arch/xen/i386/mm/init.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/config.h>
  10 #include <linux/module.h>
  11 #include <linux/signal.h>
  12 #include <linux/sched.h>
  13 #include <linux/kernel.h>
  14 #include <linux/errno.h>
  15 #include <linux/string.h>
  16 #include <linux/types.h>
  17 #include <linux/ptrace.h>
  18 #include <linux/mman.h>
  19 #include <linux/mm.h>
  20 #include <linux/hugetlb.h>
  21 #include <linux/swap.h>
  22 #include <linux/smp.h>
  23 #include <linux/init.h>
  24 #include <linux/highmem.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/slab.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/efi.h>
  30
  31 #include <asm/processor.h>
  32 #include <asm/system.h>
  33 #include <asm/uaccess.h>
  34 #include <asm/pgtable.h>
  35 #include <asm/dma.h>
  36 #include <asm/fixmap.h>
  37 #include <asm/e820.h>
  38 #include <asm/apic.h>
  39 #include <asm/tlb.h>
  40 #include <asm/tlbflush.h>
  41 #include <asm/sections.h>
  42 #include <asm-xen/hypervisor.h>
  43
  44 unsigned int __VMALLOC_RESERVE = 128 << 20;
  45
  46 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  47 unsigned long highstart_pfn, highend_pfn;
  48
  49 static int noinline do_test_wp_bit(void);
  50
  51 /*
  52  * Creates a middle page table and puts a pointer to it in the
  53  * given global directory entry. This only returns the gd entry
  54  * in non-PAE compilation mode, since the middle layer is folded.
  55  */
  56 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  57 {
  58         pud_t *pud;
  59         pmd_t *pmd_table;
  60
  61 #ifdef CONFIG_X86_PAE
  62         pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
  63         set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  64         pud = pud_offset(pgd, 0);
  65         if (pmd_table != pmd_offset(pud, 0))
  66                 BUG();
  67 #else
  68         pud = pud_offset(pgd, 0);
  69         pmd_table = pmd_offset(pud, 0);
  70 #endif
  71
  72         return pmd_table;
  73 }
  74
  75 /*
  76  * Create a page table and place a pointer to it in a middle page
  77  * directory entry.
  78  */
  79 static pte_t * __init one_page_table_init(pmd_t *pmd)
  80 {
  81         if (pmd_none(*pmd)) {
  82                 pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
  83                 make_page_readonly(page_table);
  84                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
  85                 if (page_table != pte_offset_kernel(pmd, 0))
  86                         BUG();
  87
  88                 return page_table;
  89         }
  90
  91         return pte_offset_kernel(pmd, 0);
  92 }
  93
  94 /*
  95  * This function initializes a certain range of kernel virtual memory
  96  * with new bootmem page tables, everywhere page tables are missing in
  97  * the given range.
  98  */
  99
 100 /*
 101  * NOTE: The pagetables are allocated contiguous on the physical space
 102  * so we can cache the place of the first one and move around without
 103  * checking the pgd every time.
 104  */
 105 static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base)
 106 {
 107         pgd_t *pgd;
 108         pud_t *pud;
 109         pmd_t *pmd;
 110         int pgd_idx, pmd_idx;
 111         unsigned long vaddr;
 112
 113         vaddr = start;
 114         pgd_idx = pgd_index(vaddr);
 115         pmd_idx = pmd_index(vaddr);
 116         pgd = pgd_base + pgd_idx;
 117
 118         for ( ; (pgd_idx < PTRS_PER_PGD_NO_HV) && (vaddr != end); pgd++, pgd_idx++) {
 119                 if (pgd_none(*pgd))
 120                         one_md_table_init(pgd);
 121                 pud = pud_offset(pgd, vaddr);
 122                 pmd = pmd_offset(pud, vaddr);
 123                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
 124                         if (pmd_none(*pmd))
 125                                 one_page_table_init(pmd);
 126
 127                         vaddr += PMD_SIZE;
 128                 }
 129                 pmd_idx = 0;
 130         }
 131 }
 132
 133 static inline int is_kernel_text(unsigned long addr)
 134 {
 135         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 136                 return 1;
 137         return 0;
 138 }
 139
 140 /*
 141  * This maps the physical memory to kernel virtual address space, a total
 142  * of max_low_pfn pages, by creating page tables starting from address
 143  * PAGE_OFFSET.
 144  */
 145 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
 146 {
 147         unsigned long pfn;
 148         pgd_t *pgd;
 149         pmd_t *pmd;
 150         pte_t *pte;
 151         int pgd_idx, pmd_idx, pte_ofs;
 152
 153         unsigned long max_ram_pfn = xen_start_info.nr_pages;
 154         if (max_ram_pfn > max_low_pfn)
 155                 max_ram_pfn = max_low_pfn;
 156
 157         pgd_idx = pgd_index(PAGE_OFFSET);
 158         pgd = pgd_base + pgd_idx;
 159         pfn = 0;
 160         pmd_idx = pmd_index(PAGE_OFFSET);
 161         pte_ofs = pte_index(PAGE_OFFSET);
 162
 163         for (; pgd_idx < PTRS_PER_PGD_NO_HV; pgd++, pgd_idx++) {
 164                 pmd = one_md_table_init(pgd);
 165                 if (pfn >= max_low_pfn)
 166                         continue;
 167                 pmd += pmd_idx;
 168                 for (; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
 169                         unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
 170
 171                         /* Map with big pages if possible, otherwise create normal page tables. */
 172                         if (cpu_has_pse) {
 173                                 unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
 174
 175                                 if (is_kernel_text(address) || is_kernel_text(address2))
 176                                         set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
 177                                 else
 178                                         set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
 179                                 pfn += PTRS_PER_PTE;
 180                         } else {
 181                                 pte = one_page_table_init(pmd);
 182
 183                                 pte += pte_ofs;
 184                                 for (; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) {
 185                                                 /* XEN: Only map initial RAM allocation. */
 186                                                 if ((pfn >= max_ram_pfn) || pte_present(*pte))
 187                                                         continue;
 188                                                 if (is_kernel_text(address))
 189                                                         set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
 190                                                 else
 191                                                         set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
 192                                 }
 193                                 pte_ofs = 0;
 194                         }
 195                 }
 196                 pmd_idx = 0;
 197         }
 198 }
 199
 200 static inline int page_kills_ppro(unsigned long pagenr)
 201 {
 202         if (pagenr >= 0x70000 && pagenr <= 0x7003F)
 203                 return 1;
 204         return 0;
 205 }
 206
 207 extern int is_available_memory(efi_memory_desc_t *);
 208
 209 static inline int page_is_ram(unsigned long pagenr)
 210 {
 211         int i;
 212         unsigned long addr, end;
 213
 214         if (efi_enabled) {
 215                 efi_memory_desc_t *md;
 216
 217                 for (i = 0; i < memmap.nr_map; i++) {
 218                         md = &memmap.map[i];
 219                         if (!is_available_memory(md))
 220                                 continue;
 221                         addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
 222                         end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
 223
 224                         if ((pagenr >= addr) && (pagenr < end))
 225                                 return 1;
 226                 }
 227                 return 0;
 228         }
 229
 230         for (i = 0; i < e820.nr_map; i++) {
 231
 232                 if (e820.map[i].type != E820_RAM)       /* not usable memory */
 233                         continue;
 234                 /*
 235                  *      !!!FIXME!!! Some BIOSen report areas as RAM that
 236                  *      are not. Notably the 640->1Mb area. We need a sanity
 237                  *      check here.
 238                  */
 239                 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
 240                 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
 241                 if  ((pagenr >= addr) && (pagenr < end))
 242                         return 1;
 243         }
 244         return 0;
 245 }
 246 EXPORT_SYMBOL_GPL(page_is_ram);
 247
 248 #ifdef CONFIG_HIGHMEM
 249 pte_t *kmap_pte;
 250 pgprot_t kmap_prot;
 251
 252 EXPORT_SYMBOL(kmap_prot);
 253 EXPORT_SYMBOL(kmap_pte);
 254
 255 #define kmap_get_fixmap_pte(vaddr)                                      \
 256         pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr))
 257
 258 void __init kmap_init(void)
 259 {
 260         unsigned long kmap_vstart;
 261
 262         /* cache the first kmap pte */
 263         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 264         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 265
 266         kmap_prot = PAGE_KERNEL;
 267 }
 268
 269 void __init permanent_kmaps_init(pgd_t *pgd_base)
 270 {
 271         pgd_t *pgd;
 272         pud_t *pud;
 273         pmd_t *pmd;
 274         pte_t *pte;
 275         unsigned long vaddr;
 276
 277         vaddr = PKMAP_BASE;
 278         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 279
 280         pgd = swapper_pg_dir + pgd_index(vaddr);
 281         pud = pud_offset(pgd, vaddr);
 282         pmd = pmd_offset(pud, vaddr);
 283         pte = pte_offset_kernel(pmd, vaddr);
 284         pkmap_page_table = pte;
 285 }
 286
 287 void __init one_highpage_init(struct page *page, int pfn, int bad_ppro)
 288 {
 289         if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
 290                 ClearPageReserved(page);
 291                 set_bit(PG_highmem, &page->flags);
 292                 set_page_count(page, 1);
 293                 if (pfn < xen_start_info.nr_pages)
 294                         __free_page(page);
 295                 totalhigh_pages++;
 296         } else
 297                 SetPageReserved(page);
 298 }
 299
 300 #ifndef CONFIG_DISCONTIGMEM
 301 void __init set_highmem_pages_init(int bad_ppro)
 302 {
 303         int pfn;
 304         for (pfn = highstart_pfn; pfn < highend_pfn; pfn++)
 305                 one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
 306         totalram_pages += totalhigh_pages;
 307 }
 308 #else
 309 extern void set_highmem_pages_init(int);
 310 #endif /* !CONFIG_DISCONTIGMEM */
 311
 312 #else
 313 #define kmap_init() do { } while (0)
 314 #define permanent_kmaps_init(pgd_base) do { } while (0)
 315 #define set_highmem_pages_init(bad_ppro) do { } while (0)
 316 #endif /* CONFIG_HIGHMEM */
 317
 318 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
 319 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
 320
 321 #ifndef CONFIG_DISCONTIGMEM
 322 #define remap_numa_kva() do {} while (0)
 323 #else
 324 extern void __init remap_numa_kva(void);
 325 #endif
 326
 327 static void __init pagetable_init (void)
 328 {
 329         unsigned long vaddr;
 330         pgd_t *pgd_base = swapper_pg_dir;
 331         pgd_t *old_pgd = (pgd_t *)xen_start_info.pt_base;
 332
 333 #ifdef CONFIG_X86_PAE
 334         int i;
 335         /* Init entries of the first-level page table to the zero page */
 336         for (i = 0; i < PTRS_PER_PGD; i++)
 337                 set_pgd(pgd_base + i, __pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
 338 #endif
 339
 340         /* Enable PSE if available */
 341         if (cpu_has_pse) {
 342                 set_in_cr4(X86_CR4_PSE);
 343         }
 344
 345         /* Enable PGE if available */
 346         if (cpu_has_pge) {
 347                 set_in_cr4(X86_CR4_PGE);
 348                 __PAGE_KERNEL |= _PAGE_GLOBAL;
 349                 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
 350         }
 351
 352         /*
 353          * Switch to proper mm_init page directory. Initialise from the current
 354          * page directory, write-protect the new page directory, then switch to
 355          * it. We clean up by write-enabling and then freeing the old page dir.
 356          */
 357         memcpy(pgd_base, old_pgd, PTRS_PER_PGD_NO_HV*sizeof(pgd_t));
 358         make_page_readonly(pgd_base);
 359         xen_pgd_pin(__pa(pgd_base));
 360         load_cr3(pgd_base);
 361         xen_pgd_unpin(__pa(old_pgd));
 362         make_page_writable(old_pgd);
 363         __flush_tlb_all();
 364         free_bootmem(__pa(old_pgd), PAGE_SIZE);
 365
 366         kernel_physical_mapping_init(pgd_base);
 367         remap_numa_kva();
 368
 369         /*
 370          * Fixed mappings, only the page table structure has to be
 371          * created - mappings will be set by set_fixmap():
 372          */
 373         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 374         page_table_range_init(vaddr, 0, pgd_base);
 375
 376         permanent_kmaps_init(pgd_base);
 377
 378 #ifdef CONFIG_X86_PAE
 379         /*
 380          * Add low memory identity-mappings - SMP needs it when
 381          * starting up on an AP from real-mode. In the non-PAE
 382          * case we already have these mappings through head.S.
 383          * All user-space mappings are explicitly cleared after
 384          * SMP startup.
 385          */
 386         pgd_base[0] = pgd_base[USER_PTRS_PER_PGD];
 387 #endif
 388 }
 389
 390 #if defined(CONFIG_PM_DISK) || defined(CONFIG_SOFTWARE_SUSPEND)
 391 /*
 392  * Swap suspend & friends need this for resume because things like the intel-agp
 393  * driver might have split up a kernel 4MB mapping.
 394  */
 395 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
 396         __attribute__ ((aligned (PAGE_SIZE)));
 397
 398 static inline void save_pg_dir(void)
 399 {
 400         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 401 }
 402 #else
 403 static inline void save_pg_dir(void)
 404 {
 405 }
 406 #endif
 407
 408 void zap_low_mappings (void)
 409 {
 410         int i;
 411
 412         save_pg_dir();
 413
 414         /*
 415          * Zap initial low-memory mappings.
 416          *
 417          * Note that "pgd_clear()" doesn't do it for
 418          * us, because pgd_clear() is a no-op on i386.
 419          */
 420         for (i = 0; i < USER_PTRS_PER_PGD; i++)
 421 #ifdef CONFIG_X86_PAE
 422                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 423 #else
 424                 set_pgd(swapper_pg_dir+i, __pgd(0));
 425 #endif
 426         flush_tlb_all();
 427 }
 428
 429 #ifndef CONFIG_DISCONTIGMEM
 430 void __init zone_sizes_init(void)
 431 {
 432         unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
 433         unsigned int /*max_dma,*/ high, low;
 434
 435         /*
 436          * XEN: Our notion of "DMA memory" is fake when running over Xen.
 437          * We simply put all RAM in the DMA zone so that those drivers which
 438          * needlessly specify GFP_DMA do not get starved of RAM unnecessarily.
 439          * Those drivers that *do* require lowmem are screwed anyway when
 440          * running over Xen!
 441          */
 442         /*max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;*/
 443         low = max_low_pfn;
 444         high = highend_pfn;
 445
 446         /*if (low < max_dma)*/
 447                 zones_size[ZONE_DMA] = low;
 448         /*else*/ {
 449                 /*zones_size[ZONE_DMA] = max_dma;*/
 450                 /*zones_size[ZONE_NORMAL] = low - max_dma;*/
 451 #ifdef CONFIG_HIGHMEM
 452                 zones_size[ZONE_HIGHMEM] = high - low;
 453 #endif
 454         }
 455         free_area_init(zones_size);
 456 }
 457 #else
 458 extern void zone_sizes_init(void);
 459 #endif /* !CONFIG_DISCONTIGMEM */
 460
 461 static int disable_nx __initdata = 0;
 462 u64 __supported_pte_mask = ~_PAGE_NX;
 463
 464 /*
 465  * noexec = on|off
 466  *
 467  * Control non executable mappings.
 468  *
 469  * on      Enable
 470  * off     Disable
 471  */
 472 void __init noexec_setup(const char *str)
 473 {
 474         if (!strncmp(str, "on",2) && cpu_has_nx) {
 475                 __supported_pte_mask |= _PAGE_NX;
 476                 disable_nx = 0;
 477         } else if (!strncmp(str,"off",3)) {
 478                 disable_nx = 1;
 479                 __supported_pte_mask &= ~_PAGE_NX;
 480         }
 481 }
 482
 483 int nx_enabled = 0;
 484 #ifdef CONFIG_X86_PAE
 485
 486 static void __init set_nx(void)
 487 {
 488         unsigned int v[4], l, h;
 489
 490         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 491                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 492                 if ((v[3] & (1 << 20)) && !disable_nx) {
 493                         rdmsr(MSR_EFER, l, h);
 494                         l |= EFER_NX;
 495                         wrmsr(MSR_EFER, l, h);
 496                         nx_enabled = 1;
 497                         __supported_pte_mask |= _PAGE_NX;
 498                 }
 499         }
 500 }
 501
 502 /*
 503  * Enables/disables executability of a given kernel page and
 504  * returns the previous setting.
 505  */
 506 int __init set_kernel_exec(unsigned long vaddr, int enable)
 507 {
 508         pte_t *pte;
 509         int ret = 1;
 510
 511         if (!nx_enabled)
 512                 goto out;
 513
 514         pte = lookup_address(vaddr);
 515         BUG_ON(!pte);
 516
 517         if (!pte_exec_kernel(*pte))
 518                 ret = 0;
 519
 520         if (enable)
 521                 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
 522         else
 523                 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
 524         __flush_tlb_all();
 525 out:
 526         return ret;
 527 }
 528
 529 #endif
 530
 531 /*
 532  * paging_init() sets up the page tables - note that the first 8MB are
 533  * already mapped by head.S.
 534  *
 535  * This routines also unmaps the page at virtual kernel address 0, so
 536  * that we can trap those pesky NULL-reference errors in the kernel.
 537  */
 538 void __init paging_init(void)
 539 {
 540 #ifdef CONFIG_XEN_PHYSDEV_ACCESS
 541         int i;
 542 #endif
 543
 544 #ifdef CONFIG_X86_PAE
 545         set_nx();
 546         if (nx_enabled)
 547                 printk("NX (Execute Disable) protection: active\n");
 548         else
 549 #endif
 550         if (exec_shield)
 551                 printk("Using x86 segment limits to approximate NX protection\n");
 552
 553         pagetable_init();
 554
 555 #ifdef CONFIG_X86_PAE
 556         /*
 557          * We will bail out later - printk doesn't work right now so
 558          * the user would just see a hanging kernel.
 559          */
 560         if (cpu_has_pae)
 561                 set_in_cr4(X86_CR4_PAE);
 562 #endif
 563         __flush_tlb_all();
 564
 565         kmap_init();
 566         zone_sizes_init();
 567
 568         /* Switch to the real shared_info page, and clear the dummy page. */
 569         set_fixmap_ma(FIX_SHARED_INFO, xen_start_info.shared_info);
 570         HYPERVISOR_shared_info = (shared_info_t *)fix_to_virt(FIX_SHARED_INFO);
 571         memset(empty_zero_page, 0, sizeof(empty_zero_page));
 572
 573 #ifdef CONFIG_XEN_PHYSDEV_ACCESS
 574         /* Setup mapping of lower 1st MB */
 575         for (i = 0; i < NR_FIX_ISAMAPS; i++)
 576                 if (xen_start_info.flags & SIF_PRIVILEGED)
 577                         set_fixmap_ma(FIX_ISAMAP_BEGIN - i, i * PAGE_SIZE);
 578                 else
 579                         set_fixmap_ma_ro(FIX_ISAMAP_BEGIN - i,
 580                                          virt_to_machine(empty_zero_page));
 581 #endif
 582 }
 583
 584 /*
 585  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 586  * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
 587  * used to involve black magic jumps to work around some nasty CPU bugs,
 588  * but fortunately the switch to using exceptions got rid of all that.
 589  */
 590
 591 void __init test_wp_bit(void)
 592 {
 593         printk("Checking if this processor honours the WP bit even in supervisor mode... ");
 594
 595         /* Any page-aligned address will do, the test is non-destructive */
 596         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 597         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 598         clear_fixmap(FIX_WP_TEST);
 599
 600         if (!boot_cpu_data.wp_works_ok) {
 601                 printk("No.\n");
 602 #ifdef CONFIG_X86_WP_WORKS_OK
 603                 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 604 #endif
 605         } else {
 606                 printk("Ok.\n");
 607         }
 608 }
 609
 610 #ifndef CONFIG_DISCONTIGMEM
 611 static void __init set_max_mapnr_init(void)
 612 {
 613 #ifdef CONFIG_HIGHMEM
 614         max_mapnr = num_physpages = highend_pfn;
 615 #else
 616         max_mapnr = num_physpages = max_low_pfn;
 617 #endif
 618 }
 619 #define __free_all_bootmem() free_all_bootmem()
 620 #else
 621 #define __free_all_bootmem() free_all_bootmem_node(NODE_DATA(0))
 622 extern void set_max_mapnr_init(void);
 623 #endif /* !CONFIG_DISCONTIGMEM */
 624
 625 static struct kcore_list kcore_mem, kcore_vmalloc;
 626
 627 void __init mem_init(void)
 628 {
 629         extern int ppro_with_ram_bug(void);
 630         int codesize, reservedpages, datasize, initsize;
 631         int tmp;
 632         int bad_ppro;
 633         unsigned long pfn;
 634
 635 #ifndef CONFIG_DISCONTIGMEM
 636         if (!mem_map)
 637                 BUG();
 638 #endif
 639
 640         bad_ppro = ppro_with_ram_bug();
 641
 642 #ifdef CONFIG_HIGHMEM
 643         /* check that fixmap and pkmap do not overlap */
 644         if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
 645                 printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n");
 646                 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
 647                                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START);
 648                 BUG();
 649         }
 650 #endif
 651
 652         set_max_mapnr_init();
 653
 654 #ifdef CONFIG_HIGHMEM
 655         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE);
 656 #else
 657         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 658 #endif
 659
 660         /* this will put all low memory onto the freelists */
 661         totalram_pages += __free_all_bootmem();
 662         /* XEN: init and count low-mem pages outside initial allocation. */
 663         for (pfn = xen_start_info.nr_pages; pfn < max_low_pfn; pfn++) {
 664                 ClearPageReserved(&mem_map[pfn]);
 665                 set_page_count(&mem_map[pfn], 1);
 666                 totalram_pages++;
 667         }
 668
 669         reservedpages = 0;
 670         for (tmp = 0; tmp < max_low_pfn; tmp++)
 671                 /*
 672                  * Only count reserved RAM pages
 673                  */
 674                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 675                         reservedpages++;
 676
 677         set_highmem_pages_init(bad_ppro);
 678
 679         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 680         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 681         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 682
 683         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 684         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 685                    VMALLOC_END-VMALLOC_START);
 686
 687         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
 688                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 689                 num_physpages << (PAGE_SHIFT-10),
 690                 codesize >> 10,
 691                 reservedpages << (PAGE_SHIFT-10),
 692                 datasize >> 10,
 693                 initsize >> 10,
 694                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 695                );
 696
 697 #ifdef CONFIG_X86_PAE
 698         if (!cpu_has_pae)
 699                 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
 700 #endif
 701         if (boot_cpu_data.wp_works_ok < 0)
 702                 test_wp_bit();
 703
 704         /*
 705          * Subtle. SMP is doing it's boot stuff late (because it has to
 706          * fork idle threads) - but it also needs low mappings for the
 707          * protected-mode entry to work. We zap these entries only after
 708          * the WP-bit has been tested.
 709          */
 710 #ifndef CONFIG_SMP
 711         zap_low_mappings();
 712 #endif
 713 }
 714
 715 kmem_cache_t *pgd_cache;
 716 kmem_cache_t *pmd_cache;
 717 kmem_cache_t *pte_cache;
 718
 719 void __init pgtable_cache_init(void)
 720 {
 721         pte_cache = kmem_cache_create("pte",
 722                                 PTRS_PER_PTE*sizeof(pte_t),
 723                                 PTRS_PER_PTE*sizeof(pte_t),
 724                                 0,
 725                                 pte_ctor,
 726                                 pte_dtor);
 727         if (!pte_cache)
 728                 panic("pgtable_cache_init(): Cannot create pte cache");
 729         if (PTRS_PER_PMD > 1) {
 730                 pmd_cache = kmem_cache_create("pmd",
 731                                         PTRS_PER_PMD*sizeof(pmd_t),
 732                                         PTRS_PER_PMD*sizeof(pmd_t),
 733                                         0,
 734                                         pmd_ctor,
 735                                         NULL);
 736                 if (!pmd_cache)
 737                         panic("pgtable_cache_init(): cannot create pmd cache");
 738         }
 739         pgd_cache = kmem_cache_create("pgd",
 740                                 PTRS_PER_PGD*sizeof(pgd_t),
 741                                 PTRS_PER_PGD*sizeof(pgd_t),
 742                                 0,
 743                                 pgd_ctor,
 744                                 pgd_dtor);
 745         if (!pgd_cache)
 746                 panic("pgtable_cache_init(): Cannot create pgd cache");
 747 }
 748
 749 /*
 750  * This function cannot be __init, since exceptions don't work in that
 751  * section.  Put this after the callers, so that it cannot be inlined.
 752  */
 753 static int noinline do_test_wp_bit(void)
 754 {
 755         char tmp_reg;
 756         int flag;
 757
 758         __asm__ __volatile__(
 759                 "       movb %0,%1      \n"
 760                 "1:     movb %1,%0      \n"
 761                 "       xorl %2,%2      \n"
 762                 "2:                     \n"
 763                 ".section __ex_table,\"a\"\n"
 764                 "       .align 4        \n"
 765                 "       .long 1b,2b     \n"
 766                 ".previous              \n"
 767                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 768                  "=q" (tmp_reg),
 769                  "=r" (flag)
 770                 :"2" (1)
 771                 :"memory");
 772
 773         return flag;
 774 }
 775
 776 void free_initmem(void)
 777 {
 778         unsigned long addr;
 779
 780         addr = (unsigned long)(&__init_begin);
 781         for (; addr < (unsigned long)(&__init_end); addr += PAGE_SIZE) {
 782                 ClearPageReserved(virt_to_page(addr));
 783                 set_page_count(virt_to_page(addr), 1);
 784                 memset((void *)addr, 0xcc, PAGE_SIZE);
 785                 free_page(addr);
 786                 totalram_pages++;
 787         }
 788         printk (KERN_INFO "Freeing unused kernel memory: %dk freed\n", (__init_end - __init_begin) >> 10);
 789 }
 790
 791 #ifdef CONFIG_BLK_DEV_INITRD
 792 void free_initrd_mem(unsigned long start, unsigned long end)
 793 {
 794         if (start < end)
 795                 printk (KERN_INFO "Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
 796         for (; start < end; start += PAGE_SIZE) {
 797                 ClearPageReserved(virt_to_page(start));
 798                 set_page_count(virt_to_page(start), 1);
 799                 free_page(start);
 800                 totalram_pages++;
 801         }
 802 }
 803 #endif