arch/ia64/mm/init.c

   1 /*
   2  * Initialize MMU support.
   3  *
   4  * Copyright (C) 1998-2003 Hewlett-Packard Co
   5  *      David Mosberger-Tang <davidm@hpl.hp.com>
   6  */
   7 #include <linux/kernel.h>
   8 #include <linux/init.h>
   9
  10 #include <linux/bootmem.h>
  11 #include <linux/efi.h>
  12 #include <linux/elf.h>
  13 #include <linux/mm.h>
  14 #include <linux/mmzone.h>
  15 #include <linux/module.h>
  16 #include <linux/personality.h>
  17 #include <linux/reboot.h>
  18 #include <linux/slab.h>
  19 #include <linux/swap.h>
  20 #include <linux/proc_fs.h>
  21 #include <linux/bitops.h>
  22 #include <linux/kexec.h>
  23
  24 #include <asm/dma.h>
  25 #include <asm/io.h>
  26 #include <asm/machvec.h>
  27 #include <asm/numa.h>
  28 #include <asm/patch.h>
  29 #include <asm/pgalloc.h>
  30 #include <asm/sal.h>
  31 #include <asm/sections.h>
  32 #include <asm/system.h>
  33 #include <asm/tlb.h>
  34 #include <asm/uaccess.h>
  35 #include <asm/unistd.h>
  36 #include <asm/mca.h>
  37 #include <asm/paravirt.h>
  38
  39 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  40
  41 extern void ia64_tlb_init (void);
  42
  43 unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
  44
  45 #ifdef CONFIG_VIRTUAL_MEM_MAP
  46 unsigned long VMALLOC_END = VMALLOC_END_INIT;
  47 EXPORT_SYMBOL(VMALLOC_END);
  48 struct page *vmem_map;
  49 EXPORT_SYMBOL(vmem_map);
  50 #endif
  51
  52 struct page *zero_page_memmap_ptr;      /* map entry for zero page */
  53 EXPORT_SYMBOL(zero_page_memmap_ptr);
  54
  55 void
  56 __ia64_sync_icache_dcache (pte_t pte)
  57 {
  58         unsigned long addr;
  59         struct page *page;
  60
  61         page = pte_page(pte);
  62         addr = (unsigned long) page_address(page);
  63
  64         if (test_bit(PG_arch_1, &page->flags))
  65                 return;                         /* i-cache is already coherent with d-cache */
  66
  67         flush_icache_range(addr, addr + (PAGE_SIZE << compound_order(page)));
  68         set_bit(PG_arch_1, &page->flags);       /* mark page as clean */
  69 }
  70
  71 /*
  72  * Since DMA is i-cache coherent, any (complete) pages that were written via
  73  * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
  74  * flush them when they get mapped into an executable vm-area.
  75  */
  76 void
  77 dma_mark_clean(void *addr, size_t size)
  78 {
  79         unsigned long pg_addr, end;
  80
  81         pg_addr = PAGE_ALIGN((unsigned long) addr);
  82         end = (unsigned long) addr + size;
  83         while (pg_addr + PAGE_SIZE <= end) {
  84                 struct page *page = virt_to_page(pg_addr);
  85                 set_bit(PG_arch_1, &page->flags);
  86                 pg_addr += PAGE_SIZE;
  87         }
  88 }
  89
  90 inline void
  91 ia64_set_rbs_bot (void)
  92 {
  93         unsigned long stack_size = rlimit_max(RLIMIT_STACK) & -16;
  94
  95         if (stack_size > MAX_USER_STACK_SIZE)
  96                 stack_size = MAX_USER_STACK_SIZE;
  97         current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size);
  98 }
  99
 100 /*
 101  * This performs some platform-dependent address space initialization.
 102  * On IA-64, we want to setup the VM area for the register backing
 103  * store (which grows upwards) and install the gateway page which is
 104  * used for signal trampolines, etc.
 105  */
 106 void
 107 ia64_init_addr_space (void)
 108 {
 109         struct vm_area_struct *vma;
 110
 111         ia64_set_rbs_bot();
 112
 113         /*
 114          * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
 115          * the problem.  When the process attempts to write to the register backing store
 116          * for the first time, it will get a SEGFAULT in this case.
 117          */
 118         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 119         if (vma) {
 120                 INIT_LIST_HEAD(&vma->anon_vma_chain);
 121                 vma->vm_mm = current->mm;
 122                 vma->vm_start = current->thread.rbs_bot & PAGE_MASK;
 123                 vma->vm_end = vma->vm_start + PAGE_SIZE;
 124                 vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT;
 125                 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 126                 down_write(&current->mm->mmap_sem);
 127                 if (insert_vm_struct(current->mm, vma)) {
 128                         up_write(&current->mm->mmap_sem);
 129                         kmem_cache_free(vm_area_cachep, vma);
 130                         return;
 131                 }
 132                 up_write(&current->mm->mmap_sem);
 133         }
 134
 135         /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
 136         if (!(current->personality & MMAP_PAGE_ZERO)) {
 137                 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 138                 if (vma) {
 139                         INIT_LIST_HEAD(&vma->anon_vma_chain);
 140                         vma->vm_mm = current->mm;
 141                         vma->vm_end = PAGE_SIZE;
 142                         vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT);
 143                         vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO | VM_RESERVED;
 144                         down_write(&current->mm->mmap_sem);
 145                         if (insert_vm_struct(current->mm, vma)) {
 146                                 up_write(&current->mm->mmap_sem);
 147                                 kmem_cache_free(vm_area_cachep, vma);
 148                                 return;
 149                         }
 150                         up_write(&current->mm->mmap_sem);
 151                 }
 152         }
 153 }
 154
 155 void
 156 free_initmem (void)
 157 {
 158         unsigned long addr, eaddr;
 159
 160         addr = (unsigned long) ia64_imva(__init_begin);
 161         eaddr = (unsigned long) ia64_imva(__init_end);
 162         while (addr < eaddr) {
 163                 ClearPageReserved(virt_to_page(addr));
 164                 init_page_count(virt_to_page(addr));
 165                 free_page(addr);
 166                 ++totalram_pages;
 167                 addr += PAGE_SIZE;
 168         }
 169         printk(KERN_INFO "Freeing unused kernel memory: %ldkB freed\n",
 170                (__init_end - __init_begin) >> 10);
 171 }
 172
 173 void __init
 174 free_initrd_mem (unsigned long start, unsigned long end)
 175 {
 176         struct page *page;
 177         /*
 178          * EFI uses 4KB pages while the kernel can use 4KB or bigger.
 179          * Thus EFI and the kernel may have different page sizes. It is
 180          * therefore possible to have the initrd share the same page as
 181          * the end of the kernel (given current setup).
 182          *
 183          * To avoid freeing/using the wrong page (kernel sized) we:
 184          *      - align up the beginning of initrd
 185          *      - align down the end of initrd
 186          *
 187          *  |             |
 188          *  |=============| a000
 189          *  |             |
 190          *  |             |
 191          *  |             | 9000
 192          *  |/////////////|
 193          *  |/////////////|
 194          *  |=============| 8000
 195          *  |///INITRD////|
 196          *  |/////////////|
 197          *  |/////////////| 7000
 198          *  |             |
 199          *  |KKKKKKKKKKKKK|
 200          *  |=============| 6000
 201          *  |KKKKKKKKKKKKK|
 202          *  |KKKKKKKKKKKKK|
 203          *  K=kernel using 8KB pages
 204          *
 205          * In this example, we must free page 8000 ONLY. So we must align up
 206          * initrd_start and keep initrd_end as is.
 207          */
 208         start = PAGE_ALIGN(start);
 209         end = end & PAGE_MASK;
 210
 211         if (start < end)
 212                 printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
 213
 214         for (; start < end; start += PAGE_SIZE) {
 215                 if (!virt_addr_valid(start))
 216                         continue;
 217                 page = virt_to_page(start);
 218                 ClearPageReserved(page);
 219                 init_page_count(page);
 220                 free_page(start);
 221                 ++totalram_pages;
 222         }
 223 }
 224
 225 /*
 226  * This installs a clean page in the kernel's page table.
 227  */
 228 static struct page * __init
 229 put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
 230 {
 231         pgd_t *pgd;
 232         pud_t *pud;
 233         pmd_t *pmd;
 234         pte_t *pte;
 235
 236         if (!PageReserved(page))
 237                 printk(KERN_ERR "put_kernel_page: page at 0x%p not in reserved memory\n",
 238                        page_address(page));
 239
 240         pgd = pgd_offset_k(address);            /* note: this is NOT pgd_offset()! */
 241
 242         {
 243                 pud = pud_alloc(&init_mm, pgd, address);
 244                 if (!pud)
 245                         goto out;
 246                 pmd = pmd_alloc(&init_mm, pud, address);
 247                 if (!pmd)
 248                         goto out;
 249                 pte = pte_alloc_kernel(pmd, address);
 250                 if (!pte)
 251                         goto out;
 252                 if (!pte_none(*pte))
 253                         goto out;
 254                 set_pte(pte, mk_pte(page, pgprot));
 255         }
 256   out:
 257         /* no need for flush_tlb */
 258         return page;
 259 }
 260
 261 static void __init
 262 setup_gate (void)
 263 {
 264         void *gate_section;
 265         struct page *page;
 266
 267         /*
 268          * Map the gate page twice: once read-only to export the ELF
 269          * headers etc. and once execute-only page to enable
 270          * privilege-promotion via "epc":
 271          */
 272         gate_section = paravirt_get_gate_section();
 273         page = virt_to_page(ia64_imva(gate_section));
 274         put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
 275 #ifdef HAVE_BUGGY_SEGREL
 276         page = virt_to_page(ia64_imva(gate_section + PAGE_SIZE));
 277         put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
 278 #else
 279         put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
 280         /* Fill in the holes (if any) with read-only zero pages: */
 281         {
 282                 unsigned long addr;
 283
 284                 for (addr = GATE_ADDR + PAGE_SIZE;
 285                      addr < GATE_ADDR + PERCPU_PAGE_SIZE;
 286                      addr += PAGE_SIZE)
 287                 {
 288                         put_kernel_page(ZERO_PAGE(0), addr,
 289                                         PAGE_READONLY);
 290                         put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
 291                                         PAGE_READONLY);
 292                 }
 293         }
 294 #endif
 295         ia64_patch_gate();
 296 }
 297
 298 void __devinit
 299 ia64_mmu_init (void *my_cpu_data)
 300 {
 301         unsigned long pta, impl_va_bits;
 302         extern void __devinit tlb_init (void);
 303
 304 #ifdef CONFIG_DISABLE_VHPT
 305 #       define VHPT_ENABLE_BIT  0
 306 #else
 307 #       define VHPT_ENABLE_BIT  1
 308 #endif
 309
 310         /*
 311          * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
 312          * address space.  The IA-64 architecture guarantees that at least 50 bits of
 313          * virtual address space are implemented but if we pick a large enough page size
 314          * (e.g., 64KB), the mapped address space is big enough that it will overlap with
 315          * VMLPT.  I assume that once we run on machines big enough to warrant 64KB pages,
 316          * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
 317          * problem in practice.  Alternatively, we could truncate the top of the mapped
 318          * address space to not permit mappings that would overlap with the VMLPT.
 319          * --davidm 00/12/06
 320          */
 321 #       define pte_bits                 3
 322 #       define mapped_space_bits        (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
 323         /*
 324          * The virtual page table has to cover the entire implemented address space within
 325          * a region even though not all of this space may be mappable.  The reason for
 326          * this is that the Access bit and Dirty bit fault handlers perform
 327          * non-speculative accesses to the virtual page table, so the address range of the
 328          * virtual page table itself needs to be covered by virtual page table.
 329          */
 330 #       define vmlpt_bits               (impl_va_bits - PAGE_SHIFT + pte_bits)
 331 #       define POW2(n)                  (1ULL << (n))
 332
 333         impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
 334
 335         if (impl_va_bits < 51 || impl_va_bits > 61)
 336                 panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
 337         /*
 338          * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
 339          * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
 340          * the test makes sure that our mapped space doesn't overlap the
 341          * unimplemented hole in the middle of the region.
 342          */
 343         if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
 344             (mapped_space_bits > impl_va_bits - 1))
 345                 panic("Cannot build a big enough virtual-linear page table"
 346                       " to cover mapped address space.\n"
 347                       " Try using a smaller page size.\n");
 348
 349
 350         /* place the VMLPT at the end of each page-table mapped region: */
 351         pta = POW2(61) - POW2(vmlpt_bits);
 352
 353         /*
 354          * Set the (virtually mapped linear) page table address.  Bit
 355          * 8 selects between the short and long format, bits 2-7 the
 356          * size of the table, and bit 0 whether the VHPT walker is
 357          * enabled.
 358          */
 359         ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
 360
 361         ia64_tlb_init();
 362
 363 #ifdef  CONFIG_HUGETLB_PAGE
 364         ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
 365         ia64_srlz_d();
 366 #endif
 367 }
 368
 369 #ifdef CONFIG_VIRTUAL_MEM_MAP
 370 int vmemmap_find_next_valid_pfn(int node, int i)
 371 {
 372         unsigned long end_address, hole_next_pfn;
 373         unsigned long stop_address;
 374         pg_data_t *pgdat = NODE_DATA(node);
 375
 376         end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
 377         end_address = PAGE_ALIGN(end_address);
 378
 379         stop_address = (unsigned long) &vmem_map[
 380                 pgdat->node_start_pfn + pgdat->node_spanned_pages];
 381
 382         do {
 383                 pgd_t *pgd;
 384                 pud_t *pud;
 385                 pmd_t *pmd;
 386                 pte_t *pte;
 387
 388                 pgd = pgd_offset_k(end_address);
 389                 if (pgd_none(*pgd)) {
 390                         end_address += PGDIR_SIZE;
 391                         continue;
 392                 }
 393
 394                 pud = pud_offset(pgd, end_address);
 395                 if (pud_none(*pud)) {
 396                         end_address += PUD_SIZE;
 397                         continue;
 398                 }
 399
 400                 pmd = pmd_offset(pud, end_address);
 401                 if (pmd_none(*pmd)) {
 402                         end_address += PMD_SIZE;
 403                         continue;
 404                 }
 405
 406                 pte = pte_offset_kernel(pmd, end_address);
 407 retry_pte:
 408                 if (pte_none(*pte)) {
 409                         end_address += PAGE_SIZE;
 410                         pte++;
 411                         if ((end_address < stop_address) &&
 412                             (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
 413                                 goto retry_pte;
 414                         continue;
 415                 }
 416                 /* Found next valid vmem_map page */
 417                 break;
 418         } while (end_address < stop_address);
 419
 420         end_address = min(end_address, stop_address);
 421         end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
 422         hole_next_pfn = end_address / sizeof(struct page);
 423         return hole_next_pfn - pgdat->node_start_pfn;
 424 }
 425
 426 int __init create_mem_map_page_table(u64 start, u64 end, void *arg)
 427 {
 428         unsigned long address, start_page, end_page;
 429         struct page *map_start, *map_end;
 430         int node;
 431         pgd_t *pgd;
 432         pud_t *pud;
 433         pmd_t *pmd;
 434         pte_t *pte;
 435
 436         map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
 437         map_end   = vmem_map + (__pa(end) >> PAGE_SHIFT);
 438
 439         start_page = (unsigned long) map_start & PAGE_MASK;
 440         end_page = PAGE_ALIGN((unsigned long) map_end);
 441         node = paddr_to_nid(__pa(start));
 442
 443         for (address = start_page; address < end_page; address += PAGE_SIZE) {
 444                 pgd = pgd_offset_k(address);
 445                 if (pgd_none(*pgd))
 446                         pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
 447                 pud = pud_offset(pgd, address);
 448
 449                 if (pud_none(*pud))
 450                         pud_populate(&init_mm, pud, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
 451                 pmd = pmd_offset(pud, address);
 452
 453                 if (pmd_none(*pmd))
 454                         pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
 455                 pte = pte_offset_kernel(pmd, address);
 456
 457                 if (pte_none(*pte))
 458                         set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
 459                                              PAGE_KERNEL));
 460         }
 461         return 0;
 462 }
 463
 464 struct memmap_init_callback_data {
 465         struct page *start;
 466         struct page *end;
 467         int nid;
 468         unsigned long zone;
 469 };
 470
 471 static int __meminit
 472 virtual_memmap_init(u64 start, u64 end, void *arg)
 473 {
 474         struct memmap_init_callback_data *args;
 475         struct page *map_start, *map_end;
 476
 477         args = (struct memmap_init_callback_data *) arg;
 478         map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
 479         map_end   = vmem_map + (__pa(end) >> PAGE_SHIFT);
 480
 481         if (map_start < args->start)
 482                 map_start = args->start;
 483         if (map_end > args->end)
 484                 map_end = args->end;
 485
 486         /*
 487          * We have to initialize "out of bounds" struct page elements that fit completely
 488          * on the same pages that were allocated for the "in bounds" elements because they
 489          * may be referenced later (and found to be "reserved").
 490          */
 491         map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
 492         map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
 493                     / sizeof(struct page));
 494
 495         if (map_start < map_end)
 496                 memmap_init_zone((unsigned long)(map_end - map_start),
 497                                  args->nid, args->zone, page_to_pfn(map_start),
 498                                  MEMMAP_EARLY);
 499         return 0;
 500 }
 501
 502 void __meminit
 503 memmap_init (unsigned long size, int nid, unsigned long zone,
 504              unsigned long start_pfn)
 505 {
 506         if (!vmem_map)
 507                 memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY);
 508         else {
 509                 struct page *start;
 510                 struct memmap_init_callback_data args;
 511
 512                 start = pfn_to_page(start_pfn);
 513                 args.start = start;
 514                 args.end = start + size;
 515                 args.nid = nid;
 516                 args.zone = zone;
 517
 518                 efi_memmap_walk(virtual_memmap_init, &args);
 519         }
 520 }
 521
 522 int
 523 ia64_pfn_valid (unsigned long pfn)
 524 {
 525         char byte;
 526         struct page *pg = pfn_to_page(pfn);
 527
 528         return     (__get_user(byte, (char __user *) pg) == 0)
 529                 && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
 530                         || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
 531 }
 532 EXPORT_SYMBOL(ia64_pfn_valid);
 533
 534 int __init find_largest_hole(u64 start, u64 end, void *arg)
 535 {
 536         u64 *max_gap = arg;
 537
 538         static u64 last_end = PAGE_OFFSET;
 539
 540         /* NOTE: this algorithm assumes efi memmap table is ordered */
 541
 542         if (*max_gap < (start - last_end))
 543                 *max_gap = start - last_end;
 544         last_end = end;
 545         return 0;
 546 }
 547
 548 #endif /* CONFIG_VIRTUAL_MEM_MAP */
 549
 550 int __init register_active_ranges(u64 start, u64 len, int nid)
 551 {
 552         u64 end = start + len;
 553
 554 #ifdef CONFIG_KEXEC
 555         if (start > crashk_res.start && start < crashk_res.end)
 556                 start = crashk_res.end;
 557         if (end > crashk_res.start && end < crashk_res.end)
 558                 end = crashk_res.start;
 559 #endif
 560
 561         if (start < end)
 562                 add_active_range(nid, __pa(start) >> PAGE_SHIFT,
 563                         __pa(end) >> PAGE_SHIFT);
 564         return 0;
 565 }
 566
 567 static int __init
 568 count_reserved_pages(u64 start, u64 end, void *arg)
 569 {
 570         unsigned long num_reserved = 0;
 571         unsigned long *count = arg;
 572
 573         for (; start < end; start += PAGE_SIZE)
 574                 if (PageReserved(virt_to_page(start)))
 575                         ++num_reserved;
 576         *count += num_reserved;
 577         return 0;
 578 }
 579
 580 int
 581 find_max_min_low_pfn (u64 start, u64 end, void *arg)
 582 {
 583         unsigned long pfn_start, pfn_end;
 584 #ifdef CONFIG_FLATMEM
 585         pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
 586         pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
 587 #else
 588         pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
 589         pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
 590 #endif
 591         min_low_pfn = min(min_low_pfn, pfn_start);
 592         max_low_pfn = max(max_low_pfn, pfn_end);
 593         return 0;
 594 }
 595
 596 /*
 597  * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
 598  * system call handler.  When this option is in effect, all fsyscalls will end up bubbling
 599  * down into the kernel and calling the normal (heavy-weight) syscall handler.  This is
 600  * useful for performance testing, but conceivably could also come in handy for debugging
 601  * purposes.
 602  */
 603
 604 static int nolwsys __initdata;
 605
 606 static int __init
 607 nolwsys_setup (char *s)
 608 {
 609         nolwsys = 1;
 610         return 1;
 611 }
 612
 613 __setup("nolwsys", nolwsys_setup);
 614
 615 void __init
 616 mem_init (void)
 617 {
 618         long reserved_pages, codesize, datasize, initsize;
 619         pg_data_t *pgdat;
 620         int i;
 621
 622         BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
 623         BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
 624         BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
 625
 626 #ifdef CONFIG_PCI
 627         /*
 628          * This needs to be called _after_ the command line has been parsed but _before_
 629          * any drivers that may need the PCI DMA interface are initialized or bootmem has
 630          * been freed.
 631          */
 632         platform_dma_init();
 633 #endif
 634
 635 #ifdef CONFIG_FLATMEM
 636         BUG_ON(!mem_map);
 637         max_mapnr = max_low_pfn;
 638 #endif
 639
 640         high_memory = __va(max_low_pfn * PAGE_SIZE);
 641
 642         for_each_online_pgdat(pgdat)
 643                 if (pgdat->bdata->node_bootmem_map)
 644                         totalram_pages += free_all_bootmem_node(pgdat);
 645
 646         reserved_pages = 0;
 647         efi_memmap_walk(count_reserved_pages, &reserved_pages);
 648
 649         codesize =  (unsigned long) _etext - (unsigned long) _stext;
 650         datasize =  (unsigned long) _edata - (unsigned long) _etext;
 651         initsize =  (unsigned long) __init_end - (unsigned long) __init_begin;
 652
 653         printk(KERN_INFO "Memory: %luk/%luk available (%luk code, %luk reserved, "
 654                "%luk data, %luk init)\n", nr_free_pages() << (PAGE_SHIFT - 10),
 655                num_physpages << (PAGE_SHIFT - 10), codesize >> 10,
 656                reserved_pages << (PAGE_SHIFT - 10), datasize >> 10, initsize >> 10);
 657
 658
 659         /*
 660          * For fsyscall entrpoints with no light-weight handler, use the ordinary
 661          * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
 662          * code can tell them apart.
 663          */
 664         for (i = 0; i < NR_syscalls; ++i) {
 665                 extern unsigned long sys_call_table[NR_syscalls];
 666                 unsigned long *fsyscall_table = paravirt_get_fsyscall_table();
 667
 668                 if (!fsyscall_table[i] || nolwsys)
 669                         fsyscall_table[i] = sys_call_table[i] | 1;
 670         }
 671         setup_gate();
 672 }
 673
 674 #ifdef CONFIG_MEMORY_HOTPLUG
 675 int arch_add_memory(int nid, u64 start, u64 size)
 676 {
 677         pg_data_t *pgdat;
 678         struct zone *zone;
 679         unsigned long start_pfn = start >> PAGE_SHIFT;
 680         unsigned long nr_pages = size >> PAGE_SHIFT;
 681         int ret;
 682
 683         pgdat = NODE_DATA(nid);
 684
 685         zone = pgdat->node_zones + ZONE_NORMAL;
 686         ret = __add_pages(nid, zone, start_pfn, nr_pages);
 687
 688         if (ret)
 689                 printk("%s: Problem encountered in __add_pages() as ret=%d\n",
 690                        __func__,  ret);
 691
 692         return ret;
 693 }
 694 #endif
 695
 696 /*
 697  * Even when CONFIG_IA32_SUPPORT is not enabled it is
 698  * useful to have the Linux/x86 domain registered to
 699  * avoid an attempted module load when emulators call
 700  * personality(PER_LINUX32). This saves several milliseconds
 701  * on each such call.
 702  */
 703 static struct exec_domain ia32_exec_domain;
 704
 705 static int __init
 706 per_linux32_init(void)
 707 {
 708         ia32_exec_domain.name = "Linux/x86";
 709         ia32_exec_domain.handler = NULL;
 710         ia32_exec_domain.pers_low = PER_LINUX32;
 711         ia32_exec_domain.pers_high = PER_LINUX32;
 712         ia32_exec_domain.signal_map = default_exec_domain.signal_map;
 713         ia32_exec_domain.signal_invmap = default_exec_domain.signal_invmap;
 714         register_exec_domain(&ia32_exec_domain);
 715
 716         return 0;
 717 }
 718
 719 __initcall(per_linux32_init);