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 void __init add_highpages_with_active_regions(int nid,
 427                          unsigned long start_pfn, unsigned long end_pfn)
 428 {
 429         struct range *range;
 430         int nr_range;
 431         int i;
 432
 433         nr_range = __get_free_all_memory_range(&range, nid, start_pfn, end_pfn);
 434
 435         for (i = 0; i < nr_range; i++) {
 436                 struct page *page;
 437                 int node_pfn;
 438
 439                 for (node_pfn = range[i].start; node_pfn < range[i].end;
 440                      node_pfn++) {
 441                         if (!pfn_valid(node_pfn))
 442                                 continue;
 443                         page = pfn_to_page(node_pfn);
 444                         add_one_highpage_init(page);
 445                 }
 446         }
 447 }
 448 #else
 449 static inline void permanent_kmaps_init(pgd_t *pgd_base)
 450 {
 451 }
 452 #endif /* CONFIG_HIGHMEM */
 453
 454 void __init native_pagetable_setup_start(pgd_t *base)
 455 {
 456         unsigned long pfn, va;
 457         pgd_t *pgd;
 458         pud_t *pud;
 459         pmd_t *pmd;
 460         pte_t *pte;
 461
 462         /*
 463          * Remove any mappings which extend past the end of physical
 464          * memory from the boot time page table:
 465          */
 466         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 467                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 468                 pgd = base + pgd_index(va);
 469                 if (!pgd_present(*pgd))
 470                         break;
 471
 472                 pud = pud_offset(pgd, va);
 473                 pmd = pmd_offset(pud, va);
 474                 if (!pmd_present(*pmd))
 475                         break;
 476
 477                 pte = pte_offset_kernel(pmd, va);
 478                 if (!pte_present(*pte))
 479                         break;
 480
 481                 pte_clear(NULL, va, pte);
 482         }
 483         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 484 }
 485
 486 void __init native_pagetable_setup_done(pgd_t *base)
 487 {
 488 }
 489
 490 /*
 491  * Build a proper pagetable for the kernel mappings.  Up until this
 492  * point, we've been running on some set of pagetables constructed by
 493  * the boot process.
 494  *
 495  * If we're booting on native hardware, this will be a pagetable
 496  * constructed in arch/x86/kernel/head_32.S.  The root of the
 497  * pagetable will be swapper_pg_dir.
 498  *
 499  * If we're booting paravirtualized under a hypervisor, then there are
 500  * more options: we may already be running PAE, and the pagetable may
 501  * or may not be based in swapper_pg_dir.  In any case,
 502  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 503  * appropriately for the rest of the initialization to work.
 504  *
 505  * In general, pagetable_init() assumes that the pagetable may already
 506  * be partially populated, and so it avoids stomping on any existing
 507  * mappings.
 508  */
 509 void __init early_ioremap_page_table_range_init(void)
 510 {
 511         pgd_t *pgd_base = swapper_pg_dir;
 512         unsigned long vaddr, end;
 513
 514         /*
 515          * Fixed mappings, only the page table structure has to be
 516          * created - mappings will be set by set_fixmap():
 517          */
 518         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 519         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 520         page_table_range_init(vaddr, end, pgd_base);
 521         early_ioremap_reset();
 522 }
 523
 524 static void __init pagetable_init(void)
 525 {
 526         pgd_t *pgd_base = swapper_pg_dir;
 527
 528         permanent_kmaps_init(pgd_base);
 529 }
 530
 531 #ifdef CONFIG_ACPI_SLEEP
 532 /*
 533  * ACPI suspend needs this for resume, because things like the intel-agp
 534  * driver might have split up a kernel 4MB mapping.
 535  */
 536 char swsusp_pg_dir[PAGE_SIZE]
 537         __attribute__ ((aligned(PAGE_SIZE)));
 538
 539 static inline void save_pg_dir(void)
 540 {
 541         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 542 }
 543 #else /* !CONFIG_ACPI_SLEEP */
 544 static inline void save_pg_dir(void)
 545 {
 546 }
 547 #endif /* !CONFIG_ACPI_SLEEP */
 548
 549 void zap_low_mappings(bool early)
 550 {
 551         int i;
 552
 553         /*
 554          * Zap initial low-memory mappings.
 555          *
 556          * Note that "pgd_clear()" doesn't do it for
 557          * us, because pgd_clear() is a no-op on i386.
 558          */
 559         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 560 #ifdef CONFIG_X86_PAE
 561                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 562 #else
 563                 set_pgd(swapper_pg_dir+i, __pgd(0));
 564 #endif
 565         }
 566
 567         if (early)
 568                 __flush_tlb();
 569         else
 570                 flush_tlb_all();
 571 }
 572
 573 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
 574 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 575
 576 /* user-defined highmem size */
 577 static unsigned int highmem_pages = -1;
 578
 579 /*
 580  * highmem=size forces highmem to be exactly 'size' bytes.
 581  * This works even on boxes that have no highmem otherwise.
 582  * This also works to reduce highmem size on bigger boxes.
 583  */
 584 static int __init parse_highmem(char *arg)
 585 {
 586         if (!arg)
 587                 return -EINVAL;
 588
 589         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 590         return 0;
 591 }
 592 early_param("highmem", parse_highmem);
 593
 594 #define MSG_HIGHMEM_TOO_BIG \
 595         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
 596
 597 #define MSG_LOWMEM_TOO_SMALL \
 598         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
 599 /*
 600  * All of RAM fits into lowmem - but if user wants highmem
 601  * artificially via the highmem=x boot parameter then create
 602  * it:
 603  */
 604 void __init lowmem_pfn_init(void)
 605 {
 606         /* max_low_pfn is 0, we already have early_res support */
 607         max_low_pfn = max_pfn;
 608
 609         if (highmem_pages == -1)
 610                 highmem_pages = 0;
 611 #ifdef CONFIG_HIGHMEM
 612         if (highmem_pages >= max_pfn) {
 613                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
 614                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 615                 highmem_pages = 0;
 616         }
 617         if (highmem_pages) {
 618                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
 619                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
 620                                 pages_to_mb(highmem_pages));
 621                         highmem_pages = 0;
 622                 }
 623                 max_low_pfn -= highmem_pages;
 624         }
 625 #else
 626         if (highmem_pages)
 627                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 628 #endif
 629 }
 630
 631 #define MSG_HIGHMEM_TOO_SMALL \
 632         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
 633
 634 #define MSG_HIGHMEM_TRIMMED \
 635         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
 636 /*
 637  * We have more RAM than fits into lowmem - we try to put it into
 638  * highmem, also taking the highmem=x boot parameter into account:
 639  */
 640 void __init highmem_pfn_init(void)
 641 {
 642         max_low_pfn = MAXMEM_PFN;
 643
 644         if (highmem_pages == -1)
 645                 highmem_pages = max_pfn - MAXMEM_PFN;
 646
 647         if (highmem_pages + MAXMEM_PFN < max_pfn)
 648                 max_pfn = MAXMEM_PFN + highmem_pages;
 649
 650         if (highmem_pages + MAXMEM_PFN > max_pfn) {
 651                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
 652                         pages_to_mb(max_pfn - MAXMEM_PFN),
 653                         pages_to_mb(highmem_pages));
 654                 highmem_pages = 0;
 655         }
 656 #ifndef CONFIG_HIGHMEM
 657         /* Maximum memory usable is what is directly addressable */
 658         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
 659         if (max_pfn > MAX_NONPAE_PFN)
 660                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 661         else
 662                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 663         max_pfn = MAXMEM_PFN;
 664 #else /* !CONFIG_HIGHMEM */
 665 #ifndef CONFIG_HIGHMEM64G
 666         if (max_pfn > MAX_NONPAE_PFN) {
 667                 max_pfn = MAX_NONPAE_PFN;
 668                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
 669         }
 670 #endif /* !CONFIG_HIGHMEM64G */
 671 #endif /* !CONFIG_HIGHMEM */
 672 }
 673
 674 /*
 675  * Determine low and high memory ranges:
 676  */
 677 void __init find_low_pfn_range(void)
 678 {
 679         /* it could update max_pfn */
 680
 681         if (max_pfn <= MAXMEM_PFN)
 682                 lowmem_pfn_init();
 683         else
 684                 highmem_pfn_init();
 685 }
 686
 687 #ifndef CONFIG_NEED_MULTIPLE_NODES
 688 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn,
 689                                 int acpi, int k8)
 690 {
 691 #ifdef CONFIG_HIGHMEM
 692         highstart_pfn = highend_pfn = max_pfn;
 693         if (max_pfn > max_low_pfn)
 694                 highstart_pfn = max_low_pfn;
 695         memblock_x86_register_active_regions(0, 0, highend_pfn);
 696         sparse_memory_present_with_active_regions(0);
 697         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 698                 pages_to_mb(highend_pfn - highstart_pfn));
 699         num_physpages = highend_pfn;
 700         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 701 #else
 702         memblock_x86_register_active_regions(0, 0, max_low_pfn);
 703         sparse_memory_present_with_active_regions(0);
 704         num_physpages = max_low_pfn;
 705         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 706 #endif
 707 #ifdef CONFIG_FLATMEM
 708         max_mapnr = num_physpages;
 709 #endif
 710         __vmalloc_start_set = true;
 711
 712         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 713                         pages_to_mb(max_low_pfn));
 714
 715         setup_bootmem_allocator();
 716 }
 717 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 718
 719 static void __init zone_sizes_init(void)
 720 {
 721         unsigned long max_zone_pfns[MAX_NR_ZONES];
 722         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 723         max_zone_pfns[ZONE_DMA] =
 724                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 725         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 726 #ifdef CONFIG_HIGHMEM
 727         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 728 #endif
 729
 730         free_area_init_nodes(max_zone_pfns);
 731 }
 732
 733 void __init setup_bootmem_allocator(void)
 734 {
 735         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 736                  max_pfn_mapped<<PAGE_SHIFT);
 737         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
 738
 739         after_bootmem = 1;
 740 }
 741
 742 /*
 743  * paging_init() sets up the page tables - note that the first 8MB are
 744  * already mapped by head.S.
 745  *
 746  * This routines also unmaps the page at virtual kernel address 0, so
 747  * that we can trap those pesky NULL-reference errors in the kernel.
 748  */
 749 void __init paging_init(void)
 750 {
 751         pagetable_init();
 752
 753         __flush_tlb_all();
 754
 755         kmap_init();
 756
 757         /*
 758          * NOTE: at this point the bootmem allocator is fully available.
 759          */
 760         sparse_init();
 761         zone_sizes_init();
 762 }
 763
 764 /*
 765  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 766  * and also on some strange 486's. All 586+'s are OK. This used to involve
 767  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 768  * switch to using exceptions got rid of all that.
 769  */
 770 static void __init test_wp_bit(void)
 771 {
 772         printk(KERN_INFO
 773   "Checking if this processor honours the WP bit even in supervisor mode...");
 774
 775         /* Any page-aligned address will do, the test is non-destructive */
 776         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 777         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 778         clear_fixmap(FIX_WP_TEST);
 779
 780         if (!boot_cpu_data.wp_works_ok) {
 781                 printk(KERN_CONT "No.\n");
 782 #ifdef CONFIG_X86_WP_WORKS_OK
 783                 panic(
 784   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 785 #endif
 786         } else {
 787                 printk(KERN_CONT "Ok.\n");
 788         }
 789 }
 790
 791 void __init mem_init(void)
 792 {
 793         int codesize, reservedpages, datasize, initsize;
 794         int tmp;
 795
 796         pci_iommu_alloc();
 797
 798 #ifdef CONFIG_FLATMEM
 799         BUG_ON(!mem_map);
 800 #endif
 801         /* this will put all low memory onto the freelists */
 802         totalram_pages += free_all_bootmem();
 803
 804         reservedpages = 0;
 805         for (tmp = 0; tmp < max_low_pfn; tmp++)
 806                 /*
 807                  * Only count reserved RAM pages:
 808                  */
 809                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 810                         reservedpages++;
 811
 812         set_highmem_pages_init();
 813
 814         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 815         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 816         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 817
 818         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
 819                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
 820                 nr_free_pages() << (PAGE_SHIFT-10),
 821                 num_physpages << (PAGE_SHIFT-10),
 822                 codesize >> 10,
 823                 reservedpages << (PAGE_SHIFT-10),
 824                 datasize >> 10,
 825                 initsize >> 10,
 826                 totalhigh_pages << (PAGE_SHIFT-10));
 827
 828         printk(KERN_INFO "virtual kernel memory layout:\n"
 829                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 830 #ifdef CONFIG_HIGHMEM
 831                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 832 #endif
 833                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 834                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 835                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 836                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 837                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 838                 FIXADDR_START, FIXADDR_TOP,
 839                 (FIXADDR_TOP - FIXADDR_START) >> 10,
 840
 841 #ifdef CONFIG_HIGHMEM
 842                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 843                 (LAST_PKMAP*PAGE_SIZE) >> 10,
 844 #endif
 845
 846                 VMALLOC_START, VMALLOC_END,
 847                 (VMALLOC_END - VMALLOC_START) >> 20,
 848
 849                 (unsigned long)__va(0), (unsigned long)high_memory,
 850                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 851
 852                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
 853                 ((unsigned long)&__init_end -
 854                  (unsigned long)&__init_begin) >> 10,
 855
 856                 (unsigned long)&_etext, (unsigned long)&_edata,
 857                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 858
 859                 (unsigned long)&_text, (unsigned long)&_etext,
 860                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 861
 862         /*
 863          * Check boundaries twice: Some fundamental inconsistencies can
 864          * be detected at build time already.
 865          */
 866 #define __FIXADDR_TOP (-PAGE_SIZE)
 867 #ifdef CONFIG_HIGHMEM
 868         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
 869         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
 870 #endif
 871 #define high_memory (-128UL << 20)
 872         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
 873 #undef high_memory
 874 #undef __FIXADDR_TOP
 875
 876 #ifdef CONFIG_HIGHMEM
 877         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
 878         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
 879 #endif
 880         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
 881         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
 882
 883         if (boot_cpu_data.wp_works_ok < 0)
 884                 test_wp_bit();
 885
 886         save_pg_dir();
 887         zap_low_mappings(true);
 888 }
 889
 890 #ifdef CONFIG_MEMORY_HOTPLUG
 891 int arch_add_memory(int nid, u64 start, u64 size)
 892 {
 893         struct pglist_data *pgdata = NODE_DATA(nid);
 894         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 895         unsigned long start_pfn = start >> PAGE_SHIFT;
 896         unsigned long nr_pages = size >> PAGE_SHIFT;
 897
 898         return __add_pages(nid, zone, start_pfn, nr_pages);
 899 }
 900 #endif
 901
 902 /*
 903  * This function cannot be __init, since exceptions don't work in that
 904  * section.  Put this after the callers, so that it cannot be inlined.
 905  */
 906 static noinline int do_test_wp_bit(void)
 907 {
 908         char tmp_reg;
 909         int flag;
 910
 911         __asm__ __volatile__(
 912                 "       movb %0, %1     \n"
 913                 "1:     movb %1, %0     \n"
 914                 "       xorl %2, %2     \n"
 915                 "2:                     \n"
 916                 _ASM_EXTABLE(1b,2b)
 917                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 918                  "=q" (tmp_reg),
 919                  "=r" (flag)
 920                 :"2" (1)
 921                 :"memory");
 922
 923         return flag;
 924 }
 925
 926 #ifdef CONFIG_DEBUG_RODATA
 927 const int rodata_test_data = 0xC3;
 928 EXPORT_SYMBOL_GPL(rodata_test_data);
 929
 930 int kernel_set_to_readonly __read_mostly;
 931
 932 void set_kernel_text_rw(void)
 933 {
 934         unsigned long start = PFN_ALIGN(_text);
 935         unsigned long size = PFN_ALIGN(_etext) - start;
 936
 937         if (!kernel_set_to_readonly)
 938                 return;
 939
 940         pr_debug("Set kernel text: %lx - %lx for read write\n",
 941                  start, start+size);
 942
 943         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
 944 }
 945
 946 void set_kernel_text_ro(void)
 947 {
 948         unsigned long start = PFN_ALIGN(_text);
 949         unsigned long size = PFN_ALIGN(_etext) - start;
 950
 951         if (!kernel_set_to_readonly)
 952                 return;
 953
 954         pr_debug("Set kernel text: %lx - %lx for read only\n",
 955                  start, start+size);
 956
 957         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 958 }
 959
 960 void mark_rodata_ro(void)
 961 {
 962         unsigned long start = PFN_ALIGN(_text);
 963         unsigned long size = PFN_ALIGN(_etext) - start;
 964
 965         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 966         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
 967                 size >> 10);
 968
 969         kernel_set_to_readonly = 1;
 970
 971 #ifdef CONFIG_CPA_DEBUG
 972         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
 973                 start, start+size);
 974         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
 975
 976         printk(KERN_INFO "Testing CPA: write protecting again\n");
 977         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
 978 #endif
 979
 980         start += size;
 981         size = (unsigned long)__end_rodata - start;
 982         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 983         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 984                 size >> 10);
 985         rodata_test();
 986
 987 #ifdef CONFIG_CPA_DEBUG
 988         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
 989         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
 990
 991         printk(KERN_INFO "Testing CPA: write protecting again\n");
 992         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 993 #endif
 994 }
 995 #endif
 996