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/a.out.h>
  25 #include <asm/dma.h>
  26 #include <asm/ia32.h>
  27 #include <asm/io.h>
  28 #include <asm/machvec.h>
  29 #include <asm/numa.h>
  30 #include <asm/patch.h>
  31 #include <asm/pgalloc.h>
  32 #include <asm/sal.h>
  33 #include <asm/sections.h>
  34 #include <asm/system.h>
  35 #include <asm/tlb.h>
  36 #include <asm/uaccess.h>
  37 #include <asm/unistd.h>
  38 #include <asm/mca.h>
  39
  40 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  41
  42 extern void ia64_tlb_init (void);
  43
  44 unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
  45
  46 #ifdef CONFIG_VIRTUAL_MEM_MAP
  47 unsigned long vmalloc_end = VMALLOC_END_INIT;
  48 EXPORT_SYMBOL(vmalloc_end);
  49 struct page *vmem_map;
  50 EXPORT_SYMBOL(vmem_map);
  51 #endif
  52
  53 struct page *zero_page_memmap_ptr;      /* map entry for zero page */
  54 EXPORT_SYMBOL(zero_page_memmap_ptr);
  55
  56 void
  57 lazy_mmu_prot_update (pte_t pte)
  58 {
  59         unsigned long addr;
  60         struct page *page;
  61         unsigned long order;
  62
  63         if (!pte_exec(pte))
  64                 return;                         /* not an executable page... */
  65
  66         page = pte_page(pte);
  67         addr = (unsigned long) page_address(page);
  68
  69         if (test_bit(PG_arch_1, &page->flags))
  70                 return;                         /* i-cache is already coherent with d-cache */
  71
  72         if (PageCompound(page)) {
  73                 order = compound_order(page);
  74                 flush_icache_range(addr, addr + (1UL << order << PAGE_SHIFT));
  75         }
  76         else
  77                 flush_icache_range(addr, addr + PAGE_SIZE);
  78         set_bit(PG_arch_1, &page->flags);       /* mark page as clean */
  79 }
  80
  81 /*
  82  * Since DMA is i-cache coherent, any (complete) pages that were written via
  83  * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
  84  * flush them when they get mapped into an executable vm-area.
  85  */
  86 void
  87 dma_mark_clean(void *addr, size_t size)
  88 {
  89         unsigned long pg_addr, end;
  90
  91         pg_addr = PAGE_ALIGN((unsigned long) addr);
  92         end = (unsigned long) addr + size;
  93         while (pg_addr + PAGE_SIZE <= end) {
  94                 struct page *page = virt_to_page(pg_addr);
  95                 set_bit(PG_arch_1, &page->flags);
  96                 pg_addr += PAGE_SIZE;
  97         }
  98 }
  99
 100 inline void
 101 ia64_set_rbs_bot (void)
 102 {
 103         unsigned long stack_size = current->signal->rlim[RLIMIT_STACK].rlim_max & -16;
 104
 105         if (stack_size > MAX_USER_STACK_SIZE)
 106                 stack_size = MAX_USER_STACK_SIZE;
 107         current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size);
 108 }
 109
 110 /*
 111  * This performs some platform-dependent address space initialization.
 112  * On IA-64, we want to setup the VM area for the register backing
 113  * store (which grows upwards) and install the gateway page which is
 114  * used for signal trampolines, etc.
 115  */
 116 void
 117 ia64_init_addr_space (void)
 118 {
 119         struct vm_area_struct *vma;
 120
 121         ia64_set_rbs_bot();
 122
 123         /*
 124          * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
 125          * the problem.  When the process attempts to write to the register backing store
 126          * for the first time, it will get a SEGFAULT in this case.
 127          */
 128         vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 129         if (vma) {
 130                 vma->vm_mm = current->mm;
 131                 vma->vm_start = current->thread.rbs_bot & PAGE_MASK;
 132                 vma->vm_end = vma->vm_start + PAGE_SIZE;
 133                 vma->vm_page_prot = protection_map[VM_DATA_DEFAULT_FLAGS & 0x7];
 134                 vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT;
 135                 down_write(&current->mm->mmap_sem);
 136                 if (insert_vm_struct(current->mm, vma)) {
 137                         up_write(&current->mm->mmap_sem);
 138                         kmem_cache_free(vm_area_cachep, vma);
 139                         return;
 140                 }
 141                 up_write(&current->mm->mmap_sem);
 142         }
 143
 144         /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
 145         if (!(current->personality & MMAP_PAGE_ZERO)) {
 146                 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 147                 if (vma) {
 148                         vma->vm_mm = current->mm;
 149                         vma->vm_end = PAGE_SIZE;
 150                         vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT);
 151                         vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO | VM_RESERVED;
 152                         down_write(&current->mm->mmap_sem);
 153                         if (insert_vm_struct(current->mm, vma)) {
 154                                 up_write(&current->mm->mmap_sem);
 155                                 kmem_cache_free(vm_area_cachep, vma);
 156                                 return;
 157                         }
 158                         up_write(&current->mm->mmap_sem);
 159                 }
 160         }
 161 }
 162
 163 void
 164 free_initmem (void)
 165 {
 166         unsigned long addr, eaddr;
 167
 168         addr = (unsigned long) ia64_imva(__init_begin);
 169         eaddr = (unsigned long) ia64_imva(__init_end);
 170         while (addr < eaddr) {
 171                 ClearPageReserved(virt_to_page(addr));
 172                 init_page_count(virt_to_page(addr));
 173                 free_page(addr);
 174                 ++totalram_pages;
 175                 addr += PAGE_SIZE;
 176         }
 177         printk(KERN_INFO "Freeing unused kernel memory: %ldkB freed\n",
 178                (__init_end - __init_begin) >> 10);
 179 }
 180
 181 void __init
 182 free_initrd_mem (unsigned long start, unsigned long end)
 183 {
 184         struct page *page;
 185         /*
 186          * EFI uses 4KB pages while the kernel can use 4KB or bigger.
 187          * Thus EFI and the kernel may have different page sizes. It is
 188          * therefore possible to have the initrd share the same page as
 189          * the end of the kernel (given current setup).
 190          *
 191          * To avoid freeing/using the wrong page (kernel sized) we:
 192          *      - align up the beginning of initrd
 193          *      - align down the end of initrd
 194          *
 195          *  |             |
 196          *  |=============| a000
 197          *  |             |
 198          *  |             |
 199          *  |             | 9000
 200          *  |/////////////|
 201          *  |/////////////|
 202          *  |=============| 8000
 203          *  |///INITRD////|
 204          *  |/////////////|
 205          *  |/////////////| 7000
 206          *  |             |
 207          *  |KKKKKKKKKKKKK|
 208          *  |=============| 6000
 209          *  |KKKKKKKKKKKKK|
 210          *  |KKKKKKKKKKKKK|
 211          *  K=kernel using 8KB pages
 212          *
 213          * In this example, we must free page 8000 ONLY. So we must align up
 214          * initrd_start and keep initrd_end as is.
 215          */
 216         start = PAGE_ALIGN(start);
 217         end = end & PAGE_MASK;
 218
 219         if (start < end)
 220                 printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
 221
 222         for (; start < end; start += PAGE_SIZE) {
 223                 if (!virt_addr_valid(start))
 224                         continue;
 225                 page = virt_to_page(start);
 226                 ClearPageReserved(page);
 227                 init_page_count(page);
 228                 free_page(start);
 229                 ++totalram_pages;
 230         }
 231 }
 232
 233 /*
 234  * This installs a clean page in the kernel's page table.
 235  */
 236 static struct page * __init
 237 put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
 238 {
 239         pgd_t *pgd;
 240         pud_t *pud;
 241         pmd_t *pmd;
 242         pte_t *pte;
 243
 244         if (!PageReserved(page))
 245                 printk(KERN_ERR "put_kernel_page: page at 0x%p not in reserved memory\n",
 246                        page_address(page));
 247
 248         pgd = pgd_offset_k(address);            /* note: this is NOT pgd_offset()! */
 249
 250         {
 251                 pud = pud_alloc(&init_mm, pgd, address);
 252                 if (!pud)
 253                         goto out;
 254                 pmd = pmd_alloc(&init_mm, pud, address);
 255                 if (!pmd)
 256                         goto out;
 257                 pte = pte_alloc_kernel(pmd, address);
 258                 if (!pte)
 259                         goto out;
 260                 if (!pte_none(*pte))
 261                         goto out;
 262                 set_pte(pte, mk_pte(page, pgprot));
 263         }
 264   out:
 265         /* no need for flush_tlb */
 266         return page;
 267 }
 268
 269 static void __init
 270 setup_gate (void)
 271 {
 272         struct page *page;
 273
 274         /*
 275          * Map the gate page twice: once read-only to export the ELF
 276          * headers etc. and once execute-only page to enable
 277          * privilege-promotion via "epc":
 278          */
 279         page = virt_to_page(ia64_imva(__start_gate_section));
 280         put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
 281 #ifdef HAVE_BUGGY_SEGREL
 282         page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE));
 283         put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
 284 #else
 285         put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
 286         /* Fill in the holes (if any) with read-only zero pages: */
 287         {
 288                 unsigned long addr;
 289
 290                 for (addr = GATE_ADDR + PAGE_SIZE;
 291                      addr < GATE_ADDR + PERCPU_PAGE_SIZE;
 292                      addr += PAGE_SIZE)
 293                 {
 294                         put_kernel_page(ZERO_PAGE(0), addr,
 295                                         PAGE_READONLY);
 296                         put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
 297                                         PAGE_READONLY);
 298                 }
 299         }
 300 #endif
 301         ia64_patch_gate();
 302 }
 303
 304 void __devinit
 305 ia64_mmu_init (void *my_cpu_data)
 306 {
 307         unsigned long pta, impl_va_bits;
 308         extern void __devinit tlb_init (void);
 309
 310 #ifdef CONFIG_DISABLE_VHPT
 311 #       define VHPT_ENABLE_BIT  0
 312 #else
 313 #       define VHPT_ENABLE_BIT  1
 314 #endif
 315
 316         /*
 317          * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
 318          * address space.  The IA-64 architecture guarantees that at least 50 bits of
 319          * virtual address space are implemented but if we pick a large enough page size
 320          * (e.g., 64KB), the mapped address space is big enough that it will overlap with
 321          * VMLPT.  I assume that once we run on machines big enough to warrant 64KB pages,
 322          * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
 323          * problem in practice.  Alternatively, we could truncate the top of the mapped
 324          * address space to not permit mappings that would overlap with the VMLPT.
 325          * --davidm 00/12/06
 326          */
 327 #       define pte_bits                 3
 328 #       define mapped_space_bits        (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
 329         /*
 330          * The virtual page table has to cover the entire implemented address space within
 331          * a region even though not all of this space may be mappable.  The reason for
 332          * this is that the Access bit and Dirty bit fault handlers perform
 333          * non-speculative accesses to the virtual page table, so the address range of the
 334          * virtual page table itself needs to be covered by virtual page table.
 335          */
 336 #       define vmlpt_bits               (impl_va_bits - PAGE_SHIFT + pte_bits)
 337 #       define POW2(n)                  (1ULL << (n))
 338
 339         impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
 340
 341         if (impl_va_bits < 51 || impl_va_bits > 61)
 342                 panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
 343         /*
 344          * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
 345          * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
 346          * the test makes sure that our mapped space doesn't overlap the
 347          * unimplemented hole in the middle of the region.
 348          */
 349         if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
 350             (mapped_space_bits > impl_va_bits - 1))
 351                 panic("Cannot build a big enough virtual-linear page table"
 352                       " to cover mapped address space.\n"
 353                       " Try using a smaller page size.\n");
 354
 355
 356         /* place the VMLPT at the end of each page-table mapped region: */
 357         pta = POW2(61) - POW2(vmlpt_bits);
 358
 359         /*
 360          * Set the (virtually mapped linear) page table address.  Bit
 361          * 8 selects between the short and long format, bits 2-7 the
 362          * size of the table, and bit 0 whether the VHPT walker is
 363          * enabled.
 364          */
 365         ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
 366
 367         ia64_tlb_init();
 368
 369 #ifdef  CONFIG_HUGETLB_PAGE
 370         ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
 371         ia64_srlz_d();
 372 #endif
 373 }
 374
 375 #ifdef CONFIG_VIRTUAL_MEM_MAP
 376 int vmemmap_find_next_valid_pfn(int node, int i)
 377 {
 378         unsigned long end_address, hole_next_pfn;
 379         unsigned long stop_address;
 380         pg_data_t *pgdat = NODE_DATA(node);
 381
 382         end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
 383         end_address = PAGE_ALIGN(end_address);
 384
 385         stop_address = (unsigned long) &vmem_map[
 386                 pgdat->node_start_pfn + pgdat->node_spanned_pages];
 387
 388         do {
 389                 pgd_t *pgd;
 390                 pud_t *pud;
 391                 pmd_t *pmd;
 392                 pte_t *pte;
 393
 394                 pgd = pgd_offset_k(end_address);
 395                 if (pgd_none(*pgd)) {
 396                         end_address += PGDIR_SIZE;
 397                         continue;
 398                 }
 399
 400                 pud = pud_offset(pgd, end_address);
 401                 if (pud_none(*pud)) {
 402                         end_address += PUD_SIZE;
 403                         continue;
 404                 }
 405
 406                 pmd = pmd_offset(pud, end_address);
 407                 if (pmd_none(*pmd)) {
 408                         end_address += PMD_SIZE;
 409                         continue;
 410                 }
 411
 412                 pte = pte_offset_kernel(pmd, end_address);
 413 retry_pte:
 414                 if (pte_none(*pte)) {
 415                         end_address += PAGE_SIZE;
 416                         pte++;
 417                         if ((end_address < stop_address) &&
 418                             (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
 419                                 goto retry_pte;
 420                         continue;
 421                 }
 422                 /* Found next valid vmem_map page */
 423                 break;
 424         } while (end_address < stop_address);
 425
 426         end_address = min(end_address, stop_address);
 427         end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
 428         hole_next_pfn = end_address / sizeof(struct page);
 429         return hole_next_pfn - pgdat->node_start_pfn;
 430 }
 431
 432 int __init
 433 create_mem_map_page_table (u64 start, u64 end, void *arg)
 434 {
 435         unsigned long address, start_page, end_page;
 436         struct page *map_start, *map_end;
 437         int node;
 438         pgd_t *pgd;
 439         pud_t *pud;
 440         pmd_t *pmd;
 441         pte_t *pte;
 442
 443         map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
 444         map_end   = vmem_map + (__pa(end) >> PAGE_SHIFT);
 445
 446         start_page = (unsigned long) map_start & PAGE_MASK;
 447         end_page = PAGE_ALIGN((unsigned long) map_end);
 448         node = paddr_to_nid(__pa(start));
 449
 450         for (address = start_page; address < end_page; address += PAGE_SIZE) {
 451                 pgd = pgd_offset_k(address);
 452                 if (pgd_none(*pgd))
 453                         pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
 454                 pud = pud_offset(pgd, address);
 455
 456                 if (pud_none(*pud))
 457                         pud_populate(&init_mm, pud, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
 458                 pmd = pmd_offset(pud, address);
 459
 460                 if (pmd_none(*pmd))
 461                         pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
 462                 pte = pte_offset_kernel(pmd, address);
 463
 464                 if (pte_none(*pte))
 465                         set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
 466                                              PAGE_KERNEL));
 467         }
 468         return 0;
 469 }
 470
 471 struct memmap_init_callback_data {
 472         struct page *start;
 473         struct page *end;
 474         int nid;
 475         unsigned long zone;
 476 };
 477
 478 static int
 479 virtual_memmap_init (u64 start, u64 end, void *arg)
 480 {
 481         struct memmap_init_callback_data *args;
 482         struct page *map_start, *map_end;
 483
 484         args = (struct memmap_init_callback_data *) arg;
 485         map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
 486         map_end   = vmem_map + (__pa(end) >> PAGE_SHIFT);
 487
 488         if (map_start < args->start)
 489                 map_start = args->start;
 490         if (map_end > args->end)
 491                 map_end = args->end;
 492
 493         /*
 494          * We have to initialize "out of bounds" struct page elements that fit completely
 495          * on the same pages that were allocated for the "in bounds" elements because they
 496          * may be referenced later (and found to be "reserved").
 497          */
 498         map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
 499         map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
 500                     / sizeof(struct page));
 501
 502         if (map_start < map_end)
 503                 memmap_init_zone((unsigned long)(map_end - map_start),
 504                                  args->nid, args->zone, page_to_pfn(map_start),
 505                                  MEMMAP_EARLY);
 506         return 0;
 507 }
 508
 509 void
 510 memmap_init (unsigned long size, int nid, unsigned long zone,
 511              unsigned long start_pfn)
 512 {
 513         if (!vmem_map)
 514                 memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY);
 515         else {
 516                 struct page *start;
 517                 struct memmap_init_callback_data args;
 518
 519                 start = pfn_to_page(start_pfn);
 520                 args.start = start;
 521                 args.end = start + size;
 522                 args.nid = nid;
 523                 args.zone = zone;
 524
 525                 efi_memmap_walk(virtual_memmap_init, &args);
 526         }
 527 }
 528
 529 int
 530 ia64_pfn_valid (unsigned long pfn)
 531 {
 532         char byte;
 533         struct page *pg = pfn_to_page(pfn);
 534
 535         return     (__get_user(byte, (char __user *) pg) == 0)
 536                 && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
 537                         || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
 538 }
 539 EXPORT_SYMBOL(ia64_pfn_valid);
 540
 541 int __init
 542 find_largest_hole (u64 start, u64 end, void *arg)
 543 {
 544         u64 *max_gap = arg;
 545
 546         static u64 last_end = PAGE_OFFSET;
 547
 548         /* NOTE: this algorithm assumes efi memmap table is ordered */
 549
 550         if (*max_gap < (start - last_end))
 551                 *max_gap = start - last_end;
 552         last_end = end;
 553         return 0;
 554 }
 555
 556 #endif /* CONFIG_VIRTUAL_MEM_MAP */
 557
 558 int __init
 559 register_active_ranges(u64 start, u64 end, void *arg)
 560 {
 561         int nid = paddr_to_nid(__pa(start));
 562
 563         if (nid < 0)
 564                 nid = 0;
 565 #ifdef CONFIG_KEXEC
 566         if (start > crashk_res.start && start < crashk_res.end)
 567                 start = crashk_res.end;
 568         if (end > crashk_res.start && end < crashk_res.end)
 569                 end = crashk_res.start;
 570 #endif
 571
 572         if (start < end)
 573                 add_active_range(nid, __pa(start) >> PAGE_SHIFT,
 574                         __pa(end) >> PAGE_SHIFT);
 575         return 0;
 576 }
 577
 578 static int __init
 579 count_reserved_pages (u64 start, u64 end, void *arg)
 580 {
 581         unsigned long num_reserved = 0;
 582         unsigned long *count = arg;
 583
 584         for (; start < end; start += PAGE_SIZE)
 585                 if (PageReserved(virt_to_page(start)))
 586                         ++num_reserved;
 587         *count += num_reserved;
 588         return 0;
 589 }
 590
 591 int
 592 find_max_min_low_pfn (unsigned long start, unsigned long end, void *arg)
 593 {
 594         unsigned long pfn_start, pfn_end;
 595 #ifdef CONFIG_FLATMEM
 596         pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
 597         pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
 598 #else
 599         pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
 600         pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
 601 #endif
 602         min_low_pfn = min(min_low_pfn, pfn_start);
 603         max_low_pfn = max(max_low_pfn, pfn_end);
 604         return 0;
 605 }
 606
 607 /*
 608  * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
 609  * system call handler.  When this option is in effect, all fsyscalls will end up bubbling
 610  * down into the kernel and calling the normal (heavy-weight) syscall handler.  This is
 611  * useful for performance testing, but conceivably could also come in handy for debugging
 612  * purposes.
 613  */
 614
 615 static int nolwsys __initdata;
 616
 617 static int __init
 618 nolwsys_setup (char *s)
 619 {
 620         nolwsys = 1;
 621         return 1;
 622 }
 623
 624 __setup("nolwsys", nolwsys_setup);
 625
 626 void __init
 627 mem_init (void)
 628 {
 629         long reserved_pages, codesize, datasize, initsize;
 630         pg_data_t *pgdat;
 631         int i;
 632         static struct kcore_list kcore_mem, kcore_vmem, kcore_kernel;
 633
 634         BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
 635         BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
 636         BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
 637
 638 #ifdef CONFIG_PCI
 639         /*
 640          * This needs to be called _after_ the command line has been parsed but _before_
 641          * any drivers that may need the PCI DMA interface are initialized or bootmem has
 642          * been freed.
 643          */
 644         platform_dma_init();
 645 #endif
 646
 647 #ifdef CONFIG_FLATMEM
 648         if (!mem_map)
 649                 BUG();
 650         max_mapnr = max_low_pfn;
 651 #endif
 652
 653         high_memory = __va(max_low_pfn * PAGE_SIZE);
 654
 655         kclist_add(&kcore_mem, __va(0), max_low_pfn * PAGE_SIZE);
 656         kclist_add(&kcore_vmem, (void *)VMALLOC_START, VMALLOC_END-VMALLOC_START);
 657         kclist_add(&kcore_kernel, _stext, _end - _stext);
 658
 659         for_each_online_pgdat(pgdat)
 660                 if (pgdat->bdata->node_bootmem_map)
 661                         totalram_pages += free_all_bootmem_node(pgdat);
 662
 663         reserved_pages = 0;
 664         efi_memmap_walk(count_reserved_pages, &reserved_pages);
 665
 666         codesize =  (unsigned long) _etext - (unsigned long) _stext;
 667         datasize =  (unsigned long) _edata - (unsigned long) _etext;
 668         initsize =  (unsigned long) __init_end - (unsigned long) __init_begin;
 669
 670         printk(KERN_INFO "Memory: %luk/%luk available (%luk code, %luk reserved, "
 671                "%luk data, %luk init)\n", (unsigned long) nr_free_pages() << (PAGE_SHIFT - 10),
 672                num_physpages << (PAGE_SHIFT - 10), codesize >> 10,
 673                reserved_pages << (PAGE_SHIFT - 10), datasize >> 10, initsize >> 10);
 674
 675
 676         /*
 677          * For fsyscall entrpoints with no light-weight handler, use the ordinary
 678          * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
 679          * code can tell them apart.
 680          */
 681         for (i = 0; i < NR_syscalls; ++i) {
 682                 extern unsigned long fsyscall_table[NR_syscalls];
 683                 extern unsigned long sys_call_table[NR_syscalls];
 684
 685                 if (!fsyscall_table[i] || nolwsys)
 686                         fsyscall_table[i] = sys_call_table[i] | 1;
 687         }
 688         setup_gate();
 689
 690 #ifdef CONFIG_IA32_SUPPORT
 691         ia32_mem_init();
 692 #endif
 693 }
 694
 695 #ifdef CONFIG_MEMORY_HOTPLUG
 696 void online_page(struct page *page)
 697 {
 698         ClearPageReserved(page);
 699         init_page_count(page);
 700         __free_page(page);
 701         totalram_pages++;
 702         num_physpages++;
 703 }
 704
 705 int arch_add_memory(int nid, u64 start, u64 size)
 706 {
 707         pg_data_t *pgdat;
 708         struct zone *zone;
 709         unsigned long start_pfn = start >> PAGE_SHIFT;
 710         unsigned long nr_pages = size >> PAGE_SHIFT;
 711         int ret;
 712
 713         pgdat = NODE_DATA(nid);
 714
 715         zone = pgdat->node_zones + ZONE_NORMAL;
 716         ret = __add_pages(zone, start_pfn, nr_pages);
 717
 718         if (ret)
 719                 printk("%s: Problem encountered in __add_pages() as ret=%d\n",
 720                        __FUNCTION__,  ret);
 721
 722         return ret;
 723 }
 724
 725 int remove_memory(u64 start, u64 size)
 726 {
 727         return -EINVAL;
 728 }
 729 EXPORT_SYMBOL_GPL(remove_memory);
 730 #endif