arch/i386/mm/init-xen.c

   1 /*
   2  *  linux/arch/i386/mm/init.c
   3  *
   4  *  Copyright (C) 1995  Linus Torvalds
   5  *
   6  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
   7  */
   8
   9 #include <linux/module.h>
  10 #include <linux/signal.h>
  11 #include <linux/sched.h>
  12 #include <linux/kernel.h>
  13 #include <linux/errno.h>
  14 #include <linux/string.h>
  15 #include <linux/types.h>
  16 #include <linux/ptrace.h>
  17 #include <linux/mman.h>
  18 #include <linux/mm.h>
  19 #include <linux/hugetlb.h>
  20 #include <linux/swap.h>
  21 #include <linux/smp.h>
  22 #include <linux/init.h>
  23 #include <linux/highmem.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/poison.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/slab.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/efi.h>
  30 #include <linux/memory_hotplug.h>
  31 #include <linux/initrd.h>
  32 #include <linux/cpumask.h>
  33 #include <linux/dma-mapping.h>
  34 #include <linux/scatterlist.h>
  35
  36 #include <asm/processor.h>
  37 #include <asm/system.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/pgtable.h>
  40 #include <asm/dma.h>
  41 #include <asm/fixmap.h>
  42 #include <asm/e820.h>
  43 #include <asm/apic.h>
  44 #include <asm/tlb.h>
  45 #include <asm/tlbflush.h>
  46 #include <asm/sections.h>
  47 #include <asm/hypervisor.h>
  48 #include <asm/swiotlb.h>
  49
  50 extern unsigned long *contiguous_bitmap;
  51
  52 unsigned int __VMALLOC_RESERVE = 128 << 20;
  53
  54 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  55 unsigned long highstart_pfn, highend_pfn;
  56
  57 static int noinline do_test_wp_bit(void);
  58
  59 /*
  60  * Creates a middle page table and puts a pointer to it in the
  61  * given global directory entry. This only returns the gd entry
  62  * in non-PAE compilation mode, since the middle layer is folded.
  63  */
  64 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  65 {
  66         pud_t *pud;
  67         pmd_t *pmd_table;
  68
  69 #ifdef CONFIG_X86_PAE
  70         pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
  71         make_lowmem_page_readonly(pmd_table, XENFEAT_writable_page_tables);
  72         set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  73         pud = pud_offset(pgd, 0);
  74         if (pmd_table != pmd_offset(pud, 0))
  75                 BUG();
  76 #else
  77         pud = pud_offset(pgd, 0);
  78         pmd_table = pmd_offset(pud, 0);
  79 #endif
  80
  81         return pmd_table;
  82 }
  83
  84 /*
  85  * Create a page table and place a pointer to it in a middle page
  86  * directory entry.
  87  */
  88 static pte_t * __init one_page_table_init(pmd_t *pmd)
  89 {
  90         if (pmd_none(*pmd)) {
  91                 pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
  92                 make_lowmem_page_readonly(page_table,
  93                                           XENFEAT_writable_page_tables);
  94                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
  95                 if (page_table != pte_offset_kernel(pmd, 0))
  96                         BUG();
  97
  98                 return page_table;
  99         }
 100
 101         return pte_offset_kernel(pmd, 0);
 102 }
 103
 104 /*
 105  * This function initializes a certain range of kernel virtual memory
 106  * with new bootmem page tables, everywhere page tables are missing in
 107  * the given range.
 108  */
 109
 110 /*
 111  * NOTE: The pagetables are allocated contiguous on the physical space
 112  * so we can cache the place of the first one and move around without
 113  * checking the pgd every time.
 114  */
 115 static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base)
 116 {
 117         pgd_t *pgd;
 118         pud_t *pud;
 119         pmd_t *pmd;
 120         int pgd_idx, pmd_idx;
 121         unsigned long vaddr;
 122
 123         vaddr = start;
 124         pgd_idx = pgd_index(vaddr);
 125         pmd_idx = pmd_index(vaddr);
 126         pgd = pgd_base + pgd_idx;
 127
 128         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 129                 if (pgd_none(*pgd))
 130                         one_md_table_init(pgd);
 131                 pud = pud_offset(pgd, vaddr);
 132                 pmd = pmd_offset(pud, vaddr);
 133                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
 134                         if (vaddr < HYPERVISOR_VIRT_START && pmd_none(*pmd))
 135                                 one_page_table_init(pmd);
 136
 137                         vaddr += PMD_SIZE;
 138                 }
 139                 pmd_idx = 0;
 140         }
 141 }
 142
 143 static inline int is_kernel_text(unsigned long addr)
 144 {
 145         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 146                 return 1;
 147         return 0;
 148 }
 149
 150 /*
 151  * This maps the physical memory to kernel virtual address space, a total
 152  * of max_low_pfn pages, by creating page tables starting from address
 153  * PAGE_OFFSET.
 154  */
 155 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
 156 {
 157         unsigned long pfn;
 158         pgd_t *pgd;
 159         pmd_t *pmd;
 160         pte_t *pte;
 161         int pgd_idx, pmd_idx, pte_ofs;
 162
 163         unsigned long max_ram_pfn = xen_start_info->nr_pages;
 164         if (max_ram_pfn > max_low_pfn)
 165                 max_ram_pfn = max_low_pfn;
 166
 167         pgd_idx = pgd_index(PAGE_OFFSET);
 168         pgd = pgd_base + pgd_idx;
 169         pfn = 0;
 170         pmd_idx = pmd_index(PAGE_OFFSET);
 171         pte_ofs = pte_index(PAGE_OFFSET);
 172
 173         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 174 #ifdef CONFIG_XEN
 175                 /*
 176                  * Native linux hasn't PAE-paging enabled yet at this
 177                  * point.  When running as xen domain we are in PAE
 178                  * mode already, thus we can't simply hook a empty
 179                  * pmd.  That would kill the mappings we are currently
 180                  * using ...
 181                  */
 182                 pmd = pmd_offset(pud_offset(pgd, PAGE_OFFSET), PAGE_OFFSET);
 183 #else
 184                 pmd = one_md_table_init(pgd);
 185 #endif
 186                 if (pfn >= max_low_pfn)
 187                         continue;
 188                 pmd += pmd_idx;
 189                 for (; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
 190                         unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
 191                         if (address >= HYPERVISOR_VIRT_START)
 192                                 continue;
 193
 194                         /* Map with big pages if possible, otherwise create normal page tables. */
 195                         if (cpu_has_pse) {
 196                                 unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
 197
 198                                 if (is_kernel_text(address) || is_kernel_text(address2))
 199                                         set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
 200                                 else
 201                                         set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
 202                                 pfn += PTRS_PER_PTE;
 203                         } else {
 204                                 pte = one_page_table_init(pmd);
 205
 206                                 pte += pte_ofs;
 207                                 for (; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) {
 208                                                 /* XEN: Only map initial RAM allocation. */
 209                                                 if ((pfn >= max_ram_pfn) || pte_present(*pte))
 210                                                         continue;
 211                                                 if (is_kernel_text(address))
 212                                                         set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
 213                                                 else
 214                                                         set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
 215                                 }
 216                                 pte_ofs = 0;
 217                         }
 218                 }
 219                 pmd_idx = 0;
 220         }
 221 }
 222
 223 #ifndef CONFIG_XEN
 224
 225 static inline int page_kills_ppro(unsigned long pagenr)
 226 {
 227         if (pagenr >= 0x70000 && pagenr <= 0x7003F)
 228                 return 1;
 229         return 0;
 230 }
 231
 232 #else
 233
 234 #define page_kills_ppro(p)      0
 235
 236 #endif
 237
 238 int page_is_ram(unsigned long pagenr)
 239 {
 240         int i;
 241         unsigned long addr, end;
 242
 243         if (efi_enabled) {
 244                 efi_memory_desc_t *md;
 245                 void *p;
 246
 247                 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 248                         md = p;
 249                         if (!is_available_memory(md))
 250                                 continue;
 251                         addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
 252                         end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
 253
 254                         if ((pagenr >= addr) && (pagenr < end))
 255                                 return 1;
 256                 }
 257                 return 0;
 258         }
 259
 260         for (i = 0; i < e820.nr_map; i++) {
 261
 262                 if (e820.map[i].type != E820_RAM)       /* not usable memory */
 263                         continue;
 264                 /*
 265                  *      !!!FIXME!!! Some BIOSen report areas as RAM that
 266                  *      are not. Notably the 640->1Mb area. We need a sanity
 267                  *      check here.
 268                  */
 269                 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
 270                 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
 271                 if  ((pagenr >= addr) && (pagenr < end))
 272                         return 1;
 273         }
 274         return 0;
 275 }
 276 EXPORT_SYMBOL_GPL(page_is_ram);
 277
 278 /*
 279  * devmem_is_allowed() checks to see if /dev/mem access to a certain address is
 280  * valid. The argument is a physical page number.
 281  *
 282  *
 283  * On x86, access has to be given to the first megabyte of ram because that area
 284  * contains bios code and data regions used by X and dosemu and similar apps.
 285  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
 286  * mmio resources as well as potential bios/acpi data regions.
 287  */
 288 int devmem_is_allowed(unsigned long pagenr)
 289 {
 290    if (pagenr <= 256)
 291        return 1;
 292    if (!page_is_ram(pagenr))
 293        return 1;
 294    return 0;
 295 }
 296
 297 #ifdef CONFIG_HIGHMEM
 298 pte_t *kmap_pte;
 299 pgprot_t kmap_prot;
 300
 301 #define kmap_get_fixmap_pte(vaddr)                                      \
 302         pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr))
 303
 304 static void __init kmap_init(void)
 305 {
 306         unsigned long kmap_vstart;
 307
 308         /* cache the first kmap pte */
 309         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 310         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 311
 312         kmap_prot = PAGE_KERNEL;
 313 }
 314
 315 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 316 {
 317         pgd_t *pgd;
 318         pud_t *pud;
 319         pmd_t *pmd;
 320         pte_t *pte;
 321         unsigned long vaddr;
 322
 323         vaddr = PKMAP_BASE;
 324         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 325
 326         pgd = swapper_pg_dir + pgd_index(vaddr);
 327         pud = pud_offset(pgd, vaddr);
 328         pmd = pmd_offset(pud, vaddr);
 329         pte = pte_offset_kernel(pmd, vaddr);
 330         pkmap_page_table = pte;
 331 }
 332
 333 static void __meminit free_new_highpage(struct page *page, int pfn)
 334 {
 335         init_page_count(page);
 336         if (pfn < xen_start_info->nr_pages)
 337                 __free_page(page);
 338         totalhigh_pages++;
 339 }
 340
 341 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
 342 {
 343         if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
 344                 ClearPageReserved(page);
 345                 free_new_highpage(page, pfn);
 346         } else
 347                 SetPageReserved(page);
 348 }
 349
 350 static int __meminit add_one_highpage_hotplug(struct page *page, unsigned long pfn)
 351 {
 352         free_new_highpage(page, pfn);
 353         totalram_pages++;
 354 #ifdef CONFIG_FLATMEM
 355         max_mapnr = max(pfn, max_mapnr);
 356 #endif
 357         num_physpages++;
 358         return 0;
 359 }
 360
 361 /*
 362  * Not currently handling the NUMA case.
 363  * Assuming single node and all memory that
 364  * has been added dynamically that would be
 365  * onlined here is in HIGHMEM
 366  */
 367 void __meminit online_page(struct page *page)
 368 {
 369         ClearPageReserved(page);
 370         add_one_highpage_hotplug(page, page_to_pfn(page));
 371 }
 372
 373
 374 #ifdef CONFIG_NUMA
 375 extern void set_highmem_pages_init(int);
 376 #else
 377 static void __init set_highmem_pages_init(int bad_ppro)
 378 {
 379         int pfn;
 380         for (pfn = highstart_pfn; pfn < highend_pfn; pfn++)
 381                 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
 382         totalram_pages += totalhigh_pages;
 383 }
 384 #endif /* CONFIG_FLATMEM */
 385
 386 #else
 387 #define kmap_init() do { } while (0)
 388 #define permanent_kmaps_init(pgd_base) do { } while (0)
 389 #define set_highmem_pages_init(bad_ppro) do { } while (0)
 390 #endif /* CONFIG_HIGHMEM */
 391
 392 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
 393 EXPORT_SYMBOL(__PAGE_KERNEL);
 394 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
 395
 396 #ifdef CONFIG_NUMA
 397 extern void __init remap_numa_kva(void);
 398 #else
 399 #define remap_numa_kva() do {} while (0)
 400 #endif
 401
 402 pgd_t *swapper_pg_dir;
 403
 404 static void __init pagetable_init (void)
 405 {
 406         unsigned long vaddr;
 407         pgd_t *pgd_base = (pgd_t *)xen_start_info->pt_base;
 408
 409         swapper_pg_dir = pgd_base;
 410         init_mm.pgd    = pgd_base;
 411
 412         /* Enable PSE if available */
 413         if (cpu_has_pse) {
 414                 set_in_cr4(X86_CR4_PSE);
 415         }
 416
 417         /* Enable PGE if available */
 418         if (cpu_has_pge) {
 419                 set_in_cr4(X86_CR4_PGE);
 420                 __PAGE_KERNEL |= _PAGE_GLOBAL;
 421                 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
 422         }
 423
 424         kernel_physical_mapping_init(pgd_base);
 425         remap_numa_kva();
 426
 427         /*
 428          * Fixed mappings, only the page table structure has to be
 429          * created - mappings will be set by set_fixmap():
 430          */
 431         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 432         page_table_range_init(vaddr, 0, pgd_base);
 433
 434         permanent_kmaps_init(pgd_base);
 435 }
 436
 437 #if defined(CONFIG_SOFTWARE_SUSPEND) || defined(CONFIG_ACPI_SLEEP)
 438 /*
 439  * Swap suspend & friends need this for resume because things like the intel-agp
 440  * driver might have split up a kernel 4MB mapping.
 441  */
 442 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
 443         __attribute__ ((aligned (PAGE_SIZE)));
 444
 445 static inline void save_pg_dir(void)
 446 {
 447         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 448 }
 449 #else
 450 static inline void save_pg_dir(void)
 451 {
 452 }
 453 #endif
 454
 455 void zap_low_mappings (void)
 456 {
 457         int i;
 458
 459         save_pg_dir();
 460
 461         /*
 462          * Zap initial low-memory mappings.
 463          *
 464          * Note that "pgd_clear()" doesn't do it for
 465          * us, because pgd_clear() is a no-op on i386.
 466          */
 467         for (i = 0; i < USER_PTRS_PER_PGD; i++)
 468 #if defined(CONFIG_X86_PAE) && !defined(CONFIG_XEN)
 469                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 470 #else
 471                 set_pgd(swapper_pg_dir+i, __pgd(0));
 472 #endif
 473         flush_tlb_all();
 474 }
 475
 476 static int disable_nx __initdata = 0;
 477 u64 __supported_pte_mask __read_mostly = ~_PAGE_NX;
 478 EXPORT_SYMBOL(__supported_pte_mask);
 479
 480 /*
 481  * noexec = on|off
 482  *
 483  * Control non executable mappings.
 484  *
 485  * on      Enable
 486  * off     Disable (disables exec-shield too)
 487  */
 488 static int __init noexec_setup(char *str)
 489 {
 490         if (!str || !strcmp(str, "on")) {
 491                 if (cpu_has_nx) {
 492                         __supported_pte_mask |= _PAGE_NX;
 493                         disable_nx = 0;
 494                 }
 495         } else if (!strcmp(str,"off")) {
 496                 disable_nx = 1;
 497                 __supported_pte_mask &= ~_PAGE_NX;
 498                 exec_shield = 0;
 499         } else
 500                 return -EINVAL;
 501
 502         return 0;
 503 }
 504 early_param("noexec", noexec_setup);
 505
 506 int nx_enabled = 0;
 507 #ifdef CONFIG_X86_PAE
 508
 509 static void __init set_nx(void)
 510 {
 511         unsigned int v[4], l, h;
 512
 513         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 514                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 515                 if ((v[3] & (1 << 20)) && !disable_nx) {
 516                         rdmsr(MSR_EFER, l, h);
 517                         l |= EFER_NX;
 518                         wrmsr(MSR_EFER, l, h);
 519                         nx_enabled = 1;
 520                         __supported_pte_mask |= _PAGE_NX;
 521                 }
 522         }
 523 }
 524
 525 /*
 526  * Enables/disables executability of a given kernel page and
 527  * returns the previous setting.
 528  */
 529 int __init set_kernel_exec(unsigned long vaddr, int enable)
 530 {
 531         pte_t *pte;
 532         int ret = 1;
 533
 534         if (!nx_enabled)
 535                 goto out;
 536
 537         pte = lookup_address(vaddr);
 538         BUG_ON(!pte);
 539
 540         if (!pte_exec_kernel(*pte))
 541                 ret = 0;
 542
 543         if (enable)
 544                 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
 545         else
 546                 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
 547         pte_update_defer(&init_mm, vaddr, pte);
 548         __flush_tlb_all();
 549 out:
 550         return ret;
 551 }
 552
 553 #endif
 554
 555 /*
 556  * paging_init() sets up the page tables - note that the first 8MB are
 557  * already mapped by head.S.
 558  *
 559  * This routines also unmaps the page at virtual kernel address 0, so
 560  * that we can trap those pesky NULL-reference errors in the kernel.
 561  */
 562 void __init paging_init(void)
 563 {
 564         int i;
 565
 566 #ifdef CONFIG_X86_PAE
 567         set_nx();
 568         if (nx_enabled)
 569                 printk("NX (Execute Disable) protection: active\n");
 570         else
 571 #endif
 572         if (exec_shield)
 573                 printk("Using x86 segment limits to approximate NX protection\n");
 574
 575         pagetable_init();
 576
 577 #if defined(CONFIG_X86_PAE) && !defined(CONFIG_XEN)
 578         /*
 579          * We will bail out later - printk doesn't work right now so
 580          * the user would just see a hanging kernel.
 581          * when running as xen domain we are already in PAE mode at
 582          * this point.
 583          */
 584         if (cpu_has_pae)
 585                 set_in_cr4(X86_CR4_PAE);
 586 #endif
 587         __flush_tlb_all();
 588
 589         kmap_init();
 590
 591         /* Switch to the real shared_info page, and clear the
 592          * dummy page. */
 593         set_fixmap(FIX_SHARED_INFO, xen_start_info->shared_info);
 594         HYPERVISOR_shared_info = (shared_info_t *)fix_to_virt(FIX_SHARED_INFO);
 595         memset(empty_zero_page, 0, sizeof(empty_zero_page));
 596
 597         /* Setup mapping of lower 1st MB */
 598         for (i = 0; i < NR_FIX_ISAMAPS; i++)
 599                 if (is_initial_xendomain())
 600                         set_fixmap(FIX_ISAMAP_BEGIN - i, i * PAGE_SIZE);
 601                 else
 602                         __set_fixmap(FIX_ISAMAP_BEGIN - i,
 603                                      virt_to_machine(empty_zero_page),
 604                                      PAGE_KERNEL_RO);
 605 }
 606
 607 /*
 608  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 609  * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
 610  * used to involve black magic jumps to work around some nasty CPU bugs,
 611  * but fortunately the switch to using exceptions got rid of all that.
 612  */
 613
 614 static void __init test_wp_bit(void)
 615 {
 616         printk("Checking if this processor honours the WP bit even in supervisor mode... ");
 617
 618         /* Any page-aligned address will do, the test is non-destructive */
 619         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 620         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 621         clear_fixmap(FIX_WP_TEST);
 622
 623         if (!boot_cpu_data.wp_works_ok) {
 624                 printk("No.\n");
 625 #ifdef CONFIG_X86_WP_WORKS_OK
 626                 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 627 #endif
 628         } else {
 629                 printk("Ok.\n");
 630         }
 631 }
 632
 633 static struct kcore_list kcore_mem, kcore_vmalloc;
 634
 635 void __init mem_init(void)
 636 {
 637         extern int ppro_with_ram_bug(void);
 638         int codesize, reservedpages, datasize, initsize;
 639         int tmp;
 640         int bad_ppro;
 641         unsigned long pfn;
 642
 643         contiguous_bitmap = alloc_bootmem_low_pages(
 644                 (max_low_pfn + 2*BITS_PER_LONG) >> 3);
 645         BUG_ON(!contiguous_bitmap);
 646         memset(contiguous_bitmap, 0, (max_low_pfn + 2*BITS_PER_LONG) >> 3);
 647
 648 #if defined(CONFIG_SWIOTLB)
 649         swiotlb_init();
 650 #endif
 651
 652 #ifdef CONFIG_FLATMEM
 653         BUG_ON(!mem_map);
 654 #endif
 655
 656         bad_ppro = ppro_with_ram_bug();
 657
 658 #ifdef CONFIG_HIGHMEM
 659         /* check that fixmap and pkmap do not overlap */
 660         if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
 661                 printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n");
 662                 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
 663                                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START);
 664                 BUG();
 665         }
 666 #endif
 667
 668         printk("vmalloc area: %lx-%lx, maxmem %lx\n",
 669                VMALLOC_START,VMALLOC_END,MAXMEM);
 670         BUG_ON(VMALLOC_START > VMALLOC_END);
 671
 672         /* this will put all low memory onto the freelists */
 673         totalram_pages += free_all_bootmem();
 674         /* XEN: init and count low-mem pages outside initial allocation. */
 675         for (pfn = xen_start_info->nr_pages; pfn < max_low_pfn; pfn++) {
 676                 ClearPageReserved(&mem_map[pfn]);
 677                 init_page_count(&mem_map[pfn]);
 678                 totalram_pages++;
 679         }
 680
 681         reservedpages = 0;
 682         for (tmp = 0; tmp < max_low_pfn; tmp++)
 683                 /*
 684                  * Only count reserved RAM pages
 685                  */
 686                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 687                         reservedpages++;
 688
 689         set_highmem_pages_init(bad_ppro);
 690
 691         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 692         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 693         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 694
 695         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 696         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 697                    VMALLOC_END-VMALLOC_START);
 698
 699         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
 700                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 701                 num_physpages << (PAGE_SHIFT-10),
 702                 codesize >> 10,
 703                 reservedpages << (PAGE_SHIFT-10),
 704                 datasize >> 10,
 705                 initsize >> 10,
 706                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 707                );
 708
 709 #if 1 /* double-sanity-check paranoia */
 710         printk("virtual kernel memory layout:\n"
 711                "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 712 #ifdef CONFIG_HIGHMEM
 713                "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 714 #endif
 715                "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 716                "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 717                "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 718                "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 719                "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 720                FIXADDR_START, FIXADDR_TOP,
 721                (FIXADDR_TOP - FIXADDR_START) >> 10,
 722
 723 #ifdef CONFIG_HIGHMEM
 724                PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 725                (LAST_PKMAP*PAGE_SIZE) >> 10,
 726 #endif
 727
 728                VMALLOC_START, VMALLOC_END,
 729                (VMALLOC_END - VMALLOC_START) >> 20,
 730
 731                (unsigned long)__va(0), (unsigned long)high_memory,
 732                ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 733
 734                (unsigned long)&__init_begin, (unsigned long)&__init_end,
 735                ((unsigned long)&__init_end - (unsigned long)&__init_begin) >> 10,
 736
 737                (unsigned long)&_etext, (unsigned long)&_edata,
 738                ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 739
 740                (unsigned long)&_text, (unsigned long)&_etext,
 741                ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 742
 743 #ifdef CONFIG_HIGHMEM
 744         BUG_ON(PKMAP_BASE+LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
 745         BUG_ON(VMALLOC_END                     > PKMAP_BASE);
 746 #endif
 747         BUG_ON(VMALLOC_START                   > VMALLOC_END);
 748         BUG_ON((unsigned long)high_memory      > VMALLOC_START);
 749 #endif /* double-sanity-check paranoia */
 750
 751 #ifdef CONFIG_X86_PAE
 752         if (!cpu_has_pae)
 753                 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
 754 #endif
 755         if (boot_cpu_data.wp_works_ok < 0)
 756                 test_wp_bit();
 757
 758         /*
 759          * Subtle. SMP is doing it's boot stuff late (because it has to
 760          * fork idle threads) - but it also needs low mappings for the
 761          * protected-mode entry to work. We zap these entries only after
 762          * the WP-bit has been tested.
 763          */
 764 #ifndef CONFIG_SMP
 765         zap_low_mappings();
 766 #endif
 767
 768         set_bit(PG_pinned, &virt_to_page(init_mm.pgd)->flags);
 769 }
 770
 771 #ifdef CONFIG_MEMORY_HOTPLUG
 772 int arch_add_memory(int nid, u64 start, u64 size)
 773 {
 774         struct pglist_data *pgdata = NODE_DATA(nid);
 775         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 776         unsigned long start_pfn = start >> PAGE_SHIFT;
 777         unsigned long nr_pages = size >> PAGE_SHIFT;
 778
 779         return __add_pages(zone, start_pfn, nr_pages);
 780 }
 781
 782 int remove_memory(u64 start, u64 size)
 783 {
 784         return -EINVAL;
 785 }
 786 EXPORT_SYMBOL_GPL(remove_memory);
 787 #endif
 788
 789 struct kmem_cache *pgd_cache;
 790 struct kmem_cache *pmd_cache;
 791
 792 void __init pgtable_cache_init(void)
 793 {
 794         if (PTRS_PER_PMD > 1) {
 795                 pmd_cache = kmem_cache_create("pmd",
 796                                         PTRS_PER_PMD*sizeof(pmd_t),
 797                                         PTRS_PER_PMD*sizeof(pmd_t),
 798                                         0,
 799                                         pmd_ctor,
 800                                         NULL);
 801                 if (!pmd_cache)
 802                         panic("pgtable_cache_init(): cannot create pmd cache");
 803         }
 804         pgd_cache = kmem_cache_create("pgd",
 805 #ifndef CONFIG_XEN
 806                                 PTRS_PER_PGD*sizeof(pgd_t),
 807                                 PTRS_PER_PGD*sizeof(pgd_t),
 808 #else
 809                                 PAGE_SIZE,
 810                                 PAGE_SIZE,
 811 #endif
 812                                 0,
 813                                 pgd_ctor,
 814                                 PTRS_PER_PMD == 1 ? pgd_dtor : NULL);
 815         if (!pgd_cache)
 816                 panic("pgtable_cache_init(): Cannot create pgd cache");
 817 }
 818
 819 /*
 820  * This function cannot be __init, since exceptions don't work in that
 821  * section.  Put this after the callers, so that it cannot be inlined.
 822  */
 823 static int noinline do_test_wp_bit(void)
 824 {
 825         char tmp_reg;
 826         int flag;
 827
 828         __asm__ __volatile__(
 829                 "       movb %0,%1      \n"
 830                 "1:     movb %1,%0      \n"
 831                 "       xorl %2,%2      \n"
 832                 "2:                     \n"
 833                 ".section __ex_table,\"a\"\n"
 834                 "       .align 4        \n"
 835                 "       .long 1b,2b     \n"
 836                 ".previous              \n"
 837                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 838                  "=q" (tmp_reg),
 839                  "=r" (flag)
 840                 :"2" (1)
 841                 :"memory");
 842
 843         return flag;
 844 }
 845
 846 #ifdef CONFIG_DEBUG_RODATA
 847
 848 void mark_rodata_ro(void)
 849 {
 850         unsigned long addr = (unsigned long)__start_rodata;
 851
 852         for (; addr < (unsigned long)__end_rodata; addr += PAGE_SIZE)
 853                 change_page_attr(virt_to_page(addr), 1, PAGE_KERNEL_RO);
 854
 855         printk("Write protecting the kernel read-only data: %uk\n",
 856                         (__end_rodata - __start_rodata) >> 10);
 857
 858         /*
 859          * change_page_attr() requires a global_flush_tlb() call after it.
 860          * We do this after the printk so that if something went wrong in the
 861          * change, the printk gets out at least to give a better debug hint
 862          * of who is the culprit.
 863          */
 864         global_flush_tlb();
 865 }
 866 #endif
 867
 868 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 869 {
 870         unsigned long addr;
 871
 872         for (addr = begin; addr < end; addr += PAGE_SIZE) {
 873                 ClearPageReserved(virt_to_page(addr));
 874                 init_page_count(virt_to_page(addr));
 875                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
 876                 free_page(addr);
 877                 totalram_pages++;
 878         }
 879         printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10);
 880 }
 881
 882 void free_initmem(void)
 883 {
 884         free_init_pages("unused kernel memory",
 885                         (unsigned long)(&__init_begin),
 886                         (unsigned long)(&__init_end));
 887 }
 888
 889 #ifdef CONFIG_BLK_DEV_INITRD
 890 void free_initrd_mem(unsigned long start, unsigned long end)
 891 {
 892         free_init_pages("initrd memory", start, end);
 893 }
 894 #endif
 895