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