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 unsigned long highstart_pfn, highend_pfn;
  56
  57 static noinline int do_test_wp_bit(void);
  58
  59 bool __read_mostly __vmalloc_start_set = false;
  60
  61 static __init void *alloc_low_page(void)
  62 {
  63         unsigned long pfn = e820_table_end++;
  64         void *adr;
  65
  66         if (pfn >= e820_table_top)
  67                 panic("alloc_low_page: ran out of memory");
  68
  69         adr = __va(pfn * PAGE_SIZE);
  70         memset(adr, 0, PAGE_SIZE);
  71         return adr;
  72 }
  73
  74 /*
  75  * Creates a middle page table and puts a pointer to it in the
  76  * given global directory entry. This only returns the gd entry
  77  * in non-PAE compilation mode, since the middle layer is folded.
  78  */
  79 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  80 {
  81         pud_t *pud;
  82         pmd_t *pmd_table;
  83
  84 #ifdef CONFIG_X86_PAE
  85         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
  86                 if (after_bootmem)
  87                         pmd_table = (pmd_t *)alloc_bootmem_pages(PAGE_SIZE);
  88                 else
  89                         pmd_table = (pmd_t *)alloc_low_page();
  90                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
  91                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  92                 pud = pud_offset(pgd, 0);
  93                 BUG_ON(pmd_table != pmd_offset(pud, 0));
  94
  95                 return pmd_table;
  96         }
  97 #endif
  98         pud = pud_offset(pgd, 0);
  99         pmd_table = pmd_offset(pud, 0);
 100
 101         return pmd_table;
 102 }
 103
 104 /*
 105  * Create a page table and place a pointer to it in a middle page
 106  * directory entry:
 107  */
 108 static pte_t * __init one_page_table_init(pmd_t *pmd)
 109 {
 110         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
 111                 pte_t *page_table = NULL;
 112
 113                 if (after_bootmem) {
 114 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
 115                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
 116 #endif
 117                         if (!page_table)
 118                                 page_table =
 119                                 (pte_t *)alloc_bootmem_pages(PAGE_SIZE);
 120                 } else
 121                         page_table = (pte_t *)alloc_low_page();
 122
 123                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
 124                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
 125                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
 126         }
 127
 128         return pte_offset_kernel(pmd, 0);
 129 }
 130
 131 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
 132 {
 133         int pgd_idx = pgd_index(vaddr);
 134         int pmd_idx = pmd_index(vaddr);
 135
 136         return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
 137 }
 138
 139 pte_t * __init populate_extra_pte(unsigned long vaddr)
 140 {
 141         int pte_idx = pte_index(vaddr);
 142         pmd_t *pmd;
 143
 144         pmd = populate_extra_pmd(vaddr);
 145         return one_page_table_init(pmd) + pte_idx;
 146 }
 147
 148 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
 149                                            unsigned long vaddr, pte_t *lastpte)
 150 {
 151 #ifdef CONFIG_HIGHMEM
 152         /*
 153          * Something (early fixmap) may already have put a pte
 154          * page here, which causes the page table allocation
 155          * to become nonlinear. Attempt to fix it, and if it
 156          * is still nonlinear then we have to bug.
 157          */
 158         int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
 159         int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
 160
 161         if (pmd_idx_kmap_begin != pmd_idx_kmap_end
 162             && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
 163             && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
 164             && ((__pa(pte) >> PAGE_SHIFT) < e820_table_start
 165                 || (__pa(pte) >> PAGE_SHIFT) >= e820_table_end)) {
 166                 pte_t *newpte;
 167                 int i;
 168
 169                 BUG_ON(after_bootmem);
 170                 newpte = alloc_low_page();
 171                 for (i = 0; i < PTRS_PER_PTE; i++)
 172                         set_pte(newpte + i, pte[i]);
 173
 174                 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
 175                 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
 176                 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
 177                 __flush_tlb_all();
 178
 179                 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
 180                 pte = newpte;
 181         }
 182         BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
 183                && vaddr > fix_to_virt(FIX_KMAP_END)
 184                && lastpte && lastpte + PTRS_PER_PTE != pte);
 185 #endif
 186         return pte;
 187 }
 188
 189 /*
 190  * This function initializes a certain range of kernel virtual memory
 191  * with new bootmem page tables, everywhere page tables are missing in
 192  * the given range.
 193  *
 194  * NOTE: The pagetables are allocated contiguous on the physical space
 195  * so we can cache the place of the first one and move around without
 196  * checking the pgd every time.
 197  */
 198 static void __init
 199 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
 200 {
 201         int pgd_idx, pmd_idx;
 202         unsigned long vaddr;
 203         pgd_t *pgd;
 204         pmd_t *pmd;
 205         pte_t *pte = NULL;
 206
 207         vaddr = start;
 208         pgd_idx = pgd_index(vaddr);
 209         pmd_idx = pmd_index(vaddr);
 210         pgd = pgd_base + pgd_idx;
 211
 212         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 213                 pmd = one_md_table_init(pgd);
 214                 pmd = pmd + pmd_index(vaddr);
 215                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
 216                                                         pmd++, pmd_idx++) {
 217                         pte = page_table_kmap_check(one_page_table_init(pmd),
 218                                                     pmd, vaddr, pte);
 219
 220                         vaddr += PMD_SIZE;
 221                 }
 222                 pmd_idx = 0;
 223         }
 224 }
 225
 226 static inline int is_kernel_text(unsigned long addr)
 227 {
 228         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 229                 return 1;
 230         return 0;
 231 }
 232
 233 /*
 234  * This maps the physical memory to kernel virtual address space, a total
 235  * of max_low_pfn pages, by creating page tables starting from address
 236  * PAGE_OFFSET:
 237  */
 238 unsigned long __init
 239 kernel_physical_mapping_init(unsigned long start,
 240                              unsigned long end,
 241                              unsigned long page_size_mask)
 242 {
 243         int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
 244         unsigned long start_pfn, end_pfn;
 245         pgd_t *pgd_base = swapper_pg_dir;
 246         int pgd_idx, pmd_idx, pte_ofs;
 247         unsigned long pfn;
 248         pgd_t *pgd;
 249         pmd_t *pmd;
 250         pte_t *pte;
 251         unsigned pages_2m, pages_4k;
 252         int mapping_iter;
 253
 254         start_pfn = start >> PAGE_SHIFT;
 255         end_pfn = end >> PAGE_SHIFT;
 256
 257         /*
 258          * First iteration will setup identity mapping using large/small pages
 259          * based on use_pse, with other attributes same as set by
 260          * the early code in head_32.S
 261          *
 262          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
 263          * as desired for the kernel identity mapping.
 264          *
 265          * This two pass mechanism conforms to the TLB app note which says:
 266          *
 267          *     "Software should not write to a paging-structure entry in a way
 268          *      that would change, for any linear address, both the page size
 269          *      and either the page frame or attributes."
 270          */
 271         mapping_iter = 1;
 272
 273         if (!cpu_has_pse)
 274                 use_pse = 0;
 275
 276 repeat:
 277         pages_2m = pages_4k = 0;
 278         pfn = start_pfn;
 279         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 280         pgd = pgd_base + pgd_idx;
 281         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 282                 pmd = one_md_table_init(pgd);
 283
 284                 if (pfn >= end_pfn)
 285                         continue;
 286 #ifdef CONFIG_X86_PAE
 287                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 288                 pmd += pmd_idx;
 289 #else
 290                 pmd_idx = 0;
 291 #endif
 292                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
 293                      pmd++, pmd_idx++) {
 294                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
 295
 296                         /*
 297                          * Map with big pages if possible, otherwise
 298                          * create normal page tables:
 299                          */
 300                         if (use_pse) {
 301                                 unsigned int addr2;
 302                                 pgprot_t prot = PAGE_KERNEL_LARGE;
 303                                 /*
 304                                  * first pass will use the same initial
 305                                  * identity mapping attribute + _PAGE_PSE.
 306                                  */
 307                                 pgprot_t init_prot =
 308                                         __pgprot(PTE_IDENT_ATTR |
 309                                                  _PAGE_PSE);
 310
 311                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
 312                                         PAGE_OFFSET + PAGE_SIZE-1;
 313
 314                                 if (is_kernel_text(addr) ||
 315                                     is_kernel_text(addr2))
 316                                         prot = PAGE_KERNEL_LARGE_EXEC;
 317
 318                                 pages_2m++;
 319                                 if (mapping_iter == 1)
 320                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
 321                                 else
 322                                         set_pmd(pmd, pfn_pmd(pfn, prot));
 323
 324                                 pfn += PTRS_PER_PTE;
 325                                 continue;
 326                         }
 327                         pte = one_page_table_init(pmd);
 328
 329                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 330                         pte += pte_ofs;
 331                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
 332                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
 333                                 pgprot_t prot = PAGE_KERNEL;
 334                                 /*
 335                                  * first pass will use the same initial
 336                                  * identity mapping attribute.
 337                                  */
 338                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
 339
 340                                 if (is_kernel_text(addr))
 341                                         prot = PAGE_KERNEL_EXEC;
 342
 343                                 pages_4k++;
 344                                 if (mapping_iter == 1)
 345                                         set_pte(pte, pfn_pte(pfn, init_prot));
 346                                 else
 347                                         set_pte(pte, pfn_pte(pfn, prot));
 348                         }
 349                 }
 350         }
 351         if (mapping_iter == 1) {
 352                 /*
 353                  * update direct mapping page count only in the first
 354                  * iteration.
 355                  */
 356                 update_page_count(PG_LEVEL_2M, pages_2m);
 357                 update_page_count(PG_LEVEL_4K, pages_4k);
 358
 359                 /*
 360                  * local global flush tlb, which will flush the previous
 361                  * mappings present in both small and large page TLB's.
 362                  */
 363                 __flush_tlb_all();
 364
 365                 /*
 366                  * Second iteration will set the actual desired PTE attributes.
 367                  */
 368                 mapping_iter = 2;
 369                 goto repeat;
 370         }
 371         return 0;
 372 }
 373
 374 pte_t *kmap_pte;
 375 pgprot_t kmap_prot;
 376
 377 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 378 {
 379         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 380                         vaddr), vaddr), vaddr);
 381 }
 382
 383 static void __init kmap_init(void)
 384 {
 385         unsigned long kmap_vstart;
 386
 387         /*
 388          * Cache the first kmap pte:
 389          */
 390         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 391         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 392
 393         kmap_prot = PAGE_KERNEL;
 394 }
 395
 396 #ifdef CONFIG_HIGHMEM
 397 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 398 {
 399         unsigned long vaddr;
 400         pgd_t *pgd;
 401         pud_t *pud;
 402         pmd_t *pmd;
 403         pte_t *pte;
 404
 405         vaddr = PKMAP_BASE;
 406         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 407
 408         pgd = swapper_pg_dir + pgd_index(vaddr);
 409         pud = pud_offset(pgd, vaddr);
 410         pmd = pmd_offset(pud, vaddr);
 411         pte = pte_offset_kernel(pmd, vaddr);
 412         pkmap_page_table = pte;
 413 }
 414
 415 static void __init add_one_highpage_init(struct page *page)
 416 {
 417         ClearPageReserved(page);
 418         init_page_count(page);
 419         __free_page(page);
 420         totalhigh_pages++;
 421 }
 422
 423 struct add_highpages_data {
 424         unsigned long start_pfn;
 425         unsigned long end_pfn;
 426 };
 427
 428 static int __init add_highpages_work_fn(unsigned long start_pfn,
 429                                          unsigned long end_pfn, void *datax)
 430 {
 431         int node_pfn;
 432         struct page *page;
 433         unsigned long final_start_pfn, final_end_pfn;
 434         struct add_highpages_data *data;
 435
 436         data = (struct add_highpages_data *)datax;
 437
 438         final_start_pfn = max(start_pfn, data->start_pfn);
 439         final_end_pfn = min(end_pfn, data->end_pfn);
 440         if (final_start_pfn >= final_end_pfn)
 441                 return 0;
 442
 443         for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
 444              node_pfn++) {
 445                 if (!pfn_valid(node_pfn))
 446                         continue;
 447                 page = pfn_to_page(node_pfn);
 448                 add_one_highpage_init(page);
 449         }
 450
 451         return 0;
 452
 453 }
 454
 455 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
 456                                               unsigned long end_pfn)
 457 {
 458         struct add_highpages_data data;
 459
 460         data.start_pfn = start_pfn;
 461         data.end_pfn = end_pfn;
 462
 463         work_with_active_regions(nid, add_highpages_work_fn, &data);
 464 }
 465
 466 #else
 467 static inline void permanent_kmaps_init(pgd_t *pgd_base)
 468 {
 469 }
 470 #endif /* CONFIG_HIGHMEM */
 471
 472 void __init native_pagetable_setup_start(pgd_t *base)
 473 {
 474         unsigned long pfn, va;
 475         pgd_t *pgd;
 476         pud_t *pud;
 477         pmd_t *pmd;
 478         pte_t *pte;
 479
 480         /*
 481          * Remove any mappings which extend past the end of physical
 482          * memory from the boot time page table:
 483          */
 484         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 485                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 486                 pgd = base + pgd_index(va);
 487                 if (!pgd_present(*pgd))
 488                         break;
 489
 490                 pud = pud_offset(pgd, va);
 491                 pmd = pmd_offset(pud, va);
 492                 if (!pmd_present(*pmd))
 493                         break;
 494
 495                 pte = pte_offset_kernel(pmd, va);
 496                 if (!pte_present(*pte))
 497                         break;
 498
 499                 pte_clear(NULL, va, pte);
 500         }
 501         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 502 }
 503
 504 void __init native_pagetable_setup_done(pgd_t *base)
 505 {
 506 }
 507
 508 /*
 509  * Build a proper pagetable for the kernel mappings.  Up until this
 510  * point, we've been running on some set of pagetables constructed by
 511  * the boot process.
 512  *
 513  * If we're booting on native hardware, this will be a pagetable
 514  * constructed in arch/x86/kernel/head_32.S.  The root of the
 515  * pagetable will be swapper_pg_dir.
 516  *
 517  * If we're booting paravirtualized under a hypervisor, then there are
 518  * more options: we may already be running PAE, and the pagetable may
 519  * or may not be based in swapper_pg_dir.  In any case,
 520  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 521  * appropriately for the rest of the initialization to work.
 522  *
 523  * In general, pagetable_init() assumes that the pagetable may already
 524  * be partially populated, and so it avoids stomping on any existing
 525  * mappings.
 526  */
 527 void __init early_ioremap_page_table_range_init(void)
 528 {
 529         pgd_t *pgd_base = swapper_pg_dir;
 530         unsigned long vaddr, end;
 531
 532         /*
 533          * Fixed mappings, only the page table structure has to be
 534          * created - mappings will be set by set_fixmap():
 535          */
 536         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 537         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 538         page_table_range_init(vaddr, end, pgd_base);
 539         early_ioremap_reset();
 540 }
 541
 542 static void __init pagetable_init(void)
 543 {
 544         pgd_t *pgd_base = swapper_pg_dir;
 545
 546         permanent_kmaps_init(pgd_base);
 547 }
 548
 549 #ifdef CONFIG_ACPI_SLEEP
 550 /*
 551  * ACPI suspend needs this for resume, because things like the intel-agp
 552  * driver might have split up a kernel 4MB mapping.
 553  */
 554 char swsusp_pg_dir[PAGE_SIZE]
 555         __attribute__ ((aligned(PAGE_SIZE)));
 556
 557 static inline void save_pg_dir(void)
 558 {
 559         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 560 }
 561 #else /* !CONFIG_ACPI_SLEEP */
 562 static inline void save_pg_dir(void)
 563 {
 564 }
 565 #endif /* !CONFIG_ACPI_SLEEP */
 566
 567 void zap_low_mappings(bool early)
 568 {
 569         int i;
 570
 571         /*
 572          * Zap initial low-memory mappings.
 573          *
 574          * Note that "pgd_clear()" doesn't do it for
 575          * us, because pgd_clear() is a no-op on i386.
 576          */
 577         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 578 #ifdef CONFIG_X86_PAE
 579                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 580 #else
 581                 set_pgd(swapper_pg_dir+i, __pgd(0));
 582 #endif
 583         }
 584
 585         if (early)
 586                 __flush_tlb();
 587         else
 588                 flush_tlb_all();
 589 }
 590
 591 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
 592 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 593
 594 /* user-defined highmem size */
 595 static unsigned int highmem_pages = -1;
 596
 597 /*
 598  * highmem=size forces highmem to be exactly 'size' bytes.
 599  * This works even on boxes that have no highmem otherwise.
 600  * This also works to reduce highmem size on bigger boxes.
 601  */
 602 static int __init parse_highmem(char *arg)
 603 {
 604         if (!arg)
 605                 return -EINVAL;
 606
 607         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 608         return 0;
 609 }
 610 early_param("highmem", parse_highmem);
 611
 612 #define MSG_HIGHMEM_TOO_BIG \
 613         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
 614
 615 #define MSG_LOWMEM_TOO_SMALL \
 616         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
 617 /*
 618  * All of RAM fits into lowmem - but if user wants highmem
 619  * artificially via the highmem=x boot parameter then create
 620  * it:
 621  */
 622 void __init lowmem_pfn_init(void)
 623 {
 624         /* max_low_pfn is 0, we already have early_res support */
 625         max_low_pfn = max_pfn;
 626
 627         if (highmem_pages == -1)
 628                 highmem_pages = 0;
 629 #ifdef CONFIG_HIGHMEM
 630         if (highmem_pages >= max_pfn) {
 631                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
 632                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 633                 highmem_pages = 0;
 634         }
 635         if (highmem_pages) {
 636                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
 637                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
 638                                 pages_to_mb(highmem_pages));
 639                         highmem_pages = 0;
 640                 }
 641                 max_low_pfn -= highmem_pages;
 642         }
 643 #else
 644         if (highmem_pages)
 645                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 646 #endif
 647 }
 648
 649 #define MSG_HIGHMEM_TOO_SMALL \
 650         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
 651
 652 #define MSG_HIGHMEM_TRIMMED \
 653         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
 654 /*
 655  * We have more RAM than fits into lowmem - we try to put it into
 656  * highmem, also taking the highmem=x boot parameter into account:
 657  */
 658 void __init highmem_pfn_init(void)
 659 {
 660         max_low_pfn = MAXMEM_PFN;
 661
 662         if (highmem_pages == -1)
 663                 highmem_pages = max_pfn - MAXMEM_PFN;
 664
 665         if (highmem_pages + MAXMEM_PFN < max_pfn)
 666                 max_pfn = MAXMEM_PFN + highmem_pages;
 667
 668         if (highmem_pages + MAXMEM_PFN > max_pfn) {
 669                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
 670                         pages_to_mb(max_pfn - MAXMEM_PFN),
 671                         pages_to_mb(highmem_pages));
 672                 highmem_pages = 0;
 673         }
 674 #ifndef CONFIG_HIGHMEM
 675         /* Maximum memory usable is what is directly addressable */
 676         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
 677         if (max_pfn > MAX_NONPAE_PFN)
 678                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 679         else
 680                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 681         max_pfn = MAXMEM_PFN;
 682 #else /* !CONFIG_HIGHMEM */
 683 #ifndef CONFIG_HIGHMEM64G
 684         if (max_pfn > MAX_NONPAE_PFN) {
 685                 max_pfn = MAX_NONPAE_PFN;
 686                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
 687         }
 688 #endif /* !CONFIG_HIGHMEM64G */
 689 #endif /* !CONFIG_HIGHMEM */
 690 }
 691
 692 /*
 693  * Determine low and high memory ranges:
 694  */
 695 void __init find_low_pfn_range(void)
 696 {
 697         /* it could update max_pfn */
 698
 699         if (max_pfn <= MAXMEM_PFN)
 700                 lowmem_pfn_init();
 701         else
 702                 highmem_pfn_init();
 703 }
 704
 705 #ifndef CONFIG_NEED_MULTIPLE_NODES
 706 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn,
 707                                 int acpi, int k8)
 708 {
 709 #ifdef CONFIG_HIGHMEM
 710         highstart_pfn = highend_pfn = max_pfn;
 711         if (max_pfn > max_low_pfn)
 712                 highstart_pfn = max_low_pfn;
 713         e820_register_active_regions(0, 0, highend_pfn);
 714         sparse_memory_present_with_active_regions(0);
 715         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 716                 pages_to_mb(highend_pfn - highstart_pfn));
 717         num_physpages = highend_pfn;
 718         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 719 #else
 720         e820_register_active_regions(0, 0, max_low_pfn);
 721         sparse_memory_present_with_active_regions(0);
 722         num_physpages = max_low_pfn;
 723         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 724 #endif
 725 #ifdef CONFIG_FLATMEM
 726         max_mapnr = num_physpages;
 727 #endif
 728         __vmalloc_start_set = true;
 729
 730         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 731                         pages_to_mb(max_low_pfn));
 732
 733         setup_bootmem_allocator();
 734 }
 735 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 736
 737 static void __init zone_sizes_init(void)
 738 {
 739         unsigned long max_zone_pfns[MAX_NR_ZONES];
 740         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 741         max_zone_pfns[ZONE_DMA] =
 742                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 743         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 744 #ifdef CONFIG_HIGHMEM
 745         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 746 #endif
 747
 748         free_area_init_nodes(max_zone_pfns);
 749 }
 750
 751 static unsigned long __init setup_node_bootmem(int nodeid,
 752                                  unsigned long start_pfn,
 753                                  unsigned long end_pfn,
 754                                  unsigned long bootmap)
 755 {
 756         unsigned long bootmap_size;
 757
 758         /* don't touch min_low_pfn */
 759         bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
 760                                          bootmap >> PAGE_SHIFT,
 761                                          start_pfn, end_pfn);
 762         printk(KERN_INFO "  node %d low ram: %08lx - %08lx\n",
 763                 nodeid, start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
 764         printk(KERN_INFO "  node %d bootmap %08lx - %08lx\n",
 765                  nodeid, bootmap, bootmap + bootmap_size);
 766         free_bootmem_with_active_regions(nodeid, end_pfn);
 767         early_res_to_bootmem(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
 768
 769         return bootmap + bootmap_size;
 770 }
 771
 772 void __init setup_bootmem_allocator(void)
 773 {
 774         int nodeid;
 775         unsigned long bootmap_size, bootmap;
 776         /*
 777          * Initialize the boot-time allocator (with low memory only):
 778          */
 779         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
 780         bootmap = find_e820_area(0, max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
 781                                  PAGE_SIZE);
 782         if (bootmap == -1L)
 783                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 784         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
 785
 786         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 787                  max_pfn_mapped<<PAGE_SHIFT);
 788         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
 789
 790         for_each_online_node(nodeid) {
 791                  unsigned long start_pfn, end_pfn;
 792
 793 #ifdef CONFIG_NEED_MULTIPLE_NODES
 794                 start_pfn = node_start_pfn[nodeid];
 795                 end_pfn = node_end_pfn[nodeid];
 796                 if (start_pfn > max_low_pfn)
 797                         continue;
 798                 if (end_pfn > max_low_pfn)
 799                         end_pfn = max_low_pfn;
 800 #else
 801                 start_pfn = 0;
 802                 end_pfn = max_low_pfn;
 803 #endif
 804                 bootmap = setup_node_bootmem(nodeid, start_pfn, end_pfn,
 805                                                  bootmap);
 806         }
 807
 808         after_bootmem = 1;
 809 }
 810
 811 /*
 812  * paging_init() sets up the page tables - note that the first 8MB are
 813  * already mapped by head.S.
 814  *
 815  * This routines also unmaps the page at virtual kernel address 0, so
 816  * that we can trap those pesky NULL-reference errors in the kernel.
 817  */
 818 void __init paging_init(void)
 819 {
 820         pagetable_init();
 821
 822         __flush_tlb_all();
 823
 824         kmap_init();
 825
 826         /*
 827          * NOTE: at this point the bootmem allocator is fully available.
 828          */
 829         sparse_init();
 830         zone_sizes_init();
 831 }
 832
 833 /*
 834  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 835  * and also on some strange 486's. All 586+'s are OK. This used to involve
 836  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 837  * switch to using exceptions got rid of all that.
 838  */
 839 static void __init test_wp_bit(void)
 840 {
 841         printk(KERN_INFO
 842   "Checking if this processor honours the WP bit even in supervisor mode...");
 843
 844         /* Any page-aligned address will do, the test is non-destructive */
 845         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 846         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 847         clear_fixmap(FIX_WP_TEST);
 848
 849         if (!boot_cpu_data.wp_works_ok) {
 850                 printk(KERN_CONT "No.\n");
 851 #ifdef CONFIG_X86_WP_WORKS_OK
 852                 panic(
 853   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 854 #endif
 855         } else {
 856                 printk(KERN_CONT "Ok.\n");
 857         }
 858 }
 859
 860 void __init mem_init(void)
 861 {
 862         int codesize, reservedpages, datasize, initsize;
 863         int tmp;
 864
 865         pci_iommu_alloc();
 866
 867 #ifdef CONFIG_FLATMEM
 868         BUG_ON(!mem_map);
 869 #endif
 870         /* this will put all low memory onto the freelists */
 871         totalram_pages += free_all_bootmem();
 872
 873         reservedpages = 0;
 874         for (tmp = 0; tmp < max_low_pfn; tmp++)
 875                 /*
 876                  * Only count reserved RAM pages:
 877                  */
 878                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 879                         reservedpages++;
 880
 881         set_highmem_pages_init();
 882
 883         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 884         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 885         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 886
 887         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
 888                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
 889                 nr_free_pages() << (PAGE_SHIFT-10),
 890                 num_physpages << (PAGE_SHIFT-10),
 891                 codesize >> 10,
 892                 reservedpages << (PAGE_SHIFT-10),
 893                 datasize >> 10,
 894                 initsize >> 10,
 895                 totalhigh_pages << (PAGE_SHIFT-10));
 896
 897         printk(KERN_INFO "virtual kernel memory layout:\n"
 898                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 899 #ifdef CONFIG_HIGHMEM
 900                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 901 #endif
 902                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 903                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 904                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 905                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 906                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 907                 FIXADDR_START, FIXADDR_TOP,
 908                 (FIXADDR_TOP - FIXADDR_START) >> 10,
 909
 910 #ifdef CONFIG_HIGHMEM
 911                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 912                 (LAST_PKMAP*PAGE_SIZE) >> 10,
 913 #endif
 914
 915                 VMALLOC_START, VMALLOC_END,
 916                 (VMALLOC_END - VMALLOC_START) >> 20,
 917
 918                 (unsigned long)__va(0), (unsigned long)high_memory,
 919                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 920
 921                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
 922                 ((unsigned long)&__init_end -
 923                  (unsigned long)&__init_begin) >> 10,
 924
 925                 (unsigned long)&_etext, (unsigned long)&_edata,
 926                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 927
 928                 (unsigned long)&_text, (unsigned long)&_etext,
 929                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 930
 931         /*
 932          * Check boundaries twice: Some fundamental inconsistencies can
 933          * be detected at build time already.
 934          */
 935 #define __FIXADDR_TOP (-PAGE_SIZE)
 936 #ifdef CONFIG_HIGHMEM
 937         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
 938         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
 939 #endif
 940 #define high_memory (-128UL << 20)
 941         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
 942 #undef high_memory
 943 #undef __FIXADDR_TOP
 944
 945 #ifdef CONFIG_HIGHMEM
 946         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
 947         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
 948 #endif
 949         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
 950         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
 951
 952         if (boot_cpu_data.wp_works_ok < 0)
 953                 test_wp_bit();
 954
 955         save_pg_dir();
 956         zap_low_mappings(true);
 957 }
 958
 959 #ifdef CONFIG_MEMORY_HOTPLUG
 960 int arch_add_memory(int nid, u64 start, u64 size)
 961 {
 962         struct pglist_data *pgdata = NODE_DATA(nid);
 963         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 964         unsigned long start_pfn = start >> PAGE_SHIFT;
 965         unsigned long nr_pages = size >> PAGE_SHIFT;
 966
 967         return __add_pages(nid, zone, start_pfn, nr_pages);
 968 }
 969 #endif
 970
 971 /*
 972  * This function cannot be __init, since exceptions don't work in that
 973  * section.  Put this after the callers, so that it cannot be inlined.
 974  */
 975 static noinline int do_test_wp_bit(void)
 976 {
 977         char tmp_reg;
 978         int flag;
 979
 980         __asm__ __volatile__(
 981                 "       movb %0, %1     \n"
 982                 "1:     movb %1, %0     \n"
 983                 "       xorl %2, %2     \n"
 984                 "2:                     \n"
 985                 _ASM_EXTABLE(1b,2b)
 986                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 987                  "=q" (tmp_reg),
 988                  "=r" (flag)
 989                 :"2" (1)
 990                 :"memory");
 991
 992         return flag;
 993 }
 994
 995 #ifdef CONFIG_DEBUG_RODATA
 996 const int rodata_test_data = 0xC3;
 997 EXPORT_SYMBOL_GPL(rodata_test_data);
 998
 999 int kernel_set_to_readonly __read_mostly;
1000
1001 void set_kernel_text_rw(void)
1002 {
1003         unsigned long start = PFN_ALIGN(_text);
1004         unsigned long size = PFN_ALIGN(_etext) - start;
1005
1006         if (!kernel_set_to_readonly)
1007                 return;
1008
1009         pr_debug("Set kernel text: %lx - %lx for read write\n",
1010                  start, start+size);
1011
1012         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1013 }
1014
1015 void set_kernel_text_ro(void)
1016 {
1017         unsigned long start = PFN_ALIGN(_text);
1018         unsigned long size = PFN_ALIGN(_etext) - start;
1019
1020         if (!kernel_set_to_readonly)
1021                 return;
1022
1023         pr_debug("Set kernel text: %lx - %lx for read only\n",
1024                  start, start+size);
1025
1026         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1027 }
1028
1029 void mark_rodata_ro(void)
1030 {
1031         unsigned long start = PFN_ALIGN(_text);
1032         unsigned long size = PFN_ALIGN(_etext) - start;
1033
1034         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1035         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1036                 size >> 10);
1037
1038         kernel_set_to_readonly = 1;
1039
1040 #ifdef CONFIG_CPA_DEBUG
1041         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1042                 start, start+size);
1043         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1044
1045         printk(KERN_INFO "Testing CPA: write protecting again\n");
1046         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1047 #endif
1048
1049         start += size;
1050         size = (unsigned long)__end_rodata - start;
1051         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1052         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1053                 size >> 10);
1054         rodata_test();
1055
1056 #ifdef CONFIG_CPA_DEBUG
1057         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1058         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1059
1060         printk(KERN_INFO "Testing CPA: write protecting again\n");
1061         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1062 #endif
1063 }
1064 #endif
1065
1066 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1067                                    int flags)
1068 {
1069         return reserve_bootmem(phys, len, flags);
1070 }