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