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.h>
  32 #include <linux/memory_hotplug.h>
  33 #include <linux/nmi.h>
  34 #include <linux/gfp.h>
  35
  36 #include <asm/processor.h>
  37 #include <asm/bios_ebda.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 #include <asm/uv/uv.h>
  55 #include <asm/setup.h>
  56
  57 static int __init parse_direct_gbpages_off(char *arg)
  58 {
  59         direct_gbpages = 0;
  60         return 0;
  61 }
  62 early_param("nogbpages", parse_direct_gbpages_off);
  63
  64 static int __init parse_direct_gbpages_on(char *arg)
  65 {
  66         direct_gbpages = 1;
  67         return 0;
  68 }
  69 early_param("gbpages", parse_direct_gbpages_on);
  70
  71 /*
  72  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
  73  * physical space so we can cache the place of the first one and move
  74  * around without checking the pgd every time.
  75  */
  76
  77 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
  78 EXPORT_SYMBOL_GPL(__supported_pte_mask);
  79
  80 int force_personality32;
  81
  82 /*
  83  * noexec32=on|off
  84  * Control non executable heap for 32bit processes.
  85  * To control the stack too use noexec=off
  86  *
  87  * on   PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
  88  * off  PROT_READ implies PROT_EXEC
  89  */
  90 static int __init nonx32_setup(char *str)
  91 {
  92         if (!strcmp(str, "on"))
  93                 force_personality32 &= ~READ_IMPLIES_EXEC;
  94         else if (!strcmp(str, "off"))
  95                 force_personality32 |= READ_IMPLIES_EXEC;
  96         return 1;
  97 }
  98 __setup("noexec32=", nonx32_setup);
  99
 100 /*
 101  * When memory was added/removed make sure all the processes MM have
 102  * suitable PGD entries in the local PGD level page.
 103  */
 104 void sync_global_pgds(unsigned long start, unsigned long end)
 105 {
 106         unsigned long address;
 107
 108         for (address = start; address <= end; address += PGDIR_SIZE) {
 109                 const pgd_t *pgd_ref = pgd_offset_k(address);
 110                 struct page *page;
 111
 112                 if (pgd_none(*pgd_ref))
 113                         continue;
 114
 115                 spin_lock(&pgd_lock);
 116                 list_for_each_entry(page, &pgd_list, lru) {
 117                         pgd_t *pgd;
 118                         spinlock_t *pgt_lock;
 119
 120                         pgd = (pgd_t *)page_address(page) + pgd_index(address);
 121                         /* the pgt_lock only for Xen */
 122                         pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
 123                         spin_lock(pgt_lock);
 124
 125                         if (pgd_none(*pgd))
 126                                 set_pgd(pgd, *pgd_ref);
 127                         else
 128                                 BUG_ON(pgd_page_vaddr(*pgd)
 129                                        != pgd_page_vaddr(*pgd_ref));
 130
 131                         spin_unlock(pgt_lock);
 132                 }
 133                 spin_unlock(&pgd_lock);
 134         }
 135 }
 136
 137 /*
 138  * NOTE: This function is marked __ref because it calls __init function
 139  * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
 140  */
 141 static __ref void *spp_getpage(void)
 142 {
 143         void *ptr;
 144
 145         if (after_bootmem)
 146                 ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
 147         else
 148                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 149
 150         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 151                 panic("set_pte_phys: cannot allocate page data %s\n",
 152                         after_bootmem ? "after bootmem" : "");
 153         }
 154
 155         pr_debug("spp_getpage %p\n", ptr);
 156
 157         return ptr;
 158 }
 159
 160 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
 161 {
 162         if (pgd_none(*pgd)) {
 163                 pud_t *pud = (pud_t *)spp_getpage();
 164                 pgd_populate(&init_mm, pgd, pud);
 165                 if (pud != pud_offset(pgd, 0))
 166                         printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
 167                                pud, pud_offset(pgd, 0));
 168         }
 169         return pud_offset(pgd, vaddr);
 170 }
 171
 172 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
 173 {
 174         if (pud_none(*pud)) {
 175                 pmd_t *pmd = (pmd_t *) spp_getpage();
 176                 pud_populate(&init_mm, pud, pmd);
 177                 if (pmd != pmd_offset(pud, 0))
 178                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
 179                                pmd, pmd_offset(pud, 0));
 180         }
 181         return pmd_offset(pud, vaddr);
 182 }
 183
 184 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
 185 {
 186         if (pmd_none(*pmd)) {
 187                 pte_t *pte = (pte_t *) spp_getpage();
 188                 pmd_populate_kernel(&init_mm, pmd, pte);
 189                 if (pte != pte_offset_kernel(pmd, 0))
 190                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
 191         }
 192         return pte_offset_kernel(pmd, vaddr);
 193 }
 194
 195 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
 196 {
 197         pud_t *pud;
 198         pmd_t *pmd;
 199         pte_t *pte;
 200
 201         pud = pud_page + pud_index(vaddr);
 202         pmd = fill_pmd(pud, vaddr);
 203         pte = fill_pte(pmd, vaddr);
 204
 205         set_pte(pte, new_pte);
 206
 207         /*
 208          * It's enough to flush this one mapping.
 209          * (PGE mappings get flushed as well)
 210          */
 211         __flush_tlb_one(vaddr);
 212 }
 213
 214 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
 215 {
 216         pgd_t *pgd;
 217         pud_t *pud_page;
 218
 219         pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
 220
 221         pgd = pgd_offset_k(vaddr);
 222         if (pgd_none(*pgd)) {
 223                 printk(KERN_ERR
 224                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
 225                 return;
 226         }
 227         pud_page = (pud_t*)pgd_page_vaddr(*pgd);
 228         set_pte_vaddr_pud(pud_page, vaddr, pteval);
 229 }
 230
 231 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
 232 {
 233         pgd_t *pgd;
 234         pud_t *pud;
 235
 236         pgd = pgd_offset_k(vaddr);
 237         pud = fill_pud(pgd, vaddr);
 238         return fill_pmd(pud, vaddr);
 239 }
 240
 241 pte_t * __init populate_extra_pte(unsigned long vaddr)
 242 {
 243         pmd_t *pmd;
 244
 245         pmd = populate_extra_pmd(vaddr);
 246         return fill_pte(pmd, vaddr);
 247 }
 248
 249 /*
 250  * Create large page table mappings for a range of physical addresses.
 251  */
 252 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
 253                                                 pgprot_t prot)
 254 {
 255         pgd_t *pgd;
 256         pud_t *pud;
 257         pmd_t *pmd;
 258
 259         BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
 260         for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
 261                 pgd = pgd_offset_k((unsigned long)__va(phys));
 262                 if (pgd_none(*pgd)) {
 263                         pud = (pud_t *) spp_getpage();
 264                         set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
 265                                                 _PAGE_USER));
 266                 }
 267                 pud = pud_offset(pgd, (unsigned long)__va(phys));
 268                 if (pud_none(*pud)) {
 269                         pmd = (pmd_t *) spp_getpage();
 270                         set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
 271                                                 _PAGE_USER));
 272                 }
 273                 pmd = pmd_offset(pud, phys);
 274                 BUG_ON(!pmd_none(*pmd));
 275                 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
 276         }
 277 }
 278
 279 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
 280 {
 281         __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
 282 }
 283
 284 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
 285 {
 286         __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
 287 }
 288
 289 /*
 290  * The head.S code sets up the kernel high mapping:
 291  *
 292  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
 293  *
 294  * phys_addr holds the negative offset to the kernel, which is added
 295  * to the compile time generated pmds. This results in invalid pmds up
 296  * to the point where we hit the physaddr 0 mapping.
 297  *
 298  * We limit the mappings to the region from _text to _brk_end.  _brk_end
 299  * is rounded up to the 2MB boundary. This catches the invalid pmds as
 300  * well, as they are located before _text:
 301  */
 302 void __init cleanup_highmap(void)
 303 {
 304         unsigned long vaddr = __START_KERNEL_map;
 305         unsigned long vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
 306         unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
 307         pmd_t *pmd = level2_kernel_pgt;
 308
 309         for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
 310                 if (pmd_none(*pmd))
 311                         continue;
 312                 if (vaddr < (unsigned long) _text || vaddr > end)
 313                         set_pmd(pmd, __pmd(0));
 314         }
 315 }
 316
 317 static __ref void *alloc_low_page(void)
 318 {
 319         unsigned long pfn;
 320         void *adr;
 321
 322         if (after_bootmem) {
 323                 adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
 324
 325                 return adr;
 326         }
 327
 328         if ((pgt_buf_end + 1) >= pgt_buf_top) {
 329                 unsigned long ret;
 330                 if (min_pfn_mapped >= max_pfn_mapped)
 331                         panic("alloc_low_page: ran out of memory");
 332                 ret = memblock_find_in_range(min_pfn_mapped << PAGE_SHIFT,
 333                                         max_pfn_mapped << PAGE_SHIFT,
 334                                         PAGE_SIZE, PAGE_SIZE);
 335                 if (!ret)
 336                         panic("alloc_low_page: can not alloc memory");
 337                 memblock_reserve(ret, PAGE_SIZE);
 338                 pfn = ret >> PAGE_SHIFT;
 339         } else
 340                 pfn = pgt_buf_end++;
 341
 342         adr = __va(pfn * PAGE_SIZE);
 343         clear_page(adr);
 344         return adr;
 345 }
 346
 347 static unsigned long __meminit
 348 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
 349               pgprot_t prot)
 350 {
 351         unsigned long pages = 0, next;
 352         unsigned long last_map_addr = end;
 353         int i;
 354
 355         pte_t *pte = pte_page + pte_index(addr);
 356
 357         for (i = pte_index(addr); i < PTRS_PER_PTE; i++, addr = next, pte++) {
 358                 next = (addr & PAGE_MASK) + PAGE_SIZE;
 359                 if (addr >= end) {
 360                         if (!after_bootmem &&
 361                             !e820_any_mapped(addr & PAGE_MASK, next, E820_RAM) &&
 362                             !e820_any_mapped(addr & PAGE_MASK, next, E820_RESERVED_KERN))
 363                                 set_pte(pte, __pte(0));
 364                         continue;
 365                 }
 366
 367                 /*
 368                  * We will re-use the existing mapping.
 369                  * Xen for example has some special requirements, like mapping
 370                  * pagetable pages as RO. So assume someone who pre-setup
 371                  * these mappings are more intelligent.
 372                  */
 373                 if (pte_val(*pte)) {
 374                         if (!after_bootmem)
 375                                 pages++;
 376                         continue;
 377                 }
 378
 379                 if (0)
 380                         printk("   pte=%p addr=%lx pte=%016lx\n",
 381                                pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
 382                 pages++;
 383                 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
 384                 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
 385         }
 386
 387         update_page_count(PG_LEVEL_4K, pages);
 388
 389         return last_map_addr;
 390 }
 391
 392 static unsigned long __meminit
 393 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
 394               unsigned long page_size_mask, pgprot_t prot)
 395 {
 396         unsigned long pages = 0, next;
 397         unsigned long last_map_addr = end;
 398
 399         int i = pmd_index(address);
 400
 401         for (; i < PTRS_PER_PMD; i++, address = next) {
 402                 pmd_t *pmd = pmd_page + pmd_index(address);
 403                 pte_t *pte;
 404                 pgprot_t new_prot = prot;
 405
 406                 next = (address & PMD_MASK) + PMD_SIZE;
 407                 if (address >= end) {
 408                         if (!after_bootmem &&
 409                             !e820_any_mapped(address & PMD_MASK, next, E820_RAM) &&
 410                             !e820_any_mapped(address & PMD_MASK, next, E820_RESERVED_KERN))
 411                                 set_pmd(pmd, __pmd(0));
 412                         continue;
 413                 }
 414
 415                 if (pmd_val(*pmd)) {
 416                         if (!pmd_large(*pmd)) {
 417                                 spin_lock(&init_mm.page_table_lock);
 418                                 pte = (pte_t *)pmd_page_vaddr(*pmd);
 419                                 last_map_addr = phys_pte_init(pte, address,
 420                                                                 end, prot);
 421                                 spin_unlock(&init_mm.page_table_lock);
 422                                 continue;
 423                         }
 424                         /*
 425                          * If we are ok with PG_LEVEL_2M mapping, then we will
 426                          * use the existing mapping,
 427                          *
 428                          * Otherwise, we will split the large page mapping but
 429                          * use the same existing protection bits except for
 430                          * large page, so that we don't violate Intel's TLB
 431                          * Application note (317080) which says, while changing
 432                          * the page sizes, new and old translations should
 433                          * not differ with respect to page frame and
 434                          * attributes.
 435                          */
 436                         if (page_size_mask & (1 << PG_LEVEL_2M)) {
 437                                 if (!after_bootmem)
 438                                         pages++;
 439                                 last_map_addr = next;
 440                                 continue;
 441                         }
 442                         new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
 443                 }
 444
 445                 if (page_size_mask & (1<<PG_LEVEL_2M)) {
 446                         pages++;
 447                         spin_lock(&init_mm.page_table_lock);
 448                         set_pte((pte_t *)pmd,
 449                                 pfn_pte((address & PMD_MASK) >> PAGE_SHIFT,
 450                                         __pgprot(pgprot_val(prot) | _PAGE_PSE)));
 451                         spin_unlock(&init_mm.page_table_lock);
 452                         last_map_addr = next;
 453                         continue;
 454                 }
 455
 456                 pte = alloc_low_page();
 457                 last_map_addr = phys_pte_init(pte, address, end, new_prot);
 458
 459                 spin_lock(&init_mm.page_table_lock);
 460                 pmd_populate_kernel(&init_mm, pmd, pte);
 461                 spin_unlock(&init_mm.page_table_lock);
 462         }
 463         update_page_count(PG_LEVEL_2M, pages);
 464         return last_map_addr;
 465 }
 466
 467 static unsigned long __meminit
 468 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
 469                          unsigned long page_size_mask)
 470 {
 471         unsigned long pages = 0, next;
 472         unsigned long last_map_addr = end;
 473         int i = pud_index(addr);
 474
 475         for (; i < PTRS_PER_PUD; i++, addr = next) {
 476                 pud_t *pud = pud_page + pud_index(addr);
 477                 pmd_t *pmd;
 478                 pgprot_t prot = PAGE_KERNEL;
 479
 480                 next = (addr & PUD_MASK) + PUD_SIZE;
 481                 if (addr >= end) {
 482                         if (!after_bootmem &&
 483                             !e820_any_mapped(addr & PUD_MASK, next, E820_RAM) &&
 484                             !e820_any_mapped(addr & PUD_MASK, next, E820_RESERVED_KERN))
 485                                 set_pud(pud, __pud(0));
 486                         continue;
 487                 }
 488
 489                 if (pud_val(*pud)) {
 490                         if (!pud_large(*pud)) {
 491                                 pmd = pmd_offset(pud, 0);
 492                                 last_map_addr = phys_pmd_init(pmd, addr, end,
 493                                                          page_size_mask, prot);
 494                                 __flush_tlb_all();
 495                                 continue;
 496                         }
 497                         /*
 498                          * If we are ok with PG_LEVEL_1G mapping, then we will
 499                          * use the existing mapping.
 500                          *
 501                          * Otherwise, we will split the gbpage mapping but use
 502                          * the same existing protection  bits except for large
 503                          * page, so that we don't violate Intel's TLB
 504                          * Application note (317080) which says, while changing
 505                          * the page sizes, new and old translations should
 506                          * not differ with respect to page frame and
 507                          * attributes.
 508                          */
 509                         if (page_size_mask & (1 << PG_LEVEL_1G)) {
 510                                 if (!after_bootmem)
 511                                         pages++;
 512                                 last_map_addr = next;
 513                                 continue;
 514                         }
 515                         prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
 516                 }
 517
 518                 if (page_size_mask & (1<<PG_LEVEL_1G)) {
 519                         pages++;
 520                         spin_lock(&init_mm.page_table_lock);
 521                         set_pte((pte_t *)pud,
 522                                 pfn_pte((addr & PUD_MASK) >> PAGE_SHIFT,
 523                                         PAGE_KERNEL_LARGE));
 524                         spin_unlock(&init_mm.page_table_lock);
 525                         last_map_addr = next;
 526                         continue;
 527                 }
 528
 529                 pmd = alloc_low_page();
 530                 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
 531                                               prot);
 532
 533                 spin_lock(&init_mm.page_table_lock);
 534                 pud_populate(&init_mm, pud, pmd);
 535                 spin_unlock(&init_mm.page_table_lock);
 536         }
 537         __flush_tlb_all();
 538
 539         update_page_count(PG_LEVEL_1G, pages);
 540
 541         return last_map_addr;
 542 }
 543
 544 unsigned long __meminit
 545 kernel_physical_mapping_init(unsigned long start,
 546                              unsigned long end,
 547                              unsigned long page_size_mask)
 548 {
 549         bool pgd_changed = false;
 550         unsigned long next, last_map_addr = end;
 551         unsigned long addr;
 552
 553         start = (unsigned long)__va(start);
 554         end = (unsigned long)__va(end);
 555         addr = start;
 556
 557         for (; start < end; start = next) {
 558                 pgd_t *pgd = pgd_offset_k(start);
 559                 pud_t *pud;
 560
 561                 next = (start + PGDIR_SIZE) & PGDIR_MASK;
 562                 if (next > end)
 563                         next = end;
 564
 565                 if (pgd_val(*pgd)) {
 566                         pud = (pud_t *)pgd_page_vaddr(*pgd);
 567                         last_map_addr = phys_pud_init(pud, __pa(start),
 568                                                  __pa(end), page_size_mask);
 569                         continue;
 570                 }
 571
 572                 pud = alloc_low_page();
 573                 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
 574                                                  page_size_mask);
 575
 576                 spin_lock(&init_mm.page_table_lock);
 577                 pgd_populate(&init_mm, pgd, pud);
 578                 spin_unlock(&init_mm.page_table_lock);
 579                 pgd_changed = true;
 580         }
 581
 582         if (pgd_changed)
 583                 sync_global_pgds(addr, end);
 584
 585         __flush_tlb_all();
 586
 587         return last_map_addr;
 588 }
 589
 590 #ifndef CONFIG_NUMA
 591 void __init initmem_init(void)
 592 {
 593         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, 0);
 594 }
 595 #endif
 596
 597 void __init paging_init(void)
 598 {
 599         sparse_memory_present_with_active_regions(MAX_NUMNODES);
 600         sparse_init();
 601
 602         /*
 603          * clear the default setting with node 0
 604          * note: don't use nodes_clear here, that is really clearing when
 605          *       numa support is not compiled in, and later node_set_state
 606          *       will not set it back.
 607          */
 608         node_clear_state(0, N_NORMAL_MEMORY);
 609
 610         zone_sizes_init();
 611 }
 612
 613 /*
 614  * Memory hotplug specific functions
 615  */
 616 #ifdef CONFIG_MEMORY_HOTPLUG
 617 /*
 618  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
 619  * updating.
 620  */
 621 static void  update_end_of_memory_vars(u64 start, u64 size)
 622 {
 623         unsigned long end_pfn = PFN_UP(start + size);
 624
 625         if (end_pfn > max_pfn) {
 626                 max_pfn = end_pfn;
 627                 max_low_pfn = end_pfn;
 628                 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 629         }
 630 }
 631
 632 /*
 633  * Memory is added always to NORMAL zone. This means you will never get
 634  * additional DMA/DMA32 memory.
 635  */
 636 int arch_add_memory(int nid, u64 start, u64 size)
 637 {
 638         struct pglist_data *pgdat = NODE_DATA(nid);
 639         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 640         unsigned long start_pfn = start >> PAGE_SHIFT;
 641         unsigned long nr_pages = size >> PAGE_SHIFT;
 642         int ret;
 643
 644         init_memory_mapping(start, start + size);
 645
 646         ret = __add_pages(nid, zone, start_pfn, nr_pages);
 647         WARN_ON_ONCE(ret);
 648
 649         /* update max_pfn, max_low_pfn and high_memory */
 650         update_end_of_memory_vars(start, size);
 651
 652         return ret;
 653 }
 654 EXPORT_SYMBOL_GPL(arch_add_memory);
 655
 656 #endif /* CONFIG_MEMORY_HOTPLUG */
 657
 658 static struct kcore_list kcore_vsyscall;
 659
 660 void __init mem_init(void)
 661 {
 662         long codesize, reservedpages, datasize, initsize;
 663         unsigned long absent_pages;
 664
 665         pci_iommu_alloc();
 666
 667         /* clear_bss() already clear the empty_zero_page */
 668
 669         reservedpages = 0;
 670
 671         /* this will put all low memory onto the freelists */
 672 #ifdef CONFIG_NUMA
 673         totalram_pages = numa_free_all_bootmem();
 674 #else
 675         totalram_pages = free_all_bootmem();
 676 #endif
 677
 678         absent_pages = absent_pages_in_range(0, max_pfn);
 679         reservedpages = max_pfn - totalram_pages - absent_pages;
 680         after_bootmem = 1;
 681
 682         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 683         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 684         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 685
 686         /* Register memory areas for /proc/kcore */
 687         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 688                          VSYSCALL_END - VSYSCALL_START, KCORE_OTHER);
 689
 690         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 691                          "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
 692                 nr_free_pages() << (PAGE_SHIFT-10),
 693                 max_pfn << (PAGE_SHIFT-10),
 694                 codesize >> 10,
 695                 absent_pages << (PAGE_SHIFT-10),
 696                 reservedpages << (PAGE_SHIFT-10),
 697                 datasize >> 10,
 698                 initsize >> 10);
 699 }
 700
 701 #ifdef CONFIG_DEBUG_RODATA
 702 const int rodata_test_data = 0xC3;
 703 EXPORT_SYMBOL_GPL(rodata_test_data);
 704
 705 int kernel_set_to_readonly;
 706
 707 void set_kernel_text_rw(void)
 708 {
 709         unsigned long start = PFN_ALIGN(_text);
 710         unsigned long end = PFN_ALIGN(__stop___ex_table);
 711
 712         if (!kernel_set_to_readonly)
 713                 return;
 714
 715         pr_debug("Set kernel text: %lx - %lx for read write\n",
 716                  start, end);
 717
 718         /*
 719          * Make the kernel identity mapping for text RW. Kernel text
 720          * mapping will always be RO. Refer to the comment in
 721          * static_protections() in pageattr.c
 722          */
 723         set_memory_rw(start, (end - start) >> PAGE_SHIFT);
 724 }
 725
 726 void set_kernel_text_ro(void)
 727 {
 728         unsigned long start = PFN_ALIGN(_text);
 729         unsigned long end = PFN_ALIGN(__stop___ex_table);
 730
 731         if (!kernel_set_to_readonly)
 732                 return;
 733
 734         pr_debug("Set kernel text: %lx - %lx for read only\n",
 735                  start, end);
 736
 737         /*
 738          * Set the kernel identity mapping for text RO.
 739          */
 740         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 741 }
 742
 743 void mark_rodata_ro(void)
 744 {
 745         unsigned long start = PFN_ALIGN(_text);
 746         unsigned long rodata_start =
 747                 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
 748         unsigned long end = (unsigned long) &__end_rodata_hpage_align;
 749         unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table);
 750         unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata);
 751         unsigned long data_start = (unsigned long) &_sdata;
 752
 753         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 754                (end - start) >> 10);
 755         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 756
 757         kernel_set_to_readonly = 1;
 758
 759         /*
 760          * The rodata section (but not the kernel text!) should also be
 761          * not-executable.
 762          */
 763         set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
 764
 765         rodata_test();
 766
 767 #ifdef CONFIG_CPA_DEBUG
 768         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
 769         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 770
 771         printk(KERN_INFO "Testing CPA: again\n");
 772         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 773 #endif
 774
 775         free_init_pages("unused kernel memory",
 776                         (unsigned long) page_address(virt_to_page(text_end)),
 777                         (unsigned long)
 778                                  page_address(virt_to_page(rodata_start)));
 779         free_init_pages("unused kernel memory",
 780                         (unsigned long) page_address(virt_to_page(rodata_end)),
 781                         (unsigned long) page_address(virt_to_page(data_start)));
 782 }
 783
 784 #endif
 785
 786 int kern_addr_valid(unsigned long addr)
 787 {
 788         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 789         pgd_t *pgd;
 790         pud_t *pud;
 791         pmd_t *pmd;
 792         pte_t *pte;
 793
 794         if (above != 0 && above != -1UL)
 795                 return 0;
 796
 797         pgd = pgd_offset_k(addr);
 798         if (pgd_none(*pgd))
 799                 return 0;
 800
 801         pud = pud_offset(pgd, addr);
 802         if (pud_none(*pud))
 803                 return 0;
 804
 805         pmd = pmd_offset(pud, addr);
 806         if (pmd_none(*pmd))
 807                 return 0;
 808
 809         if (pmd_large(*pmd))
 810                 return pfn_valid(pmd_pfn(*pmd));
 811
 812         pte = pte_offset_kernel(pmd, addr);
 813         if (pte_none(*pte))
 814                 return 0;
 815
 816         return pfn_valid(pte_pfn(*pte));
 817 }
 818
 819 /*
 820  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 821  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 822  * not need special handling anymore:
 823  */
 824 static struct vm_area_struct gate_vma = {
 825         .vm_start       = VSYSCALL_START,
 826         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 827         .vm_page_prot   = PAGE_READONLY_EXEC,
 828         .vm_flags       = VM_READ | VM_EXEC
 829 };
 830
 831 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
 832 {
 833 #ifdef CONFIG_IA32_EMULATION
 834         if (!mm || mm->context.ia32_compat)
 835                 return NULL;
 836 #endif
 837         return &gate_vma;
 838 }
 839
 840 int in_gate_area(struct mm_struct *mm, unsigned long addr)
 841 {
 842         struct vm_area_struct *vma = get_gate_vma(mm);
 843
 844         if (!vma)
 845                 return 0;
 846
 847         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 848 }
 849
 850 /*
 851  * Use this when you have no reliable mm, typically from interrupt
 852  * context. It is less reliable than using a task's mm and may give
 853  * false positives.
 854  */
 855 int in_gate_area_no_mm(unsigned long addr)
 856 {
 857         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 858 }
 859
 860 const char *arch_vma_name(struct vm_area_struct *vma)
 861 {
 862         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 863                 return "[vdso]";
 864         if (vma == &gate_vma)
 865                 return "[vsyscall]";
 866         return NULL;
 867 }
 868
 869 #ifdef CONFIG_X86_UV
 870 unsigned long memory_block_size_bytes(void)
 871 {
 872         if (is_uv_system()) {
 873                 printk(KERN_INFO "UV: memory block size 2GB\n");
 874                 return 2UL * 1024 * 1024 * 1024;
 875         }
 876         return MIN_MEMORY_BLOCK_SIZE;
 877 }
 878 #endif
 879
 880 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 881 /*
 882  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 883  */
 884 static long __meminitdata addr_start, addr_end;
 885 static void __meminitdata *p_start, *p_end;
 886 static int __meminitdata node_start;
 887
 888 int __meminit
 889 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 890 {
 891         unsigned long addr = (unsigned long)start_page;
 892         unsigned long end = (unsigned long)(start_page + size);
 893         unsigned long next;
 894         pgd_t *pgd;
 895         pud_t *pud;
 896         pmd_t *pmd;
 897
 898         for (; addr < end; addr = next) {
 899                 void *p = NULL;
 900
 901                 pgd = vmemmap_pgd_populate(addr, node);
 902                 if (!pgd)
 903                         return -ENOMEM;
 904
 905                 pud = vmemmap_pud_populate(pgd, addr, node);
 906                 if (!pud)
 907                         return -ENOMEM;
 908
 909                 if (!cpu_has_pse) {
 910                         next = (addr + PAGE_SIZE) & PAGE_MASK;
 911                         pmd = vmemmap_pmd_populate(pud, addr, node);
 912
 913                         if (!pmd)
 914                                 return -ENOMEM;
 915
 916                         p = vmemmap_pte_populate(pmd, addr, node);
 917
 918                         if (!p)
 919                                 return -ENOMEM;
 920
 921                         addr_end = addr + PAGE_SIZE;
 922                         p_end = p + PAGE_SIZE;
 923                 } else {
 924                         next = pmd_addr_end(addr, end);
 925
 926                         pmd = pmd_offset(pud, addr);
 927                         if (pmd_none(*pmd)) {
 928                                 pte_t entry;
 929
 930                                 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
 931                                 if (!p)
 932                                         return -ENOMEM;
 933
 934                                 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
 935                                                 PAGE_KERNEL_LARGE);
 936                                 set_pmd(pmd, __pmd(pte_val(entry)));
 937
 938                                 /* check to see if we have contiguous blocks */
 939                                 if (p_end != p || node_start != node) {
 940                                         if (p_start)
 941                                                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
 942                                                        addr_start, addr_end-1, p_start, p_end-1, node_start);
 943                                         addr_start = addr;
 944                                         node_start = node;
 945                                         p_start = p;
 946                                 }
 947
 948                                 addr_end = addr + PMD_SIZE;
 949                                 p_end = p + PMD_SIZE;
 950                         } else
 951                                 vmemmap_verify((pte_t *)pmd, node, addr, next);
 952                 }
 953
 954         }
 955         sync_global_pgds((unsigned long)start_page, end);
 956         return 0;
 957 }
 958
 959 void __meminit vmemmap_populate_print_last(void)
 960 {
 961         if (p_start) {
 962                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
 963                         addr_start, addr_end-1, p_start, p_end-1, node_start);
 964                 p_start = NULL;
 965                 p_end = NULL;
 966                 node_start = 0;
 967         }
 968 }
 969 #endif