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 /*
 175  * The head.S code sets up the kernel high mapping:
 176  *
 177  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
 178  *
 179  * phys_addr holds the negative offset to the kernel, which is added
 180  * to the compile time generated pmds. This results in invalid pmds up
 181  * to the point where we hit the physaddr 0 mapping.
 182  *
 183  * We limit the mappings to the region from _text to _end.  _end is
 184  * rounded up to the 2MB boundary. This catches the invalid pmds as
 185  * well, as they are located before _text:
 186  */
 187 void __init cleanup_highmap(void)
 188 {
 189         unsigned long vaddr = __START_KERNEL_map;
 190         unsigned long end = round_up((unsigned long)_end, PMD_SIZE) - 1;
 191         pmd_t *pmd = level2_kernel_pgt;
 192         pmd_t *last_pmd = pmd + PTRS_PER_PMD;
 193
 194         for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
 195                 if (!pmd_present(*pmd))
 196                         continue;
 197                 if (vaddr < (unsigned long) _text || vaddr > end)
 198                         set_pmd(pmd, __pmd(0));
 199         }
 200 }
 201
 202 /* NOTE: this is meant to be run only at boot */
 203 void __init
 204 __set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
 205 {
 206         unsigned long address = __fix_to_virt(idx);
 207
 208         if (idx >= __end_of_fixed_addresses) {
 209                 printk(KERN_ERR "Invalid __set_fixmap\n");
 210                 return;
 211         }
 212         set_pte_phys(address, phys, prot);
 213 }
 214
 215 static unsigned long __initdata table_start;
 216 static unsigned long __meminitdata table_end;
 217
 218 static __meminit void *alloc_low_page(unsigned long *phys)
 219 {
 220         unsigned long pfn = table_end++;
 221         void *adr;
 222
 223         if (after_bootmem) {
 224                 adr = (void *)get_zeroed_page(GFP_ATOMIC);
 225                 *phys = __pa(adr);
 226
 227                 return adr;
 228         }
 229
 230         if (pfn >= end_pfn)
 231                 panic("alloc_low_page: ran out of memory");
 232
 233         adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
 234         memset(adr, 0, PAGE_SIZE);
 235         *phys  = pfn * PAGE_SIZE;
 236         return adr;
 237 }
 238
 239 static __meminit void unmap_low_page(void *adr)
 240 {
 241         if (after_bootmem)
 242                 return;
 243
 244         early_iounmap(adr, PAGE_SIZE);
 245 }
 246
 247 /* Must run before zap_low_mappings */
 248 __meminit void *early_ioremap(unsigned long addr, unsigned long size)
 249 {
 250         pmd_t *pmd, *last_pmd;
 251         unsigned long vaddr;
 252         int i, pmds;
 253
 254         pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 255         vaddr = __START_KERNEL_map;
 256         pmd = level2_kernel_pgt;
 257         last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1;
 258
 259         for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) {
 260                 for (i = 0; i < pmds; i++) {
 261                         if (pmd_present(pmd[i]))
 262                                 goto continue_outer_loop;
 263                 }
 264                 vaddr += addr & ~PMD_MASK;
 265                 addr &= PMD_MASK;
 266
 267                 for (i = 0; i < pmds; i++, addr += PMD_SIZE)
 268                         set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
 269                 __flush_tlb_all();
 270
 271                 return (void *)vaddr;
 272 continue_outer_loop:
 273                 ;
 274         }
 275         printk(KERN_ERR "early_ioremap(0x%lx, %lu) failed\n", addr, size);
 276
 277         return NULL;
 278 }
 279
 280 /*
 281  * To avoid virtual aliases later:
 282  */
 283 __meminit void early_iounmap(void *addr, unsigned long size)
 284 {
 285         unsigned long vaddr;
 286         pmd_t *pmd;
 287         int i, pmds;
 288
 289         vaddr = (unsigned long)addr;
 290         pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 291         pmd = level2_kernel_pgt + pmd_index(vaddr);
 292
 293         for (i = 0; i < pmds; i++)
 294                 pmd_clear(pmd + i);
 295
 296         __flush_tlb_all();
 297 }
 298
 299 static void __meminit
 300 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
 301 {
 302         int i = pmd_index(address);
 303
 304         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
 305                 pmd_t *pmd = pmd_page + pmd_index(address);
 306
 307                 if (address >= end) {
 308                         if (!after_bootmem) {
 309                                 for (; i < PTRS_PER_PMD; i++, pmd++)
 310                                         set_pmd(pmd, __pmd(0));
 311                         }
 312                         break;
 313                 }
 314
 315                 if (pmd_val(*pmd))
 316                         continue;
 317
 318                 set_pte((pte_t *)pmd,
 319                         pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 320         }
 321 }
 322
 323 static void __meminit
 324 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
 325 {
 326         pmd_t *pmd = pmd_offset(pud, 0);
 327         spin_lock(&init_mm.page_table_lock);
 328         phys_pmd_init(pmd, address, end);
 329         spin_unlock(&init_mm.page_table_lock);
 330         __flush_tlb_all();
 331 }
 332
 333 static void __meminit
 334 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
 335 {
 336         int i = pud_index(addr);
 337
 338         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
 339                 unsigned long pmd_phys;
 340                 pud_t *pud = pud_page + pud_index(addr);
 341                 pmd_t *pmd;
 342
 343                 if (addr >= end)
 344                         break;
 345
 346                 if (!after_bootmem &&
 347                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
 348                         set_pud(pud, __pud(0));
 349                         continue;
 350                 }
 351
 352                 if (pud_val(*pud)) {
 353                         phys_pmd_update(pud, addr, end);
 354                         continue;
 355                 }
 356
 357                 pmd = alloc_low_page(&pmd_phys);
 358
 359                 spin_lock(&init_mm.page_table_lock);
 360                 set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE));
 361                 phys_pmd_init(pmd, addr, end);
 362                 spin_unlock(&init_mm.page_table_lock);
 363
 364                 unmap_low_page(pmd);
 365         }
 366         __flush_tlb_all();
 367 }
 368
 369 static void __init find_early_table_space(unsigned long end)
 370 {
 371         unsigned long puds, pmds, tables, start;
 372
 373         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 374         pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 375         tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) +
 376                  round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
 377
 378         /*
 379          * RED-PEN putting page tables only on node 0 could
 380          * cause a hotspot and fill up ZONE_DMA. The page tables
 381          * need roughly 0.5KB per GB.
 382          */
 383         start = 0x8000;
 384         table_start = find_e820_area(start, end, tables, PAGE_SIZE);
 385         if (table_start == -1UL)
 386                 panic("Cannot find space for the kernel page tables");
 387
 388         table_start >>= PAGE_SHIFT;
 389         table_end = table_start;
 390
 391         early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n",
 392                 end, table_start << PAGE_SHIFT,
 393                 (table_start << PAGE_SHIFT) + tables);
 394 }
 395
 396 /*
 397  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 398  * This runs before bootmem is initialized and gets pages directly from
 399  * the physical memory. To access them they are temporarily mapped.
 400  */
 401 void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
 402 {
 403         unsigned long next;
 404
 405         pr_debug("init_memory_mapping\n");
 406
 407         /*
 408          * Find space for the kernel direct mapping tables.
 409          *
 410          * Later we should allocate these tables in the local node of the
 411          * memory mapped. Unfortunately this is done currently before the
 412          * nodes are discovered.
 413          */
 414         if (!after_bootmem)
 415                 find_early_table_space(end);
 416
 417         start = (unsigned long)__va(start);
 418         end = (unsigned long)__va(end);
 419
 420         for (; start < end; start = next) {
 421                 pgd_t *pgd = pgd_offset_k(start);
 422                 unsigned long pud_phys;
 423                 pud_t *pud;
 424
 425                 if (after_bootmem)
 426                         pud = pud_offset(pgd, start & PGDIR_MASK);
 427                 else
 428                         pud = alloc_low_page(&pud_phys);
 429
 430                 next = start + PGDIR_SIZE;
 431                 if (next > end)
 432                         next = end;
 433                 phys_pud_init(pud, __pa(start), __pa(next));
 434                 if (!after_bootmem)
 435                         set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys));
 436                 unmap_low_page(pud);
 437         }
 438
 439         if (!after_bootmem)
 440                 mmu_cr4_features = read_cr4();
 441         __flush_tlb_all();
 442
 443         if (!after_bootmem)
 444                 reserve_early(table_start << PAGE_SHIFT,
 445                                  table_end << PAGE_SHIFT, "PGTABLE");
 446 }
 447
 448 #ifndef CONFIG_NUMA
 449 void __init paging_init(void)
 450 {
 451         unsigned long max_zone_pfns[MAX_NR_ZONES];
 452
 453         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 454         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 455         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 456         max_zone_pfns[ZONE_NORMAL] = end_pfn;
 457
 458         memory_present(0, 0, end_pfn);
 459         sparse_init();
 460         free_area_init_nodes(max_zone_pfns);
 461 }
 462 #endif
 463
 464 /*
 465  * Memory hotplug specific functions
 466  */
 467 void online_page(struct page *page)
 468 {
 469         ClearPageReserved(page);
 470         init_page_count(page);
 471         __free_page(page);
 472         totalram_pages++;
 473         num_physpages++;
 474 }
 475
 476 #ifdef CONFIG_MEMORY_HOTPLUG
 477 /*
 478  * Memory is added always to NORMAL zone. This means you will never get
 479  * additional DMA/DMA32 memory.
 480  */
 481 int arch_add_memory(int nid, u64 start, u64 size)
 482 {
 483         struct pglist_data *pgdat = NODE_DATA(nid);
 484         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 485         unsigned long start_pfn = start >> PAGE_SHIFT;
 486         unsigned long nr_pages = size >> PAGE_SHIFT;
 487         int ret;
 488
 489         init_memory_mapping(start, start + size-1);
 490
 491         ret = __add_pages(zone, start_pfn, nr_pages);
 492         WARN_ON(1);
 493
 494         return ret;
 495 }
 496 EXPORT_SYMBOL_GPL(arch_add_memory);
 497
 498 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
 499 int memory_add_physaddr_to_nid(u64 start)
 500 {
 501         return 0;
 502 }
 503 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 504 #endif
 505
 506 #endif /* CONFIG_MEMORY_HOTPLUG */
 507
 508 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
 509                          kcore_modules, kcore_vsyscall;
 510
 511 void __init mem_init(void)
 512 {
 513         long codesize, reservedpages, datasize, initsize;
 514
 515         pci_iommu_alloc();
 516
 517         /* clear_bss() already clear the empty_zero_page */
 518
 519         reservedpages = 0;
 520
 521         /* this will put all low memory onto the freelists */
 522 #ifdef CONFIG_NUMA
 523         totalram_pages = numa_free_all_bootmem();
 524 #else
 525         totalram_pages = free_all_bootmem();
 526 #endif
 527         reservedpages = end_pfn - totalram_pages -
 528                                         absent_pages_in_range(0, end_pfn);
 529         after_bootmem = 1;
 530
 531         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 532         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 533         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 534
 535         /* Register memory areas for /proc/kcore */
 536         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 537         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 538                    VMALLOC_END-VMALLOC_START);
 539         kclist_add(&kcore_kernel, &_stext, _end - _stext);
 540         kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
 541         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 542                                  VSYSCALL_END - VSYSCALL_START);
 543
 544         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 545                                 "%ldk reserved, %ldk data, %ldk init)\n",
 546                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 547                 end_pfn << (PAGE_SHIFT-10),
 548                 codesize >> 10,
 549                 reservedpages << (PAGE_SHIFT-10),
 550                 datasize >> 10,
 551                 initsize >> 10);
 552
 553         cpa_init();
 554 }
 555
 556 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 557 {
 558         unsigned long addr = begin;
 559
 560         if (addr >= end)
 561                 return;
 562
 563         /*
 564          * If debugging page accesses then do not free this memory but
 565          * mark them not present - any buggy init-section access will
 566          * create a kernel page fault:
 567          */
 568 #ifdef CONFIG_DEBUG_PAGEALLOC
 569         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
 570                 begin, PAGE_ALIGN(end));
 571         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 572 #else
 573         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 574
 575         for (; addr < end; addr += PAGE_SIZE) {
 576                 ClearPageReserved(virt_to_page(addr));
 577                 init_page_count(virt_to_page(addr));
 578                 memset((void *)(addr & ~(PAGE_SIZE-1)),
 579                         POISON_FREE_INITMEM, PAGE_SIZE);
 580                 free_page(addr);
 581                 totalram_pages++;
 582         }
 583 #endif
 584 }
 585
 586 void free_initmem(void)
 587 {
 588         free_init_pages("unused kernel memory",
 589                         (unsigned long)(&__init_begin),
 590                         (unsigned long)(&__init_end));
 591 }
 592
 593 #ifdef CONFIG_DEBUG_RODATA
 594 const int rodata_test_data = 0xC3;
 595 EXPORT_SYMBOL_GPL(rodata_test_data);
 596
 597 void mark_rodata_ro(void)
 598 {
 599         unsigned long start = (unsigned long)_stext, end;
 600
 601 #ifdef CONFIG_HOTPLUG_CPU
 602         /* It must still be possible to apply SMP alternatives. */
 603         if (num_possible_cpus() > 1)
 604                 start = (unsigned long)_etext;
 605 #endif
 606
 607 #ifdef CONFIG_KPROBES
 608         start = (unsigned long)__start_rodata;
 609 #endif
 610
 611         end = (unsigned long)__end_rodata;
 612         start = (start + PAGE_SIZE - 1) & PAGE_MASK;
 613         end &= PAGE_MASK;
 614         if (end <= start)
 615                 return;
 616
 617
 618         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 619                (end - start) >> 10);
 620         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 621
 622         /*
 623          * The rodata section (but not the kernel text!) should also be
 624          * not-executable.
 625          */
 626         start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
 627         set_memory_nx(start, (end - start) >> PAGE_SHIFT);
 628
 629         rodata_test();
 630
 631 #ifdef CONFIG_CPA_DEBUG
 632         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
 633         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 634
 635         printk(KERN_INFO "Testing CPA: again\n");
 636         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 637 #endif
 638 }
 639 #endif
 640
 641 #ifdef CONFIG_BLK_DEV_INITRD
 642 void free_initrd_mem(unsigned long start, unsigned long end)
 643 {
 644         free_init_pages("initrd memory", start, end);
 645 }
 646 #endif
 647
 648 void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
 649 {
 650 #ifdef CONFIG_NUMA
 651         int nid = phys_to_nid(phys);
 652 #endif
 653         unsigned long pfn = phys >> PAGE_SHIFT;
 654
 655         if (pfn >= end_pfn) {
 656                 /*
 657                  * This can happen with kdump kernels when accessing
 658                  * firmware tables:
 659                  */
 660                 if (pfn < end_pfn_map)
 661                         return;
 662
 663                 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n",
 664                                 phys, len);
 665                 return;
 666         }
 667
 668         /* Should check here against the e820 map to avoid double free */
 669 #ifdef CONFIG_NUMA
 670         reserve_bootmem_node(NODE_DATA(nid), phys, len, BOOTMEM_DEFAULT);
 671 #else
 672         reserve_bootmem(phys, len, BOOTMEM_DEFAULT);
 673 #endif
 674         if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
 675                 dma_reserve += len / PAGE_SIZE;
 676                 set_dma_reserve(dma_reserve);
 677         }
 678 }
 679
 680 int kern_addr_valid(unsigned long addr)
 681 {
 682         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 683         pgd_t *pgd;
 684         pud_t *pud;
 685         pmd_t *pmd;
 686         pte_t *pte;
 687
 688         if (above != 0 && above != -1UL)
 689                 return 0;
 690
 691         pgd = pgd_offset_k(addr);
 692         if (pgd_none(*pgd))
 693                 return 0;
 694
 695         pud = pud_offset(pgd, addr);
 696         if (pud_none(*pud))
 697                 return 0;
 698
 699         pmd = pmd_offset(pud, addr);
 700         if (pmd_none(*pmd))
 701                 return 0;
 702
 703         if (pmd_large(*pmd))
 704                 return pfn_valid(pmd_pfn(*pmd));
 705
 706         pte = pte_offset_kernel(pmd, addr);
 707         if (pte_none(*pte))
 708                 return 0;
 709
 710         return pfn_valid(pte_pfn(*pte));
 711 }
 712
 713 /*
 714  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 715  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 716  * not need special handling anymore:
 717  */
 718 static struct vm_area_struct gate_vma = {
 719         .vm_start       = VSYSCALL_START,
 720         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 721         .vm_page_prot   = PAGE_READONLY_EXEC,
 722         .vm_flags       = VM_READ | VM_EXEC
 723 };
 724
 725 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 726 {
 727 #ifdef CONFIG_IA32_EMULATION
 728         if (test_tsk_thread_flag(tsk, TIF_IA32))
 729                 return NULL;
 730 #endif
 731         return &gate_vma;
 732 }
 733
 734 int in_gate_area(struct task_struct *task, unsigned long addr)
 735 {
 736         struct vm_area_struct *vma = get_gate_vma(task);
 737
 738         if (!vma)
 739                 return 0;
 740
 741         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 742 }
 743
 744 /*
 745  * Use this when you have no reliable task/vma, typically from interrupt
 746  * context. It is less reliable than using the task's vma and may give
 747  * false positives:
 748  */
 749 int in_gate_area_no_task(unsigned long addr)
 750 {
 751         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 752 }
 753
 754 const char *arch_vma_name(struct vm_area_struct *vma)
 755 {
 756         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 757                 return "[vdso]";
 758         if (vma == &gate_vma)
 759                 return "[vsyscall]";
 760         return NULL;
 761 }
 762
 763 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 764 /*
 765  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 766  */
 767 int __meminit
 768 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 769 {
 770         unsigned long addr = (unsigned long)start_page;
 771         unsigned long end = (unsigned long)(start_page + size);
 772         unsigned long next;
 773         pgd_t *pgd;
 774         pud_t *pud;
 775         pmd_t *pmd;
 776
 777         for (; addr < end; addr = next) {
 778                 next = pmd_addr_end(addr, end);
 779
 780                 pgd = vmemmap_pgd_populate(addr, node);
 781                 if (!pgd)
 782                         return -ENOMEM;
 783
 784                 pud = vmemmap_pud_populate(pgd, addr, node);
 785                 if (!pud)
 786                         return -ENOMEM;
 787
 788                 pmd = pmd_offset(pud, addr);
 789                 if (pmd_none(*pmd)) {
 790                         pte_t entry;
 791                         void *p;
 792
 793                         p = vmemmap_alloc_block(PMD_SIZE, node);
 794                         if (!p)
 795                                 return -ENOMEM;
 796
 797                         entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
 798                                                         PAGE_KERNEL_LARGE);
 799                         set_pmd(pmd, __pmd(pte_val(entry)));
 800
 801                         printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n",
 802                                 addr, addr + PMD_SIZE - 1, p, node);
 803                 } else {
 804                         vmemmap_verify((pte_t *)pmd, node, addr, next);
 805                 }
 806         }
 807         return 0;
 808 }
 809 #endif