arch/x86/mm/init_64.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
   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/swap.h>
  19 #include <linux/smp.h>
  20 #include <linux/init.h>
  21 #include <linux/initrd.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/bootmem.h>
  24 #include <linux/memblock.h>
  25 #include <linux/proc_fs.h>
  26 #include <linux/pci.h>
  27 #include <linux/pfn.h>
  28 #include <linux/poison.h>
  29 #include <linux/dma-mapping.h>
  30 #include <linux/module.h>
  31 #include <linux/memory_hotplug.h>
  32 #include <linux/nmi.h>
  33 #include <linux/gfp.h>
  34
  35 #include <asm/processor.h>
  36 #include <asm/bios_ebda.h>
  37 #include <asm/system.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/pgtable.h>
  40 #include <asm/pgalloc.h>
  41 #include <asm/dma.h>
  42 #include <asm/fixmap.h>
  43 #include <asm/e820.h>
  44 #include <asm/apic.h>
  45 #include <asm/tlb.h>
  46 #include <asm/mmu_context.h>
  47 #include <asm/proto.h>
  48 #include <asm/smp.h>
  49 #include <asm/sections.h>
  50 #include <asm/kdebug.h>
  51 #include <asm/numa.h>
  52 #include <asm/cacheflush.h>
  53 #include <asm/init.h>
  54
  55 static int __init parse_direct_gbpages_off(char *arg)
  56 {
  57         direct_gbpages = 0;
  58         return 0;
  59 }
  60 early_param("nogbpages", parse_direct_gbpages_off);
  61
  62 static int __init parse_direct_gbpages_on(char *arg)
  63 {
  64         direct_gbpages = 1;
  65         return 0;
  66 }
  67 early_param("gbpages", parse_direct_gbpages_on);
  68
  69 /*
  70  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
  71  * physical space so we can cache the place of the first one and move
  72  * around without checking the pgd every time.
  73  */
  74
  75 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
  76 EXPORT_SYMBOL_GPL(__supported_pte_mask);
  77
  78 int force_personality32;
  79
  80 /*
  81  * noexec32=on|off
  82  * Control non executable heap for 32bit processes.
  83  * To control the stack too use noexec=off
  84  *
  85  * on   PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
  86  * off  PROT_READ implies PROT_EXEC
  87  */
  88 static int __init nonx32_setup(char *str)
  89 {
  90         if (!strcmp(str, "on"))
  91                 force_personality32 &= ~READ_IMPLIES_EXEC;
  92         else if (!strcmp(str, "off"))
  93                 force_personality32 |= READ_IMPLIES_EXEC;
  94         return 1;
  95 }
  96 __setup("noexec32=", nonx32_setup);
  97
  98 /*
  99  * When memory was added/removed make sure all the processes MM have
 100  * suitable PGD entries in the local PGD level page.
 101  */
 102 void sync_global_pgds(unsigned long start, unsigned long end)
 103 {
 104         unsigned long address;
 105
 106         for (address = start; address <= end; address += PGDIR_SIZE) {
 107                 const pgd_t *pgd_ref = pgd_offset_k(address);
 108                 unsigned long flags;
 109                 struct page *page;
 110
 111                 if (pgd_none(*pgd_ref))
 112                         continue;
 113
 114                 spin_lock_irqsave(&pgd_lock, flags);
 115                 list_for_each_entry(page, &pgd_list, lru) {
 116                         pgd_t *pgd;
 117                         spinlock_t *pgt_lock;
 118
 119                         pgd = (pgd_t *)page_address(page) + pgd_index(address);
 120                         pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
 121                         spin_lock(pgt_lock);
 122
 123                         if (pgd_none(*pgd))
 124                                 set_pgd(pgd, *pgd_ref);
 125                         else
 126                                 BUG_ON(pgd_page_vaddr(*pgd)
 127                                        != pgd_page_vaddr(*pgd_ref));
 128
 129                         spin_unlock(pgt_lock);
 130                 }
 131                 spin_unlock_irqrestore(&pgd_lock, flags);
 132         }
 133 }
 134
 135 /*
 136  * NOTE: This function is marked __ref because it calls __init function
 137  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
 138  */
 139 static __ref void *spp_getpage(void)
 140 {
 141         void *ptr;
 142
 143         if (after_bootmem)
 144                 ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
 145         else
 146                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 147
 148         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 149                 panic("set_pte_phys: cannot allocate page data %s\n",
 150                         after_bootmem ? "after bootmem" : "");
 151         }
 152
 153         pr_debug("spp_getpage %p\n", ptr);
 154
 155         return ptr;
 156 }
 157
 158 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
 159 {
 160         if (pgd_none(*pgd)) {
 161                 pud_t *pud = (pud_t *)spp_getpage();
 162                 pgd_populate(&init_mm, pgd, pud);
 163                 if (pud != pud_offset(pgd, 0))
 164                         printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
 165                                pud, pud_offset(pgd, 0));
 166         }
 167         return pud_offset(pgd, vaddr);
 168 }
 169
 170 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
 171 {
 172         if (pud_none(*pud)) {
 173                 pmd_t *pmd = (pmd_t *) spp_getpage();
 174                 pud_populate(&init_mm, pud, pmd);
 175                 if (pmd != pmd_offset(pud, 0))
 176                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
 177                                pmd, pmd_offset(pud, 0));
 178         }
 179         return pmd_offset(pud, vaddr);
 180 }
 181
 182 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
 183 {
 184         if (pmd_none(*pmd)) {
 185                 pte_t *pte = (pte_t *) spp_getpage();
 186                 pmd_populate_kernel(&init_mm, pmd, pte);
 187                 if (pte != pte_offset_kernel(pmd, 0))
 188                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
 189         }
 190         return pte_offset_kernel(pmd, vaddr);
 191 }
 192
 193 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
 194 {
 195         pud_t *pud;
 196         pmd_t *pmd;
 197         pte_t *pte;
 198
 199         pud = pud_page + pud_index(vaddr);
 200         pmd = fill_pmd(pud, vaddr);
 201         pte = fill_pte(pmd, vaddr);
 202
 203         set_pte(pte, new_pte);
 204
 205         /*
 206          * It's enough to flush this one mapping.
 207          * (PGE mappings get flushed as well)
 208          */
 209         __flush_tlb_one(vaddr);
 210 }
 211
 212 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
 213 {
 214         pgd_t *pgd;
 215         pud_t *pud_page;
 216
 217         pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
 218
 219         pgd = pgd_offset_k(vaddr);
 220         if (pgd_none(*pgd)) {
 221                 printk(KERN_ERR
 222                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
 223                 return;
 224         }
 225         pud_page = (pud_t*)pgd_page_vaddr(*pgd);
 226         set_pte_vaddr_pud(pud_page, vaddr, pteval);
 227 }
 228
 229 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
 230 {
 231         pgd_t *pgd;
 232         pud_t *pud;
 233
 234         pgd = pgd_offset_k(vaddr);
 235         pud = fill_pud(pgd, vaddr);
 236         return fill_pmd(pud, vaddr);
 237 }
 238
 239 pte_t * __init populate_extra_pte(unsigned long vaddr)
 240 {
 241         pmd_t *pmd;
 242
 243         pmd = populate_extra_pmd(vaddr);
 244         return fill_pte(pmd, vaddr);
 245 }
 246
 247 /*
 248  * Create large page table mappings for a range of physical addresses.
 249  */
 250 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
 251                                                 pgprot_t prot)
 252 {
 253         pgd_t *pgd;
 254         pud_t *pud;
 255         pmd_t *pmd;
 256
 257         BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
 258         for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
 259                 pgd = pgd_offset_k((unsigned long)__va(phys));
 260                 if (pgd_none(*pgd)) {
 261                         pud = (pud_t *) spp_getpage();
 262                         set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
 263                                                 _PAGE_USER));
 264                 }
 265                 pud = pud_offset(pgd, (unsigned long)__va(phys));
 266                 if (pud_none(*pud)) {
 267                         pmd = (pmd_t *) spp_getpage();
 268                         set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
 269                                                 _PAGE_USER));
 270                 }
 271                 pmd = pmd_offset(pud, phys);
 272                 BUG_ON(!pmd_none(*pmd));
 273                 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
 274         }
 275 }
 276
 277 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
 278 {
 279         __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
 280 }
 281
 282 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
 283 {
 284         __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
 285 }
 286
 287 /*
 288  * The head.S code sets up the kernel high mapping:
 289  *
 290  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
 291  *
 292  * phys_addr holds the negative offset to the kernel, which is added
 293  * to the compile time generated pmds. This results in invalid pmds up
 294  * to the point where we hit the physaddr 0 mapping.
 295  *
 296  * We limit the mappings to the region from _text to _end.  _end is
 297  * rounded up to the 2MB boundary. This catches the invalid pmds as
 298  * well, as they are located before _text:
 299  */
 300 void __init cleanup_highmap(void)
 301 {
 302         unsigned long vaddr = __START_KERNEL_map;
 303         unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1;
 304         pmd_t *pmd = level2_kernel_pgt;
 305         pmd_t *last_pmd = pmd + PTRS_PER_PMD;
 306
 307         for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
 308                 if (pmd_none(*pmd))
 309                         continue;
 310                 if (vaddr < (unsigned long) _text || vaddr > end)
 311                         set_pmd(pmd, __pmd(0));
 312         }
 313 }
 314
 315 static __ref void *alloc_low_page(unsigned long *phys)
 316 {
 317         unsigned long pfn = e820_table_end++;
 318         void *adr;
 319
 320         if (after_bootmem) {
 321                 adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
 322                 *phys = __pa(adr);
 323
 324                 return adr;
 325         }
 326
 327         if (pfn >= e820_table_top)
 328                 panic("alloc_low_page: ran out of memory");
 329
 330         adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
 331         clear_page(adr);
 332         *phys  = pfn * PAGE_SIZE;
 333         return adr;
 334 }
 335
 336 static __ref void *map_low_page(void *virt)
 337 {
 338         void *adr;
 339         unsigned long phys, left;
 340
 341         if (after_bootmem)
 342                 return virt;
 343
 344         phys = __pa(virt);
 345         left = phys & (PAGE_SIZE - 1);
 346         adr = early_memremap(phys & PAGE_MASK, PAGE_SIZE);
 347         adr = (void *)(((unsigned long)adr) | left);
 348
 349         return adr;
 350 }
 351
 352 static __ref void unmap_low_page(void *adr)
 353 {
 354         if (after_bootmem)
 355                 return;
 356
 357         early_iounmap((void *)((unsigned long)adr & PAGE_MASK), PAGE_SIZE);
 358 }
 359
 360 static unsigned long __meminit
 361 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
 362               pgprot_t prot)
 363 {
 364         unsigned pages = 0;
 365         unsigned long last_map_addr = end;
 366         int i;
 367
 368         pte_t *pte = pte_page + pte_index(addr);
 369
 370         for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
 371
 372                 if (addr >= end) {
 373                         if (!after_bootmem) {
 374                                 for(; i < PTRS_PER_PTE; i++, pte++)
 375                                         set_pte(pte, __pte(0));
 376                         }
 377                         break;
 378                 }
 379
 380                 /*
 381                  * We will re-use the existing mapping.
 382                  * Xen for example has some special requirements, like mapping
 383                  * pagetable pages as RO. So assume someone who pre-setup
 384                  * these mappings are more intelligent.
 385                  */
 386                 if (pte_val(*pte)) {
 387                         pages++;
 388                         continue;
 389                 }
 390
 391                 if (0)
 392                         printk("   pte=%p addr=%lx pte=%016lx\n",
 393                                pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
 394                 pages++;
 395                 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
 396                 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
 397         }
 398
 399         update_page_count(PG_LEVEL_4K, pages);
 400
 401         return last_map_addr;
 402 }
 403
 404 static unsigned long __meminit
 405 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
 406               unsigned long page_size_mask, pgprot_t prot)
 407 {
 408         unsigned long pages = 0;
 409         unsigned long last_map_addr = end;
 410
 411         int i = pmd_index(address);
 412
 413         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
 414                 unsigned long pte_phys;
 415                 pmd_t *pmd = pmd_page + pmd_index(address);
 416                 pte_t *pte;
 417                 pgprot_t new_prot = prot;
 418
 419                 if (address >= end) {
 420                         if (!after_bootmem) {
 421                                 for (; i < PTRS_PER_PMD; i++, pmd++)
 422                                         set_pmd(pmd, __pmd(0));
 423                         }
 424                         break;
 425                 }
 426
 427                 if (pmd_val(*pmd)) {
 428                         if (!pmd_large(*pmd)) {
 429                                 spin_lock(&init_mm.page_table_lock);
 430                                 pte = map_low_page((pte_t *)pmd_page_vaddr(*pmd));
 431                                 last_map_addr = phys_pte_init(pte, address,
 432                                                                 end, prot);
 433                                 unmap_low_page(pte);
 434                                 spin_unlock(&init_mm.page_table_lock);
 435                                 continue;
 436                         }
 437                         /*
 438                          * If we are ok with PG_LEVEL_2M mapping, then we will
 439                          * use the existing mapping,
 440                          *
 441                          * Otherwise, we will split the large page mapping but
 442                          * use the same existing protection bits except for
 443                          * large page, so that we don't violate Intel's TLB
 444                          * Application note (317080) which says, while changing
 445                          * the page sizes, new and old translations should
 446                          * not differ with respect to page frame and
 447                          * attributes.
 448                          */
 449                         if (page_size_mask & (1 << PG_LEVEL_2M)) {
 450                                 pages++;
 451                                 continue;
 452                         }
 453                         new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
 454                 }
 455
 456                 if (page_size_mask & (1<<PG_LEVEL_2M)) {
 457                         pages++;
 458                         spin_lock(&init_mm.page_table_lock);
 459                         set_pte((pte_t *)pmd,
 460                                 pfn_pte(address >> PAGE_SHIFT,
 461                                         __pgprot(pgprot_val(prot) | _PAGE_PSE)));
 462                         spin_unlock(&init_mm.page_table_lock);
 463                         last_map_addr = (address & PMD_MASK) + PMD_SIZE;
 464                         continue;
 465                 }
 466
 467                 pte = alloc_low_page(&pte_phys);
 468                 last_map_addr = phys_pte_init(pte, address, end, new_prot);
 469                 unmap_low_page(pte);
 470
 471                 spin_lock(&init_mm.page_table_lock);
 472                 pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
 473                 spin_unlock(&init_mm.page_table_lock);
 474         }
 475         update_page_count(PG_LEVEL_2M, pages);
 476         return last_map_addr;
 477 }
 478
 479 static unsigned long __meminit
 480 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
 481                          unsigned long page_size_mask)
 482 {
 483         unsigned long pages = 0;
 484         unsigned long last_map_addr = end;
 485         int i = pud_index(addr);
 486
 487         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
 488                 unsigned long pmd_phys;
 489                 pud_t *pud = pud_page + pud_index(addr);
 490                 pmd_t *pmd;
 491                 pgprot_t prot = PAGE_KERNEL;
 492
 493                 if (addr >= end)
 494                         break;
 495
 496                 if (!after_bootmem &&
 497                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
 498                         set_pud(pud, __pud(0));
 499                         continue;
 500                 }
 501
 502                 if (pud_val(*pud)) {
 503                         if (!pud_large(*pud)) {
 504                                 pmd = map_low_page(pmd_offset(pud, 0));
 505                                 last_map_addr = phys_pmd_init(pmd, addr, end,
 506                                                          page_size_mask, prot);
 507                                 unmap_low_page(pmd);
 508                                 __flush_tlb_all();
 509                                 continue;
 510                         }
 511                         /*
 512                          * If we are ok with PG_LEVEL_1G mapping, then we will
 513                          * use the existing mapping.
 514                          *
 515                          * Otherwise, we will split the gbpage mapping but use
 516                          * the same existing protection  bits except for large
 517                          * page, so that we don't violate Intel's TLB
 518                          * Application note (317080) which says, while changing
 519                          * the page sizes, new and old translations should
 520                          * not differ with respect to page frame and
 521                          * attributes.
 522                          */
 523                         if (page_size_mask & (1 << PG_LEVEL_1G)) {
 524                                 pages++;
 525                                 continue;
 526                         }
 527                         prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
 528                 }
 529
 530                 if (page_size_mask & (1<<PG_LEVEL_1G)) {
 531                         pages++;
 532                         spin_lock(&init_mm.page_table_lock);
 533                         set_pte((pte_t *)pud,
 534                                 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 535                         spin_unlock(&init_mm.page_table_lock);
 536                         last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
 537                         continue;
 538                 }
 539
 540                 pmd = alloc_low_page(&pmd_phys);
 541                 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
 542                                               prot);
 543                 unmap_low_page(pmd);
 544
 545                 spin_lock(&init_mm.page_table_lock);
 546                 pud_populate(&init_mm, pud, __va(pmd_phys));
 547                 spin_unlock(&init_mm.page_table_lock);
 548         }
 549         __flush_tlb_all();
 550
 551         update_page_count(PG_LEVEL_1G, pages);
 552
 553         return last_map_addr;
 554 }
 555
 556 unsigned long __meminit
 557 kernel_physical_mapping_init(unsigned long start,
 558                              unsigned long end,
 559                              unsigned long page_size_mask)
 560 {
 561         bool pgd_changed = false;
 562         unsigned long next, last_map_addr = end;
 563         unsigned long addr;
 564
 565         start = (unsigned long)__va(start);
 566         end = (unsigned long)__va(end);
 567         addr = start;
 568
 569         for (; start < end; start = next) {
 570                 pgd_t *pgd = pgd_offset_k(start);
 571                 unsigned long pud_phys;
 572                 pud_t *pud;
 573
 574                 next = (start + PGDIR_SIZE) & PGDIR_MASK;
 575                 if (next > end)
 576                         next = end;
 577
 578                 if (pgd_val(*pgd)) {
 579                         pud = map_low_page((pud_t *)pgd_page_vaddr(*pgd));
 580                         last_map_addr = phys_pud_init(pud, __pa(start),
 581                                                  __pa(end), page_size_mask);
 582                         unmap_low_page(pud);
 583                         continue;
 584                 }
 585
 586                 pud = alloc_low_page(&pud_phys);
 587                 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
 588                                                  page_size_mask);
 589                 unmap_low_page(pud);
 590
 591                 spin_lock(&init_mm.page_table_lock);
 592                 pgd_populate(&init_mm, pgd, __va(pud_phys));
 593                 spin_unlock(&init_mm.page_table_lock);
 594                 pgd_changed = true;
 595         }
 596
 597         if (pgd_changed)
 598                 sync_global_pgds(addr, end);
 599
 600         __flush_tlb_all();
 601
 602         return last_map_addr;
 603 }
 604
 605 #ifndef CONFIG_NUMA
 606 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn,
 607                                 int acpi, int k8)
 608 {
 609         memblock_x86_register_active_regions(0, start_pfn, end_pfn);
 610         init_memory_mapping_high();
 611 }
 612 #endif
 613
 614 struct mapping_work_data {
 615         unsigned long start;
 616         unsigned long end;
 617         unsigned long pfn_mapped;
 618 };
 619
 620 static int __init_refok
 621 mapping_work_fn(unsigned long start_pfn, unsigned long end_pfn, void *datax)
 622 {
 623         struct mapping_work_data *data = datax;
 624         unsigned long pfn_mapped;
 625         unsigned long final_start, final_end;
 626
 627         final_start = max_t(unsigned long, start_pfn<<PAGE_SHIFT, data->start);
 628         final_end = min_t(unsigned long, end_pfn<<PAGE_SHIFT, data->end);
 629
 630         if (final_end <= final_start)
 631                 return 0;
 632
 633         pfn_mapped = init_memory_mapping(final_start, final_end);
 634
 635         if (pfn_mapped > data->pfn_mapped)
 636                 data->pfn_mapped = pfn_mapped;
 637
 638         return 0;
 639 }
 640
 641 static unsigned long __init_refok
 642 init_memory_mapping_active_regions(unsigned long start, unsigned long end)
 643 {
 644         struct mapping_work_data data;
 645
 646         data.start = start;
 647         data.end = end;
 648         data.pfn_mapped = 0;
 649
 650         work_with_active_regions(MAX_NUMNODES, mapping_work_fn, &data);
 651
 652         return data.pfn_mapped;
 653 }
 654
 655 void __init_refok init_memory_mapping_high(void)
 656 {
 657         if (max_pfn > max_low_pfn) {
 658                 max_pfn_mapped = init_memory_mapping_active_regions(1UL<<32,
 659                                                          max_pfn<<PAGE_SHIFT);
 660                 /* can we preserve max_low_pfn ? */
 661                 max_low_pfn = max_pfn;
 662
 663                 memblock.current_limit = get_max_mapped();
 664         }
 665 }
 666
 667 void __init paging_init(void)
 668 {
 669         unsigned long max_zone_pfns[MAX_NR_ZONES];
 670
 671         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 672         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 673         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 674         max_zone_pfns[ZONE_NORMAL] = max_pfn;
 675
 676         sparse_memory_present_with_active_regions(MAX_NUMNODES);
 677         sparse_init();
 678
 679         /*
 680          * clear the default setting with node 0
 681          * note: don't use nodes_clear here, that is really clearing when
 682          *       numa support is not compiled in, and later node_set_state
 683          *       will not set it back.
 684          */
 685         node_clear_state(0, N_NORMAL_MEMORY);
 686
 687         free_area_init_nodes(max_zone_pfns);
 688 }
 689
 690 /*
 691  * Memory hotplug specific functions
 692  */
 693 #ifdef CONFIG_MEMORY_HOTPLUG
 694 /*
 695  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
 696  * updating.
 697  */
 698 static void  update_end_of_memory_vars(u64 start, u64 size)
 699 {
 700         unsigned long end_pfn = PFN_UP(start + size);
 701
 702         if (end_pfn > max_pfn) {
 703                 max_pfn = end_pfn;
 704                 max_low_pfn = end_pfn;
 705                 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 706         }
 707 }
 708
 709 /*
 710  * Memory is added always to NORMAL zone. This means you will never get
 711  * additional DMA/DMA32 memory.
 712  */
 713 int arch_add_memory(int nid, u64 start, u64 size)
 714 {
 715         struct pglist_data *pgdat = NODE_DATA(nid);
 716         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 717         unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
 718         unsigned long nr_pages = size >> PAGE_SHIFT;
 719         int ret;
 720
 721         last_mapped_pfn = init_memory_mapping(start, start + size);
 722         if (last_mapped_pfn > max_pfn_mapped)
 723                 max_pfn_mapped = last_mapped_pfn;
 724
 725         ret = __add_pages(nid, zone, start_pfn, nr_pages);
 726         WARN_ON_ONCE(ret);
 727
 728         /* update max_pfn, max_low_pfn and high_memory */
 729         update_end_of_memory_vars(start, size);
 730
 731         return ret;
 732 }
 733 EXPORT_SYMBOL_GPL(arch_add_memory);
 734
 735 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
 736 int memory_add_physaddr_to_nid(u64 start)
 737 {
 738         return 0;
 739 }
 740 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 741 #endif
 742
 743 #endif /* CONFIG_MEMORY_HOTPLUG */
 744
 745 static struct kcore_list kcore_vsyscall;
 746
 747 void __init mem_init(void)
 748 {
 749         long codesize, reservedpages, datasize, initsize;
 750         unsigned long absent_pages;
 751
 752         pci_iommu_alloc();
 753
 754         /* clear_bss() already clear the empty_zero_page */
 755
 756         reservedpages = 0;
 757
 758         /* this will put all low memory onto the freelists */
 759 #ifdef CONFIG_NUMA
 760         totalram_pages = numa_free_all_bootmem();
 761 #else
 762         totalram_pages = free_all_bootmem();
 763 #endif
 764
 765         absent_pages = absent_pages_in_range(0, max_pfn);
 766         reservedpages = max_pfn - totalram_pages - absent_pages;
 767         after_bootmem = 1;
 768
 769         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 770         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 771         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 772
 773         /* Register memory areas for /proc/kcore */
 774         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 775                          VSYSCALL_END - VSYSCALL_START, KCORE_OTHER);
 776
 777         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 778                          "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
 779                 nr_free_pages() << (PAGE_SHIFT-10),
 780                 max_pfn << (PAGE_SHIFT-10),
 781                 codesize >> 10,
 782                 absent_pages << (PAGE_SHIFT-10),
 783                 reservedpages << (PAGE_SHIFT-10),
 784                 datasize >> 10,
 785                 initsize >> 10);
 786 }
 787
 788 #ifdef CONFIG_DEBUG_RODATA
 789 const int rodata_test_data = 0xC3;
 790 EXPORT_SYMBOL_GPL(rodata_test_data);
 791
 792 int kernel_set_to_readonly;
 793
 794 void set_kernel_text_rw(void)
 795 {
 796         unsigned long start = PFN_ALIGN(_text);
 797         unsigned long end = PFN_ALIGN(__stop___ex_table);
 798
 799         if (!kernel_set_to_readonly)
 800                 return;
 801
 802         pr_debug("Set kernel text: %lx - %lx for read write\n",
 803                  start, end);
 804
 805         /*
 806          * Make the kernel identity mapping for text RW. Kernel text
 807          * mapping will always be RO. Refer to the comment in
 808          * static_protections() in pageattr.c
 809          */
 810         set_memory_rw(start, (end - start) >> PAGE_SHIFT);
 811 }
 812
 813 void set_kernel_text_ro(void)
 814 {
 815         unsigned long start = PFN_ALIGN(_text);
 816         unsigned long end = PFN_ALIGN(__stop___ex_table);
 817
 818         if (!kernel_set_to_readonly)
 819                 return;
 820
 821         pr_debug("Set kernel text: %lx - %lx for read only\n",
 822                  start, end);
 823
 824         /*
 825          * Set the kernel identity mapping for text RO.
 826          */
 827         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 828 }
 829
 830 void mark_rodata_ro(void)
 831 {
 832         unsigned long start = PFN_ALIGN(_text);
 833         unsigned long rodata_start =
 834                 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
 835         unsigned long end = (unsigned long) &__end_rodata_hpage_align;
 836         unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table);
 837         unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata);
 838         unsigned long data_start = (unsigned long) &_sdata;
 839
 840         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 841                (end - start) >> 10);
 842         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 843
 844         kernel_set_to_readonly = 1;
 845
 846         /*
 847          * The rodata section (but not the kernel text!) should also be
 848          * not-executable.
 849          */
 850         set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
 851
 852         rodata_test();
 853
 854 #ifdef CONFIG_CPA_DEBUG
 855         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
 856         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 857
 858         printk(KERN_INFO "Testing CPA: again\n");
 859         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 860 #endif
 861
 862         free_init_pages("unused kernel memory",
 863                         (unsigned long) page_address(virt_to_page(text_end)),
 864                         (unsigned long)
 865                                  page_address(virt_to_page(rodata_start)));
 866         free_init_pages("unused kernel memory",
 867                         (unsigned long) page_address(virt_to_page(rodata_end)),
 868                         (unsigned long) page_address(virt_to_page(data_start)));
 869 }
 870
 871 #endif
 872
 873 int kern_addr_valid(unsigned long addr)
 874 {
 875         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 876         pgd_t *pgd;
 877         pud_t *pud;
 878         pmd_t *pmd;
 879         pte_t *pte;
 880
 881         if (above != 0 && above != -1UL)
 882                 return 0;
 883
 884         pgd = pgd_offset_k(addr);
 885         if (pgd_none(*pgd))
 886                 return 0;
 887
 888         pud = pud_offset(pgd, addr);
 889         if (pud_none(*pud))
 890                 return 0;
 891
 892         pmd = pmd_offset(pud, addr);
 893         if (pmd_none(*pmd))
 894                 return 0;
 895
 896         if (pmd_large(*pmd))
 897                 return pfn_valid(pmd_pfn(*pmd));
 898
 899         pte = pte_offset_kernel(pmd, addr);
 900         if (pte_none(*pte))
 901                 return 0;
 902
 903         return pfn_valid(pte_pfn(*pte));
 904 }
 905
 906 /*
 907  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 908  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 909  * not need special handling anymore:
 910  */
 911 static struct vm_area_struct gate_vma = {
 912         .vm_start       = VSYSCALL_START,
 913         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 914         .vm_page_prot   = PAGE_READONLY_EXEC,
 915         .vm_flags       = VM_READ | VM_EXEC
 916 };
 917
 918 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 919 {
 920 #ifdef CONFIG_IA32_EMULATION
 921         if (test_tsk_thread_flag(tsk, TIF_IA32))
 922                 return NULL;
 923 #endif
 924         return &gate_vma;
 925 }
 926
 927 int in_gate_area(struct task_struct *task, unsigned long addr)
 928 {
 929         struct vm_area_struct *vma = get_gate_vma(task);
 930
 931         if (!vma)
 932                 return 0;
 933
 934         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 935 }
 936
 937 /*
 938  * Use this when you have no reliable task/vma, typically from interrupt
 939  * context. It is less reliable than using the task's vma and may give
 940  * false positives:
 941  */
 942 int in_gate_area_no_task(unsigned long addr)
 943 {
 944         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 945 }
 946
 947 const char *arch_vma_name(struct vm_area_struct *vma)
 948 {
 949         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 950                 return "[vdso]";
 951         if (vma == &gate_vma)
 952                 return "[vsyscall]";
 953         return NULL;
 954 }
 955
 956 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 957 /*
 958  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 959  */
 960 static long __meminitdata addr_start, addr_end;
 961 static void __meminitdata *p_start, *p_end;
 962 static int __meminitdata node_start;
 963
 964 int __meminit
 965 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 966 {
 967         unsigned long addr = (unsigned long)start_page;
 968         unsigned long end = (unsigned long)(start_page + size);
 969         unsigned long next;
 970         pgd_t *pgd;
 971         pud_t *pud;
 972         pmd_t *pmd;
 973
 974         for (; addr < end; addr = next) {
 975                 void *p = NULL;
 976
 977                 pgd = vmemmap_pgd_populate(addr, node);
 978                 if (!pgd)
 979                         return -ENOMEM;
 980
 981                 pud = vmemmap_pud_populate(pgd, addr, node);
 982                 if (!pud)
 983                         return -ENOMEM;
 984
 985                 if (!cpu_has_pse) {
 986                         next = (addr + PAGE_SIZE) & PAGE_MASK;
 987                         pmd = vmemmap_pmd_populate(pud, addr, node);
 988
 989                         if (!pmd)
 990                                 return -ENOMEM;
 991
 992                         p = vmemmap_pte_populate(pmd, addr, node);
 993
 994                         if (!p)
 995                                 return -ENOMEM;
 996
 997                         addr_end = addr + PAGE_SIZE;
 998                         p_end = p + PAGE_SIZE;
 999                 } else {
1000                         next = pmd_addr_end(addr, end);
1001
1002                         pmd = pmd_offset(pud, addr);
1003                         if (pmd_none(*pmd)) {
1004                                 pte_t entry;
1005
1006                                 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
1007                                 if (!p)
1008                                         return -ENOMEM;
1009
1010                                 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1011                                                 PAGE_KERNEL_LARGE);
1012                                 set_pmd(pmd, __pmd(pte_val(entry)));
1013
1014                                 /* check to see if we have contiguous blocks */
1015                                 if (p_end != p || node_start != node) {
1016                                         if (p_start)
1017                                                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1018                                                        addr_start, addr_end-1, p_start, p_end-1, node_start);
1019                                         addr_start = addr;
1020                                         node_start = node;
1021                                         p_start = p;
1022                                 }
1023
1024                                 addr_end = addr + PMD_SIZE;
1025                                 p_end = p + PMD_SIZE;
1026                         } else
1027                                 vmemmap_verify((pte_t *)pmd, node, addr, next);
1028                 }
1029
1030         }
1031         sync_global_pgds((unsigned long)start_page, end);
1032         return 0;
1033 }
1034
1035 void __meminit vmemmap_populate_print_last(void)
1036 {
1037         if (p_start) {
1038                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1039                         addr_start, addr_end-1, p_start, p_end-1, node_start);
1040                 p_start = NULL;
1041                 p_end = NULL;
1042                 node_start = 0;
1043         }
1044 }
1045 #endif