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/memblock.h>
  29 #include <linux/proc_fs.h>
  30 #include <linux/memory_hotplug.h>
  31 #include <linux/initrd.h>
  32 #include <linux/cpumask.h>
  33 #include <linux/gfp.h>
  34
  35 #include <asm/asm.h>
  36 #include <asm/bios_ebda.h>
  37 #include <asm/processor.h>
  38 #include <asm/system.h>
  39 #include <asm/uaccess.h>
  40 #include <asm/pgtable.h>
  41 #include <asm/dma.h>
  42 #include <asm/fixmap.h>
  43 #include <asm/e820.h>
  44 #include <asm/apic.h>
  45 #include <asm/bugs.h>
  46 #include <asm/tlb.h>
  47 #include <asm/tlbflush.h>
  48 #include <asm/pgalloc.h>
  49 #include <asm/sections.h>
  50 #include <asm/paravirt.h>
  51 #include <asm/setup.h>
  52 #include <asm/cacheflush.h>
  53 #include <asm/page_types.h>
  54 #include <asm/init.h>
  55
  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_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 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
 116                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
 117 #endif
 118                         if (!page_table)
 119                                 page_table =
 120                                 (pte_t *)alloc_bootmem_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 last_map_addr = end;
 246         unsigned long start_pfn, end_pfn;
 247         pgd_t *pgd_base = swapper_pg_dir;
 248         int pgd_idx, pmd_idx, pte_ofs;
 249         unsigned long pfn;
 250         pgd_t *pgd;
 251         pmd_t *pmd;
 252         pte_t *pte;
 253         unsigned pages_2m, pages_4k;
 254         int mapping_iter;
 255
 256         start_pfn = start >> PAGE_SHIFT;
 257         end_pfn = end >> PAGE_SHIFT;
 258
 259         /*
 260          * First iteration will setup identity mapping using large/small pages
 261          * based on use_pse, with other attributes same as set by
 262          * the early code in head_32.S
 263          *
 264          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
 265          * as desired for the kernel identity mapping.
 266          *
 267          * This two pass mechanism conforms to the TLB app note which says:
 268          *
 269          *     "Software should not write to a paging-structure entry in a way
 270          *      that would change, for any linear address, both the page size
 271          *      and either the page frame or attributes."
 272          */
 273         mapping_iter = 1;
 274
 275         if (!cpu_has_pse)
 276                 use_pse = 0;
 277
 278 repeat:
 279         pages_2m = pages_4k = 0;
 280         pfn = start_pfn;
 281         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 282         pgd = pgd_base + pgd_idx;
 283         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 284                 pmd = one_md_table_init(pgd);
 285
 286                 if (pfn >= end_pfn)
 287                         continue;
 288 #ifdef CONFIG_X86_PAE
 289                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 290                 pmd += pmd_idx;
 291 #else
 292                 pmd_idx = 0;
 293 #endif
 294                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
 295                      pmd++, pmd_idx++) {
 296                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
 297
 298                         /*
 299                          * Map with big pages if possible, otherwise
 300                          * create normal page tables:
 301                          */
 302                         if (use_pse) {
 303                                 unsigned int addr2;
 304                                 pgprot_t prot = PAGE_KERNEL_LARGE;
 305                                 /*
 306                                  * first pass will use the same initial
 307                                  * identity mapping attribute + _PAGE_PSE.
 308                                  */
 309                                 pgprot_t init_prot =
 310                                         __pgprot(PTE_IDENT_ATTR |
 311                                                  _PAGE_PSE);
 312
 313                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
 314                                         PAGE_OFFSET + PAGE_SIZE-1;
 315
 316                                 if (is_kernel_text(addr) ||
 317                                     is_kernel_text(addr2))
 318                                         prot = PAGE_KERNEL_LARGE_EXEC;
 319
 320                                 pages_2m++;
 321                                 if (mapping_iter == 1)
 322                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
 323                                 else
 324                                         set_pmd(pmd, pfn_pmd(pfn, prot));
 325
 326                                 pfn += PTRS_PER_PTE;
 327                                 continue;
 328                         }
 329                         pte = one_page_table_init(pmd);
 330
 331                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 332                         pte += pte_ofs;
 333                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
 334                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
 335                                 pgprot_t prot = PAGE_KERNEL;
 336                                 /*
 337                                  * first pass will use the same initial
 338                                  * identity mapping attribute.
 339                                  */
 340                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
 341
 342                                 if (is_kernel_text(addr))
 343                                         prot = PAGE_KERNEL_EXEC;
 344
 345                                 pages_4k++;
 346                                 if (mapping_iter == 1) {
 347                                         set_pte(pte, pfn_pte(pfn, init_prot));
 348                                         last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
 349                                 } else
 350                                         set_pte(pte, pfn_pte(pfn, prot));
 351                         }
 352                 }
 353         }
 354         if (mapping_iter == 1) {
 355                 /*
 356                  * update direct mapping page count only in the first
 357                  * iteration.
 358                  */
 359                 update_page_count(PG_LEVEL_2M, pages_2m);
 360                 update_page_count(PG_LEVEL_4K, pages_4k);
 361
 362                 /*
 363                  * local global flush tlb, which will flush the previous
 364                  * mappings present in both small and large page TLB's.
 365                  */
 366                 __flush_tlb_all();
 367
 368                 /*
 369                  * Second iteration will set the actual desired PTE attributes.
 370                  */
 371                 mapping_iter = 2;
 372                 goto repeat;
 373         }
 374         return last_map_addr;
 375 }
 376
 377 pte_t *kmap_pte;
 378 pgprot_t kmap_prot;
 379
 380 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 381 {
 382         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 383                         vaddr), vaddr), vaddr);
 384 }
 385
 386 static void __init kmap_init(void)
 387 {
 388         unsigned long kmap_vstart;
 389
 390         /*
 391          * Cache the first kmap pte:
 392          */
 393         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 394         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 395
 396         kmap_prot = PAGE_KERNEL;
 397 }
 398
 399 #ifdef CONFIG_HIGHMEM
 400 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 401 {
 402         unsigned long vaddr;
 403         pgd_t *pgd;
 404         pud_t *pud;
 405         pmd_t *pmd;
 406         pte_t *pte;
 407
 408         vaddr = PKMAP_BASE;
 409         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 410
 411         pgd = swapper_pg_dir + pgd_index(vaddr);
 412         pud = pud_offset(pgd, vaddr);
 413         pmd = pmd_offset(pud, vaddr);
 414         pte = pte_offset_kernel(pmd, vaddr);
 415         pkmap_page_table = pte;
 416 }
 417
 418 static void __init add_one_highpage_init(struct page *page)
 419 {
 420         ClearPageReserved(page);
 421         init_page_count(page);
 422         __free_page(page);
 423         totalhigh_pages++;
 424 }
 425
 426 struct add_highpages_data {
 427         unsigned long start_pfn;
 428         unsigned long end_pfn;
 429 };
 430
 431 static int __init add_highpages_work_fn(unsigned long start_pfn,
 432                                          unsigned long end_pfn, void *datax)
 433 {
 434         int node_pfn;
 435         struct page *page;
 436         unsigned long final_start_pfn, final_end_pfn;
 437         struct add_highpages_data *data;
 438
 439         data = (struct add_highpages_data *)datax;
 440
 441         final_start_pfn = max(start_pfn, data->start_pfn);
 442         final_end_pfn = min(end_pfn, data->end_pfn);
 443         if (final_start_pfn >= final_end_pfn)
 444                 return 0;
 445
 446         for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
 447              node_pfn++) {
 448                 if (!pfn_valid(node_pfn))
 449                         continue;
 450                 page = pfn_to_page(node_pfn);
 451                 add_one_highpage_init(page);
 452         }
 453
 454         return 0;
 455
 456 }
 457
 458 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
 459                                               unsigned long end_pfn)
 460 {
 461         struct add_highpages_data data;
 462
 463         data.start_pfn = start_pfn;
 464         data.end_pfn = end_pfn;
 465
 466         work_with_active_regions(nid, add_highpages_work_fn, &data);
 467 }
 468
 469 #else
 470 static inline void permanent_kmaps_init(pgd_t *pgd_base)
 471 {
 472 }
 473 #endif /* CONFIG_HIGHMEM */
 474
 475 void __init native_pagetable_setup_start(pgd_t *base)
 476 {
 477         unsigned long pfn, va;
 478         pgd_t *pgd;
 479         pud_t *pud;
 480         pmd_t *pmd;
 481         pte_t *pte;
 482
 483         /*
 484          * Remove any mappings which extend past the end of physical
 485          * memory from the boot time page table:
 486          */
 487         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 488                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 489                 pgd = base + pgd_index(va);
 490                 if (!pgd_present(*pgd))
 491                         break;
 492
 493                 pud = pud_offset(pgd, va);
 494                 pmd = pmd_offset(pud, va);
 495                 if (!pmd_present(*pmd))
 496                         break;
 497
 498                 pte = pte_offset_kernel(pmd, va);
 499                 if (!pte_present(*pte))
 500                         break;
 501
 502                 pte_clear(NULL, va, pte);
 503         }
 504         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 505 }
 506
 507 void __init native_pagetable_setup_done(pgd_t *base)
 508 {
 509 }
 510
 511 /*
 512  * Build a proper pagetable for the kernel mappings.  Up until this
 513  * point, we've been running on some set of pagetables constructed by
 514  * the boot process.
 515  *
 516  * If we're booting on native hardware, this will be a pagetable
 517  * constructed in arch/x86/kernel/head_32.S.  The root of the
 518  * pagetable will be swapper_pg_dir.
 519  *
 520  * If we're booting paravirtualized under a hypervisor, then there are
 521  * more options: we may already be running PAE, and the pagetable may
 522  * or may not be based in swapper_pg_dir.  In any case,
 523  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 524  * appropriately for the rest of the initialization to work.
 525  *
 526  * In general, pagetable_init() assumes that the pagetable may already
 527  * be partially populated, and so it avoids stomping on any existing
 528  * mappings.
 529  */
 530 void __init early_ioremap_page_table_range_init(void)
 531 {
 532         pgd_t *pgd_base = swapper_pg_dir;
 533         unsigned long vaddr, end;
 534
 535         /*
 536          * Fixed mappings, only the page table structure has to be
 537          * created - mappings will be set by set_fixmap():
 538          */
 539         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 540         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 541         page_table_range_init(vaddr, end, pgd_base);
 542         early_ioremap_reset();
 543 }
 544
 545 static void __init pagetable_init(void)
 546 {
 547         pgd_t *pgd_base = swapper_pg_dir;
 548
 549         permanent_kmaps_init(pgd_base);
 550 }
 551
 552 #ifdef CONFIG_ACPI_SLEEP
 553 /*
 554  * ACPI suspend needs this for resume, because things like the intel-agp
 555  * driver might have split up a kernel 4MB mapping.
 556  */
 557 char swsusp_pg_dir[PAGE_SIZE]
 558         __attribute__ ((aligned(PAGE_SIZE)));
 559
 560 static inline void save_pg_dir(void)
 561 {
 562         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 563 }
 564 #else /* !CONFIG_ACPI_SLEEP */
 565 static inline void save_pg_dir(void)
 566 {
 567 }
 568 #endif /* !CONFIG_ACPI_SLEEP */
 569
 570 void zap_low_mappings(bool early)
 571 {
 572         int i;
 573
 574         /*
 575          * Zap initial low-memory mappings.
 576          *
 577          * Note that "pgd_clear()" doesn't do it for
 578          * us, because pgd_clear() is a no-op on i386.
 579          */
 580         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 581 #ifdef CONFIG_X86_PAE
 582                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 583 #else
 584                 set_pgd(swapper_pg_dir+i, __pgd(0));
 585 #endif
 586         }
 587
 588         if (early)
 589                 __flush_tlb();
 590         else
 591                 flush_tlb_all();
 592 }
 593
 594 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
 595 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 596
 597 /* user-defined highmem size */
 598 static unsigned int highmem_pages = -1;
 599
 600 /*
 601  * highmem=size forces highmem to be exactly 'size' bytes.
 602  * This works even on boxes that have no highmem otherwise.
 603  * This also works to reduce highmem size on bigger boxes.
 604  */
 605 static int __init parse_highmem(char *arg)
 606 {
 607         if (!arg)
 608                 return -EINVAL;
 609
 610         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 611         return 0;
 612 }
 613 early_param("highmem", parse_highmem);
 614
 615 #define MSG_HIGHMEM_TOO_BIG \
 616         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
 617
 618 #define MSG_LOWMEM_TOO_SMALL \
 619         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
 620 /*
 621  * All of RAM fits into lowmem - but if user wants highmem
 622  * artificially via the highmem=x boot parameter then create
 623  * it:
 624  */
 625 void __init lowmem_pfn_init(void)
 626 {
 627         /* max_low_pfn is 0, we already have early_res support */
 628         max_low_pfn = max_pfn;
 629
 630         if (highmem_pages == -1)
 631                 highmem_pages = 0;
 632 #ifdef CONFIG_HIGHMEM
 633         if (highmem_pages >= max_pfn) {
 634                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
 635                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 636                 highmem_pages = 0;
 637         }
 638         if (highmem_pages) {
 639                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
 640                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
 641                                 pages_to_mb(highmem_pages));
 642                         highmem_pages = 0;
 643                 }
 644                 max_low_pfn -= highmem_pages;
 645         }
 646 #else
 647         if (highmem_pages)
 648                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 649 #endif
 650 }
 651
 652 #define MSG_HIGHMEM_TOO_SMALL \
 653         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
 654
 655 #define MSG_HIGHMEM_TRIMMED \
 656         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
 657 /*
 658  * We have more RAM than fits into lowmem - we try to put it into
 659  * highmem, also taking the highmem=x boot parameter into account:
 660  */
 661 void __init highmem_pfn_init(void)
 662 {
 663         max_low_pfn = MAXMEM_PFN;
 664
 665         if (highmem_pages == -1)
 666                 highmem_pages = max_pfn - MAXMEM_PFN;
 667
 668         if (highmem_pages + MAXMEM_PFN < max_pfn)
 669                 max_pfn = MAXMEM_PFN + highmem_pages;
 670
 671         if (highmem_pages + MAXMEM_PFN > max_pfn) {
 672                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
 673                         pages_to_mb(max_pfn - MAXMEM_PFN),
 674                         pages_to_mb(highmem_pages));
 675                 highmem_pages = 0;
 676         }
 677 #ifndef CONFIG_HIGHMEM
 678         /* Maximum memory usable is what is directly addressable */
 679         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
 680         if (max_pfn > MAX_NONPAE_PFN)
 681                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 682         else
 683                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 684         max_pfn = MAXMEM_PFN;
 685 #else /* !CONFIG_HIGHMEM */
 686 #ifndef CONFIG_HIGHMEM64G
 687         if (max_pfn > MAX_NONPAE_PFN) {
 688                 max_pfn = MAX_NONPAE_PFN;
 689                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
 690         }
 691 #endif /* !CONFIG_HIGHMEM64G */
 692 #endif /* !CONFIG_HIGHMEM */
 693 }
 694
 695 /*
 696  * Determine low and high memory ranges:
 697  */
 698 void __init find_low_pfn_range(void)
 699 {
 700         /* it could update max_pfn */
 701
 702         if (max_pfn <= MAXMEM_PFN)
 703                 lowmem_pfn_init();
 704         else
 705                 highmem_pfn_init();
 706 }
 707
 708 #ifndef CONFIG_NEED_MULTIPLE_NODES
 709 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn,
 710                                 int acpi, int k8)
 711 {
 712 #ifdef CONFIG_HIGHMEM
 713         highstart_pfn = highend_pfn = max_pfn;
 714         if (max_pfn > max_low_pfn)
 715                 highstart_pfn = max_low_pfn;
 716         memblock_x86_register_active_regions(0, 0, highend_pfn);
 717         sparse_memory_present_with_active_regions(0);
 718         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 719                 pages_to_mb(highend_pfn - highstart_pfn));
 720         num_physpages = highend_pfn;
 721         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 722 #else
 723         memblock_x86_register_active_regions(0, 0, max_low_pfn);
 724         sparse_memory_present_with_active_regions(0);
 725         num_physpages = max_low_pfn;
 726         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 727 #endif
 728 #ifdef CONFIG_FLATMEM
 729         max_mapnr = num_physpages;
 730 #endif
 731         __vmalloc_start_set = true;
 732
 733         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 734                         pages_to_mb(max_low_pfn));
 735
 736         setup_bootmem_allocator();
 737 }
 738 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 739
 740 static void __init zone_sizes_init(void)
 741 {
 742         unsigned long max_zone_pfns[MAX_NR_ZONES];
 743         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 744         max_zone_pfns[ZONE_DMA] =
 745                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 746         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 747 #ifdef CONFIG_HIGHMEM
 748         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 749 #endif
 750
 751         free_area_init_nodes(max_zone_pfns);
 752 }
 753
 754 #ifndef CONFIG_NO_BOOTMEM
 755 static unsigned long __init setup_node_bootmem(int nodeid,
 756                                  unsigned long start_pfn,
 757                                  unsigned long end_pfn,
 758                                  unsigned long bootmap)
 759 {
 760         unsigned long bootmap_size;
 761
 762         /* don't touch min_low_pfn */
 763         bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
 764                                          bootmap >> PAGE_SHIFT,
 765                                          start_pfn, end_pfn);
 766         printk(KERN_INFO "  node %d low ram: %08lx - %08lx\n",
 767                 nodeid, start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
 768         printk(KERN_INFO "  node %d bootmap %08lx - %08lx\n",
 769                  nodeid, bootmap, bootmap + bootmap_size);
 770         free_bootmem_with_active_regions(nodeid, end_pfn);
 771
 772         return bootmap + bootmap_size;
 773 }
 774 #endif
 775
 776 void __init setup_bootmem_allocator(void)
 777 {
 778 #ifndef CONFIG_NO_BOOTMEM
 779         int nodeid;
 780         phys_addr_t bootmap_size, bootmap;
 781         /*
 782          * Initialize the boot-time allocator (with low memory only):
 783          */
 784         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
 785         bootmap = memblock_find_in_range(0, max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
 786                                  PAGE_SIZE);
 787         if (bootmap == MEMBLOCK_ERROR)
 788                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 789         memblock_x86_reserve_range(bootmap, bootmap + bootmap_size, "BOOTMAP");
 790 #endif
 791
 792         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 793                  max_pfn_mapped<<PAGE_SHIFT);
 794         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
 795
 796 #ifndef CONFIG_NO_BOOTMEM
 797         for_each_online_node(nodeid) {
 798                  unsigned long start_pfn, end_pfn;
 799
 800 #ifdef CONFIG_NEED_MULTIPLE_NODES
 801                 start_pfn = node_start_pfn[nodeid];
 802                 end_pfn = node_end_pfn[nodeid];
 803                 if (start_pfn > max_low_pfn)
 804                         continue;
 805                 if (end_pfn > max_low_pfn)
 806                         end_pfn = max_low_pfn;
 807 #else
 808                 start_pfn = 0;
 809                 end_pfn = max_low_pfn;
 810 #endif
 811                 bootmap = setup_node_bootmem(nodeid, start_pfn, end_pfn,
 812                                                  bootmap);
 813         }
 814 #endif
 815
 816         after_bootmem = 1;
 817 }
 818
 819 /*
 820  * paging_init() sets up the page tables - note that the first 8MB are
 821  * already mapped by head.S.
 822  *
 823  * This routines also unmaps the page at virtual kernel address 0, so
 824  * that we can trap those pesky NULL-reference errors in the kernel.
 825  */
 826 void __init paging_init(void)
 827 {
 828         pagetable_init();
 829
 830         __flush_tlb_all();
 831
 832         kmap_init();
 833
 834         /*
 835          * NOTE: at this point the bootmem allocator is fully available.
 836          */
 837         sparse_init();
 838         zone_sizes_init();
 839 }
 840
 841 /*
 842  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 843  * and also on some strange 486's. All 586+'s are OK. This used to involve
 844  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 845  * switch to using exceptions got rid of all that.
 846  */
 847 static void __init test_wp_bit(void)
 848 {
 849         printk(KERN_INFO
 850   "Checking if this processor honours the WP bit even in supervisor mode...");
 851
 852         /* Any page-aligned address will do, the test is non-destructive */
 853         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 854         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 855         clear_fixmap(FIX_WP_TEST);
 856
 857         if (!boot_cpu_data.wp_works_ok) {
 858                 printk(KERN_CONT "No.\n");
 859 #ifdef CONFIG_X86_WP_WORKS_OK
 860                 panic(
 861   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 862 #endif
 863         } else {
 864                 printk(KERN_CONT "Ok.\n");
 865         }
 866 }
 867
 868 void __init mem_init(void)
 869 {
 870         int codesize, reservedpages, datasize, initsize;
 871         int tmp;
 872
 873         pci_iommu_alloc();
 874
 875 #ifdef CONFIG_FLATMEM
 876         BUG_ON(!mem_map);
 877 #endif
 878         /* this will put all low memory onto the freelists */
 879         totalram_pages += free_all_bootmem();
 880
 881         reservedpages = 0;
 882         for (tmp = 0; tmp < max_low_pfn; tmp++)
 883                 /*
 884                  * Only count reserved RAM pages:
 885                  */
 886                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 887                         reservedpages++;
 888
 889         set_highmem_pages_init();
 890
 891         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 892         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 893         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 894
 895         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
 896                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
 897                 nr_free_pages() << (PAGE_SHIFT-10),
 898                 num_physpages << (PAGE_SHIFT-10),
 899                 codesize >> 10,
 900                 reservedpages << (PAGE_SHIFT-10),
 901                 datasize >> 10,
 902                 initsize >> 10,
 903                 totalhigh_pages << (PAGE_SHIFT-10));
 904
 905         printk(KERN_INFO "virtual kernel memory layout:\n"
 906                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 907 #ifdef CONFIG_HIGHMEM
 908                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 909 #endif
 910                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 911                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 912                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 913                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 914                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 915                 FIXADDR_START, FIXADDR_TOP,
 916                 (FIXADDR_TOP - FIXADDR_START) >> 10,
 917
 918 #ifdef CONFIG_HIGHMEM
 919                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 920                 (LAST_PKMAP*PAGE_SIZE) >> 10,
 921 #endif
 922
 923                 VMALLOC_START, VMALLOC_END,
 924                 (VMALLOC_END - VMALLOC_START) >> 20,
 925
 926                 (unsigned long)__va(0), (unsigned long)high_memory,
 927                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 928
 929                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
 930                 ((unsigned long)&__init_end -
 931                  (unsigned long)&__init_begin) >> 10,
 932
 933                 (unsigned long)&_etext, (unsigned long)&_edata,
 934                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 935
 936                 (unsigned long)&_text, (unsigned long)&_etext,
 937                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 938
 939         /*
 940          * Check boundaries twice: Some fundamental inconsistencies can
 941          * be detected at build time already.
 942          */
 943 #define __FIXADDR_TOP (-PAGE_SIZE)
 944 #ifdef CONFIG_HIGHMEM
 945         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
 946         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
 947 #endif
 948 #define high_memory (-128UL << 20)
 949         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
 950 #undef high_memory
 951 #undef __FIXADDR_TOP
 952
 953 #ifdef CONFIG_HIGHMEM
 954         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
 955         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
 956 #endif
 957         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
 958         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
 959
 960         if (boot_cpu_data.wp_works_ok < 0)
 961                 test_wp_bit();
 962
 963         save_pg_dir();
 964         zap_low_mappings(true);
 965 }
 966
 967 #ifdef CONFIG_MEMORY_HOTPLUG
 968 int arch_add_memory(int nid, u64 start, u64 size)
 969 {
 970         struct pglist_data *pgdata = NODE_DATA(nid);
 971         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 972         unsigned long start_pfn = start >> PAGE_SHIFT;
 973         unsigned long nr_pages = size >> PAGE_SHIFT;
 974
 975         return __add_pages(nid, zone, start_pfn, nr_pages);
 976 }
 977 #endif
 978
 979 /*
 980  * This function cannot be __init, since exceptions don't work in that
 981  * section.  Put this after the callers, so that it cannot be inlined.
 982  */
 983 static noinline int do_test_wp_bit(void)
 984 {
 985         char tmp_reg;
 986         int flag;
 987
 988         __asm__ __volatile__(
 989                 "       movb %0, %1     \n"
 990                 "1:     movb %1, %0     \n"
 991                 "       xorl %2, %2     \n"
 992                 "2:                     \n"
 993                 _ASM_EXTABLE(1b,2b)
 994                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 995                  "=q" (tmp_reg),
 996                  "=r" (flag)
 997                 :"2" (1)
 998                 :"memory");
 999
1000         return flag;
1001 }
1002
1003 #ifdef CONFIG_DEBUG_RODATA
1004 const int rodata_test_data = 0xC3;
1005 EXPORT_SYMBOL_GPL(rodata_test_data);
1006
1007 int kernel_set_to_readonly __read_mostly;
1008
1009 void set_kernel_text_rw(void)
1010 {
1011         unsigned long start = PFN_ALIGN(_text);
1012         unsigned long size = PFN_ALIGN(_etext) - start;
1013
1014         if (!kernel_set_to_readonly)
1015                 return;
1016
1017         pr_debug("Set kernel text: %lx - %lx for read write\n",
1018                  start, start+size);
1019
1020         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1021 }
1022
1023 void set_kernel_text_ro(void)
1024 {
1025         unsigned long start = PFN_ALIGN(_text);
1026         unsigned long size = PFN_ALIGN(_etext) - start;
1027
1028         if (!kernel_set_to_readonly)
1029                 return;
1030
1031         pr_debug("Set kernel text: %lx - %lx for read only\n",
1032                  start, start+size);
1033
1034         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1035 }
1036
1037 void mark_rodata_ro(void)
1038 {
1039         unsigned long start = PFN_ALIGN(_text);
1040         unsigned long size = PFN_ALIGN(_etext) - start;
1041
1042         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1043         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1044                 size >> 10);
1045
1046         kernel_set_to_readonly = 1;
1047
1048 #ifdef CONFIG_CPA_DEBUG
1049         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1050                 start, start+size);
1051         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1052
1053         printk(KERN_INFO "Testing CPA: write protecting again\n");
1054         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1055 #endif
1056
1057         start += size;
1058         size = (unsigned long)__end_rodata - start;
1059         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1060         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1061                 size >> 10);
1062         rodata_test();
1063
1064 #ifdef CONFIG_CPA_DEBUG
1065         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1066         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1067
1068         printk(KERN_INFO "Testing CPA: write protecting again\n");
1069         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1070 #endif
1071 }
1072 #endif
1073