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  * PFN of last memory page.
  53  */
  54 unsigned long end_pfn;
  55
  56 /*
  57  * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
  58  * The direct mapping extends to max_pfn_mapped, so that we can directly access
  59  * apertures, ACPI and other tables without having to play with fixmaps.
  60  */
  61 unsigned long max_pfn_mapped;
  62
  63 static unsigned long dma_reserve __initdata;
  64
  65 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  66
  67 int direct_gbpages __meminitdata
  68 #ifdef CONFIG_DIRECT_GBPAGES
  69                                 = 1
  70 #endif
  71 ;
  72
  73 static int __init parse_direct_gbpages_off(char *arg)
  74 {
  75         direct_gbpages = 0;
  76         return 0;
  77 }
  78 early_param("nogbpages", parse_direct_gbpages_off);
  79
  80 static int __init parse_direct_gbpages_on(char *arg)
  81 {
  82         direct_gbpages = 1;
  83         return 0;
  84 }
  85 early_param("gbpages", parse_direct_gbpages_on);
  86
  87 /*
  88  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
  89  * physical space so we can cache the place of the first one and move
  90  * around without checking the pgd every time.
  91  */
  92
  93 void show_mem(void)
  94 {
  95         long i, total = 0, reserved = 0;
  96         long shared = 0, cached = 0;
  97         struct page *page;
  98         pg_data_t *pgdat;
  99
 100         printk(KERN_INFO "Mem-info:\n");
 101         show_free_areas();
 102         for_each_online_pgdat(pgdat) {
 103                 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
 104                         /*
 105                          * This loop can take a while with 256 GB and
 106                          * 4k pages so defer the NMI watchdog:
 107                          */
 108                         if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
 109                                 touch_nmi_watchdog();
 110
 111                         if (!pfn_valid(pgdat->node_start_pfn + i))
 112                                 continue;
 113
 114                         page = pfn_to_page(pgdat->node_start_pfn + i);
 115                         total++;
 116                         if (PageReserved(page))
 117                                 reserved++;
 118                         else if (PageSwapCache(page))
 119                                 cached++;
 120                         else if (page_count(page))
 121                                 shared += page_count(page) - 1;
 122                 }
 123         }
 124         printk(KERN_INFO "%lu pages of RAM\n",          total);
 125         printk(KERN_INFO "%lu reserved pages\n",        reserved);
 126         printk(KERN_INFO "%lu pages shared\n",          shared);
 127         printk(KERN_INFO "%lu pages swap cached\n",     cached);
 128 }
 129
 130 int after_bootmem;
 131
 132 static __init void *spp_getpage(void)
 133 {
 134         void *ptr;
 135
 136         if (after_bootmem)
 137                 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
 138         else
 139                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 140
 141         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 142                 panic("set_pte_phys: cannot allocate page data %s\n",
 143                         after_bootmem ? "after bootmem" : "");
 144         }
 145
 146         pr_debug("spp_getpage %p\n", ptr);
 147
 148         return ptr;
 149 }
 150
 151 void
 152 set_pte_vaddr(unsigned long vaddr, pte_t new_pte)
 153 {
 154         pgd_t *pgd;
 155         pud_t *pud;
 156         pmd_t *pmd;
 157         pte_t *pte;
 158
 159         pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(new_pte));
 160
 161         pgd = pgd_offset_k(vaddr);
 162         if (pgd_none(*pgd)) {
 163                 printk(KERN_ERR
 164                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
 165                 return;
 166         }
 167         pud = pud_offset(pgd, vaddr);
 168         if (pud_none(*pud)) {
 169                 pmd = (pmd_t *) spp_getpage();
 170                 pud_populate(&init_mm, pud, pmd);
 171                 if (pmd != pmd_offset(pud, 0)) {
 172                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
 173                                 pmd, pmd_offset(pud, 0));
 174                         return;
 175                 }
 176         }
 177         pmd = pmd_offset(pud, vaddr);
 178         if (pmd_none(*pmd)) {
 179                 pte = (pte_t *) spp_getpage();
 180                 pmd_populate_kernel(&init_mm, pmd, pte);
 181                 if (pte != pte_offset_kernel(pmd, 0)) {
 182                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
 183                         return;
 184                 }
 185         }
 186
 187         pte = pte_offset_kernel(pmd, vaddr);
 188         if (!pte_none(*pte) && pte_val(new_pte) &&
 189             pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask))
 190                 pte_ERROR(*pte);
 191         set_pte(pte, new_pte);
 192
 193         /*
 194          * It's enough to flush this one mapping.
 195          * (PGE mappings get flushed as well)
 196          */
 197         __flush_tlb_one(vaddr);
 198 }
 199
 200 /*
 201  * The head.S code sets up the kernel high mapping:
 202  *
 203  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
 204  *
 205  * phys_addr holds the negative offset to the kernel, which is added
 206  * to the compile time generated pmds. This results in invalid pmds up
 207  * to the point where we hit the physaddr 0 mapping.
 208  *
 209  * We limit the mappings to the region from _text to _end.  _end is
 210  * rounded up to the 2MB boundary. This catches the invalid pmds as
 211  * well, as they are located before _text:
 212  */
 213 void __init cleanup_highmap(void)
 214 {
 215         unsigned long vaddr = __START_KERNEL_map;
 216         unsigned long end = round_up((unsigned long)_end, PMD_SIZE) - 1;
 217         pmd_t *pmd = level2_kernel_pgt;
 218         pmd_t *last_pmd = pmd + PTRS_PER_PMD;
 219
 220         for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
 221                 if (pmd_none(*pmd))
 222                         continue;
 223                 if (vaddr < (unsigned long) _text || vaddr > end)
 224                         set_pmd(pmd, __pmd(0));
 225         }
 226 }
 227
 228 static unsigned long __initdata table_start;
 229 static unsigned long __meminitdata table_end;
 230
 231 static __meminit void *alloc_low_page(unsigned long *phys)
 232 {
 233         unsigned long pfn = table_end++;
 234         void *adr;
 235
 236         if (after_bootmem) {
 237                 adr = (void *)get_zeroed_page(GFP_ATOMIC);
 238                 *phys = __pa(adr);
 239
 240                 return adr;
 241         }
 242
 243         if (pfn >= end_pfn)
 244                 panic("alloc_low_page: ran out of memory");
 245
 246         adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
 247         memset(adr, 0, PAGE_SIZE);
 248         *phys  = pfn * PAGE_SIZE;
 249         return adr;
 250 }
 251
 252 static __meminit void unmap_low_page(void *adr)
 253 {
 254         if (after_bootmem)
 255                 return;
 256
 257         early_iounmap(adr, PAGE_SIZE);
 258 }
 259
 260 static unsigned long __meminit
 261 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
 262 {
 263         unsigned long pages = 0;
 264
 265         int i = pmd_index(address);
 266
 267         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
 268                 pmd_t *pmd = pmd_page + pmd_index(address);
 269
 270                 if (address >= end) {
 271                         if (!after_bootmem) {
 272                                 for (; i < PTRS_PER_PMD; i++, pmd++)
 273                                         set_pmd(pmd, __pmd(0));
 274                         }
 275                         break;
 276                 }
 277
 278                 if (pmd_val(*pmd))
 279                         continue;
 280
 281                 pages++;
 282                 set_pte((pte_t *)pmd,
 283                         pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 284         }
 285         update_page_count(PG_LEVEL_2M, pages);
 286         return address;
 287 }
 288
 289 static unsigned long __meminit
 290 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
 291 {
 292         pmd_t *pmd = pmd_offset(pud, 0);
 293         unsigned long last_map_addr;
 294
 295         spin_lock(&init_mm.page_table_lock);
 296         last_map_addr = phys_pmd_init(pmd, address, end);
 297         spin_unlock(&init_mm.page_table_lock);
 298         __flush_tlb_all();
 299         return last_map_addr;
 300 }
 301
 302 static unsigned long __meminit
 303 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
 304 {
 305         unsigned long pages = 0;
 306         unsigned long last_map_addr = end;
 307         int i = pud_index(addr);
 308
 309         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
 310                 unsigned long pmd_phys;
 311                 pud_t *pud = pud_page + pud_index(addr);
 312                 pmd_t *pmd;
 313
 314                 if (addr >= end)
 315                         break;
 316
 317                 if (!after_bootmem &&
 318                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
 319                         set_pud(pud, __pud(0));
 320                         continue;
 321                 }
 322
 323                 if (pud_val(*pud)) {
 324                         if (!pud_large(*pud))
 325                                 last_map_addr = phys_pmd_update(pud, addr, end);
 326                         continue;
 327                 }
 328
 329                 if (direct_gbpages) {
 330                         pages++;
 331                         set_pte((pte_t *)pud,
 332                                 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 333                         last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
 334                         continue;
 335                 }
 336
 337                 pmd = alloc_low_page(&pmd_phys);
 338
 339                 spin_lock(&init_mm.page_table_lock);
 340                 pud_populate(&init_mm, pud, __va(pmd_phys));
 341                 last_map_addr = phys_pmd_init(pmd, addr, end);
 342                 spin_unlock(&init_mm.page_table_lock);
 343
 344                 unmap_low_page(pmd);
 345         }
 346         __flush_tlb_all();
 347         update_page_count(PG_LEVEL_1G, pages);
 348
 349         return last_map_addr >> PAGE_SHIFT;
 350 }
 351
 352 static void __init find_early_table_space(unsigned long end)
 353 {
 354         unsigned long puds, pmds, tables, start;
 355
 356         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 357         tables = round_up(puds * sizeof(pud_t), PAGE_SIZE);
 358         if (!direct_gbpages) {
 359                 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 360                 tables += round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
 361         }
 362
 363         /*
 364          * RED-PEN putting page tables only on node 0 could
 365          * cause a hotspot and fill up ZONE_DMA. The page tables
 366          * need roughly 0.5KB per GB.
 367          */
 368         start = 0x8000;
 369         table_start = find_e820_area(start, end, tables, PAGE_SIZE);
 370         if (table_start == -1UL)
 371                 panic("Cannot find space for the kernel page tables");
 372
 373         table_start >>= PAGE_SHIFT;
 374         table_end = table_start;
 375
 376         early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n",
 377                 end, table_start << PAGE_SHIFT,
 378                 (table_start << PAGE_SHIFT) + tables);
 379 }
 380
 381 static void __init init_gbpages(void)
 382 {
 383         if (direct_gbpages && cpu_has_gbpages)
 384                 printk(KERN_INFO "Using GB pages for direct mapping\n");
 385         else
 386                 direct_gbpages = 0;
 387 }
 388
 389 #ifdef CONFIG_MEMTEST
 390
 391 static void __init memtest(unsigned long start_phys, unsigned long size,
 392                                  unsigned pattern)
 393 {
 394         unsigned long i;
 395         unsigned long *start;
 396         unsigned long start_bad;
 397         unsigned long last_bad;
 398         unsigned long val;
 399         unsigned long start_phys_aligned;
 400         unsigned long count;
 401         unsigned long incr;
 402
 403         switch (pattern) {
 404         case 0:
 405                 val = 0UL;
 406                 break;
 407         case 1:
 408                 val = -1UL;
 409                 break;
 410         case 2:
 411                 val = 0x5555555555555555UL;
 412                 break;
 413         case 3:
 414                 val = 0xaaaaaaaaaaaaaaaaUL;
 415                 break;
 416         default:
 417                 return;
 418         }
 419
 420         incr = sizeof(unsigned long);
 421         start_phys_aligned = ALIGN(start_phys, incr);
 422         count = (size - (start_phys_aligned - start_phys))/incr;
 423         start = __va(start_phys_aligned);
 424         start_bad = 0;
 425         last_bad = 0;
 426
 427         for (i = 0; i < count; i++)
 428                 start[i] = val;
 429         for (i = 0; i < count; i++, start++, start_phys_aligned += incr) {
 430                 if (*start != val) {
 431                         if (start_phys_aligned == last_bad + incr) {
 432                                 last_bad += incr;
 433                         } else {
 434                                 if (start_bad) {
 435                                         printk(KERN_CONT "\n  %016lx bad mem addr %016lx - %016lx reserved",
 436                                                 val, start_bad, last_bad + incr);
 437                                         reserve_early(start_bad, last_bad - start_bad, "BAD RAM");
 438                                 }
 439                                 start_bad = last_bad = start_phys_aligned;
 440                         }
 441                 }
 442         }
 443         if (start_bad) {
 444                 printk(KERN_CONT "\n  %016lx bad mem addr %016lx - %016lx reserved",
 445                         val, start_bad, last_bad + incr);
 446                 reserve_early(start_bad, last_bad - start_bad, "BAD RAM");
 447         }
 448
 449 }
 450
 451 /* default is disabled */
 452 static int memtest_pattern __initdata;
 453
 454 static int __init parse_memtest(char *arg)
 455 {
 456         if (arg)
 457                 memtest_pattern = simple_strtoul(arg, NULL, 0);
 458         return 0;
 459 }
 460
 461 early_param("memtest", parse_memtest);
 462
 463 static void __init early_memtest(unsigned long start, unsigned long end)
 464 {
 465         u64 t_start, t_size;
 466         unsigned pattern;
 467
 468         if (!memtest_pattern)
 469                 return;
 470
 471         printk(KERN_INFO "early_memtest: pattern num %d", memtest_pattern);
 472         for (pattern = 0; pattern < memtest_pattern; pattern++) {
 473                 t_start = start;
 474                 t_size = 0;
 475                 while (t_start < end) {
 476                         t_start = find_e820_area_size(t_start, &t_size, 1);
 477
 478                         /* done ? */
 479                         if (t_start >= end)
 480                                 break;
 481                         if (t_start + t_size > end)
 482                                 t_size = end - t_start;
 483
 484                         printk(KERN_CONT "\n  %016llx - %016llx pattern %d",
 485                                 (unsigned long long)t_start,
 486                                 (unsigned long long)t_start + t_size, pattern);
 487
 488                         memtest(t_start, t_size, pattern);
 489
 490                         t_start += t_size;
 491                 }
 492         }
 493         printk(KERN_CONT "\n");
 494 }
 495 #else
 496 static void __init early_memtest(unsigned long start, unsigned long end)
 497 {
 498 }
 499 #endif
 500
 501 /*
 502  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 503  * This runs before bootmem is initialized and gets pages directly from
 504  * the physical memory. To access them they are temporarily mapped.
 505  */
 506 unsigned long __init_refok init_memory_mapping(unsigned long start, unsigned long end)
 507 {
 508         unsigned long next, last_map_addr = end;
 509         unsigned long start_phys = start, end_phys = end;
 510
 511         printk(KERN_INFO "init_memory_mapping\n");
 512
 513         /*
 514          * Find space for the kernel direct mapping tables.
 515          *
 516          * Later we should allocate these tables in the local node of the
 517          * memory mapped. Unfortunately this is done currently before the
 518          * nodes are discovered.
 519          */
 520         if (!after_bootmem) {
 521                 init_gbpages();
 522                 find_early_table_space(end);
 523         }
 524
 525         start = (unsigned long)__va(start);
 526         end = (unsigned long)__va(end);
 527
 528         for (; start < end; start = next) {
 529                 pgd_t *pgd = pgd_offset_k(start);
 530                 unsigned long pud_phys;
 531                 pud_t *pud;
 532
 533                 if (after_bootmem)
 534                         pud = pud_offset(pgd, start & PGDIR_MASK);
 535                 else
 536                         pud = alloc_low_page(&pud_phys);
 537
 538                 next = start + PGDIR_SIZE;
 539                 if (next > end)
 540                         next = end;
 541                 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next));
 542                 if (!after_bootmem)
 543                         pgd_populate(&init_mm, pgd_offset_k(start),
 544                                      __va(pud_phys));
 545                 unmap_low_page(pud);
 546         }
 547
 548         if (!after_bootmem)
 549                 mmu_cr4_features = read_cr4();
 550         __flush_tlb_all();
 551
 552         if (!after_bootmem)
 553                 reserve_early(table_start << PAGE_SHIFT,
 554                                  table_end << PAGE_SHIFT, "PGTABLE");
 555
 556         if (!after_bootmem)
 557                 early_memtest(start_phys, end_phys);
 558
 559         return last_map_addr;
 560 }
 561
 562 #ifndef CONFIG_NUMA
 563 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn)
 564 {
 565         unsigned long bootmap_size, bootmap;
 566
 567         bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
 568         bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
 569                                  PAGE_SIZE);
 570         if (bootmap == -1L)
 571                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 572         /* don't touch min_low_pfn */
 573         bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
 574                                          0, end_pfn);
 575         e820_register_active_regions(0, start_pfn, end_pfn);
 576         free_bootmem_with_active_regions(0, end_pfn);
 577         early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
 578         reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
 579 }
 580
 581 void __init paging_init(void)
 582 {
 583         unsigned long max_zone_pfns[MAX_NR_ZONES];
 584
 585         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 586         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 587         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 588         max_zone_pfns[ZONE_NORMAL] = end_pfn;
 589
 590         memory_present(0, 0, end_pfn);
 591         sparse_init();
 592         free_area_init_nodes(max_zone_pfns);
 593 }
 594 #endif
 595
 596 /*
 597  * Memory hotplug specific functions
 598  */
 599 #ifdef CONFIG_MEMORY_HOTPLUG
 600 /*
 601  * Memory is added always to NORMAL zone. This means you will never get
 602  * additional DMA/DMA32 memory.
 603  */
 604 int arch_add_memory(int nid, u64 start, u64 size)
 605 {
 606         struct pglist_data *pgdat = NODE_DATA(nid);
 607         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 608         unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
 609         unsigned long nr_pages = size >> PAGE_SHIFT;
 610         int ret;
 611
 612         last_mapped_pfn = init_memory_mapping(start, start + size-1);
 613         if (last_mapped_pfn > max_pfn_mapped)
 614                 max_pfn_mapped = last_mapped_pfn;
 615
 616         ret = __add_pages(zone, start_pfn, nr_pages);
 617         WARN_ON(1);
 618
 619         return ret;
 620 }
 621 EXPORT_SYMBOL_GPL(arch_add_memory);
 622
 623 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
 624 int memory_add_physaddr_to_nid(u64 start)
 625 {
 626         return 0;
 627 }
 628 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 629 #endif
 630
 631 #endif /* CONFIG_MEMORY_HOTPLUG */
 632
 633 /*
 634  * devmem_is_allowed() checks to see if /dev/mem access to a certain address
 635  * is valid. The argument is a physical page number.
 636  *
 637  *
 638  * On x86, access has to be given to the first megabyte of ram because that area
 639  * contains bios code and data regions used by X and dosemu and similar apps.
 640  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
 641  * mmio resources as well as potential bios/acpi data regions.
 642  */
 643 int devmem_is_allowed(unsigned long pagenr)
 644 {
 645         if (pagenr <= 256)
 646                 return 1;
 647         if (!page_is_ram(pagenr))
 648                 return 1;
 649         return 0;
 650 }
 651
 652
 653 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
 654                          kcore_modules, kcore_vsyscall;
 655
 656 void __init mem_init(void)
 657 {
 658         long codesize, reservedpages, datasize, initsize;
 659
 660         pci_iommu_alloc();
 661
 662         /* clear_bss() already clear the empty_zero_page */
 663
 664         reservedpages = 0;
 665
 666         /* this will put all low memory onto the freelists */
 667 #ifdef CONFIG_NUMA
 668         totalram_pages = numa_free_all_bootmem();
 669 #else
 670         totalram_pages = free_all_bootmem();
 671 #endif
 672         reservedpages = end_pfn - totalram_pages -
 673                                         absent_pages_in_range(0, end_pfn);
 674         after_bootmem = 1;
 675
 676         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 677         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 678         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 679
 680         /* Register memory areas for /proc/kcore */
 681         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 682         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 683                    VMALLOC_END-VMALLOC_START);
 684         kclist_add(&kcore_kernel, &_stext, _end - _stext);
 685         kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
 686         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 687                                  VSYSCALL_END - VSYSCALL_START);
 688
 689         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 690                                 "%ldk reserved, %ldk data, %ldk init)\n",
 691                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 692                 end_pfn << (PAGE_SHIFT-10),
 693                 codesize >> 10,
 694                 reservedpages << (PAGE_SHIFT-10),
 695                 datasize >> 10,
 696                 initsize >> 10);
 697
 698         cpa_init();
 699 }
 700
 701 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 702 {
 703         unsigned long addr = begin;
 704
 705         if (addr >= end)
 706                 return;
 707
 708         /*
 709          * If debugging page accesses then do not free this memory but
 710          * mark them not present - any buggy init-section access will
 711          * create a kernel page fault:
 712          */
 713 #ifdef CONFIG_DEBUG_PAGEALLOC
 714         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
 715                 begin, PAGE_ALIGN(end));
 716         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 717 #else
 718         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 719
 720         for (; addr < end; addr += PAGE_SIZE) {
 721                 ClearPageReserved(virt_to_page(addr));
 722                 init_page_count(virt_to_page(addr));
 723                 memset((void *)(addr & ~(PAGE_SIZE-1)),
 724                         POISON_FREE_INITMEM, PAGE_SIZE);
 725                 free_page(addr);
 726                 totalram_pages++;
 727         }
 728 #endif
 729 }
 730
 731 void free_initmem(void)
 732 {
 733         free_init_pages("unused kernel memory",
 734                         (unsigned long)(&__init_begin),
 735                         (unsigned long)(&__init_end));
 736 }
 737
 738 #ifdef CONFIG_DEBUG_RODATA
 739 const int rodata_test_data = 0xC3;
 740 EXPORT_SYMBOL_GPL(rodata_test_data);
 741
 742 void mark_rodata_ro(void)
 743 {
 744         unsigned long start = PFN_ALIGN(_stext), end = PFN_ALIGN(__end_rodata);
 745
 746         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 747                (end - start) >> 10);
 748         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 749
 750         /*
 751          * The rodata section (but not the kernel text!) should also be
 752          * not-executable.
 753          */
 754         start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
 755         set_memory_nx(start, (end - start) >> PAGE_SHIFT);
 756
 757         rodata_test();
 758
 759 #ifdef CONFIG_CPA_DEBUG
 760         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
 761         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 762
 763         printk(KERN_INFO "Testing CPA: again\n");
 764         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 765 #endif
 766 }
 767
 768 #endif
 769
 770 #ifdef CONFIG_BLK_DEV_INITRD
 771 void free_initrd_mem(unsigned long start, unsigned long end)
 772 {
 773         free_init_pages("initrd memory", start, end);
 774 }
 775 #endif
 776
 777 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
 778                                    int flags)
 779 {
 780 #ifdef CONFIG_NUMA
 781         int nid, next_nid;
 782         int ret;
 783 #endif
 784         unsigned long pfn = phys >> PAGE_SHIFT;
 785
 786         if (pfn >= end_pfn) {
 787                 /*
 788                  * This can happen with kdump kernels when accessing
 789                  * firmware tables:
 790                  */
 791                 if (pfn < max_pfn_mapped)
 792                         return -EFAULT;
 793
 794                 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n",
 795                                 phys, len);
 796                 return -EFAULT;
 797         }
 798
 799         /* Should check here against the e820 map to avoid double free */
 800 #ifdef CONFIG_NUMA
 801         nid = phys_to_nid(phys);
 802         next_nid = phys_to_nid(phys + len - 1);
 803         if (nid == next_nid)
 804                 ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags);
 805         else
 806                 ret = reserve_bootmem(phys, len, flags);
 807
 808         if (ret != 0)
 809                 return ret;
 810
 811 #else
 812         reserve_bootmem(phys, len, BOOTMEM_DEFAULT);
 813 #endif
 814
 815         if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
 816                 dma_reserve += len / PAGE_SIZE;
 817                 set_dma_reserve(dma_reserve);
 818         }
 819
 820         return 0;
 821 }
 822
 823 int kern_addr_valid(unsigned long addr)
 824 {
 825         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 826         pgd_t *pgd;
 827         pud_t *pud;
 828         pmd_t *pmd;
 829         pte_t *pte;
 830
 831         if (above != 0 && above != -1UL)
 832                 return 0;
 833
 834         pgd = pgd_offset_k(addr);
 835         if (pgd_none(*pgd))
 836                 return 0;
 837
 838         pud = pud_offset(pgd, addr);
 839         if (pud_none(*pud))
 840                 return 0;
 841
 842         pmd = pmd_offset(pud, addr);
 843         if (pmd_none(*pmd))
 844                 return 0;
 845
 846         if (pmd_large(*pmd))
 847                 return pfn_valid(pmd_pfn(*pmd));
 848
 849         pte = pte_offset_kernel(pmd, addr);
 850         if (pte_none(*pte))
 851                 return 0;
 852
 853         return pfn_valid(pte_pfn(*pte));
 854 }
 855
 856 /*
 857  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 858  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 859  * not need special handling anymore:
 860  */
 861 static struct vm_area_struct gate_vma = {
 862         .vm_start       = VSYSCALL_START,
 863         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 864         .vm_page_prot   = PAGE_READONLY_EXEC,
 865         .vm_flags       = VM_READ | VM_EXEC
 866 };
 867
 868 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 869 {
 870 #ifdef CONFIG_IA32_EMULATION
 871         if (test_tsk_thread_flag(tsk, TIF_IA32))
 872                 return NULL;
 873 #endif
 874         return &gate_vma;
 875 }
 876
 877 int in_gate_area(struct task_struct *task, unsigned long addr)
 878 {
 879         struct vm_area_struct *vma = get_gate_vma(task);
 880
 881         if (!vma)
 882                 return 0;
 883
 884         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 885 }
 886
 887 /*
 888  * Use this when you have no reliable task/vma, typically from interrupt
 889  * context. It is less reliable than using the task's vma and may give
 890  * false positives:
 891  */
 892 int in_gate_area_no_task(unsigned long addr)
 893 {
 894         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 895 }
 896
 897 const char *arch_vma_name(struct vm_area_struct *vma)
 898 {
 899         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 900                 return "[vdso]";
 901         if (vma == &gate_vma)
 902                 return "[vsyscall]";
 903         return NULL;
 904 }
 905
 906 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 907 /*
 908  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 909  */
 910 static long __meminitdata addr_start, addr_end;
 911 static void __meminitdata *p_start, *p_end;
 912 static int __meminitdata node_start;
 913
 914 int __meminit
 915 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 916 {
 917         unsigned long addr = (unsigned long)start_page;
 918         unsigned long end = (unsigned long)(start_page + size);
 919         unsigned long next;
 920         pgd_t *pgd;
 921         pud_t *pud;
 922         pmd_t *pmd;
 923
 924         for (; addr < end; addr = next) {
 925                 next = pmd_addr_end(addr, end);
 926
 927                 pgd = vmemmap_pgd_populate(addr, node);
 928                 if (!pgd)
 929                         return -ENOMEM;
 930
 931                 pud = vmemmap_pud_populate(pgd, addr, node);
 932                 if (!pud)
 933                         return -ENOMEM;
 934
 935                 pmd = pmd_offset(pud, addr);
 936                 if (pmd_none(*pmd)) {
 937                         pte_t entry;
 938                         void *p;
 939
 940                         p = vmemmap_alloc_block(PMD_SIZE, node);
 941                         if (!p)
 942                                 return -ENOMEM;
 943
 944                         entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
 945                                                         PAGE_KERNEL_LARGE);
 946                         set_pmd(pmd, __pmd(pte_val(entry)));
 947
 948                         /* check to see if we have contiguous blocks */
 949                         if (p_end != p || node_start != node) {
 950                                 if (p_start)
 951                                         printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
 952                                                 addr_start, addr_end-1, p_start, p_end-1, node_start);
 953                                 addr_start = addr;
 954                                 node_start = node;
 955                                 p_start = p;
 956                         }
 957                         addr_end = addr + PMD_SIZE;
 958                         p_end = p + PMD_SIZE;
 959                 } else {
 960                         vmemmap_verify((pte_t *)pmd, node, addr, next);
 961                 }
 962         }
 963         return 0;
 964 }
 965
 966 void __meminit vmemmap_populate_print_last(void)
 967 {
 968         if (p_start) {
 969                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
 970                         addr_start, addr_end-1, p_start, p_end-1, node_start);
 971                 p_start = NULL;
 972                 p_end = NULL;
 973                 node_start = 0;
 974         }
 975 }
 976 #endif