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 int direct_gbpages __meminitdata
  58 #ifdef CONFIG_DIRECT_GBPAGES
  59                                 = 1
  60 #endif
  61 ;
  62
  63 static int __init parse_direct_gbpages_off(char *arg)
  64 {
  65         direct_gbpages = 0;
  66         return 0;
  67 }
  68 early_param("nogbpages", parse_direct_gbpages_off);
  69
  70 static int __init parse_direct_gbpages_on(char *arg)
  71 {
  72         direct_gbpages = 1;
  73         return 0;
  74 }
  75 early_param("gbpages", parse_direct_gbpages_on);
  76
  77 /*
  78  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
  79  * physical space so we can cache the place of the first one and move
  80  * around without checking the pgd every time.
  81  */
  82
  83 void show_mem(void)
  84 {
  85         long i, total = 0, reserved = 0;
  86         long shared = 0, cached = 0;
  87         struct page *page;
  88         pg_data_t *pgdat;
  89
  90         printk(KERN_INFO "Mem-info:\n");
  91         show_free_areas();
  92         printk(KERN_INFO "Free swap:       %6ldkB\n",
  93                 nr_swap_pages << (PAGE_SHIFT-10));
  94
  95         for_each_online_pgdat(pgdat) {
  96                 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
  97                         /*
  98                          * This loop can take a while with 256 GB and
  99                          * 4k pages so defer the NMI watchdog:
 100                          */
 101                         if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
 102                                 touch_nmi_watchdog();
 103
 104                         if (!pfn_valid(pgdat->node_start_pfn + i))
 105                                 continue;
 106
 107                         page = pfn_to_page(pgdat->node_start_pfn + i);
 108                         total++;
 109                         if (PageReserved(page))
 110                                 reserved++;
 111                         else if (PageSwapCache(page))
 112                                 cached++;
 113                         else if (page_count(page))
 114                                 shared += page_count(page) - 1;
 115                 }
 116         }
 117         printk(KERN_INFO "%lu pages of RAM\n",          total);
 118         printk(KERN_INFO "%lu reserved pages\n",        reserved);
 119         printk(KERN_INFO "%lu pages shared\n",          shared);
 120         printk(KERN_INFO "%lu pages swap cached\n",     cached);
 121 }
 122
 123 int after_bootmem;
 124
 125 static __init void *spp_getpage(void)
 126 {
 127         void *ptr;
 128
 129         if (after_bootmem)
 130                 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
 131         else
 132                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 133
 134         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 135                 panic("set_pte_phys: cannot allocate page data %s\n",
 136                         after_bootmem ? "after bootmem" : "");
 137         }
 138
 139         pr_debug("spp_getpage %p\n", ptr);
 140
 141         return ptr;
 142 }
 143
 144 static __init void
 145 set_pte_phys(unsigned long vaddr, unsigned long phys, pgprot_t prot)
 146 {
 147         pgd_t *pgd;
 148         pud_t *pud;
 149         pmd_t *pmd;
 150         pte_t *pte, new_pte;
 151
 152         pr_debug("set_pte_phys %lx to %lx\n", vaddr, phys);
 153
 154         pgd = pgd_offset_k(vaddr);
 155         if (pgd_none(*pgd)) {
 156                 printk(KERN_ERR
 157                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
 158                 return;
 159         }
 160         pud = pud_offset(pgd, vaddr);
 161         if (pud_none(*pud)) {
 162                 pmd = (pmd_t *) spp_getpage();
 163                 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER));
 164                 if (pmd != pmd_offset(pud, 0)) {
 165                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
 166                                 pmd, pmd_offset(pud, 0));
 167                         return;
 168                 }
 169         }
 170         pmd = pmd_offset(pud, vaddr);
 171         if (pmd_none(*pmd)) {
 172                 pte = (pte_t *) spp_getpage();
 173                 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER));
 174                 if (pte != pte_offset_kernel(pmd, 0)) {
 175                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
 176                         return;
 177                 }
 178         }
 179         new_pte = pfn_pte(phys >> PAGE_SHIFT, prot);
 180
 181         pte = pte_offset_kernel(pmd, vaddr);
 182         if (!pte_none(*pte) &&
 183             pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask))
 184                 pte_ERROR(*pte);
 185         set_pte(pte, new_pte);
 186
 187         /*
 188          * It's enough to flush this one mapping.
 189          * (PGE mappings get flushed as well)
 190          */
 191         __flush_tlb_one(vaddr);
 192 }
 193
 194 /*
 195  * The head.S code sets up the kernel high mapping:
 196  *
 197  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
 198  *
 199  * phys_addr holds the negative offset to the kernel, which is added
 200  * to the compile time generated pmds. This results in invalid pmds up
 201  * to the point where we hit the physaddr 0 mapping.
 202  *
 203  * We limit the mappings to the region from _text to _end.  _end is
 204  * rounded up to the 2MB boundary. This catches the invalid pmds as
 205  * well, as they are located before _text:
 206  */
 207 void __init cleanup_highmap(void)
 208 {
 209         unsigned long vaddr = __START_KERNEL_map;
 210         unsigned long end = round_up((unsigned long)_end, PMD_SIZE) - 1;
 211         pmd_t *pmd = level2_kernel_pgt;
 212         pmd_t *last_pmd = pmd + PTRS_PER_PMD;
 213
 214         for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
 215                 if (!pmd_present(*pmd))
 216                         continue;
 217                 if (vaddr < (unsigned long) _text || vaddr > end)
 218                         set_pmd(pmd, __pmd(0));
 219         }
 220 }
 221
 222 /* NOTE: this is meant to be run only at boot */
 223 void __init
 224 __set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
 225 {
 226         unsigned long address = __fix_to_virt(idx);
 227
 228         if (idx >= __end_of_fixed_addresses) {
 229                 printk(KERN_ERR "Invalid __set_fixmap\n");
 230                 return;
 231         }
 232         set_pte_phys(address, phys, prot);
 233 }
 234
 235 static unsigned long __initdata table_start;
 236 static unsigned long __meminitdata table_end;
 237
 238 static __meminit void *alloc_low_page(unsigned long *phys)
 239 {
 240         unsigned long pfn = table_end++;
 241         void *adr;
 242
 243         if (after_bootmem) {
 244                 adr = (void *)get_zeroed_page(GFP_ATOMIC);
 245                 *phys = __pa(adr);
 246
 247                 return adr;
 248         }
 249
 250         if (pfn >= end_pfn)
 251                 panic("alloc_low_page: ran out of memory");
 252
 253         adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
 254         memset(adr, 0, PAGE_SIZE);
 255         *phys  = pfn * PAGE_SIZE;
 256         return adr;
 257 }
 258
 259 static __meminit void unmap_low_page(void *adr)
 260 {
 261         if (after_bootmem)
 262                 return;
 263
 264         early_iounmap(adr, PAGE_SIZE);
 265 }
 266
 267 /* Must run before zap_low_mappings */
 268 __meminit void *early_ioremap(unsigned long addr, unsigned long size)
 269 {
 270         pmd_t *pmd, *last_pmd;
 271         unsigned long vaddr;
 272         int i, pmds;
 273
 274         pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 275         vaddr = __START_KERNEL_map;
 276         pmd = level2_kernel_pgt;
 277         last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1;
 278
 279         for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) {
 280                 for (i = 0; i < pmds; i++) {
 281                         if (pmd_present(pmd[i]))
 282                                 goto continue_outer_loop;
 283                 }
 284                 vaddr += addr & ~PMD_MASK;
 285                 addr &= PMD_MASK;
 286
 287                 for (i = 0; i < pmds; i++, addr += PMD_SIZE)
 288                         set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
 289                 __flush_tlb_all();
 290
 291                 return (void *)vaddr;
 292 continue_outer_loop:
 293                 ;
 294         }
 295         printk(KERN_ERR "early_ioremap(0x%lx, %lu) failed\n", addr, size);
 296
 297         return NULL;
 298 }
 299
 300 /*
 301  * To avoid virtual aliases later:
 302  */
 303 __meminit void early_iounmap(void *addr, unsigned long size)
 304 {
 305         unsigned long vaddr;
 306         pmd_t *pmd;
 307         int i, pmds;
 308
 309         vaddr = (unsigned long)addr;
 310         pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 311         pmd = level2_kernel_pgt + pmd_index(vaddr);
 312
 313         for (i = 0; i < pmds; i++)
 314                 pmd_clear(pmd + i);
 315
 316         __flush_tlb_all();
 317 }
 318
 319 static void __meminit
 320 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
 321 {
 322         int i = pmd_index(address);
 323
 324         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
 325                 pmd_t *pmd = pmd_page + pmd_index(address);
 326
 327                 if (address >= end) {
 328                         if (!after_bootmem) {
 329                                 for (; i < PTRS_PER_PMD; i++, pmd++)
 330                                         set_pmd(pmd, __pmd(0));
 331                         }
 332                         break;
 333                 }
 334
 335                 if (pmd_val(*pmd))
 336                         continue;
 337
 338                 set_pte((pte_t *)pmd,
 339                         pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 340         }
 341 }
 342
 343 static void __meminit
 344 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
 345 {
 346         pmd_t *pmd = pmd_offset(pud, 0);
 347         spin_lock(&init_mm.page_table_lock);
 348         phys_pmd_init(pmd, address, end);
 349         spin_unlock(&init_mm.page_table_lock);
 350         __flush_tlb_all();
 351 }
 352
 353 static void __meminit
 354 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
 355 {
 356         int i = pud_index(addr);
 357
 358         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
 359                 unsigned long pmd_phys;
 360                 pud_t *pud = pud_page + pud_index(addr);
 361                 pmd_t *pmd;
 362
 363                 if (addr >= end)
 364                         break;
 365
 366                 if (!after_bootmem &&
 367                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
 368                         set_pud(pud, __pud(0));
 369                         continue;
 370                 }
 371
 372                 if (pud_val(*pud)) {
 373                         if (!pud_large(*pud))
 374                                 phys_pmd_update(pud, addr, end);
 375                         continue;
 376                 }
 377
 378                 if (direct_gbpages) {
 379                         set_pte((pte_t *)pud,
 380                                 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 381                         continue;
 382                 }
 383
 384                 pmd = alloc_low_page(&pmd_phys);
 385
 386                 spin_lock(&init_mm.page_table_lock);
 387                 set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE));
 388                 phys_pmd_init(pmd, addr, end);
 389                 spin_unlock(&init_mm.page_table_lock);
 390
 391                 unmap_low_page(pmd);
 392         }
 393         __flush_tlb_all();
 394 }
 395
 396 static void __init find_early_table_space(unsigned long end)
 397 {
 398         unsigned long puds, pmds, tables, start;
 399
 400         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 401         tables = round_up(puds * sizeof(pud_t), PAGE_SIZE);
 402         if (!direct_gbpages) {
 403                 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 404                 tables += round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
 405         }
 406
 407         /*
 408          * RED-PEN putting page tables only on node 0 could
 409          * cause a hotspot and fill up ZONE_DMA. The page tables
 410          * need roughly 0.5KB per GB.
 411          */
 412         start = 0x8000;
 413         table_start = find_e820_area(start, end, tables, PAGE_SIZE);
 414         if (table_start == -1UL)
 415                 panic("Cannot find space for the kernel page tables");
 416
 417         table_start >>= PAGE_SHIFT;
 418         table_end = table_start;
 419
 420         early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n",
 421                 end, table_start << PAGE_SHIFT,
 422                 (table_start << PAGE_SHIFT) + tables);
 423 }
 424
 425 static void __init init_gbpages(void)
 426 {
 427         if (direct_gbpages && cpu_has_gbpages)
 428                 printk(KERN_INFO "Using GB pages for direct mapping\n");
 429         else
 430                 direct_gbpages = 0;
 431 }
 432
 433 /*
 434  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 435  * This runs before bootmem is initialized and gets pages directly from
 436  * the physical memory. To access them they are temporarily mapped.
 437  */
 438 void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
 439 {
 440         unsigned long next;
 441
 442         pr_debug("init_memory_mapping\n");
 443
 444         /*
 445          * Find space for the kernel direct mapping tables.
 446          *
 447          * Later we should allocate these tables in the local node of the
 448          * memory mapped. Unfortunately this is done currently before the
 449          * nodes are discovered.
 450          */
 451         if (!after_bootmem) {
 452                 init_gbpages();
 453                 find_early_table_space(end);
 454         }
 455
 456         start = (unsigned long)__va(start);
 457         end = (unsigned long)__va(end);
 458
 459         for (; start < end; start = next) {
 460                 pgd_t *pgd = pgd_offset_k(start);
 461                 unsigned long pud_phys;
 462                 pud_t *pud;
 463
 464                 if (after_bootmem)
 465                         pud = pud_offset(pgd, start & PGDIR_MASK);
 466                 else
 467                         pud = alloc_low_page(&pud_phys);
 468
 469                 next = start + PGDIR_SIZE;
 470                 if (next > end)
 471                         next = end;
 472                 phys_pud_init(pud, __pa(start), __pa(next));
 473                 if (!after_bootmem)
 474                         set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys));
 475                 unmap_low_page(pud);
 476         }
 477
 478         if (!after_bootmem)
 479                 mmu_cr4_features = read_cr4();
 480         __flush_tlb_all();
 481
 482         if (!after_bootmem)
 483                 reserve_early(table_start << PAGE_SHIFT,
 484                                  table_end << PAGE_SHIFT, "PGTABLE");
 485 }
 486
 487 #ifndef CONFIG_NUMA
 488 void __init paging_init(void)
 489 {
 490         unsigned long max_zone_pfns[MAX_NR_ZONES];
 491
 492         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 493         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 494         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 495         max_zone_pfns[ZONE_NORMAL] = end_pfn;
 496
 497         memory_present(0, 0, end_pfn);
 498         sparse_init();
 499         free_area_init_nodes(max_zone_pfns);
 500 }
 501 #endif
 502
 503 /*
 504  * Memory hotplug specific functions
 505  */
 506 void online_page(struct page *page)
 507 {
 508         ClearPageReserved(page);
 509         init_page_count(page);
 510         __free_page(page);
 511         totalram_pages++;
 512         num_physpages++;
 513 }
 514
 515 #ifdef CONFIG_MEMORY_HOTPLUG
 516 /*
 517  * Memory is added always to NORMAL zone. This means you will never get
 518  * additional DMA/DMA32 memory.
 519  */
 520 int arch_add_memory(int nid, u64 start, u64 size)
 521 {
 522         struct pglist_data *pgdat = NODE_DATA(nid);
 523         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 524         unsigned long start_pfn = start >> PAGE_SHIFT;
 525         unsigned long nr_pages = size >> PAGE_SHIFT;
 526         int ret;
 527
 528         init_memory_mapping(start, start + size-1);
 529
 530         ret = __add_pages(zone, start_pfn, nr_pages);
 531         WARN_ON(1);
 532
 533         return ret;
 534 }
 535 EXPORT_SYMBOL_GPL(arch_add_memory);
 536
 537 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
 538 int memory_add_physaddr_to_nid(u64 start)
 539 {
 540         return 0;
 541 }
 542 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 543 #endif
 544
 545 #endif /* CONFIG_MEMORY_HOTPLUG */
 546
 547 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
 548                          kcore_modules, kcore_vsyscall;
 549
 550 void __init mem_init(void)
 551 {
 552         long codesize, reservedpages, datasize, initsize;
 553
 554         pci_iommu_alloc();
 555
 556         /* clear_bss() already clear the empty_zero_page */
 557
 558         reservedpages = 0;
 559
 560         /* this will put all low memory onto the freelists */
 561 #ifdef CONFIG_NUMA
 562         totalram_pages = numa_free_all_bootmem();
 563 #else
 564         totalram_pages = free_all_bootmem();
 565 #endif
 566         reservedpages = end_pfn - totalram_pages -
 567                                         absent_pages_in_range(0, end_pfn);
 568         after_bootmem = 1;
 569
 570         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 571         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 572         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 573
 574         /* Register memory areas for /proc/kcore */
 575         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 576         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 577                    VMALLOC_END-VMALLOC_START);
 578         kclist_add(&kcore_kernel, &_stext, _end - _stext);
 579         kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
 580         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 581                                  VSYSCALL_END - VSYSCALL_START);
 582
 583         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 584                                 "%ldk reserved, %ldk data, %ldk init)\n",
 585                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 586                 end_pfn << (PAGE_SHIFT-10),
 587                 codesize >> 10,
 588                 reservedpages << (PAGE_SHIFT-10),
 589                 datasize >> 10,
 590                 initsize >> 10);
 591
 592         cpa_init();
 593 }
 594
 595 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 596 {
 597         unsigned long addr = begin;
 598
 599         if (addr >= end)
 600                 return;
 601
 602         /*
 603          * If debugging page accesses then do not free this memory but
 604          * mark them not present - any buggy init-section access will
 605          * create a kernel page fault:
 606          */
 607 #ifdef CONFIG_DEBUG_PAGEALLOC
 608         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
 609                 begin, PAGE_ALIGN(end));
 610         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 611 #else
 612         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 613
 614         for (; addr < end; addr += PAGE_SIZE) {
 615                 ClearPageReserved(virt_to_page(addr));
 616                 init_page_count(virt_to_page(addr));
 617                 memset((void *)(addr & ~(PAGE_SIZE-1)),
 618                         POISON_FREE_INITMEM, PAGE_SIZE);
 619                 free_page(addr);
 620                 totalram_pages++;
 621         }
 622 #endif
 623 }
 624
 625 void free_initmem(void)
 626 {
 627         free_init_pages("unused kernel memory",
 628                         (unsigned long)(&__init_begin),
 629                         (unsigned long)(&__init_end));
 630 }
 631
 632 #ifdef CONFIG_DEBUG_RODATA
 633 const int rodata_test_data = 0xC3;
 634 EXPORT_SYMBOL_GPL(rodata_test_data);
 635
 636 void mark_rodata_ro(void)
 637 {
 638         unsigned long start = PFN_ALIGN(_stext), end = PFN_ALIGN(__end_rodata);
 639
 640         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 641                (end - start) >> 10);
 642         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 643
 644         /*
 645          * The rodata section (but not the kernel text!) should also be
 646          * not-executable.
 647          */
 648         start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
 649         set_memory_nx(start, (end - start) >> PAGE_SHIFT);
 650
 651         rodata_test();
 652
 653 #ifdef CONFIG_CPA_DEBUG
 654         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
 655         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 656
 657         printk(KERN_INFO "Testing CPA: again\n");
 658         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 659 #endif
 660 }
 661
 662 #endif
 663
 664 #ifdef CONFIG_BLK_DEV_INITRD
 665 void free_initrd_mem(unsigned long start, unsigned long end)
 666 {
 667         free_init_pages("initrd memory", start, end);
 668 }
 669 #endif
 670
 671 void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
 672 {
 673 #ifdef CONFIG_NUMA
 674         int nid = phys_to_nid(phys);
 675 #endif
 676         unsigned long pfn = phys >> PAGE_SHIFT;
 677
 678         if (pfn >= end_pfn) {
 679                 /*
 680                  * This can happen with kdump kernels when accessing
 681                  * firmware tables:
 682                  */
 683                 if (pfn < end_pfn_map)
 684                         return;
 685
 686                 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n",
 687                                 phys, len);
 688                 return;
 689         }
 690
 691         /* Should check here against the e820 map to avoid double free */
 692 #ifdef CONFIG_NUMA
 693         reserve_bootmem_node(NODE_DATA(nid), phys, len, BOOTMEM_DEFAULT);
 694 #else
 695         reserve_bootmem(phys, len, BOOTMEM_DEFAULT);
 696 #endif
 697         if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
 698                 dma_reserve += len / PAGE_SIZE;
 699                 set_dma_reserve(dma_reserve);
 700         }
 701 }
 702
 703 int kern_addr_valid(unsigned long addr)
 704 {
 705         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 706         pgd_t *pgd;
 707         pud_t *pud;
 708         pmd_t *pmd;
 709         pte_t *pte;
 710
 711         if (above != 0 && above != -1UL)
 712                 return 0;
 713
 714         pgd = pgd_offset_k(addr);
 715         if (pgd_none(*pgd))
 716                 return 0;
 717
 718         pud = pud_offset(pgd, addr);
 719         if (pud_none(*pud))
 720                 return 0;
 721
 722         pmd = pmd_offset(pud, addr);
 723         if (pmd_none(*pmd))
 724                 return 0;
 725
 726         if (pmd_large(*pmd))
 727                 return pfn_valid(pmd_pfn(*pmd));
 728
 729         pte = pte_offset_kernel(pmd, addr);
 730         if (pte_none(*pte))
 731                 return 0;
 732
 733         return pfn_valid(pte_pfn(*pte));
 734 }
 735
 736 /*
 737  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 738  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 739  * not need special handling anymore:
 740  */
 741 static struct vm_area_struct gate_vma = {
 742         .vm_start       = VSYSCALL_START,
 743         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 744         .vm_page_prot   = PAGE_READONLY_EXEC,
 745         .vm_flags       = VM_READ | VM_EXEC
 746 };
 747
 748 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 749 {
 750 #ifdef CONFIG_IA32_EMULATION
 751         if (test_tsk_thread_flag(tsk, TIF_IA32))
 752                 return NULL;
 753 #endif
 754         return &gate_vma;
 755 }
 756
 757 int in_gate_area(struct task_struct *task, unsigned long addr)
 758 {
 759         struct vm_area_struct *vma = get_gate_vma(task);
 760
 761         if (!vma)
 762                 return 0;
 763
 764         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 765 }
 766
 767 /*
 768  * Use this when you have no reliable task/vma, typically from interrupt
 769  * context. It is less reliable than using the task's vma and may give
 770  * false positives:
 771  */
 772 int in_gate_area_no_task(unsigned long addr)
 773 {
 774         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 775 }
 776
 777 const char *arch_vma_name(struct vm_area_struct *vma)
 778 {
 779         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 780                 return "[vdso]";
 781         if (vma == &gate_vma)
 782                 return "[vsyscall]";
 783         return NULL;
 784 }
 785
 786 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 787 /*
 788  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 789  */
 790 int __meminit
 791 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 792 {
 793         unsigned long addr = (unsigned long)start_page;
 794         unsigned long end = (unsigned long)(start_page + size);
 795         unsigned long next;
 796         pgd_t *pgd;
 797         pud_t *pud;
 798         pmd_t *pmd;
 799
 800         for (; addr < end; addr = next) {
 801                 next = pmd_addr_end(addr, end);
 802
 803                 pgd = vmemmap_pgd_populate(addr, node);
 804                 if (!pgd)
 805                         return -ENOMEM;
 806
 807                 pud = vmemmap_pud_populate(pgd, addr, node);
 808                 if (!pud)
 809                         return -ENOMEM;
 810
 811                 pmd = pmd_offset(pud, addr);
 812                 if (pmd_none(*pmd)) {
 813                         pte_t entry;
 814                         void *p;
 815
 816                         p = vmemmap_alloc_block(PMD_SIZE, node);
 817                         if (!p)
 818                                 return -ENOMEM;
 819
 820                         entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
 821                                                         PAGE_KERNEL_LARGE);
 822                         set_pmd(pmd, __pmd(pte_val(entry)));
 823
 824                         printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n",
 825                                 addr, addr + PMD_SIZE - 1, p, node);
 826                 } else {
 827                         vmemmap_verify((pte_t *)pmd, node, addr, next);
 828                 }
 829         }
 830         return 0;
 831 }
 832 #endif