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@suse.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/proc_fs.h>
  25 #include <linux/pci.h>
  26 #include <linux/pfn.h>
  27 #include <linux/poison.h>
  28 #include <linux/dma-mapping.h>
  29 #include <linux/module.h>
  30 #include <linux/memory_hotplug.h>
  31 #include <linux/nmi.h>
  32
  33 #include <asm/processor.h>
  34 #include <asm/system.h>
  35 #include <asm/uaccess.h>
  36 #include <asm/pgtable.h>
  37 #include <asm/pgalloc.h>
  38 #include <asm/dma.h>
  39 #include <asm/fixmap.h>
  40 #include <asm/e820.h>
  41 #include <asm/apic.h>
  42 #include <asm/tlb.h>
  43 #include <asm/mmu_context.h>
  44 #include <asm/proto.h>
  45 #include <asm/smp.h>
  46 #include <asm/sections.h>
  47 #include <asm/kdebug.h>
  48 #include <asm/numa.h>
  49 #include <asm/cacheflush.h>
  50
  51 /*
  52  * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
  53  * The direct mapping extends to max_pfn_mapped, so that we can directly access
  54  * apertures, ACPI and other tables without having to play with fixmaps.
  55  */
  56 unsigned long max_pfn_mapped;
  57
  58 static unsigned long dma_reserve __initdata;
  59
  60 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  61
  62 int direct_gbpages __meminitdata
  63 #ifdef CONFIG_DIRECT_GBPAGES
  64                                 = 1
  65 #endif
  66 ;
  67
  68 static int __init parse_direct_gbpages_off(char *arg)
  69 {
  70         direct_gbpages = 0;
  71         return 0;
  72 }
  73 early_param("nogbpages", parse_direct_gbpages_off);
  74
  75 static int __init parse_direct_gbpages_on(char *arg)
  76 {
  77         direct_gbpages = 1;
  78         return 0;
  79 }
  80 early_param("gbpages", parse_direct_gbpages_on);
  81
  82 /*
  83  * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
  84  * physical space so we can cache the place of the first one and move
  85  * around without checking the pgd every time.
  86  */
  87
  88 void show_mem(void)
  89 {
  90         long i, total = 0, reserved = 0;
  91         long shared = 0, cached = 0;
  92         struct page *page;
  93         pg_data_t *pgdat;
  94
  95         printk(KERN_INFO "Mem-info:\n");
  96         show_free_areas();
  97         for_each_online_pgdat(pgdat) {
  98                 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
  99                         /*
 100                          * This loop can take a while with 256 GB and
 101                          * 4k pages so defer the NMI watchdog:
 102                          */
 103                         if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
 104                                 touch_nmi_watchdog();
 105
 106                         if (!pfn_valid(pgdat->node_start_pfn + i))
 107                                 continue;
 108
 109                         page = pfn_to_page(pgdat->node_start_pfn + i);
 110                         total++;
 111                         if (PageReserved(page))
 112                                 reserved++;
 113                         else if (PageSwapCache(page))
 114                                 cached++;
 115                         else if (page_count(page))
 116                                 shared += page_count(page) - 1;
 117                 }
 118         }
 119         printk(KERN_INFO "%lu pages of RAM\n",          total);
 120         printk(KERN_INFO "%lu reserved pages\n",        reserved);
 121         printk(KERN_INFO "%lu pages shared\n",          shared);
 122         printk(KERN_INFO "%lu pages swap cached\n",     cached);
 123 }
 124
 125 int after_bootmem;
 126
 127 static __init void *spp_getpage(void)
 128 {
 129         void *ptr;
 130
 131         if (after_bootmem)
 132                 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
 133         else
 134                 ptr = alloc_bootmem_pages(PAGE_SIZE);
 135
 136         if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
 137                 panic("set_pte_phys: cannot allocate page data %s\n",
 138                         after_bootmem ? "after bootmem" : "");
 139         }
 140
 141         pr_debug("spp_getpage %p\n", ptr);
 142
 143         return ptr;
 144 }
 145
 146 void
 147 set_pte_vaddr(unsigned long vaddr, pte_t new_pte)
 148 {
 149         pgd_t *pgd;
 150         pud_t *pud;
 151         pmd_t *pmd;
 152         pte_t *pte;
 153
 154         pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(new_pte));
 155
 156         pgd = pgd_offset_k(vaddr);
 157         if (pgd_none(*pgd)) {
 158                 printk(KERN_ERR
 159                         "PGD FIXMAP MISSING, it should be setup in head.S!\n");
 160                 return;
 161         }
 162         pud = pud_offset(pgd, vaddr);
 163         if (pud_none(*pud)) {
 164                 pmd = (pmd_t *) spp_getpage();
 165                 pud_populate(&init_mm, pud, pmd);
 166                 if (pmd != pmd_offset(pud, 0)) {
 167                         printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
 168                                 pmd, pmd_offset(pud, 0));
 169                         return;
 170                 }
 171         }
 172         pmd = pmd_offset(pud, vaddr);
 173         if (pmd_none(*pmd)) {
 174                 pte = (pte_t *) spp_getpage();
 175                 pmd_populate_kernel(&init_mm, pmd, pte);
 176                 if (pte != pte_offset_kernel(pmd, 0)) {
 177                         printk(KERN_ERR "PAGETABLE BUG #02!\n");
 178                         return;
 179                 }
 180         }
 181
 182         pte = pte_offset_kernel(pmd, vaddr);
 183         if (!pte_none(*pte) && pte_val(new_pte) &&
 184             pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask))
 185                 pte_ERROR(*pte);
 186         set_pte(pte, new_pte);
 187
 188         /*
 189          * It's enough to flush this one mapping.
 190          * (PGE mappings get flushed as well)
 191          */
 192         __flush_tlb_one(vaddr);
 193 }
 194
 195 /*
 196  * The head.S code sets up the kernel high mapping:
 197  *
 198  *   from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
 199  *
 200  * phys_addr holds the negative offset to the kernel, which is added
 201  * to the compile time generated pmds. This results in invalid pmds up
 202  * to the point where we hit the physaddr 0 mapping.
 203  *
 204  * We limit the mappings to the region from _text to _end.  _end is
 205  * rounded up to the 2MB boundary. This catches the invalid pmds as
 206  * well, as they are located before _text:
 207  */
 208 void __init cleanup_highmap(void)
 209 {
 210         unsigned long vaddr = __START_KERNEL_map;
 211         unsigned long end = round_up((unsigned long)_end, PMD_SIZE) - 1;
 212         pmd_t *pmd = level2_kernel_pgt;
 213         pmd_t *last_pmd = pmd + PTRS_PER_PMD;
 214
 215         for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) {
 216                 if (pmd_none(*pmd))
 217                         continue;
 218                 if (vaddr < (unsigned long) _text || vaddr > end)
 219                         set_pmd(pmd, __pmd(0));
 220         }
 221 }
 222
 223 static unsigned long __initdata table_start;
 224 static unsigned long __meminitdata table_end;
 225 static unsigned long __meminitdata table_top;
 226
 227 static __meminit void *alloc_low_page(unsigned long *phys)
 228 {
 229         unsigned long pfn = table_end++;
 230         void *adr;
 231
 232         if (after_bootmem) {
 233                 adr = (void *)get_zeroed_page(GFP_ATOMIC);
 234                 *phys = __pa(adr);
 235
 236                 return adr;
 237         }
 238
 239         if (pfn >= table_top)
 240                 panic("alloc_low_page: ran out of memory");
 241
 242         adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
 243         memset(adr, 0, PAGE_SIZE);
 244         *phys  = pfn * PAGE_SIZE;
 245         return adr;
 246 }
 247
 248 static __meminit void unmap_low_page(void *adr)
 249 {
 250         if (after_bootmem)
 251                 return;
 252
 253         early_iounmap(adr, PAGE_SIZE);
 254 }
 255
 256 static void __meminit
 257 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end)
 258 {
 259         unsigned pages = 0;
 260         int i;
 261         pte_t *pte = pte_page + pte_index(addr);
 262
 263         for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
 264
 265                 if (addr >= end) {
 266                         if (!after_bootmem) {
 267                                 for(; i < PTRS_PER_PTE; i++, pte++)
 268                                         set_pte(pte, __pte(0));
 269                         }
 270                         break;
 271                 }
 272
 273                 if (pte_val(*pte))
 274                         continue;
 275
 276                 if (0)
 277                         printk("   pte=%p addr=%lx pte=%016lx\n",
 278                                pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
 279                 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL));
 280                 pages++;
 281         }
 282         update_page_count(PG_LEVEL_4K, pages);
 283 }
 284
 285 static void __meminit
 286 phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end)
 287 {
 288         pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd);
 289
 290         phys_pte_init(pte, address, end);
 291 }
 292
 293 static unsigned long __meminit
 294 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
 295 {
 296         unsigned long pages = 0;
 297
 298         int i = pmd_index(address);
 299
 300         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
 301                 unsigned long pte_phys;
 302                 pmd_t *pmd = pmd_page + pmd_index(address);
 303                 pte_t *pte;
 304
 305                 if (address >= end) {
 306                         if (!after_bootmem) {
 307                                 for (; i < PTRS_PER_PMD; i++, pmd++)
 308                                         set_pmd(pmd, __pmd(0));
 309                         }
 310                         break;
 311                 }
 312
 313                 if (pmd_val(*pmd)) {
 314                         phys_pte_update(pmd, address, end);
 315                         continue;
 316                 }
 317
 318                 if (cpu_has_pse) {
 319                         pages++;
 320                         set_pte((pte_t *)pmd,
 321                                 pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 322                         continue;
 323                 }
 324
 325                 pte = alloc_low_page(&pte_phys);
 326                 phys_pte_init(pte, address, end);
 327                 unmap_low_page(pte);
 328
 329                 pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
 330         }
 331         update_page_count(PG_LEVEL_2M, pages);
 332         return address;
 333 }
 334
 335 static unsigned long __meminit
 336 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
 337 {
 338         pmd_t *pmd = pmd_offset(pud, 0);
 339         unsigned long last_map_addr;
 340
 341         spin_lock(&init_mm.page_table_lock);
 342         last_map_addr = phys_pmd_init(pmd, address, end);
 343         spin_unlock(&init_mm.page_table_lock);
 344         __flush_tlb_all();
 345         return last_map_addr;
 346 }
 347
 348 static unsigned long __meminit
 349 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
 350 {
 351         unsigned long pages = 0;
 352         unsigned long last_map_addr = end;
 353         int i = pud_index(addr);
 354
 355         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
 356                 unsigned long pmd_phys;
 357                 pud_t *pud = pud_page + pud_index(addr);
 358                 pmd_t *pmd;
 359
 360                 if (addr >= end)
 361                         break;
 362
 363                 if (!after_bootmem &&
 364                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
 365                         set_pud(pud, __pud(0));
 366                         continue;
 367                 }
 368
 369                 if (pud_val(*pud)) {
 370                         if (!pud_large(*pud))
 371                                 last_map_addr = phys_pmd_update(pud, addr, end);
 372                         continue;
 373                 }
 374
 375                 if (direct_gbpages) {
 376                         pages++;
 377                         set_pte((pte_t *)pud,
 378                                 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 379                         last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
 380                         continue;
 381                 }
 382
 383                 pmd = alloc_low_page(&pmd_phys);
 384
 385                 spin_lock(&init_mm.page_table_lock);
 386                 last_map_addr = phys_pmd_init(pmd, addr, end);
 387                 unmap_low_page(pmd);
 388                 pud_populate(&init_mm, pud, __va(pmd_phys));
 389                 spin_unlock(&init_mm.page_table_lock);
 390
 391         }
 392         __flush_tlb_all();
 393         update_page_count(PG_LEVEL_1G, pages);
 394
 395         return last_map_addr;
 396 }
 397
 398 static unsigned long __meminit
 399 phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end)
 400 {
 401         pud_t *pud;
 402
 403         pud = (pud_t *)pgd_page_vaddr(*pgd);
 404
 405         return phys_pud_init(pud, addr, end);
 406 }
 407
 408 static void __init find_early_table_space(unsigned long end)
 409 {
 410         unsigned long puds, tables, start;
 411
 412         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 413         tables = round_up(puds * sizeof(pud_t), PAGE_SIZE);
 414         if (!direct_gbpages) {
 415                 unsigned long pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 416                 tables += round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
 417         }
 418         if (!cpu_has_pse) {
 419                 unsigned long ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
 420                 tables += round_up(ptes * sizeof(pte_t), PAGE_SIZE);
 421         }
 422
 423         /*
 424          * RED-PEN putting page tables only on node 0 could
 425          * cause a hotspot and fill up ZONE_DMA. The page tables
 426          * need roughly 0.5KB per GB.
 427          */
 428         start = 0x8000;
 429         table_start = find_e820_area(start, end, tables, PAGE_SIZE);
 430         if (table_start == -1UL)
 431                 panic("Cannot find space for the kernel page tables");
 432
 433         table_start >>= PAGE_SHIFT;
 434         table_end = table_start;
 435         table_top = table_start + (tables >> PAGE_SHIFT);
 436
 437         printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
 438                 end, table_start << PAGE_SHIFT, table_top << PAGE_SHIFT);
 439 }
 440
 441 static void __init init_gbpages(void)
 442 {
 443         if (direct_gbpages && cpu_has_gbpages)
 444                 printk(KERN_INFO "Using GB pages for direct mapping\n");
 445         else
 446                 direct_gbpages = 0;
 447 }
 448
 449 #ifdef CONFIG_MEMTEST
 450
 451 static void __init memtest(unsigned long start_phys, unsigned long size,
 452                                  unsigned pattern)
 453 {
 454         unsigned long i;
 455         unsigned long *start;
 456         unsigned long start_bad;
 457         unsigned long last_bad;
 458         unsigned long val;
 459         unsigned long start_phys_aligned;
 460         unsigned long count;
 461         unsigned long incr;
 462
 463         switch (pattern) {
 464         case 0:
 465                 val = 0UL;
 466                 break;
 467         case 1:
 468                 val = -1UL;
 469                 break;
 470         case 2:
 471                 val = 0x5555555555555555UL;
 472                 break;
 473         case 3:
 474                 val = 0xaaaaaaaaaaaaaaaaUL;
 475                 break;
 476         default:
 477                 return;
 478         }
 479
 480         incr = sizeof(unsigned long);
 481         start_phys_aligned = ALIGN(start_phys, incr);
 482         count = (size - (start_phys_aligned - start_phys))/incr;
 483         start = __va(start_phys_aligned);
 484         start_bad = 0;
 485         last_bad = 0;
 486
 487         for (i = 0; i < count; i++)
 488                 start[i] = val;
 489         for (i = 0; i < count; i++, start++, start_phys_aligned += incr) {
 490                 if (*start != val) {
 491                         if (start_phys_aligned == last_bad + incr) {
 492                                 last_bad += incr;
 493                         } else {
 494                                 if (start_bad) {
 495                                         printk(KERN_CONT "\n  %016lx bad mem addr %016lx - %016lx reserved",
 496                                                 val, start_bad, last_bad + incr);
 497                                         reserve_early(start_bad, last_bad - start_bad, "BAD RAM");
 498                                 }
 499                                 start_bad = last_bad = start_phys_aligned;
 500                         }
 501                 }
 502         }
 503         if (start_bad) {
 504                 printk(KERN_CONT "\n  %016lx bad mem addr %016lx - %016lx reserved",
 505                         val, start_bad, last_bad + incr);
 506                 reserve_early(start_bad, last_bad - start_bad, "BAD RAM");
 507         }
 508
 509 }
 510
 511 /* default is disabled */
 512 static int memtest_pattern __initdata;
 513
 514 static int __init parse_memtest(char *arg)
 515 {
 516         if (arg)
 517                 memtest_pattern = simple_strtoul(arg, NULL, 0);
 518         return 0;
 519 }
 520
 521 early_param("memtest", parse_memtest);
 522
 523 static void __init early_memtest(unsigned long start, unsigned long end)
 524 {
 525         u64 t_start, t_size;
 526         unsigned pattern;
 527
 528         if (!memtest_pattern)
 529                 return;
 530
 531         printk(KERN_INFO "early_memtest: pattern num %d", memtest_pattern);
 532         for (pattern = 0; pattern < memtest_pattern; pattern++) {
 533                 t_start = start;
 534                 t_size = 0;
 535                 while (t_start < end) {
 536                         t_start = find_e820_area_size(t_start, &t_size, 1);
 537
 538                         /* done ? */
 539                         if (t_start >= end)
 540                                 break;
 541                         if (t_start + t_size > end)
 542                                 t_size = end - t_start;
 543
 544                         printk(KERN_CONT "\n  %016llx - %016llx pattern %d",
 545                                 (unsigned long long)t_start,
 546                                 (unsigned long long)t_start + t_size, pattern);
 547
 548                         memtest(t_start, t_size, pattern);
 549
 550                         t_start += t_size;
 551                 }
 552         }
 553         printk(KERN_CONT "\n");
 554 }
 555 #else
 556 static void __init early_memtest(unsigned long start, unsigned long end)
 557 {
 558 }
 559 #endif
 560
 561 /*
 562  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 563  * This runs before bootmem is initialized and gets pages directly from
 564  * the physical memory. To access them they are temporarily mapped.
 565  */
 566 unsigned long __init_refok init_memory_mapping(unsigned long start, unsigned long end)
 567 {
 568         unsigned long next, last_map_addr = end;
 569         unsigned long start_phys = start, end_phys = end;
 570
 571         printk(KERN_INFO "init_memory_mapping\n");
 572
 573         /*
 574          * Find space for the kernel direct mapping tables.
 575          *
 576          * Later we should allocate these tables in the local node of the
 577          * memory mapped. Unfortunately this is done currently before the
 578          * nodes are discovered.
 579          */
 580         if (!after_bootmem) {
 581                 init_gbpages();
 582                 find_early_table_space(end);
 583         }
 584
 585         start = (unsigned long)__va(start);
 586         end = (unsigned long)__va(end);
 587
 588         for (; start < end; start = next) {
 589                 pgd_t *pgd = pgd_offset_k(start);
 590                 unsigned long pud_phys;
 591                 pud_t *pud;
 592
 593                 next = start + PGDIR_SIZE;
 594                 if (next > end)
 595                         next = end;
 596
 597                 if (pgd_val(*pgd)) {
 598                         last_map_addr = phys_pud_update(pgd, __pa(start), __pa(end));
 599                         continue;
 600                 }
 601
 602                 if (after_bootmem)
 603                         pud = pud_offset(pgd, start & PGDIR_MASK);
 604                 else
 605                         pud = alloc_low_page(&pud_phys);
 606
 607                 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next));
 608                 unmap_low_page(pud);
 609                 if (!after_bootmem)
 610                         pgd_populate(&init_mm, pgd_offset_k(start),
 611                                      __va(pud_phys));
 612         }
 613
 614         if (!after_bootmem)
 615                 mmu_cr4_features = read_cr4();
 616         __flush_tlb_all();
 617
 618         if (!after_bootmem)
 619                 reserve_early(table_start << PAGE_SHIFT,
 620                                  table_end << PAGE_SHIFT, "PGTABLE");
 621
 622         if (!after_bootmem)
 623                 early_memtest(start_phys, end_phys);
 624
 625         return last_map_addr >> PAGE_SHIFT;
 626 }
 627
 628 #ifndef CONFIG_NUMA
 629 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn)
 630 {
 631         unsigned long bootmap_size, bootmap;
 632
 633         bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
 634         bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
 635                                  PAGE_SIZE);
 636         if (bootmap == -1L)
 637                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 638         /* don't touch min_low_pfn */
 639         bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
 640                                          0, end_pfn);
 641         e820_register_active_regions(0, start_pfn, end_pfn);
 642         free_bootmem_with_active_regions(0, end_pfn);
 643         early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
 644         reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
 645 }
 646
 647 void __init paging_init(void)
 648 {
 649         unsigned long max_zone_pfns[MAX_NR_ZONES];
 650
 651         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 652         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 653         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 654         max_zone_pfns[ZONE_NORMAL] = max_pfn;
 655
 656         memory_present(0, 0, max_pfn);
 657         sparse_init();
 658         free_area_init_nodes(max_zone_pfns);
 659 }
 660 #endif
 661
 662 /*
 663  * Memory hotplug specific functions
 664  */
 665 #ifdef CONFIG_MEMORY_HOTPLUG
 666 /*
 667  * Memory is added always to NORMAL zone. This means you will never get
 668  * additional DMA/DMA32 memory.
 669  */
 670 int arch_add_memory(int nid, u64 start, u64 size)
 671 {
 672         struct pglist_data *pgdat = NODE_DATA(nid);
 673         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 674         unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
 675         unsigned long nr_pages = size >> PAGE_SHIFT;
 676         int ret;
 677
 678         last_mapped_pfn = init_memory_mapping(start, start + size-1);
 679         if (last_mapped_pfn > max_pfn_mapped)
 680                 max_pfn_mapped = last_mapped_pfn;
 681
 682         ret = __add_pages(zone, start_pfn, nr_pages);
 683         WARN_ON(1);
 684
 685         return ret;
 686 }
 687 EXPORT_SYMBOL_GPL(arch_add_memory);
 688
 689 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
 690 int memory_add_physaddr_to_nid(u64 start)
 691 {
 692         return 0;
 693 }
 694 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 695 #endif
 696
 697 #endif /* CONFIG_MEMORY_HOTPLUG */
 698
 699 /*
 700  * devmem_is_allowed() checks to see if /dev/mem access to a certain address
 701  * is valid. The argument is a physical page number.
 702  *
 703  *
 704  * On x86, access has to be given to the first megabyte of ram because that area
 705  * contains bios code and data regions used by X and dosemu and similar apps.
 706  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
 707  * mmio resources as well as potential bios/acpi data regions.
 708  */
 709 int devmem_is_allowed(unsigned long pagenr)
 710 {
 711         if (pagenr <= 256)
 712                 return 1;
 713         if (!page_is_ram(pagenr))
 714                 return 1;
 715         return 0;
 716 }
 717
 718
 719 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
 720                          kcore_modules, kcore_vsyscall;
 721
 722 void __init mem_init(void)
 723 {
 724         long codesize, reservedpages, datasize, initsize;
 725
 726         pci_iommu_alloc();
 727
 728         /* clear_bss() already clear the empty_zero_page */
 729
 730         reservedpages = 0;
 731
 732         /* this will put all low memory onto the freelists */
 733 #ifdef CONFIG_NUMA
 734         totalram_pages = numa_free_all_bootmem();
 735 #else
 736         totalram_pages = free_all_bootmem();
 737 #endif
 738         reservedpages = max_pfn - totalram_pages -
 739                                         absent_pages_in_range(0, max_pfn);
 740         after_bootmem = 1;
 741
 742         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 743         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 744         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 745
 746         /* Register memory areas for /proc/kcore */
 747         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 748         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 749                    VMALLOC_END-VMALLOC_START);
 750         kclist_add(&kcore_kernel, &_stext, _end - _stext);
 751         kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
 752         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 753                                  VSYSCALL_END - VSYSCALL_START);
 754
 755         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 756                                 "%ldk reserved, %ldk data, %ldk init)\n",
 757                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 758                 max_pfn << (PAGE_SHIFT-10),
 759                 codesize >> 10,
 760                 reservedpages << (PAGE_SHIFT-10),
 761                 datasize >> 10,
 762                 initsize >> 10);
 763
 764         cpa_init();
 765 }
 766
 767 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 768 {
 769         unsigned long addr = begin;
 770
 771         if (addr >= end)
 772                 return;
 773
 774         /*
 775          * If debugging page accesses then do not free this memory but
 776          * mark them not present - any buggy init-section access will
 777          * create a kernel page fault:
 778          */
 779 #ifdef CONFIG_DEBUG_PAGEALLOC
 780         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
 781                 begin, PAGE_ALIGN(end));
 782         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 783 #else
 784         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 785
 786         for (; addr < end; addr += PAGE_SIZE) {
 787                 ClearPageReserved(virt_to_page(addr));
 788                 init_page_count(virt_to_page(addr));
 789                 memset((void *)(addr & ~(PAGE_SIZE-1)),
 790                         POISON_FREE_INITMEM, PAGE_SIZE);
 791                 free_page(addr);
 792                 totalram_pages++;
 793         }
 794 #endif
 795 }
 796
 797 void free_initmem(void)
 798 {
 799         free_init_pages("unused kernel memory",
 800                         (unsigned long)(&__init_begin),
 801                         (unsigned long)(&__init_end));
 802 }
 803
 804 #ifdef CONFIG_DEBUG_RODATA
 805 const int rodata_test_data = 0xC3;
 806 EXPORT_SYMBOL_GPL(rodata_test_data);
 807
 808 void mark_rodata_ro(void)
 809 {
 810         unsigned long start = PFN_ALIGN(_stext), end = PFN_ALIGN(__end_rodata);
 811
 812         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 813                (end - start) >> 10);
 814         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 815
 816         /*
 817          * The rodata section (but not the kernel text!) should also be
 818          * not-executable.
 819          */
 820         start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
 821         set_memory_nx(start, (end - start) >> PAGE_SHIFT);
 822
 823         rodata_test();
 824
 825 #ifdef CONFIG_CPA_DEBUG
 826         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
 827         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 828
 829         printk(KERN_INFO "Testing CPA: again\n");
 830         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 831 #endif
 832 }
 833
 834 #endif
 835
 836 #ifdef CONFIG_BLK_DEV_INITRD
 837 void free_initrd_mem(unsigned long start, unsigned long end)
 838 {
 839         free_init_pages("initrd memory", start, end);
 840 }
 841 #endif
 842
 843 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
 844                                    int flags)
 845 {
 846 #ifdef CONFIG_NUMA
 847         int nid, next_nid;
 848         int ret;
 849 #endif
 850         unsigned long pfn = phys >> PAGE_SHIFT;
 851
 852         if (pfn >= max_pfn) {
 853                 /*
 854                  * This can happen with kdump kernels when accessing
 855                  * firmware tables:
 856                  */
 857                 if (pfn < max_pfn_mapped)
 858                         return -EFAULT;
 859
 860                 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n",
 861                                 phys, len);
 862                 return -EFAULT;
 863         }
 864
 865         /* Should check here against the e820 map to avoid double free */
 866 #ifdef CONFIG_NUMA
 867         nid = phys_to_nid(phys);
 868         next_nid = phys_to_nid(phys + len - 1);
 869         if (nid == next_nid)
 870                 ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags);
 871         else
 872                 ret = reserve_bootmem(phys, len, flags);
 873
 874         if (ret != 0)
 875                 return ret;
 876
 877 #else
 878         reserve_bootmem(phys, len, BOOTMEM_DEFAULT);
 879 #endif
 880
 881         if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
 882                 dma_reserve += len / PAGE_SIZE;
 883                 set_dma_reserve(dma_reserve);
 884         }
 885
 886         return 0;
 887 }
 888
 889 int kern_addr_valid(unsigned long addr)
 890 {
 891         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 892         pgd_t *pgd;
 893         pud_t *pud;
 894         pmd_t *pmd;
 895         pte_t *pte;
 896
 897         if (above != 0 && above != -1UL)
 898                 return 0;
 899
 900         pgd = pgd_offset_k(addr);
 901         if (pgd_none(*pgd))
 902                 return 0;
 903
 904         pud = pud_offset(pgd, addr);
 905         if (pud_none(*pud))
 906                 return 0;
 907
 908         pmd = pmd_offset(pud, addr);
 909         if (pmd_none(*pmd))
 910                 return 0;
 911
 912         if (pmd_large(*pmd))
 913                 return pfn_valid(pmd_pfn(*pmd));
 914
 915         pte = pte_offset_kernel(pmd, addr);
 916         if (pte_none(*pte))
 917                 return 0;
 918
 919         return pfn_valid(pte_pfn(*pte));
 920 }
 921
 922 /*
 923  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 924  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 925  * not need special handling anymore:
 926  */
 927 static struct vm_area_struct gate_vma = {
 928         .vm_start       = VSYSCALL_START,
 929         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 930         .vm_page_prot   = PAGE_READONLY_EXEC,
 931         .vm_flags       = VM_READ | VM_EXEC
 932 };
 933
 934 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 935 {
 936 #ifdef CONFIG_IA32_EMULATION
 937         if (test_tsk_thread_flag(tsk, TIF_IA32))
 938                 return NULL;
 939 #endif
 940         return &gate_vma;
 941 }
 942
 943 int in_gate_area(struct task_struct *task, unsigned long addr)
 944 {
 945         struct vm_area_struct *vma = get_gate_vma(task);
 946
 947         if (!vma)
 948                 return 0;
 949
 950         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 951 }
 952
 953 /*
 954  * Use this when you have no reliable task/vma, typically from interrupt
 955  * context. It is less reliable than using the task's vma and may give
 956  * false positives:
 957  */
 958 int in_gate_area_no_task(unsigned long addr)
 959 {
 960         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 961 }
 962
 963 const char *arch_vma_name(struct vm_area_struct *vma)
 964 {
 965         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 966                 return "[vdso]";
 967         if (vma == &gate_vma)
 968                 return "[vsyscall]";
 969         return NULL;
 970 }
 971
 972 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 973 /*
 974  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 975  */
 976 static long __meminitdata addr_start, addr_end;
 977 static void __meminitdata *p_start, *p_end;
 978 static int __meminitdata node_start;
 979
 980 int __meminit
 981 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 982 {
 983         unsigned long addr = (unsigned long)start_page;
 984         unsigned long end = (unsigned long)(start_page + size);
 985         unsigned long next;
 986         pgd_t *pgd;
 987         pud_t *pud;
 988         pmd_t *pmd;
 989
 990         for (; addr < end; addr = next) {
 991                 void *p = NULL;
 992
 993                 pgd = vmemmap_pgd_populate(addr, node);
 994                 if (!pgd)
 995                         return -ENOMEM;
 996
 997                 pud = vmemmap_pud_populate(pgd, addr, node);
 998                 if (!pud)
 999                         return -ENOMEM;
1000
1001                 if (!cpu_has_pse) {
1002                         next = (addr + PAGE_SIZE) & PAGE_MASK;
1003                         pmd = vmemmap_pmd_populate(pud, addr, node);
1004
1005                         if (!pmd)
1006                                 return -ENOMEM;
1007
1008                         p = vmemmap_pte_populate(pmd, addr, node);
1009
1010                         if (!p)
1011                                 return -ENOMEM;
1012
1013                         addr_end = addr + PAGE_SIZE;
1014                         p_end = p + PAGE_SIZE;
1015                 } else {
1016                         next = pmd_addr_end(addr, end);
1017
1018                         pmd = pmd_offset(pud, addr);
1019                         if (pmd_none(*pmd)) {
1020                                 pte_t entry;
1021
1022                                 p = vmemmap_alloc_block(PMD_SIZE, node);
1023                                 if (!p)
1024                                         return -ENOMEM;
1025
1026                                 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1027                                                 PAGE_KERNEL_LARGE);
1028                                 set_pmd(pmd, __pmd(pte_val(entry)));
1029
1030                                 addr_end = addr + PMD_SIZE;
1031                                 p_end = p + PMD_SIZE;
1032
1033                                 /* check to see if we have contiguous blocks */
1034                                 if (p_end != p || node_start != node) {
1035                                         if (p_start)
1036                                                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1037                                                        addr_start, addr_end-1, p_start, p_end-1, node_start);
1038                                         addr_start = addr;
1039                                         node_start = node;
1040                                         p_start = p;
1041                                 }
1042                         } else
1043                                 vmemmap_verify((pte_t *)pmd, node, addr, next);
1044                 }
1045
1046         }
1047         return 0;
1048 }
1049
1050 void __meminit vmemmap_populate_print_last(void)
1051 {
1052         if (p_start) {
1053                 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1054                         addr_start, addr_end-1, p_start, p_end-1, node_start);
1055                 p_start = NULL;
1056                 p_end = NULL;
1057                 node_start = 0;
1058         }
1059 }
1060 #endif