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/uaccess.h>
  41 #include <asm/pgtable.h>
  42 #include <asm/dma.h>
  43 #include <asm/fixmap.h>
  44 #include <asm/e820.h>
  45 #include <asm/apic.h>
  46 #include <asm/bugs.h>
  47 #include <asm/tlb.h>
  48 #include <asm/tlbflush.h>
  49 #include <asm/olpc_ofw.h>
  50 #include <asm/pgalloc.h>
  51 #include <asm/sections.h>
  52 #include <asm/hypervisor.h>
  53 #include <asm/swiotlb.h>
  54 #include <asm/setup.h>
  55 #include <asm/cacheflush.h>
  56 #include <asm/page_types.h>
  57 #include <asm/init.h>
  58
  59 unsigned long highstart_pfn, highend_pfn;
  60
  61 static noinline int do_test_wp_bit(void);
  62
  63 bool __read_mostly __vmalloc_start_set = false;
  64
  65 static __init void *alloc_low_page(void)
  66 {
  67         unsigned long pfn = pgt_buf_end++;
  68         void *adr;
  69
  70         if (pfn >= pgt_buf_top)
  71                 panic("alloc_low_page: ran out of memory");
  72
  73         adr = __va(pfn * PAGE_SIZE);
  74         clear_page(adr);
  75         return adr;
  76 }
  77
  78 /*
  79  * Creates a middle page table and puts a pointer to it in the
  80  * given global directory entry. This only returns the gd entry
  81  * in non-PAE compilation mode, since the middle layer is folded.
  82  */
  83 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  84 {
  85         pud_t *pud;
  86         pmd_t *pmd_table;
  87
  88 #ifdef CONFIG_X86_PAE
  89         if (!(__pgd_val(*pgd) & _PAGE_PRESENT)) {
  90                 if (after_bootmem)
  91                         pmd_table = (pmd_t *)alloc_bootmem_pages(PAGE_SIZE);
  92                 else
  93                         pmd_table = (pmd_t *)alloc_low_page();
  94                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
  95                 make_lowmem_page_readonly(pmd_table, XENFEAT_writable_page_tables);
  96                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  97                 pud = pud_offset(pgd, 0);
  98                 BUG_ON(pmd_table != pmd_offset(pud, 0));
  99
 100                 return pmd_table;
 101         }
 102 #endif
 103         pud = pud_offset(pgd, 0);
 104         pmd_table = pmd_offset(pud, 0);
 105
 106         return pmd_table;
 107 }
 108
 109 /*
 110  * Create a page table and place a pointer to it in a middle page
 111  * directory entry:
 112  */
 113 static pte_t * __init one_page_table_init(pmd_t *pmd)
 114 {
 115 #if CONFIG_XEN_COMPAT <= 0x030002
 116         if (pmd_none(*pmd)) {
 117 #else
 118         if (!(__pmd_val(*pmd) & _PAGE_PRESENT)) {
 119 #endif
 120                 pte_t *page_table = NULL;
 121
 122                 if (after_bootmem) {
 123 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
 124                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
 125 #endif
 126                         if (!page_table)
 127                                 page_table =
 128                                 (pte_t *)alloc_bootmem_pages(PAGE_SIZE);
 129                 } else
 130                         page_table = (pte_t *)alloc_low_page();
 131
 132                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
 133                 make_lowmem_page_readonly(page_table,
 134                                           XENFEAT_writable_page_tables);
 135                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
 136                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
 137         }
 138
 139         return pte_offset_kernel(pmd, 0);
 140 }
 141
 142 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
 143 {
 144         int pgd_idx = pgd_index(vaddr);
 145         int pmd_idx = pmd_index(vaddr);
 146
 147         return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
 148 }
 149
 150 pte_t * __init populate_extra_pte(unsigned long vaddr)
 151 {
 152         int pte_idx = pte_index(vaddr);
 153         pmd_t *pmd;
 154
 155         pmd = populate_extra_pmd(vaddr);
 156         return one_page_table_init(pmd) + pte_idx;
 157 }
 158
 159 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
 160                                            unsigned long vaddr, pte_t *lastpte)
 161 {
 162 #ifdef CONFIG_HIGHMEM
 163         /*
 164          * Something (early fixmap) may already have put a pte
 165          * page here, which causes the page table allocation
 166          * to become nonlinear. Attempt to fix it, and if it
 167          * is still nonlinear then we have to bug.
 168          */
 169         int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
 170         int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
 171
 172         if (pmd_idx_kmap_begin != pmd_idx_kmap_end
 173             && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
 174             && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
 175             && ((__pa(pte) >> PAGE_SHIFT) < pgt_buf_start
 176                 || (__pa(pte) >> PAGE_SHIFT) >= pgt_buf_end)) {
 177                 pte_t *newpte;
 178                 int i;
 179
 180                 BUG_ON(after_bootmem);
 181                 newpte = alloc_low_page();
 182                 for (i = 0; i < PTRS_PER_PTE; i++)
 183                         set_pte(newpte + i, pte[i]);
 184
 185                 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
 186                 make_lowmem_page_readonly(newpte,
 187                                           XENFEAT_writable_page_tables);
 188                 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
 189                 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
 190                 __flush_tlb_all();
 191
 192                 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
 193                 make_lowmem_page_writable(pte,
 194                                           XENFEAT_writable_page_tables);
 195                 pte = newpte;
 196         }
 197         BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
 198                && vaddr > fix_to_virt(FIX_KMAP_END)
 199                && lastpte && lastpte + PTRS_PER_PTE != pte);
 200 #endif
 201         return pte;
 202 }
 203
 204 /*
 205  * This function initializes a certain range of kernel virtual memory
 206  * with new bootmem page tables, everywhere page tables are missing in
 207  * the given range.
 208  *
 209  * NOTE: The pagetables are allocated contiguous on the physical space
 210  * so we can cache the place of the first one and move around without
 211  * checking the pgd every time.
 212  */
 213 static void __init
 214 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
 215 {
 216         int pgd_idx, pmd_idx;
 217         unsigned long vaddr;
 218         pgd_t *pgd;
 219         pmd_t *pmd;
 220         pte_t *pte = NULL;
 221
 222         vaddr = start;
 223         pgd_idx = pgd_index(vaddr);
 224         pmd_idx = pmd_index(vaddr);
 225         pgd = pgd_base + pgd_idx;
 226
 227         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 228                 pmd = one_md_table_init(pgd);
 229                 pmd = pmd + pmd_index(vaddr);
 230                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
 231                                                         pmd++, pmd_idx++) {
 232                         if (vaddr >= hypervisor_virt_start)
 233                                 break;
 234                         pte = page_table_kmap_check(one_page_table_init(pmd),
 235                                                     pmd, vaddr, pte);
 236
 237                         vaddr += PMD_SIZE;
 238                 }
 239                 pmd_idx = 0;
 240         }
 241 }
 242
 243 static inline int is_kernel_text(unsigned long addr)
 244 {
 245         if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
 246                 return 1;
 247         return 0;
 248 }
 249
 250 /*
 251  * This maps the physical memory to kernel virtual address space, a total
 252  * of max_low_pfn pages, by creating page tables starting from address
 253  * PAGE_OFFSET:
 254  */
 255 unsigned long __init
 256 kernel_physical_mapping_init(unsigned long start,
 257                              unsigned long end,
 258                              unsigned long page_size_mask)
 259 {
 260         int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
 261         unsigned long last_map_addr = end;
 262         unsigned long start_pfn, end_pfn;
 263         pgd_t *pgd_base = swapper_pg_dir;
 264         int pgd_idx, pmd_idx, pte_ofs;
 265         unsigned long pfn;
 266         pgd_t *pgd;
 267         pmd_t *pmd;
 268         pte_t *pte;
 269         unsigned pages_2m, pages_4k;
 270         int mapping_iter;
 271
 272         start_pfn = start >> PAGE_SHIFT;
 273         end_pfn = end >> PAGE_SHIFT;
 274
 275         /*
 276          * First iteration will setup identity mapping using large/small pages
 277          * based on use_pse, with other attributes same as set by
 278          * the early code in head_32.S
 279          *
 280          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
 281          * as desired for the kernel identity mapping.
 282          *
 283          * This two pass mechanism conforms to the TLB app note which says:
 284          *
 285          *     "Software should not write to a paging-structure entry in a way
 286          *      that would change, for any linear address, both the page size
 287          *      and either the page frame or attributes."
 288          */
 289         mapping_iter = 1;
 290
 291         if (!cpu_has_pse) {
 292                 use_pse = 0;
 293                 mapping_iter = 0;
 294         }
 295
 296 repeat:
 297         pages_2m = pages_4k = 0;
 298         pfn = start_pfn;
 299         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 300         pgd = pgd_base + pgd_idx;
 301         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 302 #ifdef CONFIG_XEN
 303                 /*
 304                  * Native linux hasn't PAE-paging enabled yet at this
 305                  * point.  When running as xen domain we are in PAE
 306                  * mode already, thus we can't simply hook a empty
 307                  * pmd.  That would kill the mappings we are currently
 308                  * using ...
 309                  */
 310                 pmd = pmd_offset(pud_offset(pgd, PAGE_OFFSET), PAGE_OFFSET);
 311 #else
 312                 pmd = one_md_table_init(pgd);
 313 #endif
 314
 315                 if (pfn >= end_pfn)
 316                         continue;
 317 #ifdef CONFIG_X86_PAE
 318                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 319                 pmd += pmd_idx;
 320 #else
 321                 pmd_idx = 0;
 322 #endif
 323                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
 324                      pmd++, pmd_idx++) {
 325                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
 326
 327                         if (addr >= hypervisor_virt_start)
 328                                 continue;
 329
 330                         /*
 331                          * Map with big pages if possible, otherwise
 332                          * create normal page tables:
 333                          */
 334                         if (use_pse) {
 335                                 unsigned int addr2;
 336                                 pgprot_t prot = PAGE_KERNEL_LARGE;
 337                                 /*
 338                                  * first pass will use the same initial
 339                                  * identity mapping attribute + _PAGE_PSE.
 340                                  */
 341                                 pgprot_t init_prot =
 342                                         __pgprot(PTE_IDENT_ATTR |
 343                                                  _PAGE_PSE);
 344
 345                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
 346                                         PAGE_OFFSET + PAGE_SIZE-1;
 347
 348                                 if (is_kernel_text(addr) ||
 349                                     is_kernel_text(addr2))
 350                                         prot = PAGE_KERNEL_LARGE_EXEC;
 351
 352                                 pages_2m++;
 353                                 if (mapping_iter == 1)
 354                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
 355                                 else
 356                                         set_pmd(pmd, pfn_pmd(pfn, prot));
 357
 358                                 pfn += PTRS_PER_PTE;
 359                                 continue;
 360                         }
 361                         pte = one_page_table_init(pmd);
 362
 363                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 364                         pte += pte_ofs;
 365                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
 366                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
 367                                 pgprot_t prot = PAGE_KERNEL;
 368                                 /*
 369                                  * first pass will use the same initial
 370                                  * identity mapping attribute.
 371                                  */
 372                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
 373
 374                                 /* XEN: Only map initial RAM allocation. */
 375                                 if (pfn >= xen_start_info->nr_pages || pte_present(*pte))
 376                                         continue;
 377                                 if (is_kernel_text(addr))
 378                                         prot = PAGE_KERNEL_EXEC;
 379
 380                                 pages_4k++;
 381                                 if (mapping_iter == 1) {
 382                                         set_pte(pte, pfn_pte(pfn, init_prot));
 383                                         last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
 384                                 } else
 385                                         set_pte(pte, pfn_pte(pfn, prot));
 386                         }
 387                 }
 388         }
 389         if (mapping_iter <= 1) {
 390                 /*
 391                  * update direct mapping page count only in the first
 392                  * iteration.
 393                  */
 394                 update_page_count(PG_LEVEL_2M, pages_2m);
 395                 update_page_count(PG_LEVEL_4K, pages_4k);
 396         }
 397         if (mapping_iter == 1) {
 398                 /*
 399                  * local global flush tlb, which will flush the previous
 400                  * mappings present in both small and large page TLB's.
 401                  */
 402                 __flush_tlb_all();
 403
 404                 /*
 405                  * Second iteration will set the actual desired PTE attributes.
 406                  */
 407                 mapping_iter = 2;
 408                 goto repeat;
 409         }
 410         return last_map_addr;
 411 }
 412
 413 pte_t *kmap_pte;
 414 pgprot_t kmap_prot;
 415
 416 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 417 {
 418         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 419                         vaddr), vaddr), vaddr);
 420 }
 421
 422 static void __init kmap_init(void)
 423 {
 424         unsigned long kmap_vstart;
 425
 426         /*
 427          * Cache the first kmap pte:
 428          */
 429         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 430         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 431
 432         kmap_prot = PAGE_KERNEL;
 433 }
 434
 435 #ifdef CONFIG_HIGHMEM
 436 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 437 {
 438         unsigned long vaddr;
 439         pgd_t *pgd;
 440         pud_t *pud;
 441         pmd_t *pmd;
 442         pte_t *pte;
 443
 444         vaddr = PKMAP_BASE;
 445         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 446
 447         pgd = swapper_pg_dir + pgd_index(vaddr);
 448         pud = pud_offset(pgd, vaddr);
 449         pmd = pmd_offset(pud, vaddr);
 450         pte = pte_offset_kernel(pmd, vaddr);
 451         pkmap_page_table = pte;
 452 }
 453
 454 static void __init add_one_highpage_init(struct page *page)
 455 {
 456         ClearPageReserved(page);
 457         init_page_count(page);
 458         __free_page(page);
 459         totalhigh_pages++;
 460 }
 461
 462 void __init add_highpages_with_active_regions(int nid,
 463                          unsigned long start_pfn, unsigned long end_pfn)
 464 {
 465         phys_addr_t start, end;
 466         u64 i;
 467
 468         for_each_free_mem_range(i, nid, &start, &end, NULL) {
 469                 unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
 470                                             start_pfn, end_pfn);
 471                 unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
 472                                               start_pfn, end_pfn);
 473                 for ( ; pfn < e_pfn; pfn++)
 474                         if (pfn_valid(pfn))
 475                                 add_one_highpage_init(pfn_to_page(pfn));
 476         }
 477 }
 478 #else
 479 static inline void permanent_kmaps_init(pgd_t *pgd_base)
 480 {
 481 }
 482 #endif /* CONFIG_HIGHMEM */
 483
 484 pgd_t *swapper_pg_dir;
 485
 486 /*
 487  * Build a proper pagetable for the kernel mappings.  Up until this
 488  * point, we've been running on some set of pagetables constructed by
 489  * the boot process.
 490  *
 491  * If we're booting on native hardware, this will be a pagetable
 492  * constructed in arch/x86/kernel/head_32.S.  The root of the
 493  * pagetable will be swapper_pg_dir.
 494  *
 495  * If we're booting paravirtualized under a hypervisor, then there are
 496  * more options: we may already be running PAE, and the pagetable may
 497  * or may not be based in swapper_pg_dir.  In any case,
 498  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 499  * appropriately for the rest of the initialization to work.
 500  *
 501  * In general, pagetable_init() assumes that the pagetable may already
 502  * be partially populated, and so it avoids stomping on any existing
 503  * mappings.
 504  */
 505 void __init early_ioremap_page_table_range_init(void)
 506 {
 507         pgd_t *pgd_base = swapper_pg_dir;
 508         unsigned long vaddr, end;
 509
 510         /*
 511          * Fixed mappings, only the page table structure has to be
 512          * created - mappings will be set by set_fixmap():
 513          */
 514         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 515         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 516         page_table_range_init(vaddr, end, pgd_base);
 517         early_ioremap_reset();
 518 }
 519
 520 static void __init pagetable_init(void)
 521 {
 522         pgd_t *pgd_base = (pgd_t *)xen_start_info->pt_base;
 523
 524         permanent_kmaps_init(pgd_base);
 525 }
 526
 527 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL);
 528 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 529
 530 /* user-defined highmem size */
 531 static unsigned int highmem_pages = -1;
 532
 533 /*
 534  * highmem=size forces highmem to be exactly 'size' bytes.
 535  * This works even on boxes that have no highmem otherwise.
 536  * This also works to reduce highmem size on bigger boxes.
 537  */
 538 static int __init parse_highmem(char *arg)
 539 {
 540         if (!arg)
 541                 return -EINVAL;
 542
 543         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 544         return 0;
 545 }
 546 early_param("highmem", parse_highmem);
 547
 548 #define MSG_HIGHMEM_TOO_BIG \
 549         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
 550
 551 #define MSG_LOWMEM_TOO_SMALL \
 552         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
 553 /*
 554  * All of RAM fits into lowmem - but if user wants highmem
 555  * artificially via the highmem=x boot parameter then create
 556  * it:
 557  */
 558 void __init lowmem_pfn_init(void)
 559 {
 560         /* max_low_pfn is 0, we already have early_res support */
 561         max_low_pfn = max_pfn;
 562
 563         if (highmem_pages == -1)
 564                 highmem_pages = 0;
 565 #ifdef CONFIG_HIGHMEM
 566         if (highmem_pages >= max_pfn) {
 567                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
 568                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 569                 highmem_pages = 0;
 570         }
 571         if (highmem_pages) {
 572                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
 573                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
 574                                 pages_to_mb(highmem_pages));
 575                         highmem_pages = 0;
 576                 }
 577                 max_low_pfn -= highmem_pages;
 578         }
 579 #else
 580         if (highmem_pages)
 581                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 582 #endif
 583 }
 584
 585 #define MSG_HIGHMEM_TOO_SMALL \
 586         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
 587
 588 #define MSG_HIGHMEM_TRIMMED \
 589         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
 590 /*
 591  * We have more RAM than fits into lowmem - we try to put it into
 592  * highmem, also taking the highmem=x boot parameter into account:
 593  */
 594 void __init highmem_pfn_init(void)
 595 {
 596         max_low_pfn = MAXMEM_PFN;
 597
 598         if (highmem_pages == -1)
 599                 highmem_pages = max_pfn - MAXMEM_PFN;
 600
 601         if (highmem_pages + MAXMEM_PFN < max_pfn)
 602                 max_pfn = MAXMEM_PFN + highmem_pages;
 603
 604         if (highmem_pages + MAXMEM_PFN > max_pfn) {
 605                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
 606                         pages_to_mb(max_pfn - MAXMEM_PFN),
 607                         pages_to_mb(highmem_pages));
 608                 highmem_pages = 0;
 609         }
 610 #ifndef CONFIG_HIGHMEM
 611         /* Maximum memory usable is what is directly addressable */
 612         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
 613         if (max_pfn > MAX_NONPAE_PFN)
 614                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 615         else
 616                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 617         max_pfn = MAXMEM_PFN;
 618 #else /* !CONFIG_HIGHMEM */
 619 #ifndef CONFIG_HIGHMEM64G
 620         if (max_pfn > MAX_NONPAE_PFN) {
 621                 max_pfn = MAX_NONPAE_PFN;
 622                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
 623         }
 624 #endif /* !CONFIG_HIGHMEM64G */
 625 #endif /* !CONFIG_HIGHMEM */
 626 }
 627
 628 /*
 629  * Determine low and high memory ranges:
 630  */
 631 void __init find_low_pfn_range(void)
 632 {
 633         /* it could update max_pfn */
 634
 635         if (max_pfn <= MAXMEM_PFN)
 636                 lowmem_pfn_init();
 637         else
 638                 highmem_pfn_init();
 639 }
 640
 641 #ifndef CONFIG_NEED_MULTIPLE_NODES
 642 void __init initmem_init(void)
 643 {
 644 #ifdef CONFIG_HIGHMEM
 645         highstart_pfn = highend_pfn = max_pfn;
 646         if (max_pfn > max_low_pfn)
 647                 highstart_pfn = max_low_pfn;
 648         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 649                 pages_to_mb(highend_pfn - highstart_pfn));
 650         num_physpages = highend_pfn;
 651         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 652 #else
 653         num_physpages = max_low_pfn;
 654         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 655 #endif
 656
 657         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, 0);
 658         sparse_memory_present_with_active_regions(0);
 659
 660 #ifdef CONFIG_FLATMEM
 661         max_mapnr = num_physpages;
 662 #endif
 663         __vmalloc_start_set = true;
 664
 665         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 666                         pages_to_mb(max_low_pfn));
 667
 668         setup_bootmem_allocator();
 669 }
 670 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 671
 672 void __init setup_bootmem_allocator(void)
 673 {
 674         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 675                  max_pfn_mapped<<PAGE_SHIFT);
 676         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
 677
 678         after_bootmem = 1;
 679 }
 680
 681 unsigned long __init extend_init_mapping(unsigned long tables_space)
 682 {
 683         unsigned long start_pfn = (__pa(xen_start_info->pt_base) >> PAGE_SHIFT)
 684                                   + xen_start_info->nr_pt_frames;
 685         unsigned long start = start_pfn, va = (unsigned long)&_text;
 686         pgd_t *pgd;
 687         pud_t *pud;
 688         pmd_t *pmd;
 689         pte_t *pte;
 690
 691         /* Ensure init mappings cover kernel text/data and initial tables. */
 692         while (va < PAGE_OFFSET + (start_pfn << PAGE_SHIFT) + tables_space) {
 693                 pgd = pgd_offset_k(va);
 694                 pud = pud_offset(pgd, va);
 695                 pmd = pmd_offset(pud, va);
 696                 if (pmd_none(*pmd)) {
 697                         unsigned long pa = start_pfn++ << PAGE_SHIFT;
 698
 699                         clear_page(__va(pa));
 700                         make_lowmem_page_readonly(__va(pa),
 701                                                   XENFEAT_writable_page_tables);
 702                         xen_l2_entry_update(pmd, __pmd(pa | _KERNPG_TABLE));
 703                 }
 704                 pte = pte_offset_kernel(pmd, va);
 705                 if (pte_none(*pte)) {
 706                         pte_t new_pte = __pte(__pa(va) | _KERNPG_TABLE);
 707
 708                         if (HYPERVISOR_update_va_mapping(va, new_pte, 0))
 709                                 BUG();
 710                 }
 711                 va += PAGE_SIZE;
 712         }
 713
 714         /* Finally, blow away any spurious initial mappings. */
 715         while (1) {
 716                 pgd = pgd_offset_k(va);
 717                 pud = pud_offset(pgd, va);
 718                 pmd = pmd_offset(pud, va);
 719                 if (pmd_none(*pmd))
 720                         break;
 721                 if (HYPERVISOR_update_va_mapping(va, __pte_ma(0), 0))
 722                         BUG();
 723                 va += PAGE_SIZE;
 724         }
 725
 726         if (start_pfn > start)
 727                 memblock_reserve(PFN_PHYS(start), PFN_PHYS(start_pfn - start));
 728
 729         return start_pfn;
 730 }
 731
 732 /*
 733  * paging_init() sets up the page tables - note that the first 8MB are
 734  * already mapped by head.S.
 735  *
 736  * This routines also unmaps the page at virtual kernel address 0, so
 737  * that we can trap those pesky NULL-reference errors in the kernel.
 738  */
 739 void __init paging_init(void)
 740 {
 741         pagetable_init();
 742
 743         __flush_tlb_all();
 744
 745         kmap_init();
 746
 747         /*
 748          * NOTE: at this point the bootmem allocator is fully available.
 749          */
 750         olpc_dt_build_devicetree();
 751         sparse_memory_present_with_active_regions(MAX_NUMNODES);
 752         sparse_init();
 753         zone_sizes_init();
 754 }
 755
 756 /*
 757  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 758  * and also on some strange 486's. All 586+'s are OK. This used to involve
 759  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 760  * switch to using exceptions got rid of all that.
 761  */
 762 static void __init test_wp_bit(void)
 763 {
 764         printk(KERN_INFO
 765   "Checking if this processor honours the WP bit even in supervisor mode...");
 766
 767         /* Any page-aligned address will do, the test is non-destructive */
 768         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 769         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 770         clear_fixmap(FIX_WP_TEST);
 771
 772         if (!boot_cpu_data.wp_works_ok) {
 773                 printk(KERN_CONT "No.\n");
 774 #ifdef CONFIG_X86_WP_WORKS_OK
 775                 panic(
 776   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 777 #endif
 778         } else {
 779                 printk(KERN_CONT "Ok.\n");
 780         }
 781 }
 782
 783 void __init mem_init(void)
 784 {
 785         int codesize, reservedpages, datasize, initsize;
 786         int tmp;
 787         unsigned long pfn;
 788
 789         pci_iommu_alloc();
 790
 791 #ifdef CONFIG_FLATMEM
 792         BUG_ON(!mem_map);
 793 #endif
 794         /*
 795          * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
 796          * be done before free_all_bootmem(). Memblock use free low memory for
 797          * temporary data (see find_range_array()) and for this purpose can use
 798          * pages that was already passed to the buddy allocator, hence marked as
 799          * not accessible in the page tables when compiled with
 800          * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
 801          * important here.
 802          */
 803         set_highmem_pages_init();
 804
 805         /* this will put all low memory onto the freelists */
 806         totalram_pages += free_all_bootmem();
 807         /* XEN: init low-mem pages outside initial allocation. */
 808         for (pfn = xen_start_info->nr_pages; pfn < max_low_pfn; pfn++) {
 809                 ClearPageReserved(pfn_to_page(pfn));
 810                 init_page_count(pfn_to_page(pfn));
 811         }
 812
 813         reservedpages = 0;
 814         for (tmp = 0; tmp < max_low_pfn; tmp++)
 815                 /*
 816                  * Only count reserved RAM pages:
 817                  */
 818                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 819                         reservedpages++;
 820
 821         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 822         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 823         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 824
 825         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
 826                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
 827                 nr_free_pages() << (PAGE_SHIFT-10),
 828                 num_physpages << (PAGE_SHIFT-10),
 829                 codesize >> 10,
 830                 reservedpages << (PAGE_SHIFT-10),
 831                 datasize >> 10,
 832                 initsize >> 10,
 833                 totalhigh_pages << (PAGE_SHIFT-10));
 834
 835         printk(KERN_INFO "virtual kernel memory layout:\n"
 836                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 837 #ifdef CONFIG_HIGHMEM
 838                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 839 #endif
 840                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 841                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 842                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 843                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 844                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 845                 FIXADDR_START, FIXADDR_TOP,
 846                 (FIXADDR_TOP - FIXADDR_START) >> 10,
 847
 848 #ifdef CONFIG_HIGHMEM
 849                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 850                 (LAST_PKMAP*PAGE_SIZE) >> 10,
 851 #endif
 852
 853                 VMALLOC_START, VMALLOC_END,
 854                 (VMALLOC_END - VMALLOC_START) >> 20,
 855
 856                 (unsigned long)__va(0), (unsigned long)high_memory,
 857                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 858
 859                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
 860                 ((unsigned long)&__init_end -
 861                  (unsigned long)&__init_begin) >> 10,
 862
 863                 (unsigned long)&_etext, (unsigned long)&_edata,
 864                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 865
 866                 (unsigned long)&_text, (unsigned long)&_etext,
 867                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 868
 869         /*
 870          * Check boundaries twice: Some fundamental inconsistencies can
 871          * be detected at build time already.
 872          */
 873 #define __FIXADDR_TOP (-PAGE_SIZE)
 874 #ifdef CONFIG_HIGHMEM
 875         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
 876         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
 877 #endif
 878 #define high_memory (-128UL << 20)
 879         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
 880 #undef high_memory
 881 #undef __FIXADDR_TOP
 882
 883 #ifdef CONFIG_HIGHMEM
 884         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
 885         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
 886 #endif
 887         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
 888         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
 889
 890         if (boot_cpu_data.wp_works_ok < 0)
 891                 test_wp_bit();
 892 }
 893
 894 #ifdef CONFIG_MEMORY_HOTPLUG
 895 int arch_add_memory(int nid, u64 start, u64 size)
 896 {
 897         struct pglist_data *pgdata = NODE_DATA(nid);
 898         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 899         unsigned long start_pfn = start >> PAGE_SHIFT;
 900         unsigned long nr_pages = size >> PAGE_SHIFT;
 901
 902         return __add_pages(nid, zone, start_pfn, nr_pages);
 903 }
 904 #endif
 905
 906 /*
 907  * This function cannot be __init, since exceptions don't work in that
 908  * section.  Put this after the callers, so that it cannot be inlined.
 909  */
 910 static noinline int do_test_wp_bit(void)
 911 {
 912         char tmp_reg;
 913         int flag;
 914
 915         __asm__ __volatile__(
 916                 "       movb %0, %1     \n"
 917                 "1:     movb %1, %0     \n"
 918                 "       xorl %2, %2     \n"
 919                 "2:                     \n"
 920                 _ASM_EXTABLE(1b,2b)
 921                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 922                  "=q" (tmp_reg),
 923                  "=r" (flag)
 924                 :"2" (1)
 925                 :"memory");
 926
 927         return flag;
 928 }
 929
 930 #ifdef CONFIG_DEBUG_RODATA
 931 const int rodata_test_data = 0xC3;
 932 EXPORT_SYMBOL_GPL(rodata_test_data);
 933
 934 int kernel_set_to_readonly __read_mostly;
 935
 936 void set_kernel_text_rw(void)
 937 {
 938         unsigned long start = PFN_ALIGN(_text);
 939         unsigned long size = PFN_ALIGN(_etext) - start;
 940
 941         if (!kernel_set_to_readonly)
 942                 return;
 943
 944         pr_debug("Set kernel text: %lx - %lx for read write\n",
 945                  start, start+size);
 946
 947         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
 948 }
 949
 950 void set_kernel_text_ro(void)
 951 {
 952         unsigned long start = PFN_ALIGN(_text);
 953         unsigned long size = PFN_ALIGN(_etext) - start;
 954
 955         if (!kernel_set_to_readonly)
 956                 return;
 957
 958         pr_debug("Set kernel text: %lx - %lx for read only\n",
 959                  start, start+size);
 960
 961         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 962 }
 963
 964 static void mark_nxdata_nx(void)
 965 {
 966         /*
 967          * When this called, init has already been executed and released,
 968          * so everything past _etext should be NX.
 969          */
 970         unsigned long start = PFN_ALIGN(_etext);
 971         /*
 972          * This comes from is_kernel_text upper limit. Also HPAGE where used:
 973          */
 974         unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
 975
 976         if (__supported_pte_mask & _PAGE_NX)
 977                 printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
 978         set_pages_nx(virt_to_page(start), size >> PAGE_SHIFT);
 979 }
 980
 981 void mark_rodata_ro(void)
 982 {
 983         unsigned long start = PFN_ALIGN(_text);
 984         unsigned long size = PFN_ALIGN(_etext) - start;
 985
 986         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 987         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
 988                 size >> 10);
 989
 990         kernel_set_to_readonly = 1;
 991
 992 #ifdef CONFIG_CPA_DEBUG
 993         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
 994                 start, start+size);
 995         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
 996
 997         printk(KERN_INFO "Testing CPA: write protecting again\n");
 998         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
 999 #endif
1000
1001         start += size;
1002         size = (unsigned long)__end_rodata - start;
1003         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1004         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1005                 size >> 10);
1006         rodata_test();
1007
1008 #ifdef CONFIG_CPA_DEBUG
1009         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1010         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1011
1012         printk(KERN_INFO "Testing CPA: write protecting again\n");
1013         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1014 #endif
1015         mark_nxdata_nx();
1016 }
1017 #endif
1018