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(unsigned long *phys)
 318 {
 319         unsigned long pfn = pgt_buf_end++;
 320         void *adr;
 321
 322         if (after_bootmem) {
 323                 adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
 324                 *phys = __pa(adr);
 325
 326                 return adr;
 327         }
 328
 329         if (pfn >= pgt_buf_top)
 330                 panic("alloc_low_page: ran out of memory");
 331
 332         adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE);
 333         clear_page(adr);
 334         *phys  = pfn * PAGE_SIZE;
 335         return adr;
 336 }
 337
 338 static __ref void *map_low_page(void *virt)
 339 {
 340         void *adr;
 341         unsigned long phys, left;
 342
 343         if (after_bootmem)
 344                 return virt;
 345
 346         phys = __pa(virt);
 347         left = phys & (PAGE_SIZE - 1);
 348         adr = early_memremap(phys & PAGE_MASK, PAGE_SIZE);
 349         adr = (void *)(((unsigned long)adr) | left);
 350
 351         return adr;
 352 }
 353
 354 static __ref void unmap_low_page(void *adr)
 355 {
 356         if (after_bootmem)
 357                 return;
 358
 359         early_iounmap((void *)((unsigned long)adr & PAGE_MASK), PAGE_SIZE);
 360 }
 361
 362 static unsigned long __meminit
 363 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
 364               pgprot_t prot)
 365 {
 366         unsigned pages = 0;
 367         unsigned long last_map_addr = end;
 368         int i;
 369
 370         pte_t *pte = pte_page + pte_index(addr);
 371
 372         for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
 373
 374                 if (addr >= end) {
 375                         if (!after_bootmem) {
 376                                 for(; i < PTRS_PER_PTE; i++, pte++)
 377                                         set_pte(pte, __pte(0));
 378                         }
 379                         break;
 380                 }
 381
 382                 /*
 383                  * We will re-use the existing mapping.
 384                  * Xen for example has some special requirements, like mapping
 385                  * pagetable pages as RO. So assume someone who pre-setup
 386                  * these mappings are more intelligent.
 387                  */
 388                 if (pte_val(*pte)) {
 389                         pages++;
 390                         continue;
 391                 }
 392
 393                 if (0)
 394                         printk("   pte=%p addr=%lx pte=%016lx\n",
 395                                pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
 396                 pages++;
 397                 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
 398                 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
 399         }
 400
 401         update_page_count(PG_LEVEL_4K, pages);
 402
 403         return last_map_addr;
 404 }
 405
 406 static unsigned long __meminit
 407 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
 408               unsigned long page_size_mask, pgprot_t prot)
 409 {
 410         unsigned long pages = 0, next;
 411         unsigned long last_map_addr = end;
 412
 413         int i = pmd_index(address);
 414
 415         for (; i < PTRS_PER_PMD; i++, address = next) {
 416                 unsigned long pte_phys;
 417                 pmd_t *pmd = pmd_page + pmd_index(address);
 418                 pte_t *pte;
 419                 pgprot_t new_prot = prot;
 420
 421                 if (address >= end) {
 422                         if (!after_bootmem) {
 423                                 for (; i < PTRS_PER_PMD; i++, pmd++)
 424                                         set_pmd(pmd, __pmd(0));
 425                         }
 426                         break;
 427                 }
 428
 429                 next = (address & PMD_MASK) + PMD_SIZE;
 430
 431                 if (pmd_val(*pmd)) {
 432                         if (!pmd_large(*pmd)) {
 433                                 spin_lock(&init_mm.page_table_lock);
 434                                 pte = map_low_page((pte_t *)pmd_page_vaddr(*pmd));
 435                                 last_map_addr = phys_pte_init(pte, address,
 436                                                                 end, prot);
 437                                 unmap_low_page(pte);
 438                                 spin_unlock(&init_mm.page_table_lock);
 439                                 continue;
 440                         }
 441                         /*
 442                          * If we are ok with PG_LEVEL_2M mapping, then we will
 443                          * use the existing mapping,
 444                          *
 445                          * Otherwise, we will split the large page mapping but
 446                          * use the same existing protection bits except for
 447                          * large page, so that we don't violate Intel's TLB
 448                          * Application note (317080) which says, while changing
 449                          * the page sizes, new and old translations should
 450                          * not differ with respect to page frame and
 451                          * attributes.
 452                          */
 453                         if (page_size_mask & (1 << PG_LEVEL_2M)) {
 454                                 last_map_addr = next;
 455                                 continue;
 456                         }
 457                         new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
 458                 }
 459
 460                 if (page_size_mask & (1<<PG_LEVEL_2M)) {
 461                         pages++;
 462                         spin_lock(&init_mm.page_table_lock);
 463                         set_pte((pte_t *)pmd,
 464                                 pfn_pte(address >> PAGE_SHIFT,
 465                                         __pgprot(pgprot_val(prot) | _PAGE_PSE)));
 466                         spin_unlock(&init_mm.page_table_lock);
 467                         last_map_addr = next;
 468                         continue;
 469                 }
 470
 471                 pte = alloc_low_page(&pte_phys);
 472                 last_map_addr = phys_pte_init(pte, address, end, new_prot);
 473                 unmap_low_page(pte);
 474
 475                 spin_lock(&init_mm.page_table_lock);
 476                 pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
 477                 spin_unlock(&init_mm.page_table_lock);
 478         }
 479         update_page_count(PG_LEVEL_2M, pages);
 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, next;
 488         unsigned long last_map_addr = end;
 489         int i = pud_index(addr);
 490
 491         for (; i < PTRS_PER_PUD; i++, addr = next) {
 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                 next = (addr & PUD_MASK) + PUD_SIZE;
 501
 502                 if (!after_bootmem && !e820_any_mapped(addr, next, 0)) {
 503                         set_pud(pud, __pud(0));
 504                         continue;
 505                 }
 506
 507                 if (pud_val(*pud)) {
 508                         if (!pud_large(*pud)) {
 509                                 pmd = map_low_page(pmd_offset(pud, 0));
 510                                 last_map_addr = phys_pmd_init(pmd, addr, end,
 511                                                          page_size_mask, prot);
 512                                 unmap_low_page(pmd);
 513                                 __flush_tlb_all();
 514                                 continue;
 515                         }
 516                         /*
 517                          * If we are ok with PG_LEVEL_1G mapping, then we will
 518                          * use the existing mapping.
 519                          *
 520                          * Otherwise, we will split the gbpage mapping but use
 521                          * the same existing protection  bits except for large
 522                          * page, so that we don't violate Intel's TLB
 523                          * Application note (317080) which says, while changing
 524                          * the page sizes, new and old translations should
 525                          * not differ with respect to page frame and
 526                          * attributes.
 527                          */
 528                         if (page_size_mask & (1 << PG_LEVEL_1G)) {
 529                                 last_map_addr = next;
 530                                 continue;
 531                         }
 532                         prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
 533                 }
 534
 535                 if (page_size_mask & (1<<PG_LEVEL_1G)) {
 536                         pages++;
 537                         spin_lock(&init_mm.page_table_lock);
 538                         set_pte((pte_t *)pud,
 539                                 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 540                         spin_unlock(&init_mm.page_table_lock);
 541                         last_map_addr = next;
 542                         continue;
 543                 }
 544
 545                 pmd = alloc_low_page(&pmd_phys);
 546                 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
 547                                               prot);
 548                 unmap_low_page(pmd);
 549
 550                 spin_lock(&init_mm.page_table_lock);
 551                 pud_populate(&init_mm, pud, __va(pmd_phys));
 552                 spin_unlock(&init_mm.page_table_lock);
 553         }
 554         __flush_tlb_all();
 555
 556         update_page_count(PG_LEVEL_1G, pages);
 557
 558         return last_map_addr;
 559 }
 560
 561 unsigned long __meminit
 562 kernel_physical_mapping_init(unsigned long start,
 563                              unsigned long end,
 564                              unsigned long page_size_mask)
 565 {
 566         bool pgd_changed = false;
 567         unsigned long next, last_map_addr = end;
 568         unsigned long addr;
 569
 570         start = (unsigned long)__va(start);
 571         end = (unsigned long)__va(end);
 572         addr = start;
 573
 574         for (; start < end; start = next) {
 575                 pgd_t *pgd = pgd_offset_k(start);
 576                 unsigned long pud_phys;
 577                 pud_t *pud;
 578
 579                 next = (start + PGDIR_SIZE) & PGDIR_MASK;
 580                 if (next > end)
 581                         next = end;
 582
 583                 if (pgd_val(*pgd)) {
 584                         pud = map_low_page((pud_t *)pgd_page_vaddr(*pgd));
 585                         last_map_addr = phys_pud_init(pud, __pa(start),
 586                                                  __pa(end), page_size_mask);
 587                         unmap_low_page(pud);
 588                         continue;
 589                 }
 590
 591                 pud = alloc_low_page(&pud_phys);
 592                 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next),
 593                                                  page_size_mask);
 594                 unmap_low_page(pud);
 595
 596                 spin_lock(&init_mm.page_table_lock);
 597                 pgd_populate(&init_mm, pgd, __va(pud_phys));
 598                 spin_unlock(&init_mm.page_table_lock);
 599                 pgd_changed = true;
 600         }
 601
 602         if (pgd_changed)
 603                 sync_global_pgds(addr, end);
 604
 605         __flush_tlb_all();
 606
 607         return last_map_addr;
 608 }
 609
 610 #ifndef CONFIG_NUMA
 611 void __init initmem_init(void)
 612 {
 613         memblock_set_node(0, (phys_addr_t)ULLONG_MAX, 0);
 614 }
 615 #endif
 616
 617 void __init paging_init(void)
 618 {
 619         sparse_memory_present_with_active_regions(MAX_NUMNODES);
 620         sparse_init();
 621
 622         /*
 623          * clear the default setting with node 0
 624          * note: don't use nodes_clear here, that is really clearing when
 625          *       numa support is not compiled in, and later node_set_state
 626          *       will not set it back.
 627          */
 628         node_clear_state(0, N_NORMAL_MEMORY);
 629
 630         zone_sizes_init();
 631 }
 632
 633 /*
 634  * Memory hotplug specific functions
 635  */
 636 #ifdef CONFIG_MEMORY_HOTPLUG
 637 /*
 638  * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
 639  * updating.
 640  */
 641 static void  update_end_of_memory_vars(u64 start, u64 size)
 642 {
 643         unsigned long end_pfn = PFN_UP(start + size);
 644
 645         if (end_pfn > max_pfn) {
 646                 max_pfn = end_pfn;
 647                 max_low_pfn = end_pfn;
 648                 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
 649         }
 650 }
 651
 652 /*
 653  * Memory is added always to NORMAL zone. This means you will never get
 654  * additional DMA/DMA32 memory.
 655  */
 656 int arch_add_memory(int nid, u64 start, u64 size)
 657 {
 658         struct pglist_data *pgdat = NODE_DATA(nid);
 659         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 660         unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
 661         unsigned long nr_pages = size >> PAGE_SHIFT;
 662         int ret;
 663
 664         last_mapped_pfn = init_memory_mapping(start, start + size);
 665         if (last_mapped_pfn > max_pfn_mapped)
 666                 max_pfn_mapped = last_mapped_pfn;
 667
 668         ret = __add_pages(nid, zone, start_pfn, nr_pages);
 669         WARN_ON_ONCE(ret);
 670
 671         /* update max_pfn, max_low_pfn and high_memory */
 672         update_end_of_memory_vars(start, size);
 673
 674         return ret;
 675 }
 676 EXPORT_SYMBOL_GPL(arch_add_memory);
 677
 678 #endif /* CONFIG_MEMORY_HOTPLUG */
 679
 680 static struct kcore_list kcore_vsyscall;
 681
 682 void __init mem_init(void)
 683 {
 684         long codesize, reservedpages, datasize, initsize;
 685         unsigned long absent_pages;
 686
 687         pci_iommu_alloc();
 688
 689         /* clear_bss() already clear the empty_zero_page */
 690
 691         reservedpages = 0;
 692
 693         /* this will put all low memory onto the freelists */
 694 #ifdef CONFIG_NUMA
 695         totalram_pages = numa_free_all_bootmem();
 696 #else
 697         totalram_pages = free_all_bootmem();
 698 #endif
 699
 700         absent_pages = absent_pages_in_range(0, max_pfn);
 701         reservedpages = max_pfn - totalram_pages - absent_pages;
 702         after_bootmem = 1;
 703
 704         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 705         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 706         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 707
 708         /* Register memory areas for /proc/kcore */
 709         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 710                          VSYSCALL_END - VSYSCALL_START, KCORE_OTHER);
 711
 712         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 713                          "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
 714                 nr_free_pages() << (PAGE_SHIFT-10),
 715                 max_pfn << (PAGE_SHIFT-10),
 716                 codesize >> 10,
 717                 absent_pages << (PAGE_SHIFT-10),
 718                 reservedpages << (PAGE_SHIFT-10),
 719                 datasize >> 10,
 720                 initsize >> 10);
 721 }
 722
 723 #ifdef CONFIG_DEBUG_RODATA
 724 const int rodata_test_data = 0xC3;
 725 EXPORT_SYMBOL_GPL(rodata_test_data);
 726
 727 int kernel_set_to_readonly;
 728
 729 void set_kernel_text_rw(void)
 730 {
 731         unsigned long start = PFN_ALIGN(_text);
 732         unsigned long end = PFN_ALIGN(__stop___ex_table);
 733
 734         if (!kernel_set_to_readonly)
 735                 return;
 736
 737         pr_debug("Set kernel text: %lx - %lx for read write\n",
 738                  start, end);
 739
 740         /*
 741          * Make the kernel identity mapping for text RW. Kernel text
 742          * mapping will always be RO. Refer to the comment in
 743          * static_protections() in pageattr.c
 744          */
 745         set_memory_rw(start, (end - start) >> PAGE_SHIFT);
 746 }
 747
 748 void set_kernel_text_ro(void)
 749 {
 750         unsigned long start = PFN_ALIGN(_text);
 751         unsigned long end = PFN_ALIGN(__stop___ex_table);
 752
 753         if (!kernel_set_to_readonly)
 754                 return;
 755
 756         pr_debug("Set kernel text: %lx - %lx for read only\n",
 757                  start, end);
 758
 759         /*
 760          * Set the kernel identity mapping for text RO.
 761          */
 762         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 763 }
 764
 765 void mark_rodata_ro(void)
 766 {
 767         unsigned long start = PFN_ALIGN(_text);
 768         unsigned long rodata_start =
 769                 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
 770         unsigned long end = (unsigned long) &__end_rodata_hpage_align;
 771         unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table);
 772         unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata);
 773         unsigned long data_start = (unsigned long) &_sdata;
 774
 775         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 776                (end - start) >> 10);
 777         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 778
 779         kernel_set_to_readonly = 1;
 780
 781         /*
 782          * The rodata section (but not the kernel text!) should also be
 783          * not-executable.
 784          */
 785         set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT);
 786
 787         rodata_test();
 788
 789 #ifdef CONFIG_CPA_DEBUG
 790         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
 791         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 792
 793         printk(KERN_INFO "Testing CPA: again\n");
 794         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 795 #endif
 796
 797         free_init_pages("unused kernel memory",
 798                         (unsigned long) page_address(virt_to_page(text_end)),
 799                         (unsigned long)
 800                                  page_address(virt_to_page(rodata_start)));
 801         free_init_pages("unused kernel memory",
 802                         (unsigned long) page_address(virt_to_page(rodata_end)),
 803                         (unsigned long) page_address(virt_to_page(data_start)));
 804 }
 805
 806 #endif
 807
 808 int kern_addr_valid(unsigned long addr)
 809 {
 810         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 811         pgd_t *pgd;
 812         pud_t *pud;
 813         pmd_t *pmd;
 814         pte_t *pte;
 815
 816         if (above != 0 && above != -1UL)
 817                 return 0;
 818
 819         pgd = pgd_offset_k(addr);
 820         if (pgd_none(*pgd))
 821                 return 0;
 822
 823         pud = pud_offset(pgd, addr);
 824         if (pud_none(*pud))
 825                 return 0;
 826
 827         pmd = pmd_offset(pud, addr);
 828         if (pmd_none(*pmd))
 829                 return 0;
 830
 831         if (pmd_large(*pmd))
 832                 return pfn_valid(pmd_pfn(*pmd));
 833
 834         pte = pte_offset_kernel(pmd, addr);
 835         if (pte_none(*pte))
 836                 return 0;
 837
 838         return pfn_valid(pte_pfn(*pte));
 839 }
 840
 841 /*
 842  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 843  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 844  * not need special handling anymore:
 845  */
 846 static struct vm_area_struct gate_vma = {
 847         .vm_start       = VSYSCALL_START,
 848         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 849         .vm_page_prot   = PAGE_READONLY_EXEC,
 850         .vm_flags       = VM_READ | VM_EXEC
 851 };
 852
 853 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
 854 {
 855 #ifdef CONFIG_IA32_EMULATION
 856         if (!mm || mm->context.ia32_compat)
 857                 return NULL;
 858 #endif
 859         return &gate_vma;
 860 }
 861
 862 int in_gate_area(struct mm_struct *mm, unsigned long addr)
 863 {
 864         struct vm_area_struct *vma = get_gate_vma(mm);
 865
 866         if (!vma)
 867                 return 0;
 868
 869         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 870 }
 871
 872 /*
 873  * Use this when you have no reliable mm, typically from interrupt
 874  * context. It is less reliable than using a task's mm and may give
 875  * false positives.
 876  */
 877 int in_gate_area_no_mm(unsigned long addr)
 878 {
 879         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 880 }
 881
 882 const char *arch_vma_name(struct vm_area_struct *vma)
 883 {
 884         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 885                 return "[vdso]";
 886         if (vma == &gate_vma)
 887                 return "[vsyscall]";
 888         return NULL;
 889 }
 890
 891 #ifdef CONFIG_X86_UV
 892 unsigned long memory_block_size_bytes(void)
 893 {
 894         if (is_uv_system()) {
 895                 printk(KERN_INFO "UV: memory block size 2GB\n");
 896                 return 2UL * 1024 * 1024 * 1024;
 897         }
 898         return MIN_MEMORY_BLOCK_SIZE;
 899 }
 900 #endif
 901
 902 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 903 /*
 904  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 905  */
 906 static long __meminitdata addr_start, addr_end;
 907 static void __meminitdata *p_start, *p_end;
 908 static int __meminitdata node_start;
 909
 910 int __meminit
 911 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 912 {
 913         unsigned long addr = (unsigned long)start_page;
 914         unsigned long end = (unsigned long)(start_page + size);
 915         unsigned long next;
 916         pgd_t *pgd;
 917         pud_t *pud;
 918         pmd_t *pmd;
 919
 920         for (; addr < end; addr = next) {
 921                 void *p = NULL;
 922
 923                 pgd = vmemmap_pgd_populate(addr, node);
 924                 if (!pgd)
 925                         return -ENOMEM;
 926
 927                 pud = vmemmap_pud_populate(pgd, addr, node);
 928                 if (!pud)
 929                         return -ENOMEM;
 930
 931                 if (!cpu_has_pse) {
 932                         next = (addr + PAGE_SIZE) & PAGE_MASK;
 933                         pmd = vmemmap_pmd_populate(pud, addr, node);
 934
 935                         if (!pmd)
 936                                 return -ENOMEM;
 937
 938                         p = vmemmap_pte_populate(pmd, addr, node);
 939
 940                         if (!p)
 941                                 return -ENOMEM;
 942
 943                         addr_end = addr + PAGE_SIZE;
 944                         p_end = p + PAGE_SIZE;
 945                 } else {
 946                         next = pmd_addr_end(addr, end);
 947
 948                         pmd = pmd_offset(pud, addr);
 949                         if (pmd_none(*pmd)) {
 950                                 pte_t entry;
 951
 952                                 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
 953                                 if (!p)
 954                                         return -ENOMEM;
 955
 956                                 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
 957                                                 PAGE_KERNEL_LARGE);
 958                                 set_pmd(pmd, __pmd(pte_val(entry)));
 959
 960                                 /* check to see if we have contiguous blocks */
 961                                 if (p_end != p || node_start != node) {
 962                                         if (p_start)
 963                                                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
 964                                                        addr_start, addr_end-1, p_start, p_end-1, node_start);
 965                                         addr_start = addr;
 966                                         node_start = node;
 967                                         p_start = p;
 968                                 }
 969
 970                                 addr_end = addr + PMD_SIZE;
 971                                 p_end = p + PMD_SIZE;
 972                         } else
 973                                 vmemmap_verify((pte_t *)pmd, node, addr, next);
 974                 }
 975
 976         }
 977         sync_global_pgds((unsigned long)start_page, end);
 978         return 0;
 979 }
 980
 981 void __meminit vmemmap_populate_print_last(void)
 982 {
 983         if (p_start) {
 984                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
 985                         addr_start, addr_end-1, p_start, p_end-1, node_start);
 986                 p_start = NULL;
 987                 p_end = NULL;
 988                 node_start = 0;
 989         }
 990 }
 991 #endif