arch/x86/mm/init_64-xen.c

   1 /*
   2  *  linux/arch/x86_64/mm/init.c
   3  *
   4  *  Copyright (C) 1995  Linus Torvalds
   5  *  Copyright (C) 2000  Pavel Machek <pavel@ucw.cz>
   6  *  Copyright (C) 2002,2003 Andi Kleen <ak@suse.de>
   7  *
   8  *  Jun Nakajima <jun.nakajima@intel.com>
   9  *      Modified for Xen.
  10  */
  11
  12 #include <linux/signal.h>
  13 #include <linux/sched.h>
  14 #include <linux/kernel.h>
  15 #include <linux/errno.h>
  16 #include <linux/string.h>
  17 #include <linux/types.h>
  18 #include <linux/ptrace.h>
  19 #include <linux/mman.h>
  20 #include <linux/mm.h>
  21 #include <linux/swap.h>
  22 #include <linux/smp.h>
  23 #include <linux/init.h>
  24 #include <linux/initrd.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/memblock.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/pci.h>
  30 #include <linux/pfn.h>
  31 #include <linux/poison.h>
  32 #include <linux/dma-mapping.h>
  33 #include <linux/module.h>
  34 #include <linux/memory.h>
  35 #include <linux/memory_hotplug.h>
  36 #include <linux/nmi.h>
  37 #include <linux/gfp.h>
  38
  39 #include <asm/processor.h>
  40 #include <asm/bios_ebda.h>
  41 #include <asm/uaccess.h>
  42 #include <asm/pgtable.h>
  43 #include <asm/pgalloc.h>
  44 #include <asm/dma.h>
  45 #include <asm/fixmap.h>
  46 #include <asm/e820.h>
  47 #include <asm/apic.h>
  48 #include <asm/tlb.h>
  49 #include <asm/mmu_context.h>
  50 #include <asm/proto.h>
  51 #include <asm/smp.h>
  52 #include <asm/sections.h>
  53 #include <asm/kdebug.h>
  54 #include <asm/numa.h>
  55 #include <asm/cacheflush.h>
  56 #include <asm/init.h>
  57 #include <asm/setup.h>
  58
  59 #include <xen/features.h>
  60
  61 #if CONFIG_XEN_COMPAT <= 0x030002
  62 unsigned int __kernel_page_user;
  63 EXPORT_SYMBOL(__kernel_page_user);
  64 #endif
  65
  66 extern pmd_t level2_fixmap_pgt[PTRS_PER_PMD];
  67 extern pte_t level1_fixmap_pgt[PTRS_PER_PTE];
  68
  69 /*
  70  * Use this until direct mapping is established, i.e. before __va() is
  71  * available in init_memory_mapping().
  72  */
  73
  74 #define addr_to_page(addr, page)                                \
  75         (addr) &= PHYSICAL_PAGE_MASK;                           \
  76         (page) = ((unsigned long *) ((unsigned long)            \
  77         (((mfn_to_pfn((addr) >> PAGE_SHIFT)) << PAGE_SHIFT) +   \
  78         __START_KERNEL_map)))
  79
  80 pmd_t *__init early_get_pmd(unsigned long va)
  81 {
  82         unsigned long addr;
  83         unsigned long *page = (unsigned long *)init_level4_pgt;
  84
  85         addr = page[pgd_index(va)];
  86         addr_to_page(addr, page);
  87
  88         addr = page[pud_index(va)];
  89         addr_to_page(addr, page);
  90
  91         return (pmd_t *)&page[pmd_index(va)];
  92 }
  93
  94 void __meminit early_make_page_readonly(void *va, unsigned int feature)
  95 {
  96         unsigned long addr, _va = (unsigned long)va;
  97         pte_t pte, *ptep;
  98         unsigned long *page = (unsigned long *) init_level4_pgt;
  99
 100         BUG_ON(after_bootmem);
 101
 102         if (xen_feature(feature))
 103                 return;
 104
 105         addr = (unsigned long) page[pgd_index(_va)];
 106         addr_to_page(addr, page);
 107
 108         addr = page[pud_index(_va)];
 109         addr_to_page(addr, page);
 110
 111         addr = page[pmd_index(_va)];
 112         addr_to_page(addr, page);
 113
 114         ptep = (pte_t *) &page[pte_index(_va)];
 115
 116         pte.pte = ptep->pte & ~_PAGE_RW;
 117         if (HYPERVISOR_update_va_mapping(_va, pte, 0))
 118                 BUG();
 119 }
 120
 121 unsigned long __init early_arbitrary_virt_to_mfn(void *v)
 122 {
 123         unsigned long va = (unsigned long)v, addr, *page;
 124
 125         BUG_ON(va < __START_KERNEL_map);
 126
 127         page = (void *)(xen_read_cr3() + __START_KERNEL_map);
 128
 129         addr = page[pgd_index(va)];
 130         addr_to_page(addr, page);
 131
 132         addr = page[pud_index(va)];
 133         addr_to_page(addr, page);
 134
 135         addr = page[pmd_index(va)];
 136         addr_to_page(addr, page);
 137
 138         return (page[pte_index(va)] & PHYSICAL_PAGE_MASK) >> PAGE_SHIFT;
 139 }
 140
 141 #ifndef CONFIG_XEN
 142 static int __init parse_direct_gbpages_off(char *arg)
 143 {
 144         direct_gbpages = 0;
 145         return 0;
 146 }
 147 early_param("nogbpages", parse_direct_gbpages_off);
 148
 149 static int __init parse_direct_gbpages_on(char *arg)
 150 {
 151         direct_gbpages = 1;
 152         return 0;
 153 }
 154 early_param("gbpages", parse_direct_gbpages_on);
 155 #endif
 156
 157 /*
 158  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
 159  * physical space so we can cache the place of the first one and move
 160  * around without checking the pgd every time.
 161  */
 162
 163 pteval_t __supported_pte_mask __read_mostly = ~0UL;
 164 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 165
 166 int force_personality32;
 167
 168 /*
 169  * noexec32=on|off
 170  * Control non executable heap for 32bit processes.
 171  * To control the stack too use noexec=off
 172  *
 173  * on   PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
 174  * off  PROT_READ implies PROT_EXEC
 175  */
 176 static int __init nonx32_setup(char *str)
 177 {
 178         if (!strcmp(str, "on"))
 179                 force_personality32 &= ~READ_IMPLIES_EXEC;
 180         else if (!strcmp(str, "off"))
 181                 force_personality32 |= READ_IMPLIES_EXEC;
 182         return 1;
 183 }
 184 __setup("noexec32=", nonx32_setup);
 185
 186 /*
 187  * When memory was added/removed make sure all the processes MM have
 188  * suitable PGD entries in the local PGD level page.
 189  */
 190 void sync_global_pgds(unsigned long start, unsigned long end)
 191 {
 192         unsigned long address;
 193
 194         for (address = start; address <= end; address += PGDIR_SIZE) {
 195                 const pgd_t *pgd_ref = pgd_offset_k(address);
 196                 struct page *page;
 197
 198                 if (pgd_none(*pgd_ref))
 199                         continue;
 200
 201                 spin_lock(&pgd_lock);
 202                 list_for_each_entry(page, &pgd_list, lru) {
 203                         pgd_t *pgd;
 204                         spinlock_t *pgt_lock;
 205
 206                         pgd = (pgd_t *)page_address(page) + pgd_index(address);
 207                         /* the pgt_lock only for Xen */
 208                         pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
 209                         spin_lock(pgt_lock);
 210
 211                         if (pgd_none(*pgd))
 212                                 set_pgd(pgd, *pgd_ref);
 213                         else
 214                                 BUG_ON(pgd_page_vaddr(*pgd)
 215                                        != pgd_page_vaddr(*pgd_ref));
 216
 217                         spin_unlock(pgt_lock);
 218                 }
 219                 spin_unlock(&pgd_lock);
 220         }
 221 }
 222
 223 static struct reserved_pfn_range {
 224         unsigned long pfn, nr;
 225 } reserved_pfn_ranges[3] __meminitdata;
 226
 227 void __init reserve_pfn_range(unsigned long pfn, unsigned long nr)
 228 {
 229         unsigned int i;
 230
 231         for (i = 0; i < ARRAY_SIZE(reserved_pfn_ranges); ++i) {
 232                 struct reserved_pfn_range *range = reserved_pfn_ranges + i;
 233
 234                 if (!range->nr) {
 235                         range->pfn = pfn;
 236                         range->nr = nr;
 237                         break;
 238                 }
 239                 BUG_ON(range->pfn < pfn + nr && pfn < range->pfn + range->nr);
 240                 if (range->pfn > pfn) {
 241                         i = ARRAY_SIZE(reserved_pfn_ranges) - 1;
 242                         if (reserved_pfn_ranges[i].nr)
 243                                 continue;
 244                         for (; reserved_pfn_ranges + i > range; --i)
 245                                 reserved_pfn_ranges[i]
 246                                          = reserved_pfn_ranges[i - 1];
 247                         range->pfn = pfn;
 248                         range->nr = nr;
 249                         break;
 250                 }
 251         }
 252         BUG_ON(i >= ARRAY_SIZE(reserved_pfn_ranges));
 253         memblock_reserve(PFN_PHYS(pfn), PFN_PHYS(nr));
 254 }
 255
 256 void __init reserve_pgtable_low(void)
 257 {
 258         unsigned int i;
 259
 260         for (i = 0; i < ARRAY_SIZE(reserved_pfn_ranges); ++i) {
 261                 struct reserved_pfn_range *range = reserved_pfn_ranges + i;
 262
 263                 if (!range->nr)
 264                         break;
 265                 if (pgt_buf_start <= range->pfn && pgt_buf_top > range->pfn) {
 266                         x86_init.mapping.pagetable_reserve(PFN_PHYS(pgt_buf_start),
 267                                         PFN_PHYS(range->pfn));
 268                         pgt_buf_start = range->pfn + range->nr;
 269                 }
 270         }
 271 }
 272
 273 static __init unsigned long get_table_end(void)
 274 {
 275         unsigned int i;
 276
 277         BUG_ON(!pgt_buf_end);
 278         for (i = 0; i < ARRAY_SIZE(reserved_pfn_ranges); ++i) {
 279                 struct reserved_pfn_range *range = reserved_pfn_ranges + i;
 280
 281                 if (!range->nr)
 282                         break;
 283                 if (pgt_buf_end == range->pfn) {
 284                         pgt_buf_end += range->nr;
 285                         pgt_buf_top += range->nr;
 286                 }
 287         }
 288         return pgt_buf_end++;
 289 }
 290
 291 /*
 292  * NOTE: This function is marked __ref because it calls __init function
 293  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
 294  */
 295 static __ref void *spp_getpage(void)
 296 {
 297         void *ptr;
 298
 299         if (after_bootmem)
 300                 ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
 301         else if (pgt_buf_end < pgt_buf_top) {
 302                 ptr = __va(get_table_end() << PAGE_SHIFT);
 303                 clear_page(ptr);
 304         } else
 305                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 306
 307         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 308                 panic("set_pte_phys: cannot allocate page data %s\n",
 309                         after_bootmem ? "after bootmem" : "");
 310         }
 311
 312         pr_debug("spp_getpage %p\n", ptr);
 313
 314         return ptr;
 315 }
 316
 317 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
 318 {
 319         if (pgd_none(*pgd)) {
 320                 pud_t *pud = (pud_t *)spp_getpage();
 321                 if (!after_bootmem) {
 322                         make_page_readonly(pud, XENFEAT_writable_page_tables);
 323                         xen_l4_entry_update(pgd, __pgd(__pa(pud) | _PAGE_TABLE));
 324                 } else
 325                         pgd_populate(&init_mm, pgd, pud);
 326                 if (pud != pud_offset(pgd, 0))
 327                         printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
 328                                pud, pud_offset(pgd, 0));
 329         }
 330         return pud_offset(pgd, vaddr);
 331 }
 332
 333 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
 334 {
 335         if (pud_none(*pud)) {
 336                 pmd_t *pmd = (pmd_t *) spp_getpage();
 337                 if (!after_bootmem) {
 338                         make_page_readonly(pmd, XENFEAT_writable_page_tables);
 339                         xen_l3_entry_update(pud, __pud(__pa(pmd) | _PAGE_TABLE));
 340                 } else
 341                         pud_populate(&init_mm, pud, pmd);
 342                 if (pmd != pmd_offset(pud, 0))
 343                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
 344                                pmd, pmd_offset(pud, 0));
 345         }
 346         return pmd_offset(pud, vaddr);
 347 }
 348
 349 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
 350 {
 351         if (pmd_none(*pmd)) {
 352                 pte_t *pte = (pte_t *) spp_getpage();
 353                 make_page_readonly(pte, XENFEAT_writable_page_tables);
 354                 pmd_populate_kernel(&init_mm, pmd, pte);
 355                 if (pte != pte_offset_kernel(pmd, 0))
 356                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
 357         }
 358         return pte_offset_kernel(pmd, vaddr);
 359 }
 360
 361 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
 362 {
 363         pud_t *pud;
 364         pmd_t *pmd;
 365         pte_t *pte;
 366
 367         pud = pud_page + pud_index(vaddr);
 368         pmd = fill_pmd(pud, vaddr);
 369         pte = fill_pte(pmd, vaddr);
 370
 371         set_pte(pte, new_pte);
 372
 373         /*
 374          * It's enough to flush this one mapping.
 375          * (PGE mappings get flushed as well)
 376          */
 377         __flush_tlb_one(vaddr);
 378 }
 379
 380 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
 381 {
 382         pgd_t *pgd;
 383         pud_t *pud_page;
 384
 385         pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, __pte_val(pteval));
 386
 387         pgd = pgd_offset_k(vaddr);
 388         if (pgd_none(*pgd)) {
 389                 printk(KERN_ERR
 390                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
 391                 return;
 392         }
 393         pud_page = (pud_t*)pgd_page_vaddr(*pgd);
 394         set_pte_vaddr_pud(pud_page, vaddr, pteval);
 395 }
 396
 397 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
 398 {
 399         pgd_t *pgd;
 400         pud_t *pud;
 401
 402         pgd = pgd_offset_k(vaddr);
 403         pud = fill_pud(pgd, vaddr);
 404         return fill_pmd(pud, vaddr);
 405 }
 406
 407 pte_t * __init populate_extra_pte(unsigned long vaddr)
 408 {
 409         pmd_t *pmd;
 410
 411         pmd = populate_extra_pmd(vaddr);
 412         return fill_pte(pmd, vaddr);
 413 }
 414
 415 #ifndef CONFIG_XEN
 416 /*
 417  * Create large page table mappings for a range of physical addresses.
 418  */
 419 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
 420                                                 pgprot_t prot)
 421 {
 422         pgd_t *pgd;
 423         pud_t *pud;
 424         pmd_t *pmd;
 425
 426         BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
 427         for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
 428                 pgd = pgd_offset_k((unsigned long)__va(phys));
 429                 if (pgd_none(*pgd)) {
 430                         pud = (pud_t *) spp_getpage();
 431                         set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
 432                                                 _PAGE_USER));
 433                 }
 434                 pud = pud_offset(pgd, (unsigned long)__va(phys));
 435                 if (pud_none(*pud)) {
 436                         pmd = (pmd_t *) spp_getpage();
 437                         set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
 438                                                 _PAGE_USER));
 439                 }
 440                 pmd = pmd_offset(pud, phys);
 441                 BUG_ON(!pmd_none(*pmd));
 442                 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
 443         }
 444 }
 445
 446 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
 447 {
 448         __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
 449 }
 450
 451 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
 452 {
 453         __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
 454 }
 455
 456 /*
 457  * The head.S code sets up the kernel high mapping:
 458  *
 459  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
 460  *
 461  * phys_addr holds the negative offset to the kernel, which is added
 462  * to the compile time generated pmds. This results in invalid pmds up
 463  * to the point where we hit the physaddr 0 mapping.
 464  *
 465  * We limit the mappings to the region from _text to _brk_end.  _brk_end
 466  * is rounded up to the 2MB boundary. This catches the invalid pmds as
 467  * well, as they are located before _text:
 468  */
 469 void __init cleanup_highmap(void)
 470 {
 471         unsigned long vaddr = __START_KERNEL_map;
 472         unsigned long vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
 473         unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
 474         pmd_t *pmd = level2_kernel_pgt;
 475
 476         for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
 477                 if (pmd_none(*pmd))
 478                         continue;
 479                 if (vaddr < (unsigned long) _text || vaddr > end)
 480                         set_pmd(pmd, __pmd(0));
 481         }
 482 }
 483 #endif
 484
 485 static __ref void *alloc_low_page(unsigned long *phys)
 486 {
 487         unsigned long pfn;
 488         void *adr;
 489
 490         if (after_bootmem) {
 491                 adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
 492                 *phys = __pa(adr);
 493
 494                 return adr;
 495         }
 496
 497         pfn = get_table_end();
 498         if (pfn >= pgt_buf_top)
 499                 panic("alloc_low_page: ran out of memory");
 500
 501         adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
 502         clear_page(adr);
 503         *phys  = pfn * PAGE_SIZE;
 504         return adr;
 505 }
 506
 507 static __ref void *map_low_page(void *virt)
 508 {
 509         void *adr;
 510         unsigned long phys, left;
 511
 512         if (after_bootmem)
 513                 return virt;
 514
 515         phys = __pa(virt);
 516         left = phys & (PAGE_SIZE - 1);
 517         adr = early_memremap_ro(phys & PAGE_MASK, PAGE_SIZE);
 518         adr = (void *)(((unsigned long)adr) | left);
 519
 520         return adr;
 521 }
 522
 523 static __ref void unmap_low_page(void *adr)
 524 {
 525         if (after_bootmem)
 526                 return;
 527
 528         early_iounmap((void *)((unsigned long)adr & PAGE_MASK), PAGE_SIZE);
 529 }
 530
 531 static inline int __meminit make_readonly(unsigned long paddr)
 532 {
 533         int readonly = 0;
 534
 535         /* Make new page tables read-only on the first pass. */
 536         if (!xen_feature(XENFEAT_writable_page_tables)
 537             && !max_pfn_mapped
 538             && (paddr >= (pgt_buf_start << PAGE_SHIFT))) {
 539                 unsigned long top = pgt_buf_top;
 540                 unsigned int i;
 541
 542                 /* Account for the ranges get_table_end() skips. */
 543                 for (i = 0; i < ARRAY_SIZE(reserved_pfn_ranges); ++i) {
 544                         const struct reserved_pfn_range *range;
 545
 546                         range = reserved_pfn_ranges + i;
 547                         if (!range->nr)
 548                                 continue;
 549                         if (pgt_buf_end <= range->pfn && top > range->pfn) {
 550                                 if (paddr > (range->pfn << PAGE_SHIFT)
 551                                     && paddr < ((range->pfn + range->nr)
 552                                                 << PAGE_SHIFT))
 553                                         break;
 554                                 top += range->nr;
 555                         }
 556                 }
 557                 if (paddr < (top << PAGE_SHIFT))
 558                         readonly = (i >= ARRAY_SIZE(reserved_pfn_ranges));
 559         }
 560         /* Make old page tables read-only. */
 561         if (!xen_feature(XENFEAT_writable_page_tables)
 562             && (paddr >= (xen_start_info->pt_base - __START_KERNEL_map))
 563             && (paddr < (pgt_buf_end << PAGE_SHIFT)))
 564                 readonly = 1;
 565         /* Make P->M table (and its page tables) read-only. */
 566         if (!xen_feature(XENFEAT_writable_page_tables)
 567             && xen_start_info->mfn_list < __START_KERNEL_map
 568             && paddr >= (xen_start_info->first_p2m_pfn << PAGE_SHIFT)
 569             && paddr < (xen_start_info->first_p2m_pfn
 570                         + xen_start_info->nr_p2m_frames) << PAGE_SHIFT)
 571                 readonly = 1;
 572
 573         /*
 574          * No need for writable mapping of kernel image. This also ensures that
 575          * page and descriptor tables embedded inside don't have writable
 576          * mappings. The range must be in sync with that passed to
 577          * reserve_early() (as "TEXT DATA BSS"), since all other regions can be
 578          * allocated from under CONFIG_NO_BOOTMEM and thus must be writable.
 579          */
 580         if ((paddr >= __pa_symbol(&_text))
 581             && (paddr < (__pa_symbol(__bss_stop) & PAGE_MASK)))
 582                 readonly = 1;
 583
 584         return readonly;
 585 }
 586
 587 static unsigned long __meminit
 588 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
 589               pgprot_t prot)
 590 {
 591         unsigned pages = 0;
 592         unsigned long last_map_addr = end;
 593         int i;
 594
 595         pte_t *pte = pte_page + pte_index(addr);
 596
 597         for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
 598                 unsigned long pteval = addr | pgprot_val(prot);
 599
 600                 if (addr >= end ||
 601                     (!after_bootmem &&
 602                      (addr >> PAGE_SHIFT) >= xen_start_info->nr_pages))
 603                         break;
 604
 605                 /*
 606                  * We will re-use the existing mapping.
 607                  * Xen for example has some special requirements, like mapping
 608                  * pagetable pages as RO. So assume someone who pre-setup
 609                  * these mappings are more intelligent.
 610                  */
 611                 if (__pte_val(*pte)) {
 612                         pages++;
 613                         continue;
 614                 }
 615
 616                 if (make_readonly(addr))
 617                         pteval &= ~_PAGE_RW;
 618                 if (0)
 619                         printk("   pte=%p addr=%lx pte=%016lx\n",
 620                                pte, addr, pteval);
 621                 pages++;
 622                 if (!after_bootmem)
 623                         *pte = __pte(pteval & __supported_pte_mask);
 624                 else
 625                         set_pte(pte, __pte(pteval & __supported_pte_mask));
 626                 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
 627         }
 628
 629         update_page_count(PG_LEVEL_4K, pages);
 630
 631         return last_map_addr;
 632 }
 633
 634 static unsigned long __meminit
 635 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
 636               unsigned long page_size_mask, pgprot_t prot)
 637 {
 638         unsigned long pages = 0;
 639         unsigned long last_map_addr = end;
 640
 641         int i = pmd_index(address);
 642
 643         for (; i < PTRS_PER_PMD; i++, address = (address & PMD_MASK) + PMD_SIZE) {
 644                 unsigned long pte_phys;
 645                 pmd_t *pmd = pmd_page + pmd_index(address);
 646                 pte_t *pte;
 647                 pgprot_t new_prot = prot;
 648
 649                 if (address >= end)
 650                         break;
 651
 652                 if (__pmd_val(*pmd)) {
 653                         if (!pmd_large(*pmd)) {
 654                                 spin_lock(&init_mm.page_table_lock);
 655                                 pte = map_low_page((pte_t *)pmd_page_vaddr(*pmd));
 656                                 last_map_addr = phys_pte_init(pte, address,
 657                                                                 end, prot);
 658                                 unmap_low_page(pte);
 659                                 spin_unlock(&init_mm.page_table_lock);
 660                                 continue;
 661                         }
 662                         /*
 663                          * If we are ok with PG_LEVEL_2M mapping, then we will
 664                          * use the existing mapping,
 665                          *
 666                          * Otherwise, we will split the large page mapping but
 667                          * use the same existing protection bits except for
 668                          * large page, so that we don't violate Intel's TLB
 669                          * Application note (317080) which says, while changing
 670                          * the page sizes, new and old translations should
 671                          * not differ with respect to page frame and
 672                          * attributes.
 673                          */
 674                         if (page_size_mask & (1 << PG_LEVEL_2M)) {
 675                                 pages++;
 676                                 continue;
 677                         }
 678                         new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
 679                 }
 680
 681                 if (page_size_mask & (1<<PG_LEVEL_2M)) {
 682                         pages++;
 683                         spin_lock(&init_mm.page_table_lock);
 684                         set_pte((pte_t *)pmd,
 685                                 pfn_pte(address >> PAGE_SHIFT,
 686                                         __pgprot(pgprot_val(prot) | _PAGE_PSE)));
 687                         spin_unlock(&init_mm.page_table_lock);
 688                         last_map_addr = (address & PMD_MASK) + PMD_SIZE;
 689                         continue;
 690                 }
 691
 692                 pte = alloc_low_page(&pte_phys);
 693                 last_map_addr = phys_pte_init(pte, address, end, new_prot);
 694                 unmap_low_page(pte);
 695
 696                 if (!after_bootmem) {
 697                         if (max_pfn_mapped)
 698                                 make_page_readonly(__va(pte_phys),
 699                                                    XENFEAT_writable_page_tables);
 700                         if (page_size_mask & (1 << PG_LEVEL_NUM)) {
 701                                 mmu_update_t u;
 702
 703                                 u.ptr = arbitrary_virt_to_machine(pmd);
 704                                 u.val = phys_to_machine(pte_phys) | _PAGE_TABLE;
 705                                 if (HYPERVISOR_mmu_update(&u, 1, NULL,
 706                                                           DOMID_SELF) < 0)
 707                                         BUG();
 708                         } else
 709                                 *pmd = __pmd(pte_phys | _PAGE_TABLE);
 710                 } else {
 711                         spin_lock(&init_mm.page_table_lock);
 712                         pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
 713                         spin_unlock(&init_mm.page_table_lock);
 714                 }
 715         }
 716         update_page_count(PG_LEVEL_2M, pages);
 717         return last_map_addr;
 718 }
 719
 720 static unsigned long __meminit
 721 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
 722                          unsigned long page_size_mask)
 723 {
 724         unsigned long pages = 0;
 725         unsigned long last_map_addr = end;
 726         int i = pud_index(addr);
 727
 728         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
 729                 unsigned long pmd_phys;
 730                 pud_t *pud = pud_page + pud_index(addr);
 731                 pmd_t *pmd;
 732                 pgprot_t prot = PAGE_KERNEL;
 733
 734                 if (addr >= end)
 735                         break;
 736
 737                 if (__pud_val(*pud)) {
 738                         if (!pud_large(*pud)) {
 739                                 pmd = map_low_page(pmd_offset(pud, 0));
 740                                 last_map_addr = phys_pmd_init(pmd, addr, end,
 741                                         page_size_mask | (1 << PG_LEVEL_NUM),
 742                                         prot);
 743                                 unmap_low_page(pmd);
 744                                 __flush_tlb_all();
 745                                 continue;
 746                         }
 747                         /*
 748                          * If we are ok with PG_LEVEL_1G mapping, then we will
 749                          * use the existing mapping.
 750                          *
 751                          * Otherwise, we will split the gbpage mapping but use
 752                          * the same existing protection  bits except for large
 753                          * page, so that we don't violate Intel's TLB
 754                          * Application note (317080) which says, while changing
 755                          * the page sizes, new and old translations should
 756                          * not differ with respect to page frame and
 757                          * attributes.
 758                          */
 759                         if (page_size_mask & (1 << PG_LEVEL_1G)) {
 760                                 pages++;
 761                                 continue;
 762                         }
 763                         prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
 764                 }
 765
 766                 if (page_size_mask & (1<<PG_LEVEL_1G)) {
 767                         pages++;
 768                         spin_lock(&init_mm.page_table_lock);
 769                         set_pte((pte_t *)pud,
 770                                 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 771                         spin_unlock(&init_mm.page_table_lock);
 772                         last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
 773                         continue;
 774                 }
 775
 776                 pmd = alloc_low_page(&pmd_phys);
 777                 last_map_addr = phys_pmd_init(pmd, addr, end,
 778                                               page_size_mask & ~(1 << PG_LEVEL_NUM),
 779                                               prot);
 780                 unmap_low_page(pmd);
 781
 782                 if (!after_bootmem) {
 783                         if (max_pfn_mapped)
 784                                 make_page_readonly(__va(pmd_phys),
 785                                                    XENFEAT_writable_page_tables);
 786                         if (page_size_mask & (1 << PG_LEVEL_NUM)) {
 787                                 mmu_update_t u;
 788
 789                                 u.ptr = arbitrary_virt_to_machine(pud);
 790                                 u.val = phys_to_machine(pmd_phys) | _PAGE_TABLE;
 791                                 if (HYPERVISOR_mmu_update(&u, 1, NULL,
 792                                                           DOMID_SELF) < 0)
 793                                         BUG();
 794                         } else
 795                                 *pud = __pud(pmd_phys | _PAGE_TABLE);
 796                 } else {
 797                         spin_lock(&init_mm.page_table_lock);
 798                         pud_populate(&init_mm, pud, __va(pmd_phys));
 799                         spin_unlock(&init_mm.page_table_lock);
 800                 }
 801         }
 802         __flush_tlb_all();
 803
 804         update_page_count(PG_LEVEL_1G, pages);
 805
 806         return last_map_addr;
 807 }
 808
 809 void __init xen_init_pt(void)
 810 {
 811         unsigned long addr, *page;
 812
 813         /* Find the initial pte page that was built for us. */
 814         page = (unsigned long *)xen_start_info->pt_base;
 815         addr = page[pgd_index(__START_KERNEL_map)];
 816         addr_to_page(addr, page);
 817
 818 #if CONFIG_XEN_COMPAT <= 0x030002
 819         /* On Xen 3.0.2 and older we may need to explicitly specify _PAGE_USER
 820            in kernel PTEs. We check that here. */
 821         if (HYPERVISOR_xen_version(XENVER_version, NULL) <= 0x30000) {
 822                 unsigned long *pg;
 823                 pte_t pte;
 824
 825                 /* Mess with the initial mapping of page 0. It's not needed. */
 826                 BUILD_BUG_ON(__START_KERNEL <= __START_KERNEL_map);
 827                 addr = page[pud_index(__START_KERNEL_map)];
 828                 addr_to_page(addr, pg);
 829                 addr = pg[pmd_index(__START_KERNEL_map)];
 830                 addr_to_page(addr, pg);
 831                 pte.pte = pg[pte_index(__START_KERNEL_map)];
 832                 BUG_ON(!(pte.pte & _PAGE_PRESENT));
 833
 834                 /* If _PAGE_USER isn't set, we obviously do not need it. */
 835                 if (pte.pte & _PAGE_USER) {
 836                         /* _PAGE_USER is needed, but is it set implicitly? */
 837                         pte.pte &= ~_PAGE_USER;
 838                         if ((HYPERVISOR_update_va_mapping(__START_KERNEL_map,
 839                                                           pte, 0) != 0) ||
 840                             !(pg[pte_index(__START_KERNEL_map)] & _PAGE_USER))
 841                                 /* We need to explicitly specify _PAGE_USER. */
 842                                 __kernel_page_user = _PAGE_USER;
 843                 }
 844         }
 845 #endif
 846
 847         /* Construct mapping of initial pte page in our own directories. */
 848         init_level4_pgt[pgd_index(__START_KERNEL_map)] =
 849                 __pgd(__pa_symbol(level3_kernel_pgt) | _PAGE_TABLE);
 850         memcpy(level3_kernel_pgt + pud_index(__START_KERNEL_map),
 851                page + pud_index(__START_KERNEL_map),
 852                (PTRS_PER_PUD - pud_index(__START_KERNEL_map))
 853                * sizeof(*level3_kernel_pgt));
 854
 855         /* Copy the initial P->M table mappings if necessary. */
 856         addr = pgd_index(xen_start_info->mfn_list);
 857         if (addr < pgd_index(__START_KERNEL_map))
 858                 init_level4_pgt[addr] =
 859                         ((pgd_t *)xen_start_info->pt_base)[addr];
 860
 861         /* Do an early initialization of the fixmap area. */
 862         addr = __fix_to_virt(FIX_EARLYCON_MEM_BASE);
 863         if (pud_present(level3_kernel_pgt[pud_index(addr)])) {
 864                 unsigned long adr = page[pud_index(addr)];
 865
 866                 addr_to_page(adr, page);
 867                 copy_page(level2_fixmap_pgt, page);
 868         }
 869         level3_kernel_pgt[pud_index(addr)] =
 870                 __pud(__pa_symbol(level2_fixmap_pgt) | _PAGE_TABLE);
 871         level2_fixmap_pgt[pmd_index(addr)] =
 872                 __pmd(__pa_symbol(level1_fixmap_pgt) | _PAGE_TABLE);
 873
 874         early_make_page_readonly(init_level4_pgt,
 875                                  XENFEAT_writable_page_tables);
 876         early_make_page_readonly(level3_kernel_pgt,
 877                                  XENFEAT_writable_page_tables);
 878         early_make_page_readonly(level3_user_pgt,
 879                                  XENFEAT_writable_page_tables);
 880         early_make_page_readonly(level2_fixmap_pgt,
 881                                  XENFEAT_writable_page_tables);
 882         early_make_page_readonly(level1_fixmap_pgt,
 883                                  XENFEAT_writable_page_tables);
 884
 885         if (!xen_feature(XENFEAT_writable_page_tables))
 886                 xen_pgd_pin(init_level4_pgt);
 887 }
 888
 889 void __init xen_finish_init_mapping(void)
 890 {
 891         unsigned long start, end;
 892         struct mmuext_op mmuext;
 893
 894         /* Re-vector virtual addresses pointing into the initial
 895            mapping to the just-established permanent ones. */
 896         xen_start_info = __va(__pa(xen_start_info));
 897         xen_start_info->pt_base = (unsigned long)
 898                 __va(__pa(xen_start_info->pt_base));
 899         if (!xen_feature(XENFEAT_auto_translated_physmap)
 900             && xen_start_info->mfn_list >= __START_KERNEL_map)
 901                 phys_to_machine_mapping =
 902                         __va(__pa(xen_start_info->mfn_list));
 903
 904         /* Unpin the no longer used Xen provided page tables. */
 905         mmuext.cmd = MMUEXT_UNPIN_TABLE;
 906         mmuext.arg1.mfn = virt_to_mfn(xen_start_info->pt_base);
 907         if (HYPERVISOR_mmuext_op(&mmuext, 1, NULL, DOMID_SELF))
 908                 BUG();
 909
 910         /* Destroy the Xen-created mappings beyond the kernel image. */
 911         start = PAGE_ALIGN(_brk_end);
 912         end   = __START_KERNEL_map + (pgt_buf_start << PAGE_SHIFT);
 913         for (; start < end; start += PAGE_SIZE)
 914                 if (HYPERVISOR_update_va_mapping(start, __pte_ma(0), 0))
 915                         BUG();
 916
 917         WARN(pgt_buf_end != pgt_buf_top, "start=%lx cur=%lx top=%lx\n",
 918              pgt_buf_start, pgt_buf_end, pgt_buf_top);
 919         if (pgt_buf_end > pgt_buf_top)
 920                 pgt_buf_top = pgt_buf_end;
 921 }
 922
 923 unsigned long __meminit
 924 kernel_physical_mapping_init(unsigned long start,
 925                              unsigned long end,
 926                              unsigned long page_size_mask)
 927 {
 928         bool pgd_changed = false;
 929         unsigned long next, last_map_addr = end;
 930         unsigned long addr;
 931
 932         start = (unsigned long)__va(start);
 933         end = (unsigned long)__va(end);
 934         addr = start;
 935
 936         for (; start < end; start = next) {
 937                 pgd_t *pgd = pgd_offset_k(start);
 938                 unsigned long pud_phys;
 939                 pud_t *pud;
 940
 941                 next = (start + PGDIR_SIZE) & PGDIR_MASK;
 942                 if (next > end)
 943                         next = end;
 944
 945                 if (__pgd_val(*pgd)) {
 946                         pud = map_low_page((pud_t *)pgd_page_vaddr(*pgd));
 947                         last_map_addr = phys_pud_init(pud, __pa(start),
 948                                 __pa(end), page_size_mask | (1 << PG_LEVEL_NUM));
 949                         unmap_low_page(pud);
 950                         continue;
 951                 }
 952
 953                 pud = alloc_low_page(&pud_phys);
 954                 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
 955                                                  page_size_mask);
 956                 unmap_low_page(pud);
 957
 958                 if (!after_bootmem) {
 959                         if (max_pfn_mapped)
 960                                 make_page_readonly(__va(pud_phys),
 961                                                    XENFEAT_writable_page_tables);
 962                         xen_l4_entry_update(pgd, __pgd(pud_phys | _PAGE_TABLE));
 963                 } else {
 964                         spin_lock(&init_mm.page_table_lock);
 965                         pgd_populate(&init_mm, pgd, __va(pud_phys));
 966                         spin_unlock(&init_mm.page_table_lock);
 967                         pgd_changed = true;
 968                 }
 969         }
 970
 971         if (pgd_changed)
 972                 sync_global_pgds(addr, end);
 973
 974         return last_map_addr;
 975 }
 976
 977 #ifndef CONFIG_NUMA
 978 void __init initmem_init(void)
 979 {
 980         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, 0);
 981 }
 982 #endif
 983
 984 void __init paging_init(void)
 985 {
 986         sparse_memory_present_with_active_regions(MAX_NUMNODES);
 987         sparse_init();
 988
 989         /*
 990          * clear the default setting with node 0
 991          * note: don't use nodes_clear here, that is really clearing when
 992          *       numa support is not compiled in, and later node_set_state
 993          *       will not set it back.
 994          */
 995         node_clear_state(0, N_NORMAL_MEMORY);
 996
 997         zone_sizes_init();
 998 }
 999
1000 /*
1001  * Memory hotplug specific functions
1002  */
1003 #ifdef CONFIG_MEMORY_HOTPLUG
1004 /*
1005  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
1006  * updating.
1007  */
1008 static void  update_end_of_memory_vars(u64 start, u64 size)
1009 {
1010         unsigned long end_pfn = PFN_UP(start + size);
1011
1012         if (end_pfn > max_pfn) {
1013                 max_pfn = end_pfn;
1014                 max_low_pfn = end_pfn;
1015                 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1016         }
1017 }
1018
1019 /*
1020  * Memory is added always to NORMAL zone. This means you will never get
1021  * additional DMA/DMA32 memory.
1022  */
1023 int arch_add_memory(int nid, u64 start, u64 size)
1024 {
1025         struct pglist_data *pgdat = NODE_DATA(nid);
1026         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
1027         unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
1028         unsigned long nr_pages = size >> PAGE_SHIFT;
1029         int ret;
1030
1031         last_mapped_pfn = init_memory_mapping(start, start + size);
1032         if (last_mapped_pfn > max_pfn_mapped)
1033                 max_pfn_mapped = last_mapped_pfn;
1034
1035         ret = __add_pages(nid, zone, start_pfn, nr_pages);
1036         WARN_ON_ONCE(ret);
1037
1038         /* update max_pfn, max_low_pfn and high_memory */
1039         update_end_of_memory_vars(start, size);
1040
1041         return ret;
1042 }
1043 EXPORT_SYMBOL_GPL(arch_add_memory);
1044
1045 #endif /* CONFIG_MEMORY_HOTPLUG */
1046
1047 static struct kcore_list kcore_vsyscall;
1048
1049 void __init mem_init(void)
1050 {
1051         long codesize, reservedpages, datasize, initsize;
1052         unsigned long absent_pages;
1053         unsigned long pfn;
1054
1055         pci_iommu_alloc();
1056
1057         /* clear_bss() already clear the empty_zero_page */
1058
1059         reservedpages = 0;
1060
1061         /* this will put all low memory onto the freelists */
1062 #ifdef CONFIG_NUMA
1063         totalram_pages = numa_free_all_bootmem();
1064 #else
1065         totalram_pages = free_all_bootmem();
1066 #endif
1067
1068         /* XEN: init pages outside initial allocation. */
1069         for (pfn = xen_start_info->nr_pages; pfn < max_pfn; pfn++) {
1070                 ClearPageReserved(pfn_to_page(pfn));
1071                 init_page_count(pfn_to_page(pfn));
1072         }
1073
1074         absent_pages = absent_pages_in_range(0, max_pfn);
1075         reservedpages = max_pfn - totalram_pages - absent_pages;
1076         after_bootmem = 1;
1077
1078         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
1079         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
1080         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
1081
1082         /* Register memory areas for /proc/kcore */
1083         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
1084                          VSYSCALL_END - VSYSCALL_START, KCORE_OTHER);
1085
1086         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
1087                          "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
1088                 nr_free_pages() << (PAGE_SHIFT-10),
1089                 max_pfn << (PAGE_SHIFT-10),
1090                 codesize >> 10,
1091                 absent_pages << (PAGE_SHIFT-10),
1092                 reservedpages << (PAGE_SHIFT-10),
1093                 datasize >> 10,
1094                 initsize >> 10);
1095 }
1096
1097 #ifdef CONFIG_DEBUG_RODATA
1098 const int rodata_test_data = 0xC3;
1099 EXPORT_SYMBOL_GPL(rodata_test_data);
1100
1101 int kernel_set_to_readonly;
1102
1103 void set_kernel_text_rw(void)
1104 {
1105         unsigned long start = PFN_ALIGN(_text);
1106         unsigned long end = PFN_ALIGN(__stop___ex_table);
1107
1108         if (!kernel_set_to_readonly)
1109                 return;
1110
1111         pr_debug("Set kernel text: %lx - %lx for read write\n",
1112                  start, end);
1113
1114         /*
1115          * Make the kernel identity mapping for text RW. Kernel text
1116          * mapping will always be RO. Refer to the comment in
1117          * static_protections() in pageattr.c
1118          */
1119         set_memory_rw(start, (end - start) >> PAGE_SHIFT);
1120 }
1121
1122 void set_kernel_text_ro(void)
1123 {
1124         unsigned long start = PFN_ALIGN(_text);
1125         unsigned long end = PFN_ALIGN(__stop___ex_table);
1126
1127         if (!kernel_set_to_readonly)
1128                 return;
1129
1130         pr_debug("Set kernel text: %lx - %lx for read only\n",
1131                  start, end);
1132
1133         /*
1134          * Set the kernel identity mapping for text RO.
1135          */
1136         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1137 }
1138
1139 void mark_rodata_ro(void)
1140 {
1141         unsigned long start = PFN_ALIGN(_text);
1142         unsigned long rodata_start =
1143                 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
1144         unsigned long end = (unsigned long) &__end_rodata;
1145         unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table);
1146         unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata);
1147         unsigned long data_start = (unsigned long) &_sdata;
1148
1149         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1150                (end - start) >> 10);
1151         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1152
1153         kernel_set_to_readonly = 1;
1154
1155         /*
1156          * The rodata section (but not the kernel text!) should also be
1157          * not-executable.
1158          */
1159         set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
1160
1161         rodata_test();
1162
1163 #ifdef CONFIG_CPA_DEBUG
1164         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
1165         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
1166
1167         printk(KERN_INFO "Testing CPA: again\n");
1168         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
1169 #endif
1170
1171         free_init_pages("unused kernel memory",
1172                         (unsigned long) page_address(virt_to_page(text_end)),
1173                         (unsigned long)
1174                                  page_address(virt_to_page(rodata_start)));
1175         free_init_pages("unused kernel memory",
1176                         (unsigned long) page_address(virt_to_page(rodata_end)),
1177                         (unsigned long) page_address(virt_to_page(data_start)));
1178 }
1179
1180 #endif
1181
1182 int kern_addr_valid(unsigned long addr)
1183 {
1184         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
1185         pgd_t *pgd;
1186         pud_t *pud;
1187         pmd_t *pmd;
1188         pte_t *pte;
1189
1190         if (above != 0 && above != -1UL)
1191                 return 0;
1192
1193 #ifdef CONFIG_XEN
1194         /*
1195          * Don't walk page tables for hypervisor addresses, but allow
1196          * the M2P table to be accessed through e.g. /proc/kcore.
1197          */
1198         if (addr >= (unsigned long)machine_to_phys_mapping &&
1199             addr < (unsigned long)(machine_to_phys_mapping +
1200                                    machine_to_phys_nr))
1201                 return 1;
1202         if (addr >= HYPERVISOR_VIRT_START && addr < HYPERVISOR_VIRT_END)
1203                 return 0;
1204 #endif
1205
1206         pgd = pgd_offset_k(addr);
1207         if (pgd_none(*pgd))
1208                 return 0;
1209
1210         pud = pud_offset(pgd, addr);
1211         if (pud_none(*pud))
1212                 return 0;
1213
1214         pmd = pmd_offset(pud, addr);
1215         if (pmd_none(*pmd))
1216                 return 0;
1217
1218         if (pmd_large(*pmd))
1219                 return pfn_valid(pmd_pfn(*pmd));
1220
1221         pte = pte_offset_kernel(pmd, addr);
1222         if (pte_none(*pte))
1223                 return 0;
1224
1225         return pfn_valid(pte_pfn(*pte));
1226 }
1227
1228 /*
1229  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
1230  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
1231  * not need special handling anymore:
1232  */
1233 static struct vm_area_struct gate_vma = {
1234         .vm_start       = VSYSCALL_START,
1235         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
1236         .vm_page_prot   = PAGE_READONLY_EXEC,
1237         .vm_flags       = VM_READ | VM_EXEC
1238 };
1239
1240 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
1241 {
1242 #ifdef CONFIG_IA32_EMULATION
1243         if (!mm || mm->context.ia32_compat)
1244                 return NULL;
1245 #endif
1246         return &gate_vma;
1247 }
1248
1249 int in_gate_area(struct mm_struct *mm, unsigned long addr)
1250 {
1251         struct vm_area_struct *vma = get_gate_vma(mm);
1252
1253         if (!vma)
1254                 return 0;
1255
1256         return (addr >= vma->vm_start) && (addr < vma->vm_end);
1257 }
1258
1259 /*
1260  * Use this when you have no reliable mm, typically from interrupt
1261  * context. It is less reliable than using a task's mm and may give
1262  * false positives.
1263  */
1264 int in_gate_area_no_mm(unsigned long addr)
1265 {
1266         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
1267 }
1268
1269 const char *arch_vma_name(struct vm_area_struct *vma)
1270 {
1271         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
1272                 return "[vdso]";
1273         if (vma == &gate_vma)
1274                 return "[vsyscall]";
1275         return NULL;
1276 }
1277
1278 #ifdef CONFIG_X86_UV
1279 unsigned long memory_block_size_bytes(void)
1280 {
1281         if (is_uv_system()) {
1282                 printk(KERN_INFO "UV: memory block size 2GB\n");
1283                 return 2UL * 1024 * 1024 * 1024;
1284         }
1285         return MIN_MEMORY_BLOCK_SIZE;
1286 }
1287 #endif
1288
1289 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1290 /*
1291  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1292  */
1293 static long __meminitdata addr_start, addr_end;
1294 static void __meminitdata *p_start, *p_end;
1295 static int __meminitdata node_start;
1296
1297 int __meminit
1298 vmemmap_populate(struct page *start_page, unsigned long size, int node)
1299 {
1300         unsigned long addr = (unsigned long)start_page;
1301         unsigned long end = (unsigned long)(start_page + size);
1302         unsigned long next;
1303         pgd_t *pgd;
1304         pud_t *pud;
1305         pmd_t *pmd;
1306
1307         for (; addr < end; addr = next) {
1308                 void *p = NULL;
1309
1310                 pgd = vmemmap_pgd_populate(addr, node);
1311                 if (!pgd)
1312                         return -ENOMEM;
1313
1314                 pud = vmemmap_pud_populate(pgd, addr, node);
1315                 if (!pud)
1316                         return -ENOMEM;
1317
1318                 if (!cpu_has_pse) {
1319                         next = (addr + PAGE_SIZE) & PAGE_MASK;
1320                         pmd = vmemmap_pmd_populate(pud, addr, node);
1321
1322                         if (!pmd)
1323                                 return -ENOMEM;
1324
1325                         p = vmemmap_pte_populate(pmd, addr, node);
1326
1327                         if (!p)
1328                                 return -ENOMEM;
1329
1330                         addr_end = addr + PAGE_SIZE;
1331                         p_end = p + PAGE_SIZE;
1332                 } else {
1333                         next = pmd_addr_end(addr, end);
1334
1335                         pmd = pmd_offset(pud, addr);
1336                         if (pmd_none(*pmd)) {
1337                                 pte_t entry;
1338
1339                                 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
1340                                 if (!p)
1341                                         return -ENOMEM;
1342
1343                                 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1344                                                 PAGE_KERNEL_LARGE);
1345                                 set_pmd(pmd, __pmd_ma(__pte_val(entry)));
1346
1347                                 /* check to see if we have contiguous blocks */
1348                                 if (p_end != p || node_start != node) {
1349                                         if (p_start)
1350                                                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1351                                                        addr_start, addr_end-1, p_start, p_end-1, node_start);
1352                                         addr_start = addr;
1353                                         node_start = node;
1354                                         p_start = p;
1355                                 }
1356
1357                                 addr_end = addr + PMD_SIZE;
1358                                 p_end = p + PMD_SIZE;
1359                         } else
1360                                 vmemmap_verify((pte_t *)pmd, node, addr, next);
1361                 }
1362
1363         }
1364         sync_global_pgds((unsigned long)start_page, end);
1365         return 0;
1366 }
1367
1368 void __meminit vmemmap_populate_print_last(void)
1369 {
1370         if (p_start) {
1371                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1372                         addr_start, addr_end-1, p_start, p_end-1, node_start);
1373                 p_start = NULL;
1374                 p_end = NULL;
1375                 node_start = 0;
1376         }
1377 }
1378 #endif