arch/i386/mm/init.c

   1 /*
   2  *  linux/arch/i386/mm/init.c
   3  *
   4  *  Copyright (C) 1995  Linus Torvalds
   5  *
   6  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
   7  */
   8
   9 #include <linux/config.h>
  10 #include <linux/module.h>
  11 #include <linux/signal.h>
  12 #include <linux/sched.h>
  13 #include <linux/kernel.h>
  14 #include <linux/errno.h>
  15 #include <linux/string.h>
  16 #include <linux/types.h>
  17 #include <linux/ptrace.h>
  18 #include <linux/mman.h>
  19 #include <linux/mm.h>
  20 #include <linux/hugetlb.h>
  21 #include <linux/swap.h>
  22 #include <linux/smp.h>
  23 #include <linux/init.h>
  24 #include <linux/highmem.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/slab.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/efi.h>
  30 #include <linux/memory_hotplug.h>
  31 #include <linux/initrd.h>
  32
  33 #include <asm/processor.h>
  34 #include <asm/system.h>
  35 #include <asm/uaccess.h>
  36 #include <asm/pgtable.h>
  37 #include <asm/dma.h>
  38 #include <asm/fixmap.h>
  39 #include <asm/e820.h>
  40 #include <asm/apic.h>
  41 #include <asm/tlb.h>
  42 #include <asm/tlbflush.h>
  43 #include <asm/sections.h>
  44
  45 unsigned int __VMALLOC_RESERVE = 128 << 20;
  46
  47 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  48 unsigned long highstart_pfn, highend_pfn;
  49
  50 static int noinline do_test_wp_bit(void);
  51
  52 /*
  53  * Creates a middle page table and puts a pointer to it in the
  54  * given global directory entry. This only returns the gd entry
  55  * in non-PAE compilation mode, since the middle layer is folded.
  56  */
  57 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  58 {
  59         pud_t *pud;
  60         pmd_t *pmd_table;
  61
  62 #ifdef CONFIG_X86_PAE
  63         pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
  64         set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  65         pud = pud_offset(pgd, 0);
  66         if (pmd_table != pmd_offset(pud, 0))
  67                 BUG();
  68 #else
  69         pud = pud_offset(pgd, 0);
  70         pmd_table = pmd_offset(pud, 0);
  71 #endif
  72
  73         return pmd_table;
  74 }
  75
  76 /*
  77  * Create a page table and place a pointer to it in a middle page
  78  * directory entry.
  79  */
  80 static pte_t * __init one_page_table_init(pmd_t *pmd)
  81 {
  82         if (pmd_none(*pmd)) {
  83                 pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
  84                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
  85                 if (page_table != pte_offset_kernel(pmd, 0))
  86                         BUG();
  87
  88                 return page_table;
  89         }
  90
  91         return pte_offset_kernel(pmd, 0);
  92 }
  93
  94 /*
  95  * This function initializes a certain range of kernel virtual memory
  96  * with new bootmem page tables, everywhere page tables are missing in
  97  * the given range.
  98  */
  99
 100 /*
 101  * NOTE: The pagetables are allocated contiguous on the physical space
 102  * so we can cache the place of the first one and move around without
 103  * checking the pgd every time.
 104  */
 105 static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base)
 106 {
 107         pgd_t *pgd;
 108         pud_t *pud;
 109         pmd_t *pmd;
 110         int pgd_idx, pmd_idx;
 111         unsigned long vaddr;
 112
 113         vaddr = start;
 114         pgd_idx = pgd_index(vaddr);
 115         pmd_idx = pmd_index(vaddr);
 116         pgd = pgd_base + pgd_idx;
 117
 118         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 119                 if (pgd_none(*pgd))
 120                         one_md_table_init(pgd);
 121                 pud = pud_offset(pgd, vaddr);
 122                 pmd = pmd_offset(pud, vaddr);
 123                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
 124                         if (pmd_none(*pmd))
 125                                 one_page_table_init(pmd);
 126
 127                         vaddr += PMD_SIZE;
 128                 }
 129                 pmd_idx = 0;
 130         }
 131 }
 132
 133 static inline int is_kernel_text(unsigned long addr)
 134 {
 135         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 136                 return 1;
 137         return 0;
 138 }
 139
 140 /*
 141  * This maps the physical memory to kernel virtual address space, a total
 142  * of max_low_pfn pages, by creating page tables starting from address
 143  * PAGE_OFFSET.
 144  */
 145 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
 146 {
 147         unsigned long pfn;
 148         pgd_t *pgd;
 149         pmd_t *pmd;
 150         pte_t *pte;
 151         int pgd_idx, pmd_idx, pte_ofs;
 152
 153         pgd_idx = pgd_index(PAGE_OFFSET);
 154         pgd = pgd_base + pgd_idx;
 155         pfn = 0;
 156
 157         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 158                 pmd = one_md_table_init(pgd);
 159                 if (pfn >= max_low_pfn)
 160                         continue;
 161                 for (pmd_idx = 0; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
 162                         unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
 163
 164                         /* Map with big pages if possible, otherwise create normal page tables. */
 165                         if (cpu_has_pse) {
 166                                 unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
 167
 168                                 if (is_kernel_text(address) || is_kernel_text(address2))
 169                                         set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
 170                                 else
 171                                         set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
 172                                 pfn += PTRS_PER_PTE;
 173                         } else {
 174                                 pte = one_page_table_init(pmd);
 175
 176                                 for (pte_ofs = 0; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) {
 177                                                 if (is_kernel_text(address))
 178                                                         set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
 179                                                 else
 180                                                         set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
 181                                 }
 182                         }
 183                 }
 184         }
 185 }
 186
 187 static inline int page_kills_ppro(unsigned long pagenr)
 188 {
 189         if (pagenr >= 0x70000 && pagenr <= 0x7003F)
 190                 return 1;
 191         return 0;
 192 }
 193
 194 extern int is_available_memory(efi_memory_desc_t *);
 195
 196 int page_is_ram(unsigned long pagenr)
 197 {
 198         int i;
 199         unsigned long addr, end;
 200
 201         if (efi_enabled) {
 202                 efi_memory_desc_t *md;
 203                 void *p;
 204
 205                 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 206                         md = p;
 207                         if (!is_available_memory(md))
 208                                 continue;
 209                         addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
 210                         end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
 211
 212                         if ((pagenr >= addr) && (pagenr < end))
 213                                 return 1;
 214                 }
 215                 return 0;
 216         }
 217
 218         for (i = 0; i < e820.nr_map; i++) {
 219
 220                 if (e820.map[i].type != E820_RAM)       /* not usable memory */
 221                         continue;
 222                 /*
 223                  *      !!!FIXME!!! Some BIOSen report areas as RAM that
 224                  *      are not. Notably the 640->1Mb area. We need a sanity
 225                  *      check here.
 226                  */
 227                 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
 228                 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
 229                 if  ((pagenr >= addr) && (pagenr < end))
 230                         return 1;
 231         }
 232         return 0;
 233 }
 234
 235 #ifdef CONFIG_HIGHMEM
 236 pte_t *kmap_pte;
 237 pgprot_t kmap_prot;
 238
 239 #define kmap_get_fixmap_pte(vaddr)                                      \
 240         pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr))
 241
 242 static void __init kmap_init(void)
 243 {
 244         unsigned long kmap_vstart;
 245
 246         /* cache the first kmap pte */
 247         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 248         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 249
 250         kmap_prot = PAGE_KERNEL;
 251 }
 252
 253 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 254 {
 255         pgd_t *pgd;
 256         pud_t *pud;
 257         pmd_t *pmd;
 258         pte_t *pte;
 259         unsigned long vaddr;
 260
 261         vaddr = PKMAP_BASE;
 262         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 263
 264         pgd = swapper_pg_dir + pgd_index(vaddr);
 265         pud = pud_offset(pgd, vaddr);
 266         pmd = pmd_offset(pud, vaddr);
 267         pte = pte_offset_kernel(pmd, vaddr);
 268         pkmap_page_table = pte;
 269 }
 270
 271 static void __meminit free_new_highpage(struct page *page)
 272 {
 273         init_page_count(page);
 274         __free_page(page);
 275         totalhigh_pages++;
 276 }
 277
 278 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
 279 {
 280         if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
 281                 ClearPageReserved(page);
 282                 free_new_highpage(page);
 283         } else
 284                 SetPageReserved(page);
 285 }
 286
 287 static int add_one_highpage_hotplug(struct page *page, unsigned long pfn)
 288 {
 289         free_new_highpage(page);
 290         totalram_pages++;
 291 #ifdef CONFIG_FLATMEM
 292         max_mapnr = max(pfn, max_mapnr);
 293 #endif
 294         num_physpages++;
 295         return 0;
 296 }
 297
 298 /*
 299  * Not currently handling the NUMA case.
 300  * Assuming single node and all memory that
 301  * has been added dynamically that would be
 302  * onlined here is in HIGHMEM
 303  */
 304 void online_page(struct page *page)
 305 {
 306         ClearPageReserved(page);
 307         add_one_highpage_hotplug(page, page_to_pfn(page));
 308 }
 309
 310
 311 #ifdef CONFIG_NUMA
 312 extern void set_highmem_pages_init(int);
 313 #else
 314 static void __init set_highmem_pages_init(int bad_ppro)
 315 {
 316         int pfn;
 317         for (pfn = highstart_pfn; pfn < highend_pfn; pfn++)
 318                 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
 319         totalram_pages += totalhigh_pages;
 320 }
 321 #endif /* CONFIG_FLATMEM */
 322
 323 #else
 324 #define kmap_init() do { } while (0)
 325 #define permanent_kmaps_init(pgd_base) do { } while (0)
 326 #define set_highmem_pages_init(bad_ppro) do { } while (0)
 327 #endif /* CONFIG_HIGHMEM */
 328
 329 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
 330 EXPORT_SYMBOL(__PAGE_KERNEL);
 331 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
 332
 333 #ifdef CONFIG_NUMA
 334 extern void __init remap_numa_kva(void);
 335 #else
 336 #define remap_numa_kva() do {} while (0)
 337 #endif
 338
 339 static void __init pagetable_init (void)
 340 {
 341         unsigned long vaddr;
 342         pgd_t *pgd_base = swapper_pg_dir;
 343
 344 #ifdef CONFIG_X86_PAE
 345         int i;
 346         /* Init entries of the first-level page table to the zero page */
 347         for (i = 0; i < PTRS_PER_PGD; i++)
 348                 set_pgd(pgd_base + i, __pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
 349 #endif
 350
 351         /* Enable PSE if available */
 352         if (cpu_has_pse) {
 353                 set_in_cr4(X86_CR4_PSE);
 354         }
 355
 356         /* Enable PGE if available */
 357         if (cpu_has_pge) {
 358                 set_in_cr4(X86_CR4_PGE);
 359                 __PAGE_KERNEL |= _PAGE_GLOBAL;
 360                 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
 361         }
 362
 363         kernel_physical_mapping_init(pgd_base);
 364         remap_numa_kva();
 365
 366         /*
 367          * Fixed mappings, only the page table structure has to be
 368          * created - mappings will be set by set_fixmap():
 369          */
 370         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 371         page_table_range_init(vaddr, 0, pgd_base);
 372
 373         permanent_kmaps_init(pgd_base);
 374
 375 #ifdef CONFIG_X86_PAE
 376         /*
 377          * Add low memory identity-mappings - SMP needs it when
 378          * starting up on an AP from real-mode. In the non-PAE
 379          * case we already have these mappings through head.S.
 380          * All user-space mappings are explicitly cleared after
 381          * SMP startup.
 382          */
 383         set_pgd(&pgd_base[0], pgd_base[USER_PTRS_PER_PGD]);
 384 #endif
 385 }
 386
 387 #ifdef CONFIG_SOFTWARE_SUSPEND
 388 /*
 389  * Swap suspend & friends need this for resume because things like the intel-agp
 390  * driver might have split up a kernel 4MB mapping.
 391  */
 392 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
 393         __attribute__ ((aligned (PAGE_SIZE)));
 394
 395 static inline void save_pg_dir(void)
 396 {
 397         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 398 }
 399 #else
 400 static inline void save_pg_dir(void)
 401 {
 402 }
 403 #endif
 404
 405 void zap_low_mappings (void)
 406 {
 407         int i;
 408
 409         save_pg_dir();
 410
 411         /*
 412          * Zap initial low-memory mappings.
 413          *
 414          * Note that "pgd_clear()" doesn't do it for
 415          * us, because pgd_clear() is a no-op on i386.
 416          */
 417         for (i = 0; i < USER_PTRS_PER_PGD; i++)
 418 #ifdef CONFIG_X86_PAE
 419                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 420 #else
 421                 set_pgd(swapper_pg_dir+i, __pgd(0));
 422 #endif
 423         flush_tlb_all();
 424 }
 425
 426 static int disable_nx __initdata = 0;
 427 u64 __supported_pte_mask __read_mostly = ~_PAGE_NX;
 428
 429 /*
 430  * noexec = on|off
 431  *
 432  * Control non executable mappings.
 433  *
 434  * on      Enable
 435  * off     Disable
 436  */
 437 void __init noexec_setup(const char *str)
 438 {
 439         if (!strncmp(str, "on",2) && cpu_has_nx) {
 440                 __supported_pte_mask |= _PAGE_NX;
 441                 disable_nx = 0;
 442         } else if (!strncmp(str,"off",3)) {
 443                 disable_nx = 1;
 444                 __supported_pte_mask &= ~_PAGE_NX;
 445         }
 446 }
 447
 448 int nx_enabled = 0;
 449 #ifdef CONFIG_X86_PAE
 450
 451 static void __init set_nx(void)
 452 {
 453         unsigned int v[4], l, h;
 454
 455         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 456                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 457                 if ((v[3] & (1 << 20)) && !disable_nx) {
 458                         rdmsr(MSR_EFER, l, h);
 459                         l |= EFER_NX;
 460                         wrmsr(MSR_EFER, l, h);
 461                         nx_enabled = 1;
 462                         __supported_pte_mask |= _PAGE_NX;
 463                 }
 464         }
 465 }
 466
 467 /*
 468  * Enables/disables executability of a given kernel page and
 469  * returns the previous setting.
 470  */
 471 int __init set_kernel_exec(unsigned long vaddr, int enable)
 472 {
 473         pte_t *pte;
 474         int ret = 1;
 475
 476         if (!nx_enabled)
 477                 goto out;
 478
 479         pte = lookup_address(vaddr);
 480         BUG_ON(!pte);
 481
 482         if (!pte_exec_kernel(*pte))
 483                 ret = 0;
 484
 485         if (enable)
 486                 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
 487         else
 488                 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
 489         __flush_tlb_all();
 490 out:
 491         return ret;
 492 }
 493
 494 #endif
 495
 496 /*
 497  * paging_init() sets up the page tables - note that the first 8MB are
 498  * already mapped by head.S.
 499  *
 500  * This routines also unmaps the page at virtual kernel address 0, so
 501  * that we can trap those pesky NULL-reference errors in the kernel.
 502  */
 503 void __init paging_init(void)
 504 {
 505 #ifdef CONFIG_X86_PAE
 506         set_nx();
 507         if (nx_enabled)
 508                 printk("NX (Execute Disable) protection: active\n");
 509 #endif
 510
 511         pagetable_init();
 512
 513         load_cr3(swapper_pg_dir);
 514
 515 #ifdef CONFIG_X86_PAE
 516         /*
 517          * We will bail out later - printk doesn't work right now so
 518          * the user would just see a hanging kernel.
 519          */
 520         if (cpu_has_pae)
 521                 set_in_cr4(X86_CR4_PAE);
 522 #endif
 523         __flush_tlb_all();
 524
 525         kmap_init();
 526 }
 527
 528 /*
 529  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 530  * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
 531  * used to involve black magic jumps to work around some nasty CPU bugs,
 532  * but fortunately the switch to using exceptions got rid of all that.
 533  */
 534
 535 static void __init test_wp_bit(void)
 536 {
 537         printk("Checking if this processor honours the WP bit even in supervisor mode... ");
 538
 539         /* Any page-aligned address will do, the test is non-destructive */
 540         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 541         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 542         clear_fixmap(FIX_WP_TEST);
 543
 544         if (!boot_cpu_data.wp_works_ok) {
 545                 printk("No.\n");
 546 #ifdef CONFIG_X86_WP_WORKS_OK
 547                 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 548 #endif
 549         } else {
 550                 printk("Ok.\n");
 551         }
 552 }
 553
 554 static void __init set_max_mapnr_init(void)
 555 {
 556 #ifdef CONFIG_HIGHMEM
 557         num_physpages = highend_pfn;
 558 #else
 559         num_physpages = max_low_pfn;
 560 #endif
 561 #ifdef CONFIG_FLATMEM
 562         max_mapnr = num_physpages;
 563 #endif
 564 }
 565
 566 static struct kcore_list kcore_mem, kcore_vmalloc;
 567
 568 void __init mem_init(void)
 569 {
 570         extern int ppro_with_ram_bug(void);
 571         int codesize, reservedpages, datasize, initsize;
 572         int tmp;
 573         int bad_ppro;
 574
 575 #ifdef CONFIG_FLATMEM
 576         if (!mem_map)
 577                 BUG();
 578 #endif
 579
 580         bad_ppro = ppro_with_ram_bug();
 581
 582 #ifdef CONFIG_HIGHMEM
 583         /* check that fixmap and pkmap do not overlap */
 584         if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
 585                 printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n");
 586                 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
 587                                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START);
 588                 BUG();
 589         }
 590 #endif
 591
 592         set_max_mapnr_init();
 593
 594 #ifdef CONFIG_HIGHMEM
 595         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 596 #else
 597         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 598 #endif
 599
 600         /* this will put all low memory onto the freelists */
 601         totalram_pages += free_all_bootmem();
 602
 603         reservedpages = 0;
 604         for (tmp = 0; tmp < max_low_pfn; tmp++)
 605                 /*
 606                  * Only count reserved RAM pages
 607                  */
 608                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 609                         reservedpages++;
 610
 611         set_highmem_pages_init(bad_ppro);
 612
 613         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 614         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 615         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 616
 617         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 618         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 619                    VMALLOC_END-VMALLOC_START);
 620
 621         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
 622                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 623                 num_physpages << (PAGE_SHIFT-10),
 624                 codesize >> 10,
 625                 reservedpages << (PAGE_SHIFT-10),
 626                 datasize >> 10,
 627                 initsize >> 10,
 628                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 629                );
 630
 631 #ifdef CONFIG_X86_PAE
 632         if (!cpu_has_pae)
 633                 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
 634 #endif
 635         if (boot_cpu_data.wp_works_ok < 0)
 636                 test_wp_bit();
 637
 638         /*
 639          * Subtle. SMP is doing it's boot stuff late (because it has to
 640          * fork idle threads) - but it also needs low mappings for the
 641          * protected-mode entry to work. We zap these entries only after
 642          * the WP-bit has been tested.
 643          */
 644 #ifndef CONFIG_SMP
 645         zap_low_mappings();
 646 #endif
 647 }
 648
 649 /*
 650  * this is for the non-NUMA, single node SMP system case.
 651  * Specifically, in the case of x86, we will always add
 652  * memory to the highmem for now.
 653  */
 654 #ifdef CONFIG_MEMORY_HOTPLUG
 655 #ifndef CONFIG_NEED_MULTIPLE_NODES
 656 int add_memory(u64 start, u64 size)
 657 {
 658         struct pglist_data *pgdata = &contig_page_data;
 659         struct zone *zone = pgdata->node_zones + MAX_NR_ZONES-1;
 660         unsigned long start_pfn = start >> PAGE_SHIFT;
 661         unsigned long nr_pages = size >> PAGE_SHIFT;
 662
 663         return __add_pages(zone, start_pfn, nr_pages);
 664 }
 665
 666 int remove_memory(u64 start, u64 size)
 667 {
 668         return -EINVAL;
 669 }
 670 #endif
 671 #endif
 672
 673 kmem_cache_t *pgd_cache;
 674 kmem_cache_t *pmd_cache;
 675
 676 void __init pgtable_cache_init(void)
 677 {
 678         if (PTRS_PER_PMD > 1) {
 679                 pmd_cache = kmem_cache_create("pmd",
 680                                         PTRS_PER_PMD*sizeof(pmd_t),
 681                                         PTRS_PER_PMD*sizeof(pmd_t),
 682                                         0,
 683                                         pmd_ctor,
 684                                         NULL);
 685                 if (!pmd_cache)
 686                         panic("pgtable_cache_init(): cannot create pmd cache");
 687         }
 688         pgd_cache = kmem_cache_create("pgd",
 689                                 PTRS_PER_PGD*sizeof(pgd_t),
 690                                 PTRS_PER_PGD*sizeof(pgd_t),
 691                                 0,
 692                                 pgd_ctor,
 693                                 PTRS_PER_PMD == 1 ? pgd_dtor : NULL);
 694         if (!pgd_cache)
 695                 panic("pgtable_cache_init(): Cannot create pgd cache");
 696 }
 697
 698 /*
 699  * This function cannot be __init, since exceptions don't work in that
 700  * section.  Put this after the callers, so that it cannot be inlined.
 701  */
 702 static int noinline do_test_wp_bit(void)
 703 {
 704         char tmp_reg;
 705         int flag;
 706
 707         __asm__ __volatile__(
 708                 "       movb %0,%1      \n"
 709                 "1:     movb %1,%0      \n"
 710                 "       xorl %2,%2      \n"
 711                 "2:                     \n"
 712                 ".section __ex_table,\"a\"\n"
 713                 "       .align 4        \n"
 714                 "       .long 1b,2b     \n"
 715                 ".previous              \n"
 716                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 717                  "=q" (tmp_reg),
 718                  "=r" (flag)
 719                 :"2" (1)
 720                 :"memory");
 721
 722         return flag;
 723 }
 724
 725 #ifdef CONFIG_DEBUG_RODATA
 726
 727 extern char __start_rodata, __end_rodata;
 728 void mark_rodata_ro(void)
 729 {
 730         unsigned long addr = (unsigned long)&__start_rodata;
 731
 732         for (; addr < (unsigned long)&__end_rodata; addr += PAGE_SIZE)
 733                 change_page_attr(virt_to_page(addr), 1, PAGE_KERNEL_RO);
 734
 735         printk ("Write protecting the kernel read-only data: %luk\n",
 736                         (unsigned long)(&__end_rodata - &__start_rodata) >> 10);
 737
 738         /*
 739          * change_page_attr() requires a global_flush_tlb() call after it.
 740          * We do this after the printk so that if something went wrong in the
 741          * change, the printk gets out at least to give a better debug hint
 742          * of who is the culprit.
 743          */
 744         global_flush_tlb();
 745 }
 746 #endif
 747
 748 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 749 {
 750         unsigned long addr;
 751
 752         for (addr = begin; addr < end; addr += PAGE_SIZE) {
 753                 ClearPageReserved(virt_to_page(addr));
 754                 init_page_count(virt_to_page(addr));
 755                 memset((void *)addr, 0xcc, PAGE_SIZE);
 756                 free_page(addr);
 757                 totalram_pages++;
 758         }
 759         printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10);
 760 }
 761
 762 void free_initmem(void)
 763 {
 764         free_init_pages("unused kernel memory",
 765                         (unsigned long)(&__init_begin),
 766                         (unsigned long)(&__init_end));
 767 }
 768
 769 #ifdef CONFIG_BLK_DEV_INITRD
 770 void free_initrd_mem(unsigned long start, unsigned long end)
 771 {
 772         free_init_pages("initrd memory", start, end);
 773 }
 774 #endif
 775