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/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 #include <asm/kdebug.h>
  47 #include <asm/numa.h>
  48 #include <asm/cacheflush.h>
  49
  50 const struct dma_mapping_ops *dma_ops;
  51 EXPORT_SYMBOL(dma_ops);
  52
  53 static unsigned long dma_reserve __initdata;
  54
  55 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  56
  57 /*
  58  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
  59  * physical space so we can cache the place of the first one and move
  60  * around without checking the pgd every time.
  61  */
  62
  63 void show_mem(void)
  64 {
  65         long i, total = 0, reserved = 0;
  66         long shared = 0, cached = 0;
  67         struct page *page;
  68         pg_data_t *pgdat;
  69
  70         printk(KERN_INFO "Mem-info:\n");
  71         show_free_areas();
  72         printk(KERN_INFO "Free swap:       %6ldkB\n",
  73                 nr_swap_pages << (PAGE_SHIFT-10));
  74
  75         for_each_online_pgdat(pgdat) {
  76                 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
  77                         /*
  78                          * This loop can take a while with 256 GB and
  79                          * 4k pages so defer the NMI watchdog:
  80                          */
  81                         if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
  82                                 touch_nmi_watchdog();
  83
  84                         if (!pfn_valid(pgdat->node_start_pfn + i))
  85                                 continue;
  86
  87                         page = pfn_to_page(pgdat->node_start_pfn + i);
  88                         total++;
  89                         if (PageReserved(page))
  90                                 reserved++;
  91                         else if (PageSwapCache(page))
  92                                 cached++;
  93                         else if (page_count(page))
  94                                 shared += page_count(page) - 1;
  95                 }
  96         }
  97         printk(KERN_INFO "%lu pages of RAM\n",          total);
  98         printk(KERN_INFO "%lu reserved pages\n",        reserved);
  99         printk(KERN_INFO "%lu pages shared\n",          shared);
 100         printk(KERN_INFO "%lu pages swap cached\n",     cached);
 101 }
 102
 103 int after_bootmem;
 104
 105 static __init void *spp_getpage(void)
 106 {
 107         void *ptr;
 108
 109         if (after_bootmem)
 110                 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
 111         else
 112                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 113
 114         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 115                 panic("set_pte_phys: cannot allocate page data %s\n",
 116                         after_bootmem ? "after bootmem" : "");
 117         }
 118
 119         pr_debug("spp_getpage %p\n", ptr);
 120
 121         return ptr;
 122 }
 123
 124 static __init void
 125 set_pte_phys(unsigned long vaddr, unsigned long phys, pgprot_t prot)
 126 {
 127         pgd_t *pgd;
 128         pud_t *pud;
 129         pmd_t *pmd;
 130         pte_t *pte, new_pte;
 131
 132         pr_debug("set_pte_phys %lx to %lx\n", vaddr, phys);
 133
 134         pgd = pgd_offset_k(vaddr);
 135         if (pgd_none(*pgd)) {
 136                 printk(KERN_ERR
 137                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
 138                 return;
 139         }
 140         pud = pud_offset(pgd, vaddr);
 141         if (pud_none(*pud)) {
 142                 pmd = (pmd_t *) spp_getpage();
 143                 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER));
 144                 if (pmd != pmd_offset(pud, 0)) {
 145                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
 146                                 pmd, pmd_offset(pud, 0));
 147                         return;
 148                 }
 149         }
 150         pmd = pmd_offset(pud, vaddr);
 151         if (pmd_none(*pmd)) {
 152                 pte = (pte_t *) spp_getpage();
 153                 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER));
 154                 if (pte != pte_offset_kernel(pmd, 0)) {
 155                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
 156                         return;
 157                 }
 158         }
 159         new_pte = pfn_pte(phys >> PAGE_SHIFT, prot);
 160
 161         pte = pte_offset_kernel(pmd, vaddr);
 162         if (!pte_none(*pte) &&
 163             pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask))
 164                 pte_ERROR(*pte);
 165         set_pte(pte, new_pte);
 166
 167         /*
 168          * It's enough to flush this one mapping.
 169          * (PGE mappings get flushed as well)
 170          */
 171         __flush_tlb_one(vaddr);
 172 }
 173
 174 /* NOTE: this is meant to be run only at boot */
 175 void __init
 176 __set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
 177 {
 178         unsigned long address = __fix_to_virt(idx);
 179
 180         if (idx >= __end_of_fixed_addresses) {
 181                 printk(KERN_ERR "Invalid __set_fixmap\n");
 182                 return;
 183         }
 184         set_pte_phys(address, phys, prot);
 185 }
 186
 187 static unsigned long __initdata table_start;
 188 static unsigned long __meminitdata table_end;
 189
 190 static __meminit void *alloc_low_page(unsigned long *phys)
 191 {
 192         unsigned long pfn = table_end++;
 193         void *adr;
 194
 195         if (after_bootmem) {
 196                 adr = (void *)get_zeroed_page(GFP_ATOMIC);
 197                 *phys = __pa(adr);
 198
 199                 return adr;
 200         }
 201
 202         if (pfn >= end_pfn)
 203                 panic("alloc_low_page: ran out of memory");
 204
 205         adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
 206         memset(adr, 0, PAGE_SIZE);
 207         *phys  = pfn * PAGE_SIZE;
 208         return adr;
 209 }
 210
 211 static __meminit void unmap_low_page(void *adr)
 212 {
 213         if (after_bootmem)
 214                 return;
 215
 216         early_iounmap(adr, PAGE_SIZE);
 217 }
 218
 219 /* Must run before zap_low_mappings */
 220 __meminit void *early_ioremap(unsigned long addr, unsigned long size)
 221 {
 222         pmd_t *pmd, *last_pmd;
 223         unsigned long vaddr;
 224         int i, pmds;
 225
 226         pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 227         vaddr = __START_KERNEL_map;
 228         pmd = level2_kernel_pgt;
 229         last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1;
 230
 231         for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) {
 232                 for (i = 0; i < pmds; i++) {
 233                         if (pmd_present(pmd[i]))
 234                                 goto continue_outer_loop;
 235                 }
 236                 vaddr += addr & ~PMD_MASK;
 237                 addr &= PMD_MASK;
 238
 239                 for (i = 0; i < pmds; i++, addr += PMD_SIZE)
 240                         set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
 241                 __flush_tlb_all();
 242
 243                 return (void *)vaddr;
 244 continue_outer_loop:
 245                 ;
 246         }
 247         printk(KERN_ERR "early_ioremap(0x%lx, %lu) failed\n", addr, size);
 248
 249         return NULL;
 250 }
 251
 252 /*
 253  * To avoid virtual aliases later:
 254  */
 255 __meminit void early_iounmap(void *addr, unsigned long size)
 256 {
 257         unsigned long vaddr;
 258         pmd_t *pmd;
 259         int i, pmds;
 260
 261         vaddr = (unsigned long)addr;
 262         pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 263         pmd = level2_kernel_pgt + pmd_index(vaddr);
 264
 265         for (i = 0; i < pmds; i++)
 266                 pmd_clear(pmd + i);
 267
 268         __flush_tlb_all();
 269 }
 270
 271 static void __meminit
 272 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
 273 {
 274         int i = pmd_index(address);
 275
 276         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
 277                 pmd_t *pmd = pmd_page + pmd_index(address);
 278
 279                 if (address >= end) {
 280                         if (!after_bootmem) {
 281                                 for (; i < PTRS_PER_PMD; i++, pmd++)
 282                                         set_pmd(pmd, __pmd(0));
 283                         }
 284                         break;
 285                 }
 286
 287                 if (pmd_val(*pmd))
 288                         continue;
 289
 290                 set_pte((pte_t *)pmd,
 291                         pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 292         }
 293 }
 294
 295 static void __meminit
 296 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
 297 {
 298         pmd_t *pmd = pmd_offset(pud, 0);
 299         spin_lock(&init_mm.page_table_lock);
 300         phys_pmd_init(pmd, address, end);
 301         spin_unlock(&init_mm.page_table_lock);
 302         __flush_tlb_all();
 303 }
 304
 305 static void __meminit
 306 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
 307 {
 308         int i = pud_index(addr);
 309
 310         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
 311                 unsigned long pmd_phys;
 312                 pud_t *pud = pud_page + pud_index(addr);
 313                 pmd_t *pmd;
 314
 315                 if (addr >= end)
 316                         break;
 317
 318                 if (!after_bootmem &&
 319                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
 320                         set_pud(pud, __pud(0));
 321                         continue;
 322                 }
 323
 324                 if (pud_val(*pud)) {
 325                         phys_pmd_update(pud, addr, end);
 326                         continue;
 327                 }
 328
 329                 pmd = alloc_low_page(&pmd_phys);
 330
 331                 spin_lock(&init_mm.page_table_lock);
 332                 set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE));
 333                 phys_pmd_init(pmd, addr, end);
 334                 spin_unlock(&init_mm.page_table_lock);
 335
 336                 unmap_low_page(pmd);
 337         }
 338         __flush_tlb_all();
 339 }
 340
 341 static void __init find_early_table_space(unsigned long end)
 342 {
 343         unsigned long puds, pmds, tables, start;
 344
 345         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 346         pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 347         tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) +
 348                  round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
 349
 350         /*
 351          * RED-PEN putting page tables only on node 0 could
 352          * cause a hotspot and fill up ZONE_DMA. The page tables
 353          * need roughly 0.5KB per GB.
 354          */
 355         start = 0x8000;
 356         table_start = find_e820_area(start, end, tables, PAGE_SIZE);
 357         if (table_start == -1UL)
 358                 panic("Cannot find space for the kernel page tables");
 359
 360         table_start >>= PAGE_SHIFT;
 361         table_end = table_start;
 362
 363         early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n",
 364                 end, table_start << PAGE_SHIFT,
 365                 (table_start << PAGE_SHIFT) + tables);
 366 }
 367
 368 /*
 369  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 370  * This runs before bootmem is initialized and gets pages directly from
 371  * the physical memory. To access them they are temporarily mapped.
 372  */
 373 void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
 374 {
 375         unsigned long next;
 376
 377         pr_debug("init_memory_mapping\n");
 378
 379         /*
 380          * Find space for the kernel direct mapping tables.
 381          *
 382          * Later we should allocate these tables in the local node of the
 383          * memory mapped. Unfortunately this is done currently before the
 384          * nodes are discovered.
 385          */
 386         if (!after_bootmem)
 387                 find_early_table_space(end);
 388
 389         start = (unsigned long)__va(start);
 390         end = (unsigned long)__va(end);
 391
 392         for (; start < end; start = next) {
 393                 pgd_t *pgd = pgd_offset_k(start);
 394                 unsigned long pud_phys;
 395                 pud_t *pud;
 396
 397                 if (after_bootmem)
 398                         pud = pud_offset(pgd, start & PGDIR_MASK);
 399                 else
 400                         pud = alloc_low_page(&pud_phys);
 401
 402                 next = start + PGDIR_SIZE;
 403                 if (next > end)
 404                         next = end;
 405                 phys_pud_init(pud, __pa(start), __pa(next));
 406                 if (!after_bootmem)
 407                         set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys));
 408                 unmap_low_page(pud);
 409         }
 410
 411         if (!after_bootmem)
 412                 mmu_cr4_features = read_cr4();
 413         __flush_tlb_all();
 414
 415         if (!after_bootmem)
 416                 reserve_early(table_start << PAGE_SHIFT,
 417                                  table_end << PAGE_SHIFT, "PGTABLE");
 418 }
 419
 420 #ifndef CONFIG_NUMA
 421 void __init paging_init(void)
 422 {
 423         unsigned long max_zone_pfns[MAX_NR_ZONES];
 424
 425         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 426         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 427         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 428         max_zone_pfns[ZONE_NORMAL] = end_pfn;
 429
 430         memory_present(0, 0, end_pfn);
 431         sparse_init();
 432         free_area_init_nodes(max_zone_pfns);
 433 }
 434 #endif
 435
 436 /*
 437  * Memory hotplug specific functions
 438  */
 439 void online_page(struct page *page)
 440 {
 441         ClearPageReserved(page);
 442         init_page_count(page);
 443         __free_page(page);
 444         totalram_pages++;
 445         num_physpages++;
 446 }
 447
 448 #ifdef CONFIG_MEMORY_HOTPLUG
 449 /*
 450  * Memory is added always to NORMAL zone. This means you will never get
 451  * additional DMA/DMA32 memory.
 452  */
 453 int arch_add_memory(int nid, u64 start, u64 size)
 454 {
 455         struct pglist_data *pgdat = NODE_DATA(nid);
 456         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 457         unsigned long start_pfn = start >> PAGE_SHIFT;
 458         unsigned long nr_pages = size >> PAGE_SHIFT;
 459         int ret;
 460
 461         init_memory_mapping(start, start + size-1);
 462
 463         ret = __add_pages(zone, start_pfn, nr_pages);
 464         WARN_ON(1);
 465
 466         return ret;
 467 }
 468 EXPORT_SYMBOL_GPL(arch_add_memory);
 469
 470 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
 471 int memory_add_physaddr_to_nid(u64 start)
 472 {
 473         return 0;
 474 }
 475 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 476 #endif
 477
 478 #endif /* CONFIG_MEMORY_HOTPLUG */
 479
 480 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
 481                          kcore_modules, kcore_vsyscall;
 482
 483 void __init mem_init(void)
 484 {
 485         long codesize, reservedpages, datasize, initsize;
 486
 487         pci_iommu_alloc();
 488
 489         /* clear_bss() already clear the empty_zero_page */
 490
 491         /* temporary debugging - double check it's true: */
 492         {
 493                 int i;
 494
 495                 for (i = 0; i < 1024; i++)
 496                         WARN_ON_ONCE(empty_zero_page[i]);
 497         }
 498
 499         reservedpages = 0;
 500
 501         /* this will put all low memory onto the freelists */
 502 #ifdef CONFIG_NUMA
 503         totalram_pages = numa_free_all_bootmem();
 504 #else
 505         totalram_pages = free_all_bootmem();
 506 #endif
 507         reservedpages = end_pfn - totalram_pages -
 508                                         absent_pages_in_range(0, end_pfn);
 509         after_bootmem = 1;
 510
 511         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 512         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 513         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 514
 515         /* Register memory areas for /proc/kcore */
 516         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 517         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 518                    VMALLOC_END-VMALLOC_START);
 519         kclist_add(&kcore_kernel, &_stext, _end - _stext);
 520         kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
 521         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 522                                  VSYSCALL_END - VSYSCALL_START);
 523
 524         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 525                                 "%ldk reserved, %ldk data, %ldk init)\n",
 526                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 527                 end_pfn << (PAGE_SHIFT-10),
 528                 codesize >> 10,
 529                 reservedpages << (PAGE_SHIFT-10),
 530                 datasize >> 10,
 531                 initsize >> 10);
 532
 533         cpa_init();
 534 }
 535
 536 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 537 {
 538         unsigned long addr = begin;
 539
 540         if (addr >= end)
 541                 return;
 542
 543         /*
 544          * If debugging page accesses then do not free this memory but
 545          * mark them not present - any buggy init-section access will
 546          * create a kernel page fault:
 547          */
 548 #ifdef CONFIG_DEBUG_PAGEALLOC
 549         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
 550                 begin, PAGE_ALIGN(end));
 551         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 552 #else
 553         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 554
 555         for (; addr < end; addr += PAGE_SIZE) {
 556                 ClearPageReserved(virt_to_page(addr));
 557                 init_page_count(virt_to_page(addr));
 558                 memset((void *)(addr & ~(PAGE_SIZE-1)),
 559                         POISON_FREE_INITMEM, PAGE_SIZE);
 560                 free_page(addr);
 561                 totalram_pages++;
 562         }
 563 #endif
 564 }
 565
 566 void free_initmem(void)
 567 {
 568         free_init_pages("unused kernel memory",
 569                         (unsigned long)(&__init_begin),
 570                         (unsigned long)(&__init_end));
 571 }
 572
 573 #ifdef CONFIG_DEBUG_RODATA
 574 const int rodata_test_data = 0xC3;
 575 EXPORT_SYMBOL_GPL(rodata_test_data);
 576
 577 void mark_rodata_ro(void)
 578 {
 579         unsigned long start = (unsigned long)_stext, end;
 580
 581 #ifdef CONFIG_HOTPLUG_CPU
 582         /* It must still be possible to apply SMP alternatives. */
 583         if (num_possible_cpus() > 1)
 584                 start = (unsigned long)_etext;
 585 #endif
 586
 587 #ifdef CONFIG_KPROBES
 588         start = (unsigned long)__start_rodata;
 589 #endif
 590
 591         end = (unsigned long)__end_rodata;
 592         start = (start + PAGE_SIZE - 1) & PAGE_MASK;
 593         end &= PAGE_MASK;
 594         if (end <= start)
 595                 return;
 596
 597
 598         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 599                (end - start) >> 10);
 600         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 601
 602         /*
 603          * The rodata section (but not the kernel text!) should also be
 604          * not-executable.
 605          */
 606         start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
 607         set_memory_nx(start, (end - start) >> PAGE_SHIFT);
 608
 609         rodata_test();
 610
 611 #ifdef CONFIG_CPA_DEBUG
 612         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
 613         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 614
 615         printk(KERN_INFO "Testing CPA: again\n");
 616         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 617 #endif
 618 }
 619 #endif
 620
 621 #ifdef CONFIG_BLK_DEV_INITRD
 622 void free_initrd_mem(unsigned long start, unsigned long end)
 623 {
 624         free_init_pages("initrd memory", start, end);
 625 }
 626 #endif
 627
 628 void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
 629 {
 630 #ifdef CONFIG_NUMA
 631         int nid = phys_to_nid(phys);
 632 #endif
 633         unsigned long pfn = phys >> PAGE_SHIFT;
 634
 635         if (pfn >= end_pfn) {
 636                 /*
 637                  * This can happen with kdump kernels when accessing
 638                  * firmware tables:
 639                  */
 640                 if (pfn < end_pfn_map)
 641                         return;
 642
 643                 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n",
 644                                 phys, len);
 645                 return;
 646         }
 647
 648         /* Should check here against the e820 map to avoid double free */
 649 #ifdef CONFIG_NUMA
 650         reserve_bootmem_node(NODE_DATA(nid), phys, len, BOOTMEM_DEFAULT);
 651 #else
 652         reserve_bootmem(phys, len, BOOTMEM_DEFAULT);
 653 #endif
 654         if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
 655                 dma_reserve += len / PAGE_SIZE;
 656                 set_dma_reserve(dma_reserve);
 657         }
 658 }
 659
 660 int kern_addr_valid(unsigned long addr)
 661 {
 662         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 663         pgd_t *pgd;
 664         pud_t *pud;
 665         pmd_t *pmd;
 666         pte_t *pte;
 667
 668         if (above != 0 && above != -1UL)
 669                 return 0;
 670
 671         pgd = pgd_offset_k(addr);
 672         if (pgd_none(*pgd))
 673                 return 0;
 674
 675         pud = pud_offset(pgd, addr);
 676         if (pud_none(*pud))
 677                 return 0;
 678
 679         pmd = pmd_offset(pud, addr);
 680         if (pmd_none(*pmd))
 681                 return 0;
 682
 683         if (pmd_large(*pmd))
 684                 return pfn_valid(pmd_pfn(*pmd));
 685
 686         pte = pte_offset_kernel(pmd, addr);
 687         if (pte_none(*pte))
 688                 return 0;
 689
 690         return pfn_valid(pte_pfn(*pte));
 691 }
 692
 693 /*
 694  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 695  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 696  * not need special handling anymore:
 697  */
 698 static struct vm_area_struct gate_vma = {
 699         .vm_start       = VSYSCALL_START,
 700         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 701         .vm_page_prot   = PAGE_READONLY_EXEC,
 702         .vm_flags       = VM_READ | VM_EXEC
 703 };
 704
 705 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 706 {
 707 #ifdef CONFIG_IA32_EMULATION
 708         if (test_tsk_thread_flag(tsk, TIF_IA32))
 709                 return NULL;
 710 #endif
 711         return &gate_vma;
 712 }
 713
 714 int in_gate_area(struct task_struct *task, unsigned long addr)
 715 {
 716         struct vm_area_struct *vma = get_gate_vma(task);
 717
 718         if (!vma)
 719                 return 0;
 720
 721         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 722 }
 723
 724 /*
 725  * Use this when you have no reliable task/vma, typically from interrupt
 726  * context. It is less reliable than using the task's vma and may give
 727  * false positives:
 728  */
 729 int in_gate_area_no_task(unsigned long addr)
 730 {
 731         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 732 }
 733
 734 const char *arch_vma_name(struct vm_area_struct *vma)
 735 {
 736         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 737                 return "[vdso]";
 738         if (vma == &gate_vma)
 739                 return "[vsyscall]";
 740         return NULL;
 741 }
 742
 743 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 744 /*
 745  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 746  */
 747 int __meminit
 748 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 749 {
 750         unsigned long addr = (unsigned long)start_page;
 751         unsigned long end = (unsigned long)(start_page + size);
 752         unsigned long next;
 753         pgd_t *pgd;
 754         pud_t *pud;
 755         pmd_t *pmd;
 756
 757         for (; addr < end; addr = next) {
 758                 next = pmd_addr_end(addr, end);
 759
 760                 pgd = vmemmap_pgd_populate(addr, node);
 761                 if (!pgd)
 762                         return -ENOMEM;
 763
 764                 pud = vmemmap_pud_populate(pgd, addr, node);
 765                 if (!pud)
 766                         return -ENOMEM;
 767
 768                 pmd = pmd_offset(pud, addr);
 769                 if (pmd_none(*pmd)) {
 770                         pte_t entry;
 771                         void *p;
 772
 773                         p = vmemmap_alloc_block(PMD_SIZE, node);
 774                         if (!p)
 775                                 return -ENOMEM;
 776
 777                         entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
 778                                                         PAGE_KERNEL_LARGE);
 779                         set_pmd(pmd, __pmd(pte_val(entry)));
 780
 781                         printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n",
 782                                 addr, addr + PMD_SIZE - 1, p, node);
 783                 } else {
 784                         vmemmap_verify((pte_t *)pmd, node, addr, next);
 785                 }
 786         }
 787         return 0;
 788 }
 789 #endif