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