arch/x86/mm/init_32-xen.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 #include <linux/dma-mapping.h>
  35 #include <linux/scatterlist.h>
  36
  37 #include <asm/asm.h>
  38 #include <asm/bios_ebda.h>
  39 #include <asm/processor.h>
  40 #include <asm/system.h>
  41 #include <asm/uaccess.h>
  42 #include <asm/pgtable.h>
  43 #include <asm/dma.h>
  44 #include <asm/fixmap.h>
  45 #include <asm/e820.h>
  46 #include <asm/apic.h>
  47 #include <asm/bugs.h>
  48 #include <asm/tlb.h>
  49 #include <asm/tlbflush.h>
  50 #include <asm/olpc_ofw.h>
  51 #include <asm/pgalloc.h>
  52 #include <asm/sections.h>
  53 #include <asm/hypervisor.h>
  54 #include <asm/swiotlb.h>
  55 #include <asm/setup.h>
  56 #include <asm/cacheflush.h>
  57 #include <asm/page_types.h>
  58 #include <asm/init.h>
  59
  60 unsigned long highstart_pfn, highend_pfn;
  61
  62 static noinline int do_test_wp_bit(void);
  63
  64 bool __read_mostly __vmalloc_start_set = false;
  65
  66 static __init void *alloc_low_page(void)
  67 {
  68         unsigned long pfn = pgt_buf_end++;
  69         void *adr;
  70
  71         if (pfn >= pgt_buf_top)
  72                 panic("alloc_low_page: ran out of memory");
  73
  74         adr = __va(pfn * PAGE_SIZE);
  75         clear_page(adr);
  76         return adr;
  77 }
  78
  79 /*
  80  * Creates a middle page table and puts a pointer to it in the
  81  * given global directory entry. This only returns the gd entry
  82  * in non-PAE compilation mode, since the middle layer is folded.
  83  */
  84 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  85 {
  86         pud_t *pud;
  87         pmd_t *pmd_table;
  88
  89 #ifdef CONFIG_X86_PAE
  90         if (!(__pgd_val(*pgd) & _PAGE_PRESENT)) {
  91                 if (after_bootmem)
  92                         pmd_table = (pmd_t *)alloc_bootmem_pages(PAGE_SIZE);
  93                 else
  94                         pmd_table = (pmd_t *)alloc_low_page();
  95                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
  96                 make_lowmem_page_readonly(pmd_table, XENFEAT_writable_page_tables);
  97                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  98                 pud = pud_offset(pgd, 0);
  99                 BUG_ON(pmd_table != pmd_offset(pud, 0));
 100
 101                 return pmd_table;
 102         }
 103 #endif
 104         pud = pud_offset(pgd, 0);
 105         pmd_table = pmd_offset(pud, 0);
 106
 107         return pmd_table;
 108 }
 109
 110 /*
 111  * Create a page table and place a pointer to it in a middle page
 112  * directory entry:
 113  */
 114 static pte_t * __init one_page_table_init(pmd_t *pmd)
 115 {
 116 #if CONFIG_XEN_COMPAT <= 0x030002
 117         if (pmd_none(*pmd)) {
 118 #else
 119         if (!(__pmd_val(*pmd) & _PAGE_PRESENT)) {
 120 #endif
 121                 pte_t *page_table = NULL;
 122
 123                 if (after_bootmem) {
 124 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
 125                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
 126 #endif
 127                         if (!page_table)
 128                                 page_table =
 129                                 (pte_t *)alloc_bootmem_pages(PAGE_SIZE);
 130                 } else
 131                         page_table = (pte_t *)alloc_low_page();
 132
 133                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
 134                 make_lowmem_page_readonly(page_table,
 135                                           XENFEAT_writable_page_tables);
 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                 make_lowmem_page_readonly(newpte,
 188                                           XENFEAT_writable_page_tables);
 189                 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
 190                 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
 191                 __flush_tlb_all();
 192
 193                 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
 194                 make_lowmem_page_writable(pte,
 195                                           XENFEAT_writable_page_tables);
 196                 pte = newpte;
 197         }
 198         BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
 199                && vaddr > fix_to_virt(FIX_KMAP_END)
 200                && lastpte && lastpte + PTRS_PER_PTE != pte);
 201 #endif
 202         return pte;
 203 }
 204
 205 /*
 206  * This function initializes a certain range of kernel virtual memory
 207  * with new bootmem page tables, everywhere page tables are missing in
 208  * the given range.
 209  *
 210  * NOTE: The pagetables are allocated contiguous on the physical space
 211  * so we can cache the place of the first one and move around without
 212  * checking the pgd every time.
 213  */
 214 static void __init
 215 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
 216 {
 217         int pgd_idx, pmd_idx;
 218         unsigned long vaddr;
 219         pgd_t *pgd;
 220         pmd_t *pmd;
 221         pte_t *pte = NULL;
 222
 223         vaddr = start;
 224         pgd_idx = pgd_index(vaddr);
 225         pmd_idx = pmd_index(vaddr);
 226         pgd = pgd_base + pgd_idx;
 227
 228         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 229                 pmd = one_md_table_init(pgd);
 230                 pmd = pmd + pmd_index(vaddr);
 231                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
 232                                                         pmd++, pmd_idx++) {
 233                         if (vaddr >= hypervisor_virt_start)
 234                                 break;
 235                         pte = page_table_kmap_check(one_page_table_init(pmd),
 236                                                     pmd, vaddr, pte);
 237
 238                         vaddr += PMD_SIZE;
 239                 }
 240                 pmd_idx = 0;
 241         }
 242 }
 243
 244 static inline int is_kernel_text(unsigned long addr)
 245 {
 246         if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
 247                 return 1;
 248         return 0;
 249 }
 250
 251 /*
 252  * This maps the physical memory to kernel virtual address space, a total
 253  * of max_low_pfn pages, by creating page tables starting from address
 254  * PAGE_OFFSET:
 255  */
 256 unsigned long __init
 257 kernel_physical_mapping_init(unsigned long start,
 258                              unsigned long end,
 259                              unsigned long page_size_mask)
 260 {
 261         int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
 262         unsigned long last_map_addr = end;
 263         unsigned long start_pfn, end_pfn;
 264         pgd_t *pgd_base = swapper_pg_dir;
 265         int pgd_idx, pmd_idx, pte_ofs;
 266         unsigned long pfn;
 267         pgd_t *pgd;
 268         pmd_t *pmd;
 269         pte_t *pte;
 270         unsigned pages_2m, pages_4k;
 271         int mapping_iter;
 272
 273         start_pfn = start >> PAGE_SHIFT;
 274         end_pfn = end >> PAGE_SHIFT;
 275
 276         /*
 277          * First iteration will setup identity mapping using large/small pages
 278          * based on use_pse, with other attributes same as set by
 279          * the early code in head_32.S
 280          *
 281          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
 282          * as desired for the kernel identity mapping.
 283          *
 284          * This two pass mechanism conforms to the TLB app note which says:
 285          *
 286          *     "Software should not write to a paging-structure entry in a way
 287          *      that would change, for any linear address, both the page size
 288          *      and either the page frame or attributes."
 289          */
 290         mapping_iter = 1;
 291
 292         if (!cpu_has_pse) {
 293                 use_pse = 0;
 294                 mapping_iter = 0;
 295         }
 296
 297 repeat:
 298         pages_2m = pages_4k = 0;
 299         pfn = start_pfn;
 300         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 301         pgd = pgd_base + pgd_idx;
 302         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 303 #ifdef CONFIG_XEN
 304                 /*
 305                  * Native linux hasn't PAE-paging enabled yet at this
 306                  * point.  When running as xen domain we are in PAE
 307                  * mode already, thus we can't simply hook a empty
 308                  * pmd.  That would kill the mappings we are currently
 309                  * using ...
 310                  */
 311                 pmd = pmd_offset(pud_offset(pgd, PAGE_OFFSET), PAGE_OFFSET);
 312 #else
 313                 pmd = one_md_table_init(pgd);
 314 #endif
 315
 316                 if (pfn >= end_pfn)
 317                         continue;
 318 #ifdef CONFIG_X86_PAE
 319                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 320                 pmd += pmd_idx;
 321 #else
 322                 pmd_idx = 0;
 323 #endif
 324                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
 325                      pmd++, pmd_idx++) {
 326                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
 327
 328                         if (addr >= hypervisor_virt_start)
 329                                 continue;
 330
 331                         /*
 332                          * Map with big pages if possible, otherwise
 333                          * create normal page tables:
 334                          */
 335                         if (use_pse) {
 336                                 unsigned int addr2;
 337                                 pgprot_t prot = PAGE_KERNEL_LARGE;
 338                                 /*
 339                                  * first pass will use the same initial
 340                                  * identity mapping attribute + _PAGE_PSE.
 341                                  */
 342                                 pgprot_t init_prot =
 343                                         __pgprot(PTE_IDENT_ATTR |
 344                                                  _PAGE_PSE);
 345
 346                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
 347                                         PAGE_OFFSET + PAGE_SIZE-1;
 348
 349                                 if (is_kernel_text(addr) ||
 350                                     is_kernel_text(addr2))
 351                                         prot = PAGE_KERNEL_LARGE_EXEC;
 352
 353                                 pages_2m++;
 354                                 if (mapping_iter == 1)
 355                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
 356                                 else
 357                                         set_pmd(pmd, pfn_pmd(pfn, prot));
 358
 359                                 pfn += PTRS_PER_PTE;
 360                                 continue;
 361                         }
 362                         pte = one_page_table_init(pmd);
 363
 364                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 365                         pte += pte_ofs;
 366                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
 367                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
 368                                 pgprot_t prot = PAGE_KERNEL;
 369                                 /*
 370                                  * first pass will use the same initial
 371                                  * identity mapping attribute.
 372                                  */
 373                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
 374
 375                                 /* XEN: Only map initial RAM allocation. */
 376                                 if (pfn >= xen_start_info->nr_pages || pte_present(*pte))
 377                                         continue;
 378                                 if (is_kernel_text(addr))
 379                                         prot = PAGE_KERNEL_EXEC;
 380
 381                                 pages_4k++;
 382                                 if (mapping_iter == 1) {
 383                                         set_pte(pte, pfn_pte(pfn, init_prot));
 384                                         last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
 385                                 } else
 386                                         set_pte(pte, pfn_pte(pfn, prot));
 387                         }
 388                 }
 389         }
 390         if (mapping_iter <= 1) {
 391                 /*
 392                  * update direct mapping page count only in the first
 393                  * iteration.
 394                  */
 395                 update_page_count(PG_LEVEL_2M, pages_2m);
 396                 update_page_count(PG_LEVEL_4K, pages_4k);
 397         }
 398         if (mapping_iter == 1) {
 399                 /*
 400                  * local global flush tlb, which will flush the previous
 401                  * mappings present in both small and large page TLB's.
 402                  */
 403                 __flush_tlb_all();
 404
 405                 /*
 406                  * Second iteration will set the actual desired PTE attributes.
 407                  */
 408                 mapping_iter = 2;
 409                 goto repeat;
 410         }
 411         return last_map_addr;
 412 }
 413
 414 pte_t *kmap_pte;
 415 pgprot_t kmap_prot;
 416
 417 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 418 {
 419         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 420                         vaddr), vaddr), vaddr);
 421 }
 422
 423 static void __init kmap_init(void)
 424 {
 425         unsigned long kmap_vstart;
 426
 427         /*
 428          * Cache the first kmap pte:
 429          */
 430         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 431         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 432
 433         kmap_prot = PAGE_KERNEL;
 434 }
 435
 436 #ifdef CONFIG_HIGHMEM
 437 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 438 {
 439         unsigned long vaddr;
 440         pgd_t *pgd;
 441         pud_t *pud;
 442         pmd_t *pmd;
 443         pte_t *pte;
 444
 445         vaddr = PKMAP_BASE;
 446         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 447
 448         pgd = swapper_pg_dir + pgd_index(vaddr);
 449         pud = pud_offset(pgd, vaddr);
 450         pmd = pmd_offset(pud, vaddr);
 451         pte = pte_offset_kernel(pmd, vaddr);
 452         pkmap_page_table = pte;
 453 }
 454
 455 static void __init add_one_highpage_init(struct page *page)
 456 {
 457         ClearPageReserved(page);
 458         init_page_count(page);
 459         __free_page(page);
 460         totalhigh_pages++;
 461 }
 462
 463 void __init add_highpages_with_active_regions(int nid,
 464                          unsigned long start_pfn, unsigned long end_pfn)
 465 {
 466         phys_addr_t start, end;
 467         u64 i;
 468
 469         for_each_free_mem_range(i, nid, &start, &end, NULL) {
 470                 unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
 471                                             start_pfn, end_pfn);
 472                 unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
 473                                               start_pfn, end_pfn);
 474                 for ( ; pfn < e_pfn; pfn++)
 475                         if (pfn_valid(pfn))
 476                                 add_one_highpage_init(pfn_to_page(pfn));
 477         }
 478 }
 479 #else
 480 static inline void permanent_kmaps_init(pgd_t *pgd_base)
 481 {
 482 }
 483 #endif /* CONFIG_HIGHMEM */
 484
 485 pgd_t *swapper_pg_dir;
 486
 487 /*
 488  * Build a proper pagetable for the kernel mappings.  Up until this
 489  * point, we've been running on some set of pagetables constructed by
 490  * the boot process.
 491  *
 492  * If we're booting on native hardware, this will be a pagetable
 493  * constructed in arch/x86/kernel/head_32.S.  The root of the
 494  * pagetable will be swapper_pg_dir.
 495  *
 496  * If we're booting paravirtualized under a hypervisor, then there are
 497  * more options: we may already be running PAE, and the pagetable may
 498  * or may not be based in swapper_pg_dir.  In any case,
 499  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 500  * appropriately for the rest of the initialization to work.
 501  *
 502  * In general, pagetable_init() assumes that the pagetable may already
 503  * be partially populated, and so it avoids stomping on any existing
 504  * mappings.
 505  */
 506 void __init early_ioremap_page_table_range_init(void)
 507 {
 508         pgd_t *pgd_base = swapper_pg_dir;
 509         unsigned long vaddr, end;
 510
 511         /*
 512          * Fixed mappings, only the page table structure has to be
 513          * created - mappings will be set by set_fixmap():
 514          */
 515         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 516         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 517         page_table_range_init(vaddr, end, pgd_base);
 518         early_ioremap_reset();
 519 }
 520
 521 static void __init pagetable_init(void)
 522 {
 523         pgd_t *pgd_base = (pgd_t *)xen_start_info->pt_base;
 524
 525         permanent_kmaps_init(pgd_base);
 526 }
 527
 528 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL);
 529 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 530
 531 /* user-defined highmem size */
 532 static unsigned int highmem_pages = -1;
 533
 534 /*
 535  * highmem=size forces highmem to be exactly 'size' bytes.
 536  * This works even on boxes that have no highmem otherwise.
 537  * This also works to reduce highmem size on bigger boxes.
 538  */
 539 static int __init parse_highmem(char *arg)
 540 {
 541         if (!arg)
 542                 return -EINVAL;
 543
 544         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 545         return 0;
 546 }
 547 early_param("highmem", parse_highmem);
 548
 549 #define MSG_HIGHMEM_TOO_BIG \
 550         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
 551
 552 #define MSG_LOWMEM_TOO_SMALL \
 553         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
 554 /*
 555  * All of RAM fits into lowmem - but if user wants highmem
 556  * artificially via the highmem=x boot parameter then create
 557  * it:
 558  */
 559 void __init lowmem_pfn_init(void)
 560 {
 561         /* max_low_pfn is 0, we already have early_res support */
 562         max_low_pfn = max_pfn;
 563
 564         if (highmem_pages == -1)
 565                 highmem_pages = 0;
 566 #ifdef CONFIG_HIGHMEM
 567         if (highmem_pages >= max_pfn) {
 568                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
 569                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 570                 highmem_pages = 0;
 571         }
 572         if (highmem_pages) {
 573                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
 574                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
 575                                 pages_to_mb(highmem_pages));
 576                         highmem_pages = 0;
 577                 }
 578                 max_low_pfn -= highmem_pages;
 579         }
 580 #else
 581         if (highmem_pages)
 582                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 583 #endif
 584 }
 585
 586 #define MSG_HIGHMEM_TOO_SMALL \
 587         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
 588
 589 #define MSG_HIGHMEM_TRIMMED \
 590         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
 591 /*
 592  * We have more RAM than fits into lowmem - we try to put it into
 593  * highmem, also taking the highmem=x boot parameter into account:
 594  */
 595 void __init highmem_pfn_init(void)
 596 {
 597         max_low_pfn = MAXMEM_PFN;
 598
 599         if (highmem_pages == -1)
 600                 highmem_pages = max_pfn - MAXMEM_PFN;
 601
 602         if (highmem_pages + MAXMEM_PFN < max_pfn)
 603                 max_pfn = MAXMEM_PFN + highmem_pages;
 604
 605         if (highmem_pages + MAXMEM_PFN > max_pfn) {
 606                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
 607                         pages_to_mb(max_pfn - MAXMEM_PFN),
 608                         pages_to_mb(highmem_pages));
 609                 highmem_pages = 0;
 610         }
 611 #ifndef CONFIG_HIGHMEM
 612         /* Maximum memory usable is what is directly addressable */
 613         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
 614         if (max_pfn > MAX_NONPAE_PFN)
 615                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 616         else
 617                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 618         max_pfn = MAXMEM_PFN;
 619 #else /* !CONFIG_HIGHMEM */
 620 #ifndef CONFIG_HIGHMEM64G
 621         if (max_pfn > MAX_NONPAE_PFN) {
 622                 max_pfn = MAX_NONPAE_PFN;
 623                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
 624         }
 625 #endif /* !CONFIG_HIGHMEM64G */
 626 #endif /* !CONFIG_HIGHMEM */
 627 }
 628
 629 /*
 630  * Determine low and high memory ranges:
 631  */
 632 void __init find_low_pfn_range(void)
 633 {
 634         /* it could update max_pfn */
 635
 636         if (max_pfn <= MAXMEM_PFN)
 637                 lowmem_pfn_init();
 638         else
 639                 highmem_pfn_init();
 640 }
 641
 642 #ifndef CONFIG_NEED_MULTIPLE_NODES
 643 void __init initmem_init(void)
 644 {
 645 #ifdef CONFIG_HIGHMEM
 646         highstart_pfn = highend_pfn = max_pfn;
 647         if (max_pfn > max_low_pfn)
 648                 highstart_pfn = max_low_pfn;
 649         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 650                 pages_to_mb(highend_pfn - highstart_pfn));
 651         num_physpages = highend_pfn;
 652         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 653 #else
 654         num_physpages = max_low_pfn;
 655         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 656 #endif
 657
 658         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, 0);
 659         sparse_memory_present_with_active_regions(0);
 660
 661 #ifdef CONFIG_FLATMEM
 662         max_mapnr = num_physpages;
 663 #endif
 664         __vmalloc_start_set = true;
 665
 666         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 667                         pages_to_mb(max_low_pfn));
 668
 669         setup_bootmem_allocator();
 670 }
 671 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 672
 673 void __init setup_bootmem_allocator(void)
 674 {
 675         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 676                  max_pfn_mapped<<PAGE_SHIFT);
 677         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
 678
 679         after_bootmem = 1;
 680 }
 681
 682 unsigned long __init extend_init_mapping(unsigned long tables_space)
 683 {
 684         unsigned long start_pfn = (__pa(xen_start_info->pt_base) >> PAGE_SHIFT)
 685                                   + xen_start_info->nr_pt_frames;
 686         unsigned long start = start_pfn, va = (unsigned long)&_text;
 687         pgd_t *pgd;
 688         pud_t *pud;
 689         pmd_t *pmd;
 690         pte_t *pte;
 691
 692         /* Ensure init mappings cover kernel text/data and initial tables. */
 693         while (va < PAGE_OFFSET + (start_pfn << PAGE_SHIFT) + tables_space) {
 694                 pgd = pgd_offset_k(va);
 695                 pud = pud_offset(pgd, va);
 696                 pmd = pmd_offset(pud, va);
 697                 if (pmd_none(*pmd)) {
 698                         unsigned long pa = start_pfn++ << PAGE_SHIFT;
 699
 700                         clear_page(__va(pa));
 701                         make_lowmem_page_readonly(__va(pa),
 702                                                   XENFEAT_writable_page_tables);
 703                         xen_l2_entry_update(pmd, __pmd(pa | _KERNPG_TABLE));
 704                 }
 705                 pte = pte_offset_kernel(pmd, va);
 706                 if (pte_none(*pte)) {
 707                         pte_t new_pte = __pte(__pa(va) | _KERNPG_TABLE);
 708
 709                         if (HYPERVISOR_update_va_mapping(va, new_pte, 0))
 710                                 BUG();
 711                 }
 712                 va += PAGE_SIZE;
 713         }
 714
 715         /* Finally, blow away any spurious initial mappings. */
 716         while (1) {
 717                 pgd = pgd_offset_k(va);
 718                 pud = pud_offset(pgd, va);
 719                 pmd = pmd_offset(pud, va);
 720                 if (pmd_none(*pmd))
 721                         break;
 722                 if (HYPERVISOR_update_va_mapping(va, __pte_ma(0), 0))
 723                         BUG();
 724                 va += PAGE_SIZE;
 725         }
 726
 727         if (start_pfn > start)
 728                 memblock_reserve(PFN_PHYS(start), PFN_PHYS(start_pfn - start));
 729
 730         return start_pfn;
 731 }
 732
 733 /*
 734  * paging_init() sets up the page tables - note that the first 8MB are
 735  * already mapped by head.S.
 736  *
 737  * This routines also unmaps the page at virtual kernel address 0, so
 738  * that we can trap those pesky NULL-reference errors in the kernel.
 739  */
 740 void __init paging_init(void)
 741 {
 742         pagetable_init();
 743
 744         __flush_tlb_all();
 745
 746         kmap_init();
 747
 748         /*
 749          * NOTE: at this point the bootmem allocator is fully available.
 750          */
 751         olpc_dt_build_devicetree();
 752         sparse_memory_present_with_active_regions(MAX_NUMNODES);
 753         sparse_init();
 754         zone_sizes_init();
 755 }
 756
 757 /*
 758  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 759  * and also on some strange 486's. All 586+'s are OK. This used to involve
 760  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 761  * switch to using exceptions got rid of all that.
 762  */
 763 static void __init test_wp_bit(void)
 764 {
 765         printk(KERN_INFO
 766   "Checking if this processor honours the WP bit even in supervisor mode...");
 767
 768         /* Any page-aligned address will do, the test is non-destructive */
 769         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 770         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 771         clear_fixmap(FIX_WP_TEST);
 772
 773         if (!boot_cpu_data.wp_works_ok) {
 774                 printk(KERN_CONT "No.\n");
 775 #ifdef CONFIG_X86_WP_WORKS_OK
 776                 panic(
 777   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 778 #endif
 779         } else {
 780                 printk(KERN_CONT "Ok.\n");
 781         }
 782 }
 783
 784 void __init mem_init(void)
 785 {
 786         int codesize, reservedpages, datasize, initsize;
 787         int tmp;
 788         unsigned long pfn;
 789
 790         pci_iommu_alloc();
 791
 792 #ifdef CONFIG_FLATMEM
 793         BUG_ON(!mem_map);
 794 #endif
 795         /*
 796          * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
 797          * be done before free_all_bootmem(). Memblock use free low memory for
 798          * temporary data (see find_range_array()) and for this purpose can use
 799          * pages that was already passed to the buddy allocator, hence marked as
 800          * not accessible in the page tables when compiled with
 801          * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
 802          * important here.
 803          */
 804         set_highmem_pages_init();
 805
 806         /* this will put all low memory onto the freelists */
 807         totalram_pages += free_all_bootmem();
 808         /* XEN: init low-mem pages outside initial allocation. */
 809         for (pfn = xen_start_info->nr_pages; pfn < max_low_pfn; pfn++) {
 810                 ClearPageReserved(pfn_to_page(pfn));
 811                 init_page_count(pfn_to_page(pfn));
 812         }
 813
 814         reservedpages = 0;
 815         for (tmp = 0; tmp < max_low_pfn; tmp++)
 816                 /*
 817                  * Only count reserved RAM pages:
 818                  */
 819                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 820                         reservedpages++;
 821
 822         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 823         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 824         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 825
 826         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
 827                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
 828                 nr_free_pages() << (PAGE_SHIFT-10),
 829                 num_physpages << (PAGE_SHIFT-10),
 830                 codesize >> 10,
 831                 reservedpages << (PAGE_SHIFT-10),
 832                 datasize >> 10,
 833                 initsize >> 10,
 834                 totalhigh_pages << (PAGE_SHIFT-10));
 835
 836         printk(KERN_INFO "virtual kernel memory layout:\n"
 837                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 838 #ifdef CONFIG_HIGHMEM
 839                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 840 #endif
 841                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 842                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 843                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 844                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 845                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 846                 FIXADDR_START, FIXADDR_TOP,
 847                 (FIXADDR_TOP - FIXADDR_START) >> 10,
 848
 849 #ifdef CONFIG_HIGHMEM
 850                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 851                 (LAST_PKMAP*PAGE_SIZE) >> 10,
 852 #endif
 853
 854                 VMALLOC_START, VMALLOC_END,
 855                 (VMALLOC_END - VMALLOC_START) >> 20,
 856
 857                 (unsigned long)__va(0), (unsigned long)high_memory,
 858                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 859
 860                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
 861                 ((unsigned long)&__init_end -
 862                  (unsigned long)&__init_begin) >> 10,
 863
 864                 (unsigned long)&_etext, (unsigned long)&_edata,
 865                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 866
 867                 (unsigned long)&_text, (unsigned long)&_etext,
 868                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 869
 870         /*
 871          * Check boundaries twice: Some fundamental inconsistencies can
 872          * be detected at build time already.
 873          */
 874 #define __FIXADDR_TOP (-PAGE_SIZE)
 875 #ifdef CONFIG_HIGHMEM
 876         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
 877         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
 878 #endif
 879 #define high_memory (-128UL << 20)
 880         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
 881 #undef high_memory
 882 #undef __FIXADDR_TOP
 883
 884 #ifdef CONFIG_HIGHMEM
 885         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
 886         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
 887 #endif
 888         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
 889         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
 890
 891         if (boot_cpu_data.wp_works_ok < 0)
 892                 test_wp_bit();
 893 }
 894
 895 #ifdef CONFIG_MEMORY_HOTPLUG
 896 int arch_add_memory(int nid, u64 start, u64 size)
 897 {
 898         struct pglist_data *pgdata = NODE_DATA(nid);
 899         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 900         unsigned long start_pfn = start >> PAGE_SHIFT;
 901         unsigned long nr_pages = size >> PAGE_SHIFT;
 902
 903         return __add_pages(nid, zone, start_pfn, nr_pages);
 904 }
 905 #endif
 906
 907 /*
 908  * This function cannot be __init, since exceptions don't work in that
 909  * section.  Put this after the callers, so that it cannot be inlined.
 910  */
 911 static noinline int do_test_wp_bit(void)
 912 {
 913         char tmp_reg;
 914         int flag;
 915
 916         __asm__ __volatile__(
 917                 "       movb %0, %1     \n"
 918                 "1:     movb %1, %0     \n"
 919                 "       xorl %2, %2     \n"
 920                 "2:                     \n"
 921                 _ASM_EXTABLE(1b,2b)
 922                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 923                  "=q" (tmp_reg),
 924                  "=r" (flag)
 925                 :"2" (1)
 926                 :"memory");
 927
 928         return flag;
 929 }
 930
 931 #ifdef CONFIG_DEBUG_RODATA
 932 const int rodata_test_data = 0xC3;
 933 EXPORT_SYMBOL_GPL(rodata_test_data);
 934
 935 int kernel_set_to_readonly __read_mostly;
 936
 937 void set_kernel_text_rw(void)
 938 {
 939         unsigned long start = PFN_ALIGN(_text);
 940         unsigned long size = PFN_ALIGN(_etext) - start;
 941
 942         if (!kernel_set_to_readonly)
 943                 return;
 944
 945         pr_debug("Set kernel text: %lx - %lx for read write\n",
 946                  start, start+size);
 947
 948         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
 949 }
 950
 951 void set_kernel_text_ro(void)
 952 {
 953         unsigned long start = PFN_ALIGN(_text);
 954         unsigned long size = PFN_ALIGN(_etext) - start;
 955
 956         if (!kernel_set_to_readonly)
 957                 return;
 958
 959         pr_debug("Set kernel text: %lx - %lx for read only\n",
 960                  start, start+size);
 961
 962         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 963 }
 964
 965 static void mark_nxdata_nx(void)
 966 {
 967         /*
 968          * When this called, init has already been executed and released,
 969          * so everything past _etext should be NX.
 970          */
 971         unsigned long start = PFN_ALIGN(_etext);
 972         /*
 973          * This comes from is_kernel_text upper limit. Also HPAGE where used:
 974          */
 975         unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
 976
 977         if (__supported_pte_mask & _PAGE_NX)
 978                 printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
 979         set_pages_nx(virt_to_page(start), size >> PAGE_SHIFT);
 980 }
 981
 982 void mark_rodata_ro(void)
 983 {
 984         unsigned long start = PFN_ALIGN(_text);
 985         unsigned long size = PFN_ALIGN(_etext) - start;
 986
 987         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 988         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
 989                 size >> 10);
 990
 991         kernel_set_to_readonly = 1;
 992
 993 #ifdef CONFIG_CPA_DEBUG
 994         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
 995                 start, start+size);
 996         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
 997
 998         printk(KERN_INFO "Testing CPA: write protecting again\n");
 999         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1000 #endif
1001
1002         start += size;
1003         size = (unsigned long)__end_rodata - start;
1004         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1005         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1006                 size >> 10);
1007         rodata_test();
1008
1009 #ifdef CONFIG_CPA_DEBUG
1010         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1011         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1012
1013         printk(KERN_INFO "Testing CPA: write protecting again\n");
1014         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1015 #endif
1016         mark_nxdata_nx();
1017 }
1018 #endif
1019