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