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