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 #ifndef Dprintk
  50 # define Dprintk(x...)
  51 #endif
  52
  53 const struct dma_mapping_ops *dma_ops;
  54 EXPORT_SYMBOL(dma_ops);
  55
  56 static unsigned long dma_reserve __initdata;
  57
  58 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  59
  60 /*
  61  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
  62  * physical space so we can cache the place of the first one and move
  63  * around without checking the pgd every time.
  64  */
  65
  66 void show_mem(void)
  67 {
  68         long i, total = 0, reserved = 0;
  69         long shared = 0, cached = 0;
  70         struct page *page;
  71         pg_data_t *pgdat;
  72
  73         printk(KERN_INFO "Mem-info:\n");
  74         show_free_areas();
  75         printk(KERN_INFO "Free swap:       %6ldkB\n",
  76                 nr_swap_pages << (PAGE_SHIFT-10));
  77
  78         for_each_online_pgdat(pgdat) {
  79                 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
  80                         /*
  81                          * This loop can take a while with 256 GB and
  82                          * 4k pages so defer the NMI watchdog:
  83                          */
  84                         if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
  85                                 touch_nmi_watchdog();
  86
  87                         if (!pfn_valid(pgdat->node_start_pfn + i))
  88                                 continue;
  89
  90                         page = pfn_to_page(pgdat->node_start_pfn + i);
  91                         total++;
  92                         if (PageReserved(page))
  93                                 reserved++;
  94                         else if (PageSwapCache(page))
  95                                 cached++;
  96                         else if (page_count(page))
  97                                 shared += page_count(page) - 1;
  98                 }
  99         }
 100         printk(KERN_INFO "%lu pages of RAM\n",          total);
 101         printk(KERN_INFO "%lu reserved pages\n",        reserved);
 102         printk(KERN_INFO "%lu pages shared\n",          shared);
 103         printk(KERN_INFO "%lu pages swap cached\n",     cached);
 104 }
 105
 106 int after_bootmem;
 107
 108 static __init void *spp_getpage(void)
 109 {
 110         void *ptr;
 111
 112         if (after_bootmem)
 113                 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
 114         else
 115                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 116
 117         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 118                 panic("set_pte_phys: cannot allocate page data %s\n",
 119                         after_bootmem ? "after bootmem" : "");
 120         }
 121
 122         Dprintk("spp_getpage %p\n", ptr);
 123
 124         return ptr;
 125 }
 126
 127 static __init void
 128 set_pte_phys(unsigned long vaddr, unsigned long phys, pgprot_t prot)
 129 {
 130         pgd_t *pgd;
 131         pud_t *pud;
 132         pmd_t *pmd;
 133         pte_t *pte, new_pte;
 134
 135         Dprintk("set_pte_phys %lx to %lx\n", vaddr, phys);
 136
 137         pgd = pgd_offset_k(vaddr);
 138         if (pgd_none(*pgd)) {
 139                 printk("PGD FIXMAP MISSING, it should be setup in head.S!\n");
 140                 return;
 141         }
 142         pud = pud_offset(pgd, vaddr);
 143         if (pud_none(*pud)) {
 144                 pmd = (pmd_t *) spp_getpage();
 145                 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER));
 146                 if (pmd != pmd_offset(pud, 0)) {
 147                         printk("PAGETABLE BUG #01! %p <-> %p\n",
 148                                 pmd, pmd_offset(pud, 0));
 149                         return;
 150                 }
 151         }
 152         pmd = pmd_offset(pud, vaddr);
 153         if (pmd_none(*pmd)) {
 154                 pte = (pte_t *) spp_getpage();
 155                 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER));
 156                 if (pte != pte_offset_kernel(pmd, 0)) {
 157                         printk("PAGETABLE BUG #02!\n");
 158                         return;
 159                 }
 160         }
 161         new_pte = pfn_pte(phys >> PAGE_SHIFT, prot);
 162
 163         pte = pte_offset_kernel(pmd, vaddr);
 164         if (!pte_none(*pte) &&
 165             pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask))
 166                 pte_ERROR(*pte);
 167         set_pte(pte, new_pte);
 168
 169         /*
 170          * It's enough to flush this one mapping.
 171          * (PGE mappings get flushed as well)
 172          */
 173         __flush_tlb_one(vaddr);
 174 }
 175
 176 /* NOTE: this is meant to be run only at boot */
 177 void __init
 178 __set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
 179 {
 180         unsigned long address = __fix_to_virt(idx);
 181
 182         if (idx >= __end_of_fixed_addresses) {
 183                 printk("Invalid __set_fixmap\n");
 184                 return;
 185         }
 186         set_pte_phys(address, phys, prot);
 187 }
 188
 189 static unsigned long __initdata table_start;
 190 static unsigned long __meminitdata table_end;
 191
 192 static __meminit void *alloc_low_page(unsigned long *phys)
 193 {
 194         unsigned long pfn = table_end++;
 195         void *adr;
 196
 197         if (after_bootmem) {
 198                 adr = (void *)get_zeroed_page(GFP_ATOMIC);
 199                 *phys = __pa(adr);
 200
 201                 return adr;
 202         }
 203
 204         if (pfn >= end_pfn)
 205                 panic("alloc_low_page: ran out of memory");
 206
 207         adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
 208         memset(adr, 0, PAGE_SIZE);
 209         *phys  = pfn * PAGE_SIZE;
 210         return adr;
 211 }
 212
 213 static __meminit void unmap_low_page(void *adr)
 214 {
 215         if (after_bootmem)
 216                 return;
 217
 218         early_iounmap(adr, PAGE_SIZE);
 219 }
 220
 221 /* Must run before zap_low_mappings */
 222 __meminit void *early_ioremap(unsigned long addr, unsigned long size)
 223 {
 224         pmd_t *pmd, *last_pmd;
 225         unsigned long vaddr;
 226         int i, pmds;
 227
 228         pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 229         vaddr = __START_KERNEL_map;
 230         pmd = level2_kernel_pgt;
 231         last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1;
 232
 233         for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) {
 234                 for (i = 0; i < pmds; i++) {
 235                         if (pmd_present(pmd[i]))
 236                                 goto continue_outer_loop;
 237                 }
 238                 vaddr += addr & ~PMD_MASK;
 239                 addr &= PMD_MASK;
 240
 241                 for (i = 0; i < pmds; i++, addr += PMD_SIZE)
 242                         set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
 243                 __flush_tlb_all();
 244
 245                 return (void *)vaddr;
 246 continue_outer_loop:
 247                 ;
 248         }
 249         printk("early_ioremap(0x%lx, %lu) failed\n", addr, size);
 250
 251         return NULL;
 252 }
 253
 254 /*
 255  * To avoid virtual aliases later:
 256  */
 257 __meminit void early_iounmap(void *addr, unsigned long size)
 258 {
 259         unsigned long vaddr;
 260         pmd_t *pmd;
 261         int i, pmds;
 262
 263         vaddr = (unsigned long)addr;
 264         pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 265         pmd = level2_kernel_pgt + pmd_index(vaddr);
 266
 267         for (i = 0; i < pmds; i++)
 268                 pmd_clear(pmd + i);
 269
 270         __flush_tlb_all();
 271 }
 272
 273 static void __meminit
 274 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
 275 {
 276         int i = pmd_index(address);
 277
 278         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
 279                 unsigned long entry;
 280                 pmd_t *pmd = pmd_page + pmd_index(address);
 281
 282                 if (address >= end) {
 283                         if (!after_bootmem) {
 284                                 for (; i < PTRS_PER_PMD; i++, pmd++)
 285                                         set_pmd(pmd, __pmd(0));
 286                         }
 287                         break;
 288                 }
 289
 290                 if (pmd_val(*pmd))
 291                         continue;
 292
 293                 entry = __PAGE_KERNEL_LARGE|_PAGE_GLOBAL|address;
 294                 entry &= __supported_pte_mask;
 295                 set_pmd(pmd, __pmd(entry));
 296         }
 297 }
 298
 299 static void __meminit
 300 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
 301 {
 302         pmd_t *pmd = pmd_offset(pud, 0);
 303         spin_lock(&init_mm.page_table_lock);
 304         phys_pmd_init(pmd, address, end);
 305         spin_unlock(&init_mm.page_table_lock);
 306         __flush_tlb_all();
 307 }
 308
 309 static void __meminit
 310 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
 311 {
 312         int i = pud_index(addr);
 313
 314         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
 315                 unsigned long pmd_phys;
 316                 pud_t *pud = pud_page + pud_index(addr);
 317                 pmd_t *pmd;
 318
 319                 if (addr >= end)
 320                         break;
 321
 322                 if (!after_bootmem &&
 323                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
 324                         set_pud(pud, __pud(0));
 325                         continue;
 326                 }
 327
 328                 if (pud_val(*pud)) {
 329                         phys_pmd_update(pud, addr, end);
 330                         continue;
 331                 }
 332
 333                 pmd = alloc_low_page(&pmd_phys);
 334
 335                 spin_lock(&init_mm.page_table_lock);
 336                 set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE));
 337                 phys_pmd_init(pmd, addr, end);
 338                 spin_unlock(&init_mm.page_table_lock);
 339
 340                 unmap_low_page(pmd);
 341         }
 342         __flush_tlb_all();
 343 }
 344
 345 static void __init find_early_table_space(unsigned long end)
 346 {
 347         unsigned long puds, pmds, tables, start;
 348
 349         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 350         pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 351         tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) +
 352                  round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
 353
 354         /*
 355          * RED-PEN putting page tables only on node 0 could
 356          * cause a hotspot and fill up ZONE_DMA. The page tables
 357          * need roughly 0.5KB per GB.
 358          */
 359         start = 0x8000;
 360         table_start = find_e820_area(start, end, tables);
 361         if (table_start == -1UL)
 362                 panic("Cannot find space for the kernel page tables");
 363
 364         table_start >>= PAGE_SHIFT;
 365         table_end = table_start;
 366
 367         early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n",
 368                 end, table_start << PAGE_SHIFT,
 369                 (table_start << PAGE_SHIFT) + tables);
 370 }
 371
 372 /*
 373  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 374  * This runs before bootmem is initialized and gets pages directly from
 375  * the physical memory. To access them they are temporarily mapped.
 376  */
 377 void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
 378 {
 379         unsigned long next;
 380
 381         Dprintk("init_memory_mapping\n");
 382
 383         /*
 384          * Find space for the kernel direct mapping tables.
 385          *
 386          * Later we should allocate these tables in the local node of the
 387          * memory mapped. Unfortunately this is done currently before the
 388          * nodes are discovered.
 389          */
 390         if (!after_bootmem)
 391                 find_early_table_space(end);
 392
 393         start = (unsigned long)__va(start);
 394         end = (unsigned long)__va(end);
 395
 396         for (; start < end; start = next) {
 397                 pgd_t *pgd = pgd_offset_k(start);
 398                 unsigned long pud_phys;
 399                 pud_t *pud;
 400
 401                 if (after_bootmem)
 402                         pud = pud_offset(pgd, start & PGDIR_MASK);
 403                 else
 404                         pud = alloc_low_page(&pud_phys);
 405
 406                 next = start + PGDIR_SIZE;
 407                 if (next > end)
 408                         next = end;
 409                 phys_pud_init(pud, __pa(start), __pa(next));
 410                 if (!after_bootmem)
 411                         set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys));
 412                 unmap_low_page(pud);
 413         }
 414
 415         if (!after_bootmem)
 416                 mmu_cr4_features = read_cr4();
 417         __flush_tlb_all();
 418
 419         reserve_early(table_start << PAGE_SHIFT, table_end << PAGE_SHIFT);
 420 }
 421
 422 #ifndef CONFIG_NUMA
 423 void __init paging_init(void)
 424 {
 425         unsigned long max_zone_pfns[MAX_NR_ZONES];
 426
 427         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 428         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 429         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 430         max_zone_pfns[ZONE_NORMAL] = end_pfn;
 431
 432         memory_present(0, 0, end_pfn);
 433         sparse_init();
 434         free_area_init_nodes(max_zone_pfns);
 435 }
 436 #endif
 437
 438 /*
 439  * Unmap a kernel mapping if it exists. This is useful to avoid
 440  * prefetches from the CPU leading to inconsistent cache lines.
 441  * address and size must be aligned to 2MB boundaries.
 442  * Does nothing when the mapping doesn't exist.
 443  */
 444 void __init clear_kernel_mapping(unsigned long address, unsigned long size)
 445 {
 446         unsigned long end = address + size;
 447
 448         BUG_ON(address & ~LARGE_PAGE_MASK);
 449         BUG_ON(size & ~LARGE_PAGE_MASK);
 450
 451         for (; address < end; address += LARGE_PAGE_SIZE) {
 452                 pgd_t *pgd = pgd_offset_k(address);
 453                 pud_t *pud;
 454                 pmd_t *pmd;
 455
 456                 if (pgd_none(*pgd))
 457                         continue;
 458
 459                 pud = pud_offset(pgd, address);
 460                 if (pud_none(*pud))
 461                         continue;
 462
 463                 pmd = pmd_offset(pud, address);
 464                 if (!pmd || pmd_none(*pmd))
 465                         continue;
 466
 467                 if (!(pmd_val(*pmd) & _PAGE_PSE)) {
 468                         /*
 469                          * Could handle this, but it should not happen
 470                          * currently:
 471                          */
 472                         printk(KERN_ERR "clear_kernel_mapping: "
 473                                 "mapping has been split. will leak memory\n");
 474                         pmd_ERROR(*pmd);
 475                 }
 476                 set_pmd(pmd, __pmd(0));
 477         }
 478         __flush_tlb_all();
 479 }
 480
 481 /*
 482  * Memory hotplug specific functions
 483  */
 484 void online_page(struct page *page)
 485 {
 486         ClearPageReserved(page);
 487         init_page_count(page);
 488         __free_page(page);
 489         totalram_pages++;
 490         num_physpages++;
 491 }
 492
 493 #ifdef CONFIG_MEMORY_HOTPLUG
 494 /*
 495  * Memory is added always to NORMAL zone. This means you will never get
 496  * additional DMA/DMA32 memory.
 497  */
 498 int arch_add_memory(int nid, u64 start, u64 size)
 499 {
 500         struct pglist_data *pgdat = NODE_DATA(nid);
 501         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 502         unsigned long start_pfn = start >> PAGE_SHIFT;
 503         unsigned long nr_pages = size >> PAGE_SHIFT;
 504         int ret;
 505
 506         init_memory_mapping(start, start + size-1);
 507
 508         ret = __add_pages(zone, start_pfn, nr_pages);
 509         if (ret)
 510                 printk("%s: Problem encountered in __add_pages!\n", __func__);
 511
 512         return ret;
 513 }
 514 EXPORT_SYMBOL_GPL(arch_add_memory);
 515
 516 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
 517 int memory_add_physaddr_to_nid(u64 start)
 518 {
 519         return 0;
 520 }
 521 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 522 #endif
 523
 524 #endif /* CONFIG_MEMORY_HOTPLUG */
 525
 526 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
 527                          kcore_modules, kcore_vsyscall;
 528
 529 void __init mem_init(void)
 530 {
 531         long codesize, reservedpages, datasize, initsize;
 532
 533         pci_iommu_alloc();
 534
 535         /* clear_bss() already clear the empty_zero_page */
 536
 537         /* temporary debugging - double check it's true: */
 538         {
 539                 int i;
 540
 541                 for (i = 0; i < 1024; i++)
 542                         WARN_ON_ONCE(empty_zero_page[i]);
 543         }
 544
 545         reservedpages = 0;
 546
 547         /* this will put all low memory onto the freelists */
 548 #ifdef CONFIG_NUMA
 549         totalram_pages = numa_free_all_bootmem();
 550 #else
 551         totalram_pages = free_all_bootmem();
 552 #endif
 553         reservedpages = end_pfn - totalram_pages -
 554                                         absent_pages_in_range(0, end_pfn);
 555         after_bootmem = 1;
 556
 557         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 558         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 559         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 560
 561         /* Register memory areas for /proc/kcore */
 562         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 563         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 564                    VMALLOC_END-VMALLOC_START);
 565         kclist_add(&kcore_kernel, &_stext, _end - _stext);
 566         kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
 567         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 568                                  VSYSCALL_END - VSYSCALL_START);
 569
 570         printk("Memory: %luk/%luk available (%ldk kernel code, "
 571                                 "%ldk reserved, %ldk data, %ldk init)\n",
 572                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 573                 end_pfn << (PAGE_SHIFT-10),
 574                 codesize >> 10,
 575                 reservedpages << (PAGE_SHIFT-10),
 576                 datasize >> 10,
 577                 initsize >> 10);
 578 }
 579
 580 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 581 {
 582         unsigned long addr;
 583
 584         if (begin >= end)
 585                 return;
 586
 587         /*
 588          * If debugging page accesses then do not free this memory but
 589          * mark them not present - any buggy init-section access will
 590          * create a kernel page fault:
 591          */
 592 #ifdef CONFIG_DEBUG_PAGEALLOC
 593         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
 594                 begin, PAGE_ALIGN(end));
 595         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 596 #else
 597         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 598
 599         for (addr = begin; addr < end; addr += PAGE_SIZE) {
 600                 ClearPageReserved(virt_to_page(addr));
 601                 init_page_count(virt_to_page(addr));
 602                 memset((void *)(addr & ~(PAGE_SIZE-1)),
 603                         POISON_FREE_INITMEM, PAGE_SIZE);
 604                 free_page(addr);
 605                 totalram_pages++;
 606         }
 607 #endif
 608 }
 609
 610 void free_initmem(void)
 611 {
 612         free_init_pages("unused kernel memory",
 613                         (unsigned long)(&__init_begin),
 614                         (unsigned long)(&__init_end));
 615 }
 616
 617 #ifdef CONFIG_DEBUG_RODATA
 618 const int rodata_test_data = 0xC3;
 619 EXPORT_SYMBOL_GPL(rodata_test_data);
 620
 621 void mark_rodata_ro(void)
 622 {
 623         unsigned long start = (unsigned long)_stext, end;
 624
 625 #ifdef CONFIG_HOTPLUG_CPU
 626         /* It must still be possible to apply SMP alternatives. */
 627         if (num_possible_cpus() > 1)
 628                 start = (unsigned long)_etext;
 629 #endif
 630
 631 #ifdef CONFIG_KPROBES
 632         start = (unsigned long)__start_rodata;
 633 #endif
 634
 635         end = (unsigned long)__end_rodata;
 636         start = (start + PAGE_SIZE - 1) & PAGE_MASK;
 637         end &= PAGE_MASK;
 638         if (end <= start)
 639                 return;
 640
 641         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 642
 643         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 644                (end - start) >> 10);
 645
 646         rodata_test();
 647
 648 #ifdef CONFIG_CPA_DEBUG
 649         printk("Testing CPA: undo %lx-%lx\n", start, end);
 650         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 651
 652         printk("Testing CPA: again\n");
 653         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 654 #endif
 655 }
 656 #endif
 657
 658 #ifdef CONFIG_BLK_DEV_INITRD
 659 void free_initrd_mem(unsigned long start, unsigned long end)
 660 {
 661         free_init_pages("initrd memory", start, end);
 662 }
 663 #endif
 664
 665 void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
 666 {
 667 #ifdef CONFIG_NUMA
 668         int nid = phys_to_nid(phys);
 669 #endif
 670         unsigned long pfn = phys >> PAGE_SHIFT;
 671
 672         if (pfn >= end_pfn) {
 673                 /*
 674                  * This can happen with kdump kernels when accessing
 675                  * firmware tables:
 676                  */
 677                 if (pfn < end_pfn_map)
 678                         return;
 679
 680                 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n",
 681                                 phys, len);
 682                 return;
 683         }
 684
 685         /* Should check here against the e820 map to avoid double free */
 686 #ifdef CONFIG_NUMA
 687         reserve_bootmem_node(NODE_DATA(nid), phys, len);
 688 #else
 689         reserve_bootmem(phys, len);
 690 #endif
 691         if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
 692                 dma_reserve += len / PAGE_SIZE;
 693                 set_dma_reserve(dma_reserve);
 694         }
 695 }
 696
 697 int kern_addr_valid(unsigned long addr)
 698 {
 699         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 700         pgd_t *pgd;
 701         pud_t *pud;
 702         pmd_t *pmd;
 703         pte_t *pte;
 704
 705         if (above != 0 && above != -1UL)
 706                 return 0;
 707
 708         pgd = pgd_offset_k(addr);
 709         if (pgd_none(*pgd))
 710                 return 0;
 711
 712         pud = pud_offset(pgd, addr);
 713         if (pud_none(*pud))
 714                 return 0;
 715
 716         pmd = pmd_offset(pud, addr);
 717         if (pmd_none(*pmd))
 718                 return 0;
 719
 720         if (pmd_large(*pmd))
 721                 return pfn_valid(pmd_pfn(*pmd));
 722
 723         pte = pte_offset_kernel(pmd, addr);
 724         if (pte_none(*pte))
 725                 return 0;
 726
 727         return pfn_valid(pte_pfn(*pte));
 728 }
 729
 730 /*
 731  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 732  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 733  * not need special handling anymore:
 734  */
 735 static struct vm_area_struct gate_vma = {
 736         .vm_start       = VSYSCALL_START,
 737         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 738         .vm_page_prot   = PAGE_READONLY_EXEC,
 739         .vm_flags       = VM_READ | VM_EXEC
 740 };
 741
 742 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 743 {
 744 #ifdef CONFIG_IA32_EMULATION
 745         if (test_tsk_thread_flag(tsk, TIF_IA32))
 746                 return NULL;
 747 #endif
 748         return &gate_vma;
 749 }
 750
 751 int in_gate_area(struct task_struct *task, unsigned long addr)
 752 {
 753         struct vm_area_struct *vma = get_gate_vma(task);
 754
 755         if (!vma)
 756                 return 0;
 757
 758         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 759 }
 760
 761 /*
 762  * Use this when you have no reliable task/vma, typically from interrupt
 763  * context. It is less reliable than using the task's vma and may give
 764  * false positives:
 765  */
 766 int in_gate_area_no_task(unsigned long addr)
 767 {
 768         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 769 }
 770
 771 const char *arch_vma_name(struct vm_area_struct *vma)
 772 {
 773         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 774                 return "[vdso]";
 775         if (vma == &gate_vma)
 776                 return "[vsyscall]";
 777         return NULL;
 778 }
 779
 780 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 781 /*
 782  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 783  */
 784 int __meminit
 785 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 786 {
 787         unsigned long addr = (unsigned long)start_page;
 788         unsigned long end = (unsigned long)(start_page + size);
 789         unsigned long next;
 790         pgd_t *pgd;
 791         pud_t *pud;
 792         pmd_t *pmd;
 793
 794         for (; addr < end; addr = next) {
 795                 next = pmd_addr_end(addr, end);
 796
 797                 pgd = vmemmap_pgd_populate(addr, node);
 798                 if (!pgd)
 799                         return -ENOMEM;
 800
 801                 pud = vmemmap_pud_populate(pgd, addr, node);
 802                 if (!pud)
 803                         return -ENOMEM;
 804
 805                 pmd = pmd_offset(pud, addr);
 806                 if (pmd_none(*pmd)) {
 807                         pte_t entry;
 808                         void *p;
 809
 810                         p = vmemmap_alloc_block(PMD_SIZE, node);
 811                         if (!p)
 812                                 return -ENOMEM;
 813
 814                         entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
 815                                                         PAGE_KERNEL_LARGE);
 816                         set_pmd(pmd, __pmd(pte_val(entry)));
 817
 818                         printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n",
 819                                 addr, addr + PMD_SIZE - 1, p, node);
 820                 } else {
 821                         vmemmap_verify((pte_t *)pmd, node, addr, next);
 822                 }
 823         }
 824         return 0;
 825 }
 826 #endif