err.no Git - linux-2.6/blob - arch/x86_64/mm/init.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/pagemap.h>
  22 #include <linux/bootmem.h>
  23 #include <linux/proc_fs.h>
  24 #include <linux/pci.h>
  25 #include <linux/pfn.h>
  26 #include <linux/poison.h>
  27 #include <linux/dma-mapping.h>
  28 #include <linux/module.h>
  29 #include <linux/memory_hotplug.h>
  30 #include <linux/nmi.h>
  31
  32 #include <asm/processor.h>
  33 #include <asm/system.h>
  34 #include <asm/uaccess.h>
  35 #include <asm/pgtable.h>
  36 #include <asm/pgalloc.h>
  37 #include <asm/dma.h>
  38 #include <asm/fixmap.h>
  39 #include <asm/e820.h>
  40 #include <asm/apic.h>
  41 #include <asm/tlb.h>
  42 #include <asm/mmu_context.h>
  43 #include <asm/proto.h>
  44 #include <asm/smp.h>
  45 #include <asm/sections.h>
  46
  47 #ifndef Dprintk
  48 #define Dprintk(x...)
  49 #endif
  50
  51 const struct dma_mapping_ops* dma_ops;
  52 EXPORT_SYMBOL(dma_ops);
  53
  54 static unsigned long dma_reserve __initdata;
  55
  56 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  57
  58 /*
  59  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
  60  * physical space so we can cache the place of the first one and move
  61  * around without checking the pgd every time.
  62  */
  63
  64 void show_mem(void)
  65 {
  66         long i, total = 0, reserved = 0;
  67         long shared = 0, cached = 0;
  68         pg_data_t *pgdat;
  69         struct page *page;
  70
  71         printk(KERN_INFO "Mem-info:\n");
  72         show_free_areas();
  73         printk(KERN_INFO "Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
  74
  75         for_each_online_pgdat(pgdat) {
  76                for (i = 0; i < pgdat->node_spanned_pages; ++i) {
  77                         /* this loop can take a while with 256 GB and 4k pages
  78                            so update the NMI watchdog */
  79                         if (unlikely(i % MAX_ORDER_NR_PAGES == 0)) {
  80                                 touch_nmi_watchdog();
  81                         }
  82                         if (!pfn_valid(pgdat->node_start_pfn + i))
  83                                 continue;
  84                         page = pfn_to_page(pgdat->node_start_pfn + i);
  85                         total++;
  86                         if (PageReserved(page))
  87                                 reserved++;
  88                         else if (PageSwapCache(page))
  89                                 cached++;
  90                         else if (page_count(page))
  91                                 shared += page_count(page) - 1;
  92                }
  93         }
  94         printk(KERN_INFO "%lu pages of RAM\n", total);
  95         printk(KERN_INFO "%lu reserved pages\n",reserved);
  96         printk(KERN_INFO "%lu pages shared\n",shared);
  97         printk(KERN_INFO "%lu pages swap cached\n",cached);
  98 }
  99
 100 int after_bootmem;
 101
 102 static __init void *spp_getpage(void)
 103 {
 104         void *ptr;
 105         if (after_bootmem)
 106                 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
 107         else
 108                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 109         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK))
 110                 panic("set_pte_phys: cannot allocate page data %s\n", after_bootmem?"after bootmem":"");
 111
 112         Dprintk("spp_getpage %p\n", ptr);
 113         return ptr;
 114 }
 115
 116 static __init void set_pte_phys(unsigned long vaddr,
 117                          unsigned long phys, pgprot_t prot)
 118 {
 119         pgd_t *pgd;
 120         pud_t *pud;
 121         pmd_t *pmd;
 122         pte_t *pte, new_pte;
 123
 124         Dprintk("set_pte_phys %lx to %lx\n", vaddr, phys);
 125
 126         pgd = pgd_offset_k(vaddr);
 127         if (pgd_none(*pgd)) {
 128                 printk("PGD FIXMAP MISSING, it should be setup in head.S!\n");
 129                 return;
 130         }
 131         pud = pud_offset(pgd, vaddr);
 132         if (pud_none(*pud)) {
 133                 pmd = (pmd_t *) spp_getpage();
 134                 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER));
 135                 if (pmd != pmd_offset(pud, 0)) {
 136                         printk("PAGETABLE BUG #01! %p <-> %p\n", pmd, pmd_offset(pud,0));
 137                         return;
 138                 }
 139         }
 140         pmd = pmd_offset(pud, vaddr);
 141         if (pmd_none(*pmd)) {
 142                 pte = (pte_t *) spp_getpage();
 143                 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER));
 144                 if (pte != pte_offset_kernel(pmd, 0)) {
 145                         printk("PAGETABLE BUG #02!\n");
 146                         return;
 147                 }
 148         }
 149         new_pte = pfn_pte(phys >> PAGE_SHIFT, prot);
 150
 151         pte = pte_offset_kernel(pmd, vaddr);
 152         if (!pte_none(*pte) &&
 153             pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask))
 154                 pte_ERROR(*pte);
 155         set_pte(pte, new_pte);
 156
 157         /*
 158          * It's enough to flush this one mapping.
 159          * (PGE mappings get flushed as well)
 160          */
 161         __flush_tlb_one(vaddr);
 162 }
 163
 164 /* NOTE: this is meant to be run only at boot */
 165 void __init
 166 __set_fixmap (enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
 167 {
 168         unsigned long address = __fix_to_virt(idx);
 169
 170         if (idx >= __end_of_fixed_addresses) {
 171                 printk("Invalid __set_fixmap\n");
 172                 return;
 173         }
 174         set_pte_phys(address, phys, prot);
 175 }
 176
 177 unsigned long __meminitdata table_start, table_end;
 178
 179 static __meminit void *alloc_low_page(unsigned long *phys)
 180 {
 181         unsigned long pfn = table_end++;
 182         void *adr;
 183
 184         if (after_bootmem) {
 185                 adr = (void *)get_zeroed_page(GFP_ATOMIC);
 186                 *phys = __pa(adr);
 187                 return adr;
 188         }
 189
 190         if (pfn >= end_pfn)
 191                 panic("alloc_low_page: ran out of memory");
 192
 193         adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
 194         memset(adr, 0, PAGE_SIZE);
 195         *phys  = pfn * PAGE_SIZE;
 196         return adr;
 197 }
 198
 199 static __meminit void unmap_low_page(void *adr)
 200 {
 201
 202         if (after_bootmem)
 203                 return;
 204
 205         early_iounmap(adr, PAGE_SIZE);
 206 }
 207
 208 /* Must run before zap_low_mappings */
 209 __meminit void *early_ioremap(unsigned long addr, unsigned long size)
 210 {
 211         unsigned long vaddr;
 212         pmd_t *pmd, *last_pmd;
 213         int i, pmds;
 214
 215         pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 216         vaddr = __START_KERNEL_map;
 217         pmd = level2_kernel_pgt;
 218         last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1;
 219         for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) {
 220                 for (i = 0; i < pmds; i++) {
 221                         if (pmd_present(pmd[i]))
 222                                 goto next;
 223                 }
 224                 vaddr += addr & ~PMD_MASK;
 225                 addr &= PMD_MASK;
 226                 for (i = 0; i < pmds; i++, addr += PMD_SIZE)
 227                         set_pmd(pmd + i,__pmd(addr | _KERNPG_TABLE | _PAGE_PSE));
 228                 __flush_tlb();
 229                 return (void *)vaddr;
 230         next:
 231                 ;
 232         }
 233         printk("early_ioremap(0x%lx, %lu) failed\n", addr, size);
 234         return NULL;
 235 }
 236
 237 /* To avoid virtual aliases later */
 238 __meminit void early_iounmap(void *addr, unsigned long size)
 239 {
 240         unsigned long vaddr;
 241         pmd_t *pmd;
 242         int i, pmds;
 243
 244         vaddr = (unsigned long)addr;
 245         pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 246         pmd = level2_kernel_pgt + pmd_index(vaddr);
 247         for (i = 0; i < pmds; i++)
 248                 pmd_clear(pmd + i);
 249         __flush_tlb();
 250 }
 251
 252 static void __meminit
 253 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
 254 {
 255         int i = pmd_index(address);
 256
 257         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
 258                 unsigned long entry;
 259                 pmd_t *pmd = pmd_page + pmd_index(address);
 260
 261                 if (address >= end) {
 262                         if (!after_bootmem)
 263                                 for (; i < PTRS_PER_PMD; i++, pmd++)
 264                                         set_pmd(pmd, __pmd(0));
 265                         break;
 266                 }
 267
 268                 if (pmd_val(*pmd))
 269                         continue;
 270
 271                 entry = _PAGE_NX|_PAGE_PSE|_KERNPG_TABLE|_PAGE_GLOBAL|address;
 272                 entry &= __supported_pte_mask;
 273                 set_pmd(pmd, __pmd(entry));
 274         }
 275 }
 276
 277 static void __meminit
 278 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
 279 {
 280         pmd_t *pmd = pmd_offset(pud,0);
 281         spin_lock(&init_mm.page_table_lock);
 282         phys_pmd_init(pmd, address, end);
 283         spin_unlock(&init_mm.page_table_lock);
 284         __flush_tlb_all();
 285 }
 286
 287 static void __meminit phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
 288 {
 289         int i = pud_index(addr);
 290
 291
 292         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE ) {
 293                 unsigned long pmd_phys;
 294                 pud_t *pud = pud_page + pud_index(addr);
 295                 pmd_t *pmd;
 296
 297                 if (addr >= end)
 298                         break;
 299
 300                 if (!after_bootmem && !e820_any_mapped(addr,addr+PUD_SIZE,0)) {
 301                         set_pud(pud, __pud(0));
 302                         continue;
 303                 }
 304
 305                 if (pud_val(*pud)) {
 306                         phys_pmd_update(pud, addr, end);
 307                         continue;
 308                 }
 309
 310                 pmd = alloc_low_page(&pmd_phys);
 311                 spin_lock(&init_mm.page_table_lock);
 312                 set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE));
 313                 phys_pmd_init(pmd, addr, end);
 314                 spin_unlock(&init_mm.page_table_lock);
 315                 unmap_low_page(pmd);
 316         }
 317         __flush_tlb();
 318 }
 319
 320 static void __init find_early_table_space(unsigned long end)
 321 {
 322         unsigned long puds, pmds, tables, start;
 323
 324         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 325         pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 326         tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) +
 327                  round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
 328
 329         /* RED-PEN putting page tables only on node 0 could
 330            cause a hotspot and fill up ZONE_DMA. The page tables
 331            need roughly 0.5KB per GB. */
 332         start = 0x8000;
 333         table_start = find_e820_area(start, end, tables);
 334         if (table_start == -1UL)
 335                 panic("Cannot find space for the kernel page tables");
 336
 337         table_start >>= PAGE_SHIFT;
 338         table_end = table_start;
 339
 340         early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n",
 341                 end, table_start << PAGE_SHIFT,
 342                 (table_start << PAGE_SHIFT) + tables);
 343 }
 344
 345 /* Setup the direct mapping of the physical memory at PAGE_OFFSET.
 346    This runs before bootmem is initialized and gets pages directly from the
 347    physical memory. To access them they are temporarily mapped. */
 348 void __meminit init_memory_mapping(unsigned long start, unsigned long end)
 349 {
 350         unsigned long next;
 351
 352         Dprintk("init_memory_mapping\n");
 353
 354         /*
 355          * Find space for the kernel direct mapping tables.
 356          * Later we should allocate these tables in the local node of the memory
 357          * mapped.  Unfortunately this is done currently before the nodes are
 358          * discovered.
 359          */
 360         if (!after_bootmem)
 361                 find_early_table_space(end);
 362
 363         start = (unsigned long)__va(start);
 364         end = (unsigned long)__va(end);
 365
 366         for (; start < end; start = next) {
 367                 unsigned long pud_phys;
 368                 pgd_t *pgd = pgd_offset_k(start);
 369                 pud_t *pud;
 370
 371                 if (after_bootmem)
 372                         pud = pud_offset(pgd, start & PGDIR_MASK);
 373                 else
 374                         pud = alloc_low_page(&pud_phys);
 375
 376                 next = start + PGDIR_SIZE;
 377                 if (next > end)
 378                         next = end;
 379                 phys_pud_init(pud, __pa(start), __pa(next));
 380                 if (!after_bootmem)
 381                         set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys));
 382                 unmap_low_page(pud);
 383         }
 384
 385         if (!after_bootmem)
 386                 mmu_cr4_features = read_cr4();
 387         __flush_tlb_all();
 388 }
 389
 390 #ifndef CONFIG_NUMA
 391 void __init paging_init(void)
 392 {
 393         unsigned long max_zone_pfns[MAX_NR_ZONES];
 394         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 395         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 396         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 397         max_zone_pfns[ZONE_NORMAL] = end_pfn;
 398
 399         memory_present(0, 0, end_pfn);
 400         sparse_init();
 401         free_area_init_nodes(max_zone_pfns);
 402 }
 403 #endif
 404
 405 /* Unmap a kernel mapping if it exists. This is useful to avoid prefetches
 406    from the CPU leading to inconsistent cache lines. address and size
 407    must be aligned to 2MB boundaries.
 408    Does nothing when the mapping doesn't exist. */
 409 void __init clear_kernel_mapping(unsigned long address, unsigned long size)
 410 {
 411         unsigned long end = address + size;
 412
 413         BUG_ON(address & ~LARGE_PAGE_MASK);
 414         BUG_ON(size & ~LARGE_PAGE_MASK);
 415
 416         for (; address < end; address += LARGE_PAGE_SIZE) {
 417                 pgd_t *pgd = pgd_offset_k(address);
 418                 pud_t *pud;
 419                 pmd_t *pmd;
 420                 if (pgd_none(*pgd))
 421                         continue;
 422                 pud = pud_offset(pgd, address);
 423                 if (pud_none(*pud))
 424                         continue;
 425                 pmd = pmd_offset(pud, address);
 426                 if (!pmd || pmd_none(*pmd))
 427                         continue;
 428                 if (0 == (pmd_val(*pmd) & _PAGE_PSE)) {
 429                         /* Could handle this, but it should not happen currently. */
 430                         printk(KERN_ERR
 431                "clear_kernel_mapping: mapping has been split. will leak memory\n");
 432                         pmd_ERROR(*pmd);
 433                 }
 434                 set_pmd(pmd, __pmd(0));
 435         }
 436         __flush_tlb_all();
 437 }
 438
 439 /*
 440  * Memory hotplug specific functions
 441  */
 442 void online_page(struct page *page)
 443 {
 444         ClearPageReserved(page);
 445         init_page_count(page);
 446         __free_page(page);
 447         totalram_pages++;
 448         num_physpages++;
 449 }
 450
 451 #ifdef CONFIG_MEMORY_HOTPLUG
 452 /*
 453  * Memory is added always to NORMAL zone. This means you will never get
 454  * additional DMA/DMA32 memory.
 455  */
 456 int arch_add_memory(int nid, u64 start, u64 size)
 457 {
 458         struct pglist_data *pgdat = NODE_DATA(nid);
 459         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 460         unsigned long start_pfn = start >> PAGE_SHIFT;
 461         unsigned long nr_pages = size >> PAGE_SHIFT;
 462         int ret;
 463
 464         init_memory_mapping(start, (start + size -1));
 465
 466         ret = __add_pages(zone, start_pfn, nr_pages);
 467         if (ret)
 468                 goto error;
 469
 470         return ret;
 471 error:
 472         printk("%s: Problem encountered in __add_pages!\n", __func__);
 473         return ret;
 474 }
 475 EXPORT_SYMBOL_GPL(arch_add_memory);
 476
 477 int remove_memory(u64 start, u64 size)
 478 {
 479         return -EINVAL;
 480 }
 481 EXPORT_SYMBOL_GPL(remove_memory);
 482
 483 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
 484 int memory_add_physaddr_to_nid(u64 start)
 485 {
 486         return 0;
 487 }
 488 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 489 #endif
 490
 491 #endif /* CONFIG_MEMORY_HOTPLUG */
 492
 493 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
 494 /*
 495  * Memory Hotadd without sparsemem. The mem_maps have been allocated in advance,
 496  * just online the pages.
 497  */
 498 int __add_pages(struct zone *z, unsigned long start_pfn, unsigned long nr_pages)
 499 {
 500         int err = -EIO;
 501         unsigned long pfn;
 502         unsigned long total = 0, mem = 0;
 503         for (pfn = start_pfn; pfn < start_pfn + nr_pages; pfn++) {
 504                 if (pfn_valid(pfn)) {
 505                         online_page(pfn_to_page(pfn));
 506                         err = 0;
 507                         mem++;
 508                 }
 509                 total++;
 510         }
 511         if (!err) {
 512                 z->spanned_pages += total;
 513                 z->present_pages += mem;
 514                 z->zone_pgdat->node_spanned_pages += total;
 515                 z->zone_pgdat->node_present_pages += mem;
 516         }
 517         return err;
 518 }
 519 #endif
 520
 521 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel, kcore_modules,
 522                          kcore_vsyscall;
 523
 524 void __init mem_init(void)
 525 {
 526         long codesize, reservedpages, datasize, initsize;
 527
 528         pci_iommu_alloc();
 529
 530         /* clear the zero-page */
 531         memset(empty_zero_page, 0, PAGE_SIZE);
 532
 533         reservedpages = 0;
 534
 535         /* this will put all low memory onto the freelists */
 536 #ifdef CONFIG_NUMA
 537         totalram_pages = numa_free_all_bootmem();
 538 #else
 539         totalram_pages = free_all_bootmem();
 540 #endif
 541         reservedpages = end_pfn - totalram_pages -
 542                                         absent_pages_in_range(0, end_pfn);
 543
 544         after_bootmem = 1;
 545
 546         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 547         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 548         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 549
 550         /* Register memory areas for /proc/kcore */
 551         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 552         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 553                    VMALLOC_END-VMALLOC_START);
 554         kclist_add(&kcore_kernel, &_stext, _end - _stext);
 555         kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
 556         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 557                                  VSYSCALL_END - VSYSCALL_START);
 558
 559         printk("Memory: %luk/%luk available (%ldk kernel code, %ldk reserved, %ldk data, %ldk init)\n",
 560                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 561                 end_pfn << (PAGE_SHIFT-10),
 562                 codesize >> 10,
 563                 reservedpages << (PAGE_SHIFT-10),
 564                 datasize >> 10,
 565                 initsize >> 10);
 566 }
 567
 568 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 569 {
 570         unsigned long addr;
 571
 572         if (begin >= end)
 573                 return;
 574
 575         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 576         for (addr = begin; addr < end; addr += PAGE_SIZE) {
 577                 ClearPageReserved(virt_to_page(addr));
 578                 init_page_count(virt_to_page(addr));
 579                 memset((void *)(addr & ~(PAGE_SIZE-1)),
 580                         POISON_FREE_INITMEM, PAGE_SIZE);
 581                 if (addr >= __START_KERNEL_map)
 582                         change_page_attr_addr(addr, 1, __pgprot(0));
 583                 free_page(addr);
 584                 totalram_pages++;
 585         }
 586         if (addr > __START_KERNEL_map)
 587                 global_flush_tlb();
 588 }
 589
 590 void free_initmem(void)
 591 {
 592         free_init_pages("unused kernel memory",
 593                         (unsigned long)(&__init_begin),
 594                         (unsigned long)(&__init_end));
 595 }
 596
 597 #ifdef CONFIG_DEBUG_RODATA
 598
 599 void mark_rodata_ro(void)
 600 {
 601         unsigned long start = (unsigned long)_stext, end;
 602
 603         end = (unsigned long)__end_rodata;
 604         start = (start + PAGE_SIZE - 1) & PAGE_MASK;
 605         end &= PAGE_MASK;
 606         if (end <= start)
 607                 return;
 608
 609         change_page_attr_addr(start, (end - start) >> PAGE_SHIFT, PAGE_KERNEL_RO);
 610
 611         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 612                (end - start) >> 10);
 613
 614         /*
 615          * change_page_attr_addr() requires a global_flush_tlb() call after it.
 616          * We do this after the printk so that if something went wrong in the
 617          * change, the printk gets out at least to give a better debug hint
 618          * of who is the culprit.
 619          */
 620         global_flush_tlb();
 621 }
 622 #endif
 623
 624 #ifdef CONFIG_BLK_DEV_INITRD
 625 void free_initrd_mem(unsigned long start, unsigned long end)
 626 {
 627         free_init_pages("initrd memory", start, end);
 628 }
 629 #endif
 630
 631 void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
 632 {
 633 #ifdef CONFIG_NUMA
 634         int nid = phys_to_nid(phys);
 635 #endif
 636         unsigned long pfn = phys >> PAGE_SHIFT;
 637         if (pfn >= end_pfn) {
 638                 /* This can happen with kdump kernels when accessing firmware
 639                    tables. */
 640                 if (pfn < end_pfn_map)
 641                         return;
 642                 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n",
 643                                 phys, len);
 644                 return;
 645         }
 646
 647         /* Should check here against the e820 map to avoid double free */
 648 #ifdef CONFIG_NUMA
 649         reserve_bootmem_node(NODE_DATA(nid), phys, len);
 650 #else
 651         reserve_bootmem(phys, len);
 652 #endif
 653         if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
 654                 dma_reserve += len / PAGE_SIZE;
 655                 set_dma_reserve(dma_reserve);
 656         }
 657 }
 658
 659 int kern_addr_valid(unsigned long addr)
 660 {
 661         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 662        pgd_t *pgd;
 663        pud_t *pud;
 664        pmd_t *pmd;
 665        pte_t *pte;
 666
 667         if (above != 0 && above != -1UL)
 668                 return 0;
 669
 670         pgd = pgd_offset_k(addr);
 671         if (pgd_none(*pgd))
 672                 return 0;
 673
 674         pud = pud_offset(pgd, addr);
 675         if (pud_none(*pud))
 676                 return 0;
 677
 678         pmd = pmd_offset(pud, addr);
 679         if (pmd_none(*pmd))
 680                 return 0;
 681         if (pmd_large(*pmd))
 682                 return pfn_valid(pmd_pfn(*pmd));
 683
 684         pte = pte_offset_kernel(pmd, addr);
 685         if (pte_none(*pte))
 686                 return 0;
 687         return pfn_valid(pte_pfn(*pte));
 688 }
 689
 690 /* A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 691    covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 692    not need special handling anymore. */
 693
 694 static struct vm_area_struct gate_vma = {
 695         .vm_start = VSYSCALL_START,
 696         .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES << PAGE_SHIFT),
 697         .vm_page_prot = PAGE_READONLY_EXEC,
 698         .vm_flags = VM_READ | VM_EXEC
 699 };
 700
 701 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 702 {
 703 #ifdef CONFIG_IA32_EMULATION
 704         if (test_tsk_thread_flag(tsk, TIF_IA32))
 705                 return NULL;
 706 #endif
 707         return &gate_vma;
 708 }
 709
 710 int in_gate_area(struct task_struct *task, unsigned long addr)
 711 {
 712         struct vm_area_struct *vma = get_gate_vma(task);
 713         if (!vma)
 714                 return 0;
 715         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 716 }
 717
 718 /* Use this when you have no reliable task/vma, typically from interrupt
 719  * context.  It is less reliable than using the task's vma and may give
 720  * false positives.
 721  */
 722 int in_gate_area_no_task(unsigned long addr)
 723 {
 724         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 725 }
 726
 727 void * __init alloc_bootmem_high_node(pg_data_t *pgdat, unsigned long size)
 728 {
 729         return __alloc_bootmem_core(pgdat->bdata, size,
 730                         SMP_CACHE_BYTES, (4UL*1024*1024*1024), 0);
 731 }
 732
 733 const char *arch_vma_name(struct vm_area_struct *vma)
 734 {
 735         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 736                 return "[vdso]";
 737         if (vma == &gate_vma)
 738                 return "[vsyscall]";
 739         return NULL;
 740 }