git.oblomov.eu Git - linux-2.6/blob - arch/x86/mm/init_32.c

   1 /*
   2  *
   3  *  Copyright (C) 1995  Linus Torvalds
   4  *
   5  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
   6  */
   7
   8 #include <linux/module.h>
   9 #include <linux/signal.h>
  10 #include <linux/sched.h>
  11 #include <linux/kernel.h>
  12 #include <linux/errno.h>
  13 #include <linux/string.h>
  14 #include <linux/types.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/mman.h>
  17 #include <linux/mm.h>
  18 #include <linux/hugetlb.h>
  19 #include <linux/swap.h>
  20 #include <linux/smp.h>
  21 #include <linux/init.h>
  22 #include <linux/highmem.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/pci.h>
  25 #include <linux/pfn.h>
  26 #include <linux/poison.h>
  27 #include <linux/bootmem.h>
  28 #include <linux/slab.h>
  29 #include <linux/proc_fs.h>
  30 #include <linux/memory_hotplug.h>
  31 #include <linux/initrd.h>
  32 #include <linux/cpumask.h>
  33
  34 #include <asm/asm.h>
  35 #include <asm/bios_ebda.h>
  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/bugs.h>
  45 #include <asm/tlb.h>
  46 #include <asm/tlbflush.h>
  47 #include <asm/pgalloc.h>
  48 #include <asm/sections.h>
  49 #include <asm/paravirt.h>
  50 #include <asm/setup.h>
  51 #include <asm/cacheflush.h>
  52 #include <asm/page_types.h>
  53 #include <asm/init.h>
  54
  55 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  56 unsigned long highstart_pfn, highend_pfn;
  57
  58 static noinline int do_test_wp_bit(void);
  59
  60 bool __read_mostly __vmalloc_start_set = false;
  61
  62 static __init void *alloc_low_page(void)
  63 {
  64         unsigned long pfn = e820_table_end++;
  65         void *adr;
  66
  67         if (pfn >= e820_table_top)
  68                 panic("alloc_low_page: ran out of memory");
  69
  70         adr = __va(pfn * PAGE_SIZE);
  71         memset(adr, 0, PAGE_SIZE);
  72         return adr;
  73 }
  74
  75 /*
  76  * Creates a middle page table and puts a pointer to it in the
  77  * given global directory entry. This only returns the gd entry
  78  * in non-PAE compilation mode, since the middle layer is folded.
  79  */
  80 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  81 {
  82         pud_t *pud;
  83         pmd_t *pmd_table;
  84
  85 #ifdef CONFIG_X86_PAE
  86         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
  87                 if (after_bootmem)
  88                         pmd_table = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
  89                 else
  90                         pmd_table = (pmd_t *)alloc_low_page();
  91                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
  92                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  93                 pud = pud_offset(pgd, 0);
  94                 BUG_ON(pmd_table != pmd_offset(pud, 0));
  95
  96                 return pmd_table;
  97         }
  98 #endif
  99         pud = pud_offset(pgd, 0);
 100         pmd_table = pmd_offset(pud, 0);
 101
 102         return pmd_table;
 103 }
 104
 105 /*
 106  * Create a page table and place a pointer to it in a middle page
 107  * directory entry:
 108  */
 109 static pte_t * __init one_page_table_init(pmd_t *pmd)
 110 {
 111         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
 112                 pte_t *page_table = NULL;
 113
 114                 if (after_bootmem) {
 115 #ifdef CONFIG_DEBUG_PAGEALLOC
 116                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
 117 #endif
 118                         if (!page_table)
 119                                 page_table =
 120                                 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
 121                 } else
 122                         page_table = (pte_t *)alloc_low_page();
 123
 124                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
 125                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
 126                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
 127         }
 128
 129         return pte_offset_kernel(pmd, 0);
 130 }
 131
 132 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
 133 {
 134         int pgd_idx = pgd_index(vaddr);
 135         int pmd_idx = pmd_index(vaddr);
 136
 137         return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
 138 }
 139
 140 pte_t * __init populate_extra_pte(unsigned long vaddr)
 141 {
 142         int pte_idx = pte_index(vaddr);
 143         pmd_t *pmd;
 144
 145         pmd = populate_extra_pmd(vaddr);
 146         return one_page_table_init(pmd) + pte_idx;
 147 }
 148
 149 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
 150                                            unsigned long vaddr, pte_t *lastpte)
 151 {
 152 #ifdef CONFIG_HIGHMEM
 153         /*
 154          * Something (early fixmap) may already have put a pte
 155          * page here, which causes the page table allocation
 156          * to become nonlinear. Attempt to fix it, and if it
 157          * is still nonlinear then we have to bug.
 158          */
 159         int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
 160         int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
 161
 162         if (pmd_idx_kmap_begin != pmd_idx_kmap_end
 163             && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
 164             && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
 165             && ((__pa(pte) >> PAGE_SHIFT) < e820_table_start
 166                 || (__pa(pte) >> PAGE_SHIFT) >= e820_table_end)) {
 167                 pte_t *newpte;
 168                 int i;
 169
 170                 BUG_ON(after_bootmem);
 171                 newpte = alloc_low_page();
 172                 for (i = 0; i < PTRS_PER_PTE; i++)
 173                         set_pte(newpte + i, pte[i]);
 174
 175                 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
 176                 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
 177                 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
 178                 __flush_tlb_all();
 179
 180                 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
 181                 pte = newpte;
 182         }
 183         BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
 184                && vaddr > fix_to_virt(FIX_KMAP_END)
 185                && lastpte && lastpte + PTRS_PER_PTE != pte);
 186 #endif
 187         return pte;
 188 }
 189
 190 /*
 191  * This function initializes a certain range of kernel virtual memory
 192  * with new bootmem page tables, everywhere page tables are missing in
 193  * the given range.
 194  *
 195  * NOTE: The pagetables are allocated contiguous on the physical space
 196  * so we can cache the place of the first one and move around without
 197  * checking the pgd every time.
 198  */
 199 static void __init
 200 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
 201 {
 202         int pgd_idx, pmd_idx;
 203         unsigned long vaddr;
 204         pgd_t *pgd;
 205         pmd_t *pmd;
 206         pte_t *pte = NULL;
 207
 208         vaddr = start;
 209         pgd_idx = pgd_index(vaddr);
 210         pmd_idx = pmd_index(vaddr);
 211         pgd = pgd_base + pgd_idx;
 212
 213         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 214                 pmd = one_md_table_init(pgd);
 215                 pmd = pmd + pmd_index(vaddr);
 216                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
 217                                                         pmd++, pmd_idx++) {
 218                         pte = page_table_kmap_check(one_page_table_init(pmd),
 219                                                     pmd, vaddr, pte);
 220
 221                         vaddr += PMD_SIZE;
 222                 }
 223                 pmd_idx = 0;
 224         }
 225 }
 226
 227 static inline int is_kernel_text(unsigned long addr)
 228 {
 229         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 230                 return 1;
 231         return 0;
 232 }
 233
 234 /*
 235  * This maps the physical memory to kernel virtual address space, a total
 236  * of max_low_pfn pages, by creating page tables starting from address
 237  * PAGE_OFFSET:
 238  */
 239 unsigned long __init
 240 kernel_physical_mapping_init(unsigned long start,
 241                              unsigned long end,
 242                              unsigned long page_size_mask)
 243 {
 244         int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
 245         unsigned long start_pfn, end_pfn;
 246         pgd_t *pgd_base = swapper_pg_dir;
 247         int pgd_idx, pmd_idx, pte_ofs;
 248         unsigned long pfn;
 249         pgd_t *pgd;
 250         pmd_t *pmd;
 251         pte_t *pte;
 252         unsigned pages_2m, pages_4k;
 253         int mapping_iter;
 254
 255         start_pfn = start >> PAGE_SHIFT;
 256         end_pfn = end >> PAGE_SHIFT;
 257
 258         /*
 259          * First iteration will setup identity mapping using large/small pages
 260          * based on use_pse, with other attributes same as set by
 261          * the early code in head_32.S
 262          *
 263          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
 264          * as desired for the kernel identity mapping.
 265          *
 266          * This two pass mechanism conforms to the TLB app note which says:
 267          *
 268          *     "Software should not write to a paging-structure entry in a way
 269          *      that would change, for any linear address, both the page size
 270          *      and either the page frame or attributes."
 271          */
 272         mapping_iter = 1;
 273
 274         if (!cpu_has_pse)
 275                 use_pse = 0;
 276
 277 repeat:
 278         pages_2m = pages_4k = 0;
 279         pfn = start_pfn;
 280         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 281         pgd = pgd_base + pgd_idx;
 282         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 283                 pmd = one_md_table_init(pgd);
 284
 285                 if (pfn >= end_pfn)
 286                         continue;
 287 #ifdef CONFIG_X86_PAE
 288                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 289                 pmd += pmd_idx;
 290 #else
 291                 pmd_idx = 0;
 292 #endif
 293                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
 294                      pmd++, pmd_idx++) {
 295                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
 296
 297                         /*
 298                          * Map with big pages if possible, otherwise
 299                          * create normal page tables:
 300                          */
 301                         if (use_pse) {
 302                                 unsigned int addr2;
 303                                 pgprot_t prot = PAGE_KERNEL_LARGE;
 304                                 /*
 305                                  * first pass will use the same initial
 306                                  * identity mapping attribute + _PAGE_PSE.
 307                                  */
 308                                 pgprot_t init_prot =
 309                                         __pgprot(PTE_IDENT_ATTR |
 310                                                  _PAGE_PSE);
 311
 312                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
 313                                         PAGE_OFFSET + PAGE_SIZE-1;
 314
 315                                 if (is_kernel_text(addr) ||
 316                                     is_kernel_text(addr2))
 317                                         prot = PAGE_KERNEL_LARGE_EXEC;
 318
 319                                 pages_2m++;
 320                                 if (mapping_iter == 1)
 321                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
 322                                 else
 323                                         set_pmd(pmd, pfn_pmd(pfn, prot));
 324
 325                                 pfn += PTRS_PER_PTE;
 326                                 continue;
 327                         }
 328                         pte = one_page_table_init(pmd);
 329
 330                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 331                         pte += pte_ofs;
 332                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
 333                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
 334                                 pgprot_t prot = PAGE_KERNEL;
 335                                 /*
 336                                  * first pass will use the same initial
 337                                  * identity mapping attribute.
 338                                  */
 339                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
 340
 341                                 if (is_kernel_text(addr))
 342                                         prot = PAGE_KERNEL_EXEC;
 343
 344                                 pages_4k++;
 345                                 if (mapping_iter == 1)
 346                                         set_pte(pte, pfn_pte(pfn, init_prot));
 347                                 else
 348                                         set_pte(pte, pfn_pte(pfn, prot));
 349                         }
 350                 }
 351         }
 352         if (mapping_iter == 1) {
 353                 /*
 354                  * update direct mapping page count only in the first
 355                  * iteration.
 356                  */
 357                 update_page_count(PG_LEVEL_2M, pages_2m);
 358                 update_page_count(PG_LEVEL_4K, pages_4k);
 359
 360                 /*
 361                  * local global flush tlb, which will flush the previous
 362                  * mappings present in both small and large page TLB's.
 363                  */
 364                 __flush_tlb_all();
 365
 366                 /*
 367                  * Second iteration will set the actual desired PTE attributes.
 368                  */
 369                 mapping_iter = 2;
 370                 goto repeat;
 371         }
 372         return 0;
 373 }
 374
 375 pte_t *kmap_pte;
 376 pgprot_t kmap_prot;
 377
 378 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 379 {
 380         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 381                         vaddr), vaddr), vaddr);
 382 }
 383
 384 static void __init kmap_init(void)
 385 {
 386         unsigned long kmap_vstart;
 387
 388         /*
 389          * Cache the first kmap pte:
 390          */
 391         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 392         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 393
 394         kmap_prot = PAGE_KERNEL;
 395 }
 396
 397 #ifdef CONFIG_HIGHMEM
 398 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 399 {
 400         unsigned long vaddr;
 401         pgd_t *pgd;
 402         pud_t *pud;
 403         pmd_t *pmd;
 404         pte_t *pte;
 405
 406         vaddr = PKMAP_BASE;
 407         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 408
 409         pgd = swapper_pg_dir + pgd_index(vaddr);
 410         pud = pud_offset(pgd, vaddr);
 411         pmd = pmd_offset(pud, vaddr);
 412         pte = pte_offset_kernel(pmd, vaddr);
 413         pkmap_page_table = pte;
 414 }
 415
 416 static void __init add_one_highpage_init(struct page *page, int pfn)
 417 {
 418         ClearPageReserved(page);
 419         init_page_count(page);
 420         __free_page(page);
 421         totalhigh_pages++;
 422 }
 423
 424 struct add_highpages_data {
 425         unsigned long start_pfn;
 426         unsigned long end_pfn;
 427 };
 428
 429 static int __init add_highpages_work_fn(unsigned long start_pfn,
 430                                          unsigned long end_pfn, void *datax)
 431 {
 432         int node_pfn;
 433         struct page *page;
 434         unsigned long final_start_pfn, final_end_pfn;
 435         struct add_highpages_data *data;
 436
 437         data = (struct add_highpages_data *)datax;
 438
 439         final_start_pfn = max(start_pfn, data->start_pfn);
 440         final_end_pfn = min(end_pfn, data->end_pfn);
 441         if (final_start_pfn >= final_end_pfn)
 442                 return 0;
 443
 444         for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
 445              node_pfn++) {
 446                 if (!pfn_valid(node_pfn))
 447                         continue;
 448                 page = pfn_to_page(node_pfn);
 449                 add_one_highpage_init(page, node_pfn);
 450         }
 451
 452         return 0;
 453
 454 }
 455
 456 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
 457                                               unsigned long end_pfn)
 458 {
 459         struct add_highpages_data data;
 460
 461         data.start_pfn = start_pfn;
 462         data.end_pfn = end_pfn;
 463
 464         work_with_active_regions(nid, add_highpages_work_fn, &data);
 465 }
 466
 467 #else
 468 static inline void permanent_kmaps_init(pgd_t *pgd_base)
 469 {
 470 }
 471 #endif /* CONFIG_HIGHMEM */
 472
 473 void __init native_pagetable_setup_start(pgd_t *base)
 474 {
 475         unsigned long pfn, va;
 476         pgd_t *pgd;
 477         pud_t *pud;
 478         pmd_t *pmd;
 479         pte_t *pte;
 480
 481         /*
 482          * Remove any mappings which extend past the end of physical
 483          * memory from the boot time page table:
 484          */
 485         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 486                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 487                 pgd = base + pgd_index(va);
 488                 if (!pgd_present(*pgd))
 489                         break;
 490
 491                 pud = pud_offset(pgd, va);
 492                 pmd = pmd_offset(pud, va);
 493                 if (!pmd_present(*pmd))
 494                         break;
 495
 496                 pte = pte_offset_kernel(pmd, va);
 497                 if (!pte_present(*pte))
 498                         break;
 499
 500                 pte_clear(NULL, va, pte);
 501         }
 502         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 503 }
 504
 505 void __init native_pagetable_setup_done(pgd_t *base)
 506 {
 507 }
 508
 509 /*
 510  * Build a proper pagetable for the kernel mappings.  Up until this
 511  * point, we've been running on some set of pagetables constructed by
 512  * the boot process.
 513  *
 514  * If we're booting on native hardware, this will be a pagetable
 515  * constructed in arch/x86/kernel/head_32.S.  The root of the
 516  * pagetable will be swapper_pg_dir.
 517  *
 518  * If we're booting paravirtualized under a hypervisor, then there are
 519  * more options: we may already be running PAE, and the pagetable may
 520  * or may not be based in swapper_pg_dir.  In any case,
 521  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 522  * appropriately for the rest of the initialization to work.
 523  *
 524  * In general, pagetable_init() assumes that the pagetable may already
 525  * be partially populated, and so it avoids stomping on any existing
 526  * mappings.
 527  */
 528 void __init early_ioremap_page_table_range_init(void)
 529 {
 530         pgd_t *pgd_base = swapper_pg_dir;
 531         unsigned long vaddr, end;
 532
 533         /*
 534          * Fixed mappings, only the page table structure has to be
 535          * created - mappings will be set by set_fixmap():
 536          */
 537         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 538         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 539         page_table_range_init(vaddr, end, pgd_base);
 540         early_ioremap_reset();
 541 }
 542
 543 static void __init pagetable_init(void)
 544 {
 545         pgd_t *pgd_base = swapper_pg_dir;
 546
 547         permanent_kmaps_init(pgd_base);
 548 }
 549
 550 #ifdef CONFIG_ACPI_SLEEP
 551 /*
 552  * ACPI suspend needs this for resume, because things like the intel-agp
 553  * driver might have split up a kernel 4MB mapping.
 554  */
 555 char swsusp_pg_dir[PAGE_SIZE]
 556         __attribute__ ((aligned(PAGE_SIZE)));
 557
 558 static inline void save_pg_dir(void)
 559 {
 560         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 561 }
 562 #else /* !CONFIG_ACPI_SLEEP */
 563 static inline void save_pg_dir(void)
 564 {
 565 }
 566 #endif /* !CONFIG_ACPI_SLEEP */
 567
 568 void zap_low_mappings(void)
 569 {
 570         int i;
 571
 572         /*
 573          * Zap initial low-memory mappings.
 574          *
 575          * Note that "pgd_clear()" doesn't do it for
 576          * us, because pgd_clear() is a no-op on i386.
 577          */
 578         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 579 #ifdef CONFIG_X86_PAE
 580                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 581 #else
 582                 set_pgd(swapper_pg_dir+i, __pgd(0));
 583 #endif
 584         }
 585         flush_tlb_all();
 586 }
 587
 588 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
 589 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 590
 591 /* user-defined highmem size */
 592 static unsigned int highmem_pages = -1;
 593
 594 /*
 595  * highmem=size forces highmem to be exactly 'size' bytes.
 596  * This works even on boxes that have no highmem otherwise.
 597  * This also works to reduce highmem size on bigger boxes.
 598  */
 599 static int __init parse_highmem(char *arg)
 600 {
 601         if (!arg)
 602                 return -EINVAL;
 603
 604         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 605         return 0;
 606 }
 607 early_param("highmem", parse_highmem);
 608
 609 #define MSG_HIGHMEM_TOO_BIG \
 610         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
 611
 612 #define MSG_LOWMEM_TOO_SMALL \
 613         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
 614 /*
 615  * All of RAM fits into lowmem - but if user wants highmem
 616  * artificially via the highmem=x boot parameter then create
 617  * it:
 618  */
 619 void __init lowmem_pfn_init(void)
 620 {
 621         /* max_low_pfn is 0, we already have early_res support */
 622         max_low_pfn = max_pfn;
 623
 624         if (highmem_pages == -1)
 625                 highmem_pages = 0;
 626 #ifdef CONFIG_HIGHMEM
 627         if (highmem_pages >= max_pfn) {
 628                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
 629                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 630                 highmem_pages = 0;
 631         }
 632         if (highmem_pages) {
 633                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
 634                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
 635                                 pages_to_mb(highmem_pages));
 636                         highmem_pages = 0;
 637                 }
 638                 max_low_pfn -= highmem_pages;
 639         }
 640 #else
 641         if (highmem_pages)
 642                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 643 #endif
 644 }
 645
 646 #define MSG_HIGHMEM_TOO_SMALL \
 647         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
 648
 649 #define MSG_HIGHMEM_TRIMMED \
 650         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
 651 /*
 652  * We have more RAM than fits into lowmem - we try to put it into
 653  * highmem, also taking the highmem=x boot parameter into account:
 654  */
 655 void __init highmem_pfn_init(void)
 656 {
 657         max_low_pfn = MAXMEM_PFN;
 658
 659         if (highmem_pages == -1)
 660                 highmem_pages = max_pfn - MAXMEM_PFN;
 661
 662         if (highmem_pages + MAXMEM_PFN < max_pfn)
 663                 max_pfn = MAXMEM_PFN + highmem_pages;
 664
 665         if (highmem_pages + MAXMEM_PFN > max_pfn) {
 666                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
 667                         pages_to_mb(max_pfn - MAXMEM_PFN),
 668                         pages_to_mb(highmem_pages));
 669                 highmem_pages = 0;
 670         }
 671 #ifndef CONFIG_HIGHMEM
 672         /* Maximum memory usable is what is directly addressable */
 673         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
 674         if (max_pfn > MAX_NONPAE_PFN)
 675                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 676         else
 677                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 678         max_pfn = MAXMEM_PFN;
 679 #else /* !CONFIG_HIGHMEM */
 680 #ifndef CONFIG_HIGHMEM64G
 681         if (max_pfn > MAX_NONPAE_PFN) {
 682                 max_pfn = MAX_NONPAE_PFN;
 683                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
 684         }
 685 #endif /* !CONFIG_HIGHMEM64G */
 686 #endif /* !CONFIG_HIGHMEM */
 687 }
 688
 689 /*
 690  * Determine low and high memory ranges:
 691  */
 692 void __init find_low_pfn_range(void)
 693 {
 694         /* it could update max_pfn */
 695
 696         if (max_pfn <= MAXMEM_PFN)
 697                 lowmem_pfn_init();
 698         else
 699                 highmem_pfn_init();
 700 }
 701
 702 #ifndef CONFIG_NEED_MULTIPLE_NODES
 703 void __init initmem_init(unsigned long start_pfn,
 704                                   unsigned long end_pfn)
 705 {
 706 #ifdef CONFIG_HIGHMEM
 707         highstart_pfn = highend_pfn = max_pfn;
 708         if (max_pfn > max_low_pfn)
 709                 highstart_pfn = max_low_pfn;
 710         memory_present(0, 0, highend_pfn);
 711         e820_register_active_regions(0, 0, highend_pfn);
 712         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 713                 pages_to_mb(highend_pfn - highstart_pfn));
 714         num_physpages = highend_pfn;
 715         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 716 #else
 717         memory_present(0, 0, max_low_pfn);
 718         e820_register_active_regions(0, 0, max_low_pfn);
 719         num_physpages = max_low_pfn;
 720         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 721 #endif
 722 #ifdef CONFIG_FLATMEM
 723         max_mapnr = num_physpages;
 724 #endif
 725         __vmalloc_start_set = true;
 726
 727         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 728                         pages_to_mb(max_low_pfn));
 729
 730         setup_bootmem_allocator();
 731 }
 732 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 733
 734 static void __init zone_sizes_init(void)
 735 {
 736         unsigned long max_zone_pfns[MAX_NR_ZONES];
 737         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 738         max_zone_pfns[ZONE_DMA] =
 739                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 740         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 741 #ifdef CONFIG_HIGHMEM
 742         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 743 #endif
 744
 745         free_area_init_nodes(max_zone_pfns);
 746 }
 747
 748 static unsigned long __init setup_node_bootmem(int nodeid,
 749                                  unsigned long start_pfn,
 750                                  unsigned long end_pfn,
 751                                  unsigned long bootmap)
 752 {
 753         unsigned long bootmap_size;
 754
 755         /* don't touch min_low_pfn */
 756         bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
 757                                          bootmap >> PAGE_SHIFT,
 758                                          start_pfn, end_pfn);
 759         printk(KERN_INFO "  node %d low ram: %08lx - %08lx\n",
 760                 nodeid, start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
 761         printk(KERN_INFO "  node %d bootmap %08lx - %08lx\n",
 762                  nodeid, bootmap, bootmap + bootmap_size);
 763         free_bootmem_with_active_regions(nodeid, end_pfn);
 764         early_res_to_bootmem(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
 765
 766         return bootmap + bootmap_size;
 767 }
 768
 769 void __init setup_bootmem_allocator(void)
 770 {
 771         int nodeid;
 772         unsigned long bootmap_size, bootmap;
 773         /*
 774          * Initialize the boot-time allocator (with low memory only):
 775          */
 776         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
 777         bootmap = find_e820_area(0, max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
 778                                  PAGE_SIZE);
 779         if (bootmap == -1L)
 780                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 781         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
 782
 783         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 784                  max_pfn_mapped<<PAGE_SHIFT);
 785         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
 786
 787         for_each_online_node(nodeid) {
 788                  unsigned long start_pfn, end_pfn;
 789
 790 #ifdef CONFIG_NEED_MULTIPLE_NODES
 791                 start_pfn = node_start_pfn[nodeid];
 792                 end_pfn = node_end_pfn[nodeid];
 793                 if (start_pfn > max_low_pfn)
 794                         continue;
 795                 if (end_pfn > max_low_pfn)
 796                         end_pfn = max_low_pfn;
 797 #else
 798                 start_pfn = 0;
 799                 end_pfn = max_low_pfn;
 800 #endif
 801                 bootmap = setup_node_bootmem(nodeid, start_pfn, end_pfn,
 802                                                  bootmap);
 803         }
 804
 805         after_bootmem = 1;
 806 }
 807
 808 /*
 809  * paging_init() sets up the page tables - note that the first 8MB are
 810  * already mapped by head.S.
 811  *
 812  * This routines also unmaps the page at virtual kernel address 0, so
 813  * that we can trap those pesky NULL-reference errors in the kernel.
 814  */
 815 void __init paging_init(void)
 816 {
 817         pagetable_init();
 818
 819         __flush_tlb_all();
 820
 821         kmap_init();
 822
 823         /*
 824          * NOTE: at this point the bootmem allocator is fully available.
 825          */
 826         sparse_init();
 827         zone_sizes_init();
 828 }
 829
 830 /*
 831  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 832  * and also on some strange 486's. All 586+'s are OK. This used to involve
 833  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 834  * switch to using exceptions got rid of all that.
 835  */
 836 static void __init test_wp_bit(void)
 837 {
 838         printk(KERN_INFO
 839   "Checking if this processor honours the WP bit even in supervisor mode...");
 840
 841         /* Any page-aligned address will do, the test is non-destructive */
 842         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 843         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 844         clear_fixmap(FIX_WP_TEST);
 845
 846         if (!boot_cpu_data.wp_works_ok) {
 847                 printk(KERN_CONT "No.\n");
 848 #ifdef CONFIG_X86_WP_WORKS_OK
 849                 panic(
 850   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 851 #endif
 852         } else {
 853                 printk(KERN_CONT "Ok.\n");
 854         }
 855 }
 856
 857 static struct kcore_list kcore_mem, kcore_vmalloc;
 858
 859 void __init mem_init(void)
 860 {
 861         int codesize, reservedpages, datasize, initsize;
 862         int tmp;
 863
 864         pci_iommu_alloc();
 865
 866 #ifdef CONFIG_FLATMEM
 867         BUG_ON(!mem_map);
 868 #endif
 869         /* this will put all low memory onto the freelists */
 870         totalram_pages += free_all_bootmem();
 871
 872         reservedpages = 0;
 873         for (tmp = 0; tmp < max_low_pfn; tmp++)
 874                 /*
 875                  * Only count reserved RAM pages:
 876                  */
 877                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 878                         reservedpages++;
 879
 880         set_highmem_pages_init();
 881
 882         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 883         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 884         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 885
 886         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 887         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 888                    VMALLOC_END-VMALLOC_START);
 889
 890         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
 891                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
 892                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 893                 num_physpages << (PAGE_SHIFT-10),
 894                 codesize >> 10,
 895                 reservedpages << (PAGE_SHIFT-10),
 896                 datasize >> 10,
 897                 initsize >> 10,
 898                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 899                );
 900
 901         printk(KERN_INFO "virtual kernel memory layout:\n"
 902                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 903 #ifdef CONFIG_HIGHMEM
 904                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 905 #endif
 906                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 907                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 908                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 909                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 910                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 911                 FIXADDR_START, FIXADDR_TOP,
 912                 (FIXADDR_TOP - FIXADDR_START) >> 10,
 913
 914 #ifdef CONFIG_HIGHMEM
 915                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 916                 (LAST_PKMAP*PAGE_SIZE) >> 10,
 917 #endif
 918
 919                 VMALLOC_START, VMALLOC_END,
 920                 (VMALLOC_END - VMALLOC_START) >> 20,
 921
 922                 (unsigned long)__va(0), (unsigned long)high_memory,
 923                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 924
 925                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
 926                 ((unsigned long)&__init_end -
 927                  (unsigned long)&__init_begin) >> 10,
 928
 929                 (unsigned long)&_etext, (unsigned long)&_edata,
 930                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 931
 932                 (unsigned long)&_text, (unsigned long)&_etext,
 933                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 934
 935         /*
 936          * Check boundaries twice: Some fundamental inconsistencies can
 937          * be detected at build time already.
 938          */
 939 #define __FIXADDR_TOP (-PAGE_SIZE)
 940 #ifdef CONFIG_HIGHMEM
 941         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
 942         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
 943 #endif
 944 #define high_memory (-128UL << 20)
 945         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
 946 #undef high_memory
 947 #undef __FIXADDR_TOP
 948
 949 #ifdef CONFIG_HIGHMEM
 950         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
 951         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
 952 #endif
 953         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
 954         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
 955
 956         if (boot_cpu_data.wp_works_ok < 0)
 957                 test_wp_bit();
 958
 959         save_pg_dir();
 960         zap_low_mappings();
 961 }
 962
 963 #ifdef CONFIG_MEMORY_HOTPLUG
 964 int arch_add_memory(int nid, u64 start, u64 size)
 965 {
 966         struct pglist_data *pgdata = NODE_DATA(nid);
 967         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 968         unsigned long start_pfn = start >> PAGE_SHIFT;
 969         unsigned long nr_pages = size >> PAGE_SHIFT;
 970
 971         return __add_pages(nid, zone, start_pfn, nr_pages);
 972 }
 973 #endif
 974
 975 /*
 976  * This function cannot be __init, since exceptions don't work in that
 977  * section.  Put this after the callers, so that it cannot be inlined.
 978  */
 979 static noinline int do_test_wp_bit(void)
 980 {
 981         char tmp_reg;
 982         int flag;
 983
 984         __asm__ __volatile__(
 985                 "       movb %0, %1     \n"
 986                 "1:     movb %1, %0     \n"
 987                 "       xorl %2, %2     \n"
 988                 "2:                     \n"
 989                 _ASM_EXTABLE(1b,2b)
 990                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 991                  "=q" (tmp_reg),
 992                  "=r" (flag)
 993                 :"2" (1)
 994                 :"memory");
 995
 996         return flag;
 997 }
 998
 999 #ifdef CONFIG_DEBUG_RODATA
1000 const int rodata_test_data = 0xC3;
1001 EXPORT_SYMBOL_GPL(rodata_test_data);
1002
1003 static int kernel_set_to_readonly;
1004
1005 void set_kernel_text_rw(void)
1006 {
1007         unsigned long start = PFN_ALIGN(_text);
1008         unsigned long size = PFN_ALIGN(_etext) - start;
1009
1010         if (!kernel_set_to_readonly)
1011                 return;
1012
1013         pr_debug("Set kernel text: %lx - %lx for read write\n",
1014                  start, start+size);
1015
1016         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1017 }
1018
1019 void set_kernel_text_ro(void)
1020 {
1021         unsigned long start = PFN_ALIGN(_text);
1022         unsigned long size = PFN_ALIGN(_etext) - start;
1023
1024         if (!kernel_set_to_readonly)
1025                 return;
1026
1027         pr_debug("Set kernel text: %lx - %lx for read only\n",
1028                  start, start+size);
1029
1030         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1031 }
1032
1033 void mark_rodata_ro(void)
1034 {
1035         unsigned long start = PFN_ALIGN(_text);
1036         unsigned long size = PFN_ALIGN(_etext) - start;
1037
1038         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1039         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1040                 size >> 10);
1041
1042         kernel_set_to_readonly = 1;
1043
1044 #ifdef CONFIG_CPA_DEBUG
1045         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1046                 start, start+size);
1047         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1048
1049         printk(KERN_INFO "Testing CPA: write protecting again\n");
1050         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1051 #endif
1052
1053         start += size;
1054         size = (unsigned long)__end_rodata - start;
1055         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1056         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1057                 size >> 10);
1058         rodata_test();
1059
1060 #ifdef CONFIG_CPA_DEBUG
1061         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1062         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1063
1064         printk(KERN_INFO "Testing CPA: write protecting again\n");
1065         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1066 #endif
1067 }
1068 #endif
1069
1070 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1071                                    int flags)
1072 {
1073         return reserve_bootmem(phys, len, flags);
1074 }