__va(end);
 523
 524         for (; start < end; start = next) {
 525                 pgd_t *pgd = pgd_offset_k(start);
 526                 unsigned long pud_phys;
 527                 pud_t *pud;
 528
 529                 if (after_bootmem)
 530                         pud = pud_offset(pgd, start & PGDIR_MASK);
 531                 else
 532                         pud = alloc_low_page(&pud_phys);
 533
 534                 next = start + PGDIR_SIZE;
 535                 if (next > end)
 536                         next = end;
 537                 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next));
 538                 if (!after_bootmem)
 539                         pgd_populate(&init_mm, pgd_offset_k(start),
 540                                      __va(pud_phys));
 541                 unmap_low_page(pud);
 542         }
 543
 544         if (!after_bootmem)
 545                 mmu_cr4_features = read_cr4();
 546         __flush_tlb_all();
 547
 548         if (!after_bootmem)
 549                 reserve_early(table_start << PAGE_SHIFT,
 550                                  table_end << PAGE_SHIFT, "PGTABLE");
 551
 552         if (!after_bootmem)
 553                 early_memtest(start_phys, end_phys);
 554
 555         return last_map_addr >> PAGE_SHIFT;
 556 }
 557
 558 #ifndef CONFIG_NUMA
 559 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn)
 560 {
 561         unsigned long bootmap_size, bootmap;
 562
 563         bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
 564         bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
 565                                  PAGE_SIZE);
 566         if (bootmap == -1L)
 567                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 568         /* don't touch min_low_pfn */
 569         bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
 570                                          0, end_pfn);
 571         e820_register_active_regions(0, start_pfn, end_pfn);
 572         free_bootmem_with_active_regions(0, end_pfn);
 573         early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
 574         reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
 575 }
 576
 577 void __init paging_init(void)
 578 {
 579         unsigned long max_zone_pfns[MAX_NR_ZONES];
 580
 581         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 582         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 583         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 584         max_zone_pfns[ZONE_NORMAL] = max_pfn;
 585
 586         memory_present(0, 0, max_pfn);
 587         sparse_init();
 588         free_area_init_nodes(max_zone_pfns);
 589 }
 590 #endif
 591
 592 /*
 593  * Memory hotplug specific functions
 594  */
 595 #ifdef CONFIG_MEMORY_HOTPLUG
 596 /*
 597  * Memory is added always to NORMAL zone. This means you will never get
 598  * additional DMA/DMA32 memory.
 599  */
 600 int arch_add_memory(int nid, u64 start, u64 size)
 601 {
 602         struct pglist_data *pgdat = NODE_DATA(nid);
 603         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 604         unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
 605         unsigned long nr_pages = size >> PAGE_SHIFT;
 606         int ret;
 607
 608         last_mapped_pfn = init_memory_mapping(start, start + size-1);
 609         if (last_mapped_pfn > max_pfn_mapped)
 610                 max_pfn_mapped = last_mapped_pfn;
 611
 612         ret = __add_pages(zone, start_pfn, nr_pages);
 613         WARN_ON(1);
 614
 615         return ret;
 616 }
 617 EXPORT_SYMBOL_GPL(arch_add_memory);
 618
 619 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
 620 int memory_add_physaddr_to_nid(u64 start)
 621 {
 622         return 0;
 623 }
 624 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 625 #endif
 626
 627 #endif /* CONFIG_MEMORY_HOTPLUG */
 628
 629 /*
 630  * devmem_is_allowed() checks to see if /dev/mem access to a certain address
 631  * is valid. The argument is a physical page number.
 632  *
 633  *
 634  * On x86, access has to be given to the first megabyte of ram because that area
 635  * contains bios code and data regions used by X and dosemu and similar apps.
 636  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
 637  * mmio resources as well as potential bios/acpi data regions.
 638  */
 639 int devmem_is_allowed(unsigned long pagenr)
 640 {
 641         if (pagenr <= 256)
 642                 return 1;
 643         if (!page_is_ram(pagenr))
 644                 return 1;
 645         return 0;
 646 }
 647
 648
 649 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
 650                          kcore_modules, kcore_vsyscall;
 651
 652 void __init mem_init(void)
 653 {
 654         long codesize, reservedpages, datasize, initsize;
 655
 656         pci_iommu_alloc();
 657
 658         /* clear_bss() already clear the empty_zero_page */
 659
 660         reservedpages = 0;
 661
 662         /* this will put all low memory onto the freelists */
 663 #ifdef CONFIG_NUMA
 664         totalram_pages = numa_free_all_bootmem();
 665 #else
 666         totalram_pages = free_all_bootmem();
 667 #endif
 668         reservedpages = max_pfn - totalram_pages -
 669                                         absent_pages_in_range(0, max_pfn);
 670         after_bootmem = 1;
 671
 672         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 673         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 674         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 675
 676         /* Register memory areas for /proc/kcore */
 677         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 678         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 679                    VMALLOC_END-VMALLOC_START);
 680         kclist_add(&kcore_kernel, &_stext, _end - _stext);
 681         kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
 682         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 683                                  VSYSCALL_END - VSYSCALL_START);
 684
 685         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 686                                 "%ldk reserved, %ldk data, %ldk init)\n",
 687                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 688                 max_pfn << (PAGE_SHIFT-10),
 689                 codesize >> 10,
 690                 reservedpages << (PAGE_SHIFT-10),
 691                 datasize >> 10,
 692                 initsize >> 10);
 693
 694         cpa_init();
 695 }
 696
 697 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 698 {
 699         unsigned long addr = begin;
 700
 701         if (addr >= end)
 702                 return;
 703
 704         /*
 705          * If debugging page accesses then do not free this memory but
 706          * mark them not present - any buggy init-section access will
 707          * create a kernel page fault:
 708          */
 709 #ifdef CONFIG_DEBUG_PAGEALLOC
 710         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
 711                 begin, PAGE_ALIGN(end));
 712         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 713 #else
 714         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 715
 716         for (; addr < end; addr += PAGE_SIZE) {
 717                 ClearPageReserved(virt_to_page(addr));
 718                 init_page_count(virt_to_page(addr));
 719                 memset((void *)(addr & ~(PAGE_SIZE-1)),
 720                         POISON_FREE_INITMEM, PAGE_SIZE);
 721                 free_page(addr);
 722                 totalram_pages++;
 723         }
 724 #endif
 725 }
 726
 727 void free_initmem(void)
 728 {
 729         free_init_pages("unused kernel memory",
 730                         (unsigned long)(&__init_begin),
 731                         (unsigned long)(&__init_end));
 732 }
 733
 734 #ifdef CONFIG_DEBUG_RODATA
 735 const int rodata_test_data = 0xC3;
 736 EXPORT_SYMBOL_GPL(rodata_test_data);
 737
 738 void mark_rodata_ro(void)
 739 {
 740         unsigned long start = PFN_ALIGN(_stext), end = PFN_ALIGN(__end_rodata);
 741
 742         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 743                (end - start) >> 10);
 744         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 745
 746         /*
 747          * The rodata section (but not the kernel text!) should also be
 748          * not-executable.
 749          */
 750         start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
 751         set_memory_nx(start, (end - start) >> PAGE_SHIFT);
 752
 753         rodata_test();
 754
 755 #ifdef CONFIG_CPA_DEBUG
 756         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
 757         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 758
 759         printk(KERN_INFO "Testing CPA: again\n");
 760         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 761 #endif
 762 }
 763
 764 #endif
 765
 766 #ifdef CONFIG_BLK_DEV_INITRD
 767 void free_initrd_mem(unsigned long start, unsigned long end)
 768 {
 769         free_init_pages("initrd memory", start, end);
 770 }
 771 #endif
 772
 773 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
 774                                    int flags)
 775 {
 776 #ifdef CONFIG_NUMA
 777         int nid, next_nid;
 778         int ret;
 779 #endif
 780         unsigned long pfn = phys >> PAGE_SHIFT;
 781
 782         if (pfn >= max_pfn) {
 783                 /*
 784                  * This can happen with kdump kernels when accessing
 785                  * firmware tables:
 786                  */
 787                 if (pfn < max_pfn_mapped)
 788                         return -EFAULT;
 789
 790                 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n",
 791                                 phys, len);
 792                 return -EFAULT;
 793         }
 794
 795         /* Should check here against the e820 map to avoid double free */
 796 #ifdef CONFIG_NUMA
 797         nid = phys_to_nid(phys);
 798         next_nid = phys_to_nid(phys + len - 1);
 799         if (nid == next_nid)
 800                 ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags);
 801         else
 802                 ret = reserve_bootmem(phys, len, flags);
 803
 804         if (ret != 0)
 805                 return ret;
 806
 807 #else
 808         reserve_bootmem(phys, len, BOOTMEM_DEFAULT);
 809 #endif
 810
 811         if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
 812                 dma_reserve += len / PAGE_SIZE;
 813                 set_dma_reserve(dma_reserve);
 814         }
 815
 816         return 0;
 817 }
 818
 819 int kern_addr_valid(unsigned long addr)
 820 {
 821         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 822         pgd_t *pgd;
 823         pud_t *pud;
 824         pmd_t *pmd;
 825         pte_t *pte;
 826
 827         if (above != 0 && above != -1UL)
 828                 return 0;
 829
 830         pgd = pgd_offset_k(addr);
 831         if (pgd_none(*pgd))
 832                 return 0;
 833
 834         pud = pud_offset(pgd, addr);
 835         if (pud_none(*pud))
 836                 return 0;
 837
 838         pmd = pmd_offset(pud, addr);
 839         if (pmd_none(*pmd))
 840                 return 0;
 841
 842         if (pmd_large(*pmd))
 843                 return pfn_valid(pmd_pfn(*pmd));
 844
 845         pte = pte_offset_kernel(pmd, addr);
 846         if (pte_none(*pte))
 847                 return 0;
 848
 849         return pfn_valid(pte_pfn(*pte));
 850 }
 851
 852 /*
 853  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 854  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 855  * not need special handling anymore:
 856  */
 857 static struct vm_area_struct gate_vma = {
 858         .vm_start       = VSYSCALL_START,
 859         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 860         .vm_page_prot   = PAGE_READONLY_EXEC,
 861         .vm_flags       = VM_READ | VM_EXEC
 862 };
 863
 864 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 865 {
 866 #ifdef CONFIG_IA32_EMULATION
 867         if (test_tsk_thread_flag(tsk, TIF_IA32))
 868                 return NULL;
 869 #endif
 870         return &gate_vma;
 871 }
 872
 873 int in_gate_area(struct task_struct *task, unsigned long addr)
 874 {
 875         struct vm_area_struct *vma = get_gate_vma(task);
 876
 877         if (!vma)
 878                 return 0;
 879
 880         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 881 }
 882
 883 /*
 884  * Use this when you have no reliable task/vma, typically from interrupt
 885  * context. It is less reliable than using the task's vma and may give
 886  * false positives:
 887  */
 888 int in_gate_area_no_task(unsigned long addr)
 889 {
 890         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 891 }
 892
 893 const char *arch_vma_name(struct vm_area_struct *vma)
 894 {
 895         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 896                 return "[vdso]";
 897         if (vma == &gate_vma)
 898                 return "[vsyscall]";
 899         return NULL;
 900 }
 901
 902 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 903 /*
 904  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 905  */
 906 static long __meminitdata addr_start, addr_end;
 907 static void __meminitdata *p_start, *p_end;
 908 static int __meminitdata node_start;
 909
 910 int __meminit
 911 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 912 {
 913         unsigned long addr = (unsigned long)start_page;
 914         unsigned long end = (unsigned long)(start_page + size);
 915         unsigned long next;
 916         pgd_t *pgd;
 917         pud_t *pud;
 918         pmd_t *pmd;
 919
 920         for (; addr < end; addr = next) {
 921                 next = pmd_addr_end(addr, end);
 922
 923                 pgd = vmemmap_pgd_populate(addr, node);
 924                 if (!pgd)
 925                         return -ENOMEM;
 926
 927                 pud = vmemmap_pud_populate(pgd, addr, node);
 928                 if (!pud)
 929                         return -ENOMEM;
 930
 931                 pmd = pmd_offset(pud, addr);
 932                 if (pmd_none(*pmd)) {
 933                         pte_t entry;
 934                         void *p;
 935
 936                         p = vmemmap_alloc_block(PMD_SIZE, node);
 937                         if (!p)
 938                                 return -ENOMEM;
 939
 940                         entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
 941                                                         PAGE_KERNEL_LARGE);
 942                         set_pmd(pmd, __pmd(pte_val(entry)));
 943
 944                         /* check to see if we have contiguous blocks */
 945                         if (p_end != p || node_start != node) {
 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                                 addr_start = addr;
 950                                 node_start = node;
 951                                 p_start = p;
 952                         }
 953                         addr_end = addr + PMD_SIZE;
 954                         p_end = p + PMD_SIZE;
 955                 } else {
 956                         vmemmap_verify((pte_t *)pmd, node, addr, next);
 957                 }
 958         }
 959         return 0;
 960 }
 961
 962 void __meminit vmemmap_populate_print_last(void)
 963 {
 964         if (p_start) {
 965                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
 966                         addr_start, addr_end-1, p_start, p_end-1, node_start);
 967                 p_start = NULL;
 968                 p_end = NULL;
 969                 node_start = 0;
 970         }
 971 }
 972 #endif