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