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
  49 const struct dma_mapping_ops *dma_ops;
  50 EXPORT_SYMBOL(dma_ops);
  51
  52 static unsigned long dma_reserve __initdata;
  53
  54 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  55
  56 /*
  57  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
  58  * physical space so we can cache the place of the first one and move
  59  * around without checking the pgd every time.
  60  */
  61
  62 void show_mem(void)
  63 {
  64         long i, total = 0, reserved = 0;
  65         long shared = 0, cached = 0;
  66         struct page *page;
  67         pg_data_t *pgdat;
  68
  69         printk(KERN_INFO "Mem-info:\n");
  70         show_free_areas();
  71         printk(KERN_INFO "Free swap:       %6ldkB\n",
  72                 nr_swap_pages << (PAGE_SHIFT-10));
  73
  74         for_each_online_pgdat(pgdat) {
  75                 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
  76                         /*
  77                          * This loop can take a while with 256 GB and
  78                          * 4k pages so defer the NMI watchdog:
  79                          */
  80                         if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
  81                                 touch_nmi_watchdog();
  82
  83                         if (!pfn_valid(pgdat->node_start_pfn + i))
  84                                 continue;
  85
  86                         page = pfn_to_page(pgdat->node_start_pfn + i);
  87                         total++;
  88                         if (PageReserved(page))
  89                                 reserved++;
  90                         else if (PageSwapCache(page))
  91                                 cached++;
  92                         else if (page_count(page))
  93                                 shared += page_count(page) - 1;
  94                 }
  95         }
  96         printk(KERN_INFO "%lu pages of RAM\n",          total);
  97         printk(KERN_INFO "%lu reserved pages\n",        reserved);
  98         printk(KERN_INFO "%lu pages shared\n",          shared);
  99         printk(KERN_INFO "%lu pages swap cached\n",     cached);
 100 }
 101
 102 int after_bootmem;
 103
 104 static __init void *spp_getpage(void)
 105 {
 106         void *ptr;
 107
 108         if (after_bootmem)
 109                 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
 110         else
 111                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 112
 113         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 114                 panic("set_pte_phys: cannot allocate page data %s\n",
 115                         after_bootmem ? "after bootmem" : "");
 116         }
 117
 118         pr_debug("spp_getpage %p\n", ptr);
 119
 120         return ptr;
 121 }
 122
 123 static __init void
 124 set_pte_phys(unsigned long vaddr, unsigned long phys, pgprot_t prot)
 125 {
 126         pgd_t *pgd;
 127         pud_t *pud;
 128         pmd_t *pmd;
 129         pte_t *pte, new_pte;
 130
 131         pr_debug("set_pte_phys %lx to %lx\n", vaddr, phys);
 132
 133         pgd = pgd_offset_k(vaddr);
 134         if (pgd_none(*pgd)) {
 135                 printk(KERN_ERR
 136                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
 137                 return;
 138         }
 139         pud = pud_offset(pgd, vaddr);
 140         if (pud_none(*pud)) {
 141                 pmd = (pmd_t *) spp_getpage();
 142                 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER));
 143                 if (pmd != pmd_offset(pud, 0)) {
 144                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
 145                                 pmd, pmd_offset(pud, 0));
 146                         return;
 147                 }
 148         }
 149         pmd = pmd_offset(pud, vaddr);
 150         if (pmd_none(*pmd)) {
 151                 pte = (pte_t *) spp_getpage();
 152                 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER));
 153                 if (pte != pte_offset_kernel(pmd, 0)) {
 154                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
 155                         return;
 156                 }
 157         }
 158         new_pte = pfn_pte(phys >> PAGE_SHIFT, prot);
 159
 160         pte = pte_offset_kernel(pmd, vaddr);
 161         if (!pte_none(*pte) &&
 162             pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask))
 163                 pte_ERROR(*pte);
 164         set_pte(pte, new_pte);
 165
 166         /*
 167          * It's enough to flush this one mapping.
 168          * (PGE mappings get flushed as well)
 169          */
 170         __flush_tlb_one(vaddr);
 171 }
 172
 173 /* NOTE: this is meant to be run only at boot */
 174 void __init
 175 __set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
 176 {
 177         unsigned long address = __fix_to_virt(idx);
 178
 179         if (idx >= __end_of_fixed_addresses) {
 180                 printk(KERN_ERR "Invalid __set_fixmap\n");
 181                 return;
 182         }
 183         set_pte_phys(address, phys, prot);
 184 }
 185
 186 static unsigned long __initdata table_start;
 187 static unsigned long __meminitdata table_end;
 188
 189 static __meminit void *alloc_low_page(unsigned long *phys)
 190 {
 191         unsigned long pfn = table_end++;
 192         void *adr;
 193
 194         if (after_bootmem) {
 195                 adr = (void *)get_zeroed_page(GFP_ATOMIC);
 196                 *phys = __pa(adr);
 197
 198                 return adr;
 199         }
 200
 201         if (pfn >= end_pfn)
 202                 panic("alloc_low_page: ran out of memory");
 203
 204         adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
 205         memset(adr, 0, PAGE_SIZE);
 206         *phys  = pfn * PAGE_SIZE;
 207         return adr;
 208 }
 209
 210 static __meminit void unmap_low_page(void *adr)
 211 {
 212         if (after_bootmem)
 213                 return;
 214
 215         early_iounmap(adr, PAGE_SIZE);
 216 }
 217
 218 /* Must run before zap_low_mappings */
 219 __meminit void *early_ioremap(unsigned long addr, unsigned long size)
 220 {
 221         pmd_t *pmd, *last_pmd;
 222         unsigned long vaddr;
 223         int i, pmds;
 224
 225         pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 226         vaddr = __START_KERNEL_map;
 227         pmd = level2_kernel_pgt;
 228         last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1;
 229
 230         for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) {
 231                 for (i = 0; i < pmds; i++) {
 232                         if (pmd_present(pmd[i]))
 233                                 goto continue_outer_loop;
 234                 }
 235                 vaddr += addr & ~PMD_MASK;
 236                 addr &= PMD_MASK;
 237
 238                 for (i = 0; i < pmds; i++, addr += PMD_SIZE)
 239                         set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
 240                 __flush_tlb_all();
 241
 242                 return (void *)vaddr;
 243 continue_outer_loop:
 244                 ;
 245         }
 246         printk(KERN_ERR "early_ioremap(0x%lx, %lu) failed\n", addr, size);
 247
 248         return NULL;
 249 }
 250
 251 /*
 252  * To avoid virtual aliases later:
 253  */
 254 __meminit void early_iounmap(void *addr, unsigned long size)
 255 {
 256         unsigned long vaddr;
 257         pmd_t *pmd;
 258         int i, pmds;
 259
 260         vaddr = (unsigned long)addr;
 261         pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 262         pmd = level2_kernel_pgt + pmd_index(vaddr);
 263
 264         for (i = 0; i < pmds; i++)
 265                 pmd_clear(pmd + i);
 266
 267         __flush_tlb_all();
 268 }
 269
 270 static void __meminit
 271 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
 272 {
 273         int i = pmd_index(address);
 274
 275         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
 276                 unsigned long entry;
 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                 entry = __PAGE_KERNEL_LARGE|_PAGE_GLOBAL|address;
 291                 entry &= __supported_pte_mask;
 292                 set_pmd(pmd, __pmd(entry));
 293         }
 294 }
 295
 296 static void __meminit
 297 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
 298 {
 299         pmd_t *pmd = pmd_offset(pud, 0);
 300         spin_lock(&init_mm.page_table_lock);
 301         phys_pmd_init(pmd, address, end);
 302         spin_unlock(&init_mm.page_table_lock);
 303         __flush_tlb_all();
 304 }
 305
 306 static void __meminit
 307 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
 308 {
 309         int i = pud_index(addr);
 310
 311         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
 312                 unsigned long pmd_phys;
 313                 pud_t *pud = pud_page + pud_index(addr);
 314                 pmd_t *pmd;
 315
 316                 if (addr >= end)
 317                         break;
 318
 319                 if (!after_bootmem &&
 320                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
 321                         set_pud(pud, __pud(0));
 322                         continue;
 323                 }
 324
 325                 if (pud_val(*pud)) {
 326                         phys_pmd_update(pud, addr, end);
 327                         continue;
 328                 }
 329
 330                 pmd = alloc_low_page(&pmd_phys);
 331
 332                 spin_lock(&init_mm.page_table_lock);
 333                 set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE));
 334                 phys_pmd_init(pmd, addr, end);
 335                 spin_unlock(&init_mm.page_table_lock);
 336
 337                 unmap_low_page(pmd);
 338         }
 339         __flush_tlb_all();
 340 }
 341
 342 static void __init find_early_table_space(unsigned long end)
 343 {
 344         unsigned long puds, pmds, tables, start;
 345
 346         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 347         pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 348         tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) +
 349                  round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
 350
 351         /*
 352          * RED-PEN putting page tables only on node 0 could
 353          * cause a hotspot and fill up ZONE_DMA. The page tables
 354          * need roughly 0.5KB per GB.
 355          */
 356         start = 0x8000;
 357         table_start = find_e820_area(start, end, tables, PAGE_SIZE);
 358         if (table_start == -1UL)
 359                 panic("Cannot find space for the kernel page tables");
 360
 361         table_start >>= PAGE_SHIFT;
 362         table_end = table_start;
 363
 364         early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n",
 365                 end, table_start << PAGE_SHIFT,
 366                 (table_start << PAGE_SHIFT) + tables);
 367 }
 368
 369 /*
 370  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 371  * This runs before bootmem is initialized and gets pages directly from
 372  * the physical memory. To access them they are temporarily mapped.
 373  */
 374 void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
 375 {
 376         unsigned long next;
 377
 378         pr_debug("init_memory_mapping\n");
 379
 380         /*
 381          * Find space for the kernel direct mapping tables.
 382          *
 383          * Later we should allocate these tables in the local node of the
 384          * memory mapped. Unfortunately this is done currently before the
 385          * nodes are discovered.
 386          */
 387         if (!after_bootmem)
 388                 find_early_table_space(end);
 389
 390         start = (unsigned long)__va(start);
 391         end = (unsigned long)__va(end);
 392
 393         for (; start < end; start = next) {
 394                 pgd_t *pgd = pgd_offset_k(start);
 395                 unsigned long pud_phys;
 396                 pud_t *pud;
 397
 398                 if (after_bootmem)
 399                         pud = pud_offset(pgd, start & PGDIR_MASK);
 400                 else
 401                         pud = alloc_low_page(&pud_phys);
 402
 403                 next = start + PGDIR_SIZE;
 404                 if (next > end)
 405                         next = end;
 406                 phys_pud_init(pud, __pa(start), __pa(next));
 407                 if (!after_bootmem)
 408                         set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys));
 409                 unmap_low_page(pud);
 410         }
 411
 412         if (!after_bootmem)
 413                 mmu_cr4_features = read_cr4();
 414         __flush_tlb_all();
 415
 416         if (!after_bootmem)
 417                 reserve_early(table_start << PAGE_SHIFT,
 418                                  table_end << PAGE_SHIFT, "PGTABLE");
 419 }
 420
 421 #ifndef CONFIG_NUMA
 422 void __init paging_init(void)
 423 {
 424         unsigned long max_zone_pfns[MAX_NR_ZONES];
 425
 426         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 427         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 428         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 429         max_zone_pfns[ZONE_NORMAL] = end_pfn;
 430
 431         memory_present(0, 0, end_pfn);
 432         sparse_init();
 433         free_area_init_nodes(max_zone_pfns);
 434 }
 435 #endif
 436
 437 /*
 438  * Unmap a kernel mapping if it exists. This is useful to avoid
 439  * prefetches from the CPU leading to inconsistent cache lines.
 440  * address and size must be aligned to 2MB boundaries.
 441  * Does nothing when the mapping doesn't exist.
 442  */
 443 void __init clear_kernel_mapping(unsigned long address, unsigned long size)
 444 {
 445         unsigned long end = address + size;
 446
 447         BUG_ON(address & ~LARGE_PAGE_MASK);
 448         BUG_ON(size & ~LARGE_PAGE_MASK);
 449
 450         for (; address < end; address += LARGE_PAGE_SIZE) {
 451                 pgd_t *pgd = pgd_offset_k(address);
 452                 pud_t *pud;
 453                 pmd_t *pmd;
 454
 455                 if (pgd_none(*pgd))
 456                         continue;
 457
 458                 pud = pud_offset(pgd, address);
 459                 if (pud_none(*pud))
 460                         continue;
 461
 462                 pmd = pmd_offset(pud, address);
 463                 if (!pmd || pmd_none(*pmd))
 464                         continue;
 465
 466                 if (!(pmd_val(*pmd) & _PAGE_PSE)) {
 467                         /*
 468                          * Could handle this, but it should not happen
 469                          * currently:
 470                          */
 471                         printk(KERN_ERR "clear_kernel_mapping: "
 472                                 "mapping has been split. will leak memory\n");
 473                         pmd_ERROR(*pmd);
 474                 }
 475                 set_pmd(pmd, __pmd(0));
 476         }
 477         __flush_tlb_all();
 478 }
 479
 480 /*
 481  * Memory hotplug specific functions
 482  */
 483 void online_page(struct page *page)
 484 {
 485         ClearPageReserved(page);
 486         init_page_count(page);
 487         __free_page(page);
 488         totalram_pages++;
 489         num_physpages++;
 490 }
 491
 492 #ifdef CONFIG_MEMORY_HOTPLUG
 493 /*
 494  * Memory is added always to NORMAL zone. This means you will never get
 495  * additional DMA/DMA32 memory.
 496  */
 497 int arch_add_memory(int nid, u64 start, u64 size)
 498 {
 499         struct pglist_data *pgdat = NODE_DATA(nid);
 500         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 501         unsigned long start_pfn = start >> PAGE_SHIFT;
 502         unsigned long nr_pages = size >> PAGE_SHIFT;
 503         int ret;
 504
 505         init_memory_mapping(start, start + size-1);
 506
 507         ret = __add_pages(zone, start_pfn, nr_pages);
 508         WARN_ON(1);
 509
 510         return ret;
 511 }
 512 EXPORT_SYMBOL_GPL(arch_add_memory);
 513
 514 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
 515 int memory_add_physaddr_to_nid(u64 start)
 516 {
 517         return 0;
 518 }
 519 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 520 #endif
 521
 522 #endif /* CONFIG_MEMORY_HOTPLUG */
 523
 524 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
 525                          kcore_modules, kcore_vsyscall;
 526
 527 void __init mem_init(void)
 528 {
 529         long codesize, reservedpages, datasize, initsize;
 530
 531         pci_iommu_alloc();
 532
 533         /* clear_bss() already clear the empty_zero_page */
 534
 535         /* temporary debugging - double check it's true: */
 536         {
 537                 int i;
 538
 539                 for (i = 0; i < 1024; i++)
 540                         WARN_ON_ONCE(empty_zero_page[i]);
 541         }
 542
 543         reservedpages = 0;
 544
 545         /* this will put all low memory onto the freelists */
 546 #ifdef CONFIG_NUMA
 547         totalram_pages = numa_free_all_bootmem();
 548 #else
 549         totalram_pages = free_all_bootmem();
 550 #endif
 551         reservedpages = end_pfn - totalram_pages -
 552                                         absent_pages_in_range(0, end_pfn);
 553         after_bootmem = 1;
 554
 555         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 556         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 557         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 558
 559         /* Register memory areas for /proc/kcore */
 560         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 561         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 562                    VMALLOC_END-VMALLOC_START);
 563         kclist_add(&kcore_kernel, &_stext, _end - _stext);
 564         kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
 565         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 566                                  VSYSCALL_END - VSYSCALL_START);
 567
 568         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 569                                 "%ldk reserved, %ldk data, %ldk init)\n",
 570                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 571                 end_pfn << (PAGE_SHIFT-10),
 572                 codesize >> 10,
 573                 reservedpages << (PAGE_SHIFT-10),
 574                 datasize >> 10,
 575                 initsize >> 10);
 576 }
 577
 578 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 579 {
 580         unsigned long addr;
 581
 582         if (begin >= end)
 583                 return;
 584
 585         /*
 586          * If debugging page accesses then do not free this memory but
 587          * mark them not present - any buggy init-section access will
 588          * create a kernel page fault:
 589          */
 590 #ifdef CONFIG_DEBUG_PAGEALLOC
 591         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
 592                 begin, PAGE_ALIGN(end));
 593         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 594 #else
 595         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 596
 597         for (addr = begin; addr < end; addr += PAGE_SIZE) {
 598                 ClearPageReserved(virt_to_page(addr));
 599                 init_page_count(virt_to_page(addr));
 600                 memset((void *)(addr & ~(PAGE_SIZE-1)),
 601                         POISON_FREE_INITMEM, PAGE_SIZE);
 602                 free_page(addr);
 603                 totalram_pages++;
 604         }
 605 #endif
 606 }
 607
 608 void free_initmem(void)
 609 {
 610         free_init_pages("unused kernel memory",
 611                         (unsigned long)(&__init_begin),
 612                         (unsigned long)(&__init_end));
 613 }
 614
 615 #ifdef CONFIG_DEBUG_RODATA
 616 const int rodata_test_data = 0xC3;
 617 EXPORT_SYMBOL_GPL(rodata_test_data);
 618
 619 void mark_rodata_ro(void)
 620 {
 621         unsigned long start = (unsigned long)_stext, end;
 622
 623 #ifdef CONFIG_HOTPLUG_CPU
 624         /* It must still be possible to apply SMP alternatives. */
 625         if (num_possible_cpus() > 1)
 626                 start = (unsigned long)_etext;
 627 #endif
 628
 629 #ifdef CONFIG_KPROBES
 630         start = (unsigned long)__start_rodata;
 631 #endif
 632
 633         end = (unsigned long)__end_rodata;
 634         start = (start + PAGE_SIZE - 1) & PAGE_MASK;
 635         end &= PAGE_MASK;
 636         if (end <= start)
 637                 return;
 638
 639         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 640
 641         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 642                (end - start) >> 10);
 643
 644         rodata_test();
 645
 646 #ifdef CONFIG_CPA_DEBUG
 647         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
 648         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 649
 650         printk(KERN_INFO "Testing CPA: again\n");
 651         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 652 #endif
 653 }
 654 #endif
 655
 656 #ifdef CONFIG_BLK_DEV_INITRD
 657 void free_initrd_mem(unsigned long start, unsigned long end)
 658 {
 659         free_init_pages("initrd memory", start, end);
 660 }
 661 #endif
 662
 663 void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
 664 {
 665 #ifdef CONFIG_NUMA
 666         int nid = phys_to_nid(phys);
 667 #endif
 668         unsigned long pfn = phys >> PAGE_SHIFT;
 669
 670         if (pfn >= end_pfn) {
 671                 /*
 672                  * This can happen with kdump kernels when accessing
 673                  * firmware tables:
 674                  */
 675                 if (pfn < end_pfn_map)
 676                         return;
 677
 678                 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n",
 679                                 phys, len);
 680                 return;
 681         }
 682
 683         /* Should check here against the e820 map to avoid double free */
 684 #ifdef CONFIG_NUMA
 685         reserve_bootmem_node(NODE_DATA(nid), phys, len);
 686 #else
 687         reserve_bootmem(phys, len);
 688 #endif
 689         if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
 690                 dma_reserve += len / PAGE_SIZE;
 691                 set_dma_reserve(dma_reserve);
 692         }
 693 }
 694
 695 int kern_addr_valid(unsigned long addr)
 696 {
 697         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 698         pgd_t *pgd;
 699         pud_t *pud;
 700         pmd_t *pmd;
 701         pte_t *pte;
 702
 703         if (above != 0 && above != -1UL)
 704                 return 0;
 705
 706         pgd = pgd_offset_k(addr);
 707         if (pgd_none(*pgd))
 708                 return 0;
 709
 710         pud = pud_offset(pgd, addr);
 711         if (pud_none(*pud))
 712                 return 0;
 713
 714         pmd = pmd_offset(pud, addr);
 715         if (pmd_none(*pmd))
 716                 return 0;
 717
 718         if (pmd_large(*pmd))
 719                 return pfn_valid(pmd_pfn(*pmd));
 720
 721         pte = pte_offset_kernel(pmd, addr);
 722         if (pte_none(*pte))
 723                 return 0;
 724
 725         return pfn_valid(pte_pfn(*pte));
 726 }
 727
 728 /*
 729  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 730  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 731  * not need special handling anymore:
 732  */
 733 static struct vm_area_struct gate_vma = {
 734         .vm_start       = VSYSCALL_START,
 735         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 736         .vm_page_prot   = PAGE_READONLY_EXEC,
 737         .vm_flags       = VM_READ | VM_EXEC
 738 };
 739
 740 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 741 {
 742 #ifdef CONFIG_IA32_EMULATION
 743         if (test_tsk_thread_flag(tsk, TIF_IA32))
 744                 return NULL;
 745 #endif
 746         return &gate_vma;
 747 }
 748
 749 int in_gate_area(struct task_struct *task, unsigned long addr)
 750 {
 751         struct vm_area_struct *vma = get_gate_vma(task);
 752
 753         if (!vma)
 754                 return 0;
 755
 756         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 757 }
 758
 759 /*
 760  * Use this when you have no reliable task/vma, typically from interrupt
 761  * context. It is less reliable than using the task's vma and may give
 762  * false positives:
 763  */
 764 int in_gate_area_no_task(unsigned long addr)
 765 {
 766         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 767 }
 768
 769 const char *arch_vma_name(struct vm_area_struct *vma)
 770 {
 771         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 772                 return "[vdso]";
 773         if (vma == &gate_vma)
 774                 return "[vsyscall]";
 775         return NULL;
 776 }
 777
 778 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 779 /*
 780  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 781  */
 782 int __meminit
 783 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 784 {
 785         unsigned long addr = (unsigned long)start_page;
 786         unsigned long end = (unsigned long)(start_page + size);
 787         unsigned long next;
 788         pgd_t *pgd;
 789         pud_t *pud;
 790         pmd_t *pmd;
 791
 792         for (; addr < end; addr = next) {
 793                 next = pmd_addr_end(addr, end);
 794
 795                 pgd = vmemmap_pgd_populate(addr, node);
 796                 if (!pgd)
 797                         return -ENOMEM;
 798
 799                 pud = vmemmap_pud_populate(pgd, addr, node);
 800                 if (!pud)
 801                         return -ENOMEM;
 802
 803                 pmd = pmd_offset(pud, addr);
 804                 if (pmd_none(*pmd)) {
 805                         pte_t entry;
 806                         void *p;
 807
 808                         p = vmemmap_alloc_block(PMD_SIZE, node);
 809                         if (!p)
 810                                 return -ENOMEM;
 811
 812                         entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
 813                                                         PAGE_KERNEL_LARGE);
 814                         set_pmd(pmd, __pmd(pte_val(entry)));
 815
 816                         printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n",
 817                                 addr, addr + PMD_SIZE - 1, p, node);
 818                 } else {
 819                         vmemmap_verify((pte_t *)pmd, node, addr, next);
 820                 }
 821         }
 822         return 0;
 823 }
 824 #endif