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 __init 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) &&
 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 __init
 218 __set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
 219 {
 220         unsigned long address = __fix_to_virt(idx);
 221
 222         if (idx >= __end_of_fixed_addresses) {
 223                 printk(KERN_ERR "Invalid __set_fixmap\n");
 224                 return;
 225         }
 226         set_pte_phys(address, phys, prot);
 227 }
 228
 229 static unsigned long __initdata table_start;
 230 static unsigned long __meminitdata table_end;
 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 >= end_pfn)
 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 /* Must run before zap_low_mappings */
 262 __meminit void *early_ioremap(unsigned long addr, unsigned long size)
 263 {
 264         pmd_t *pmd, *last_pmd;
 265         unsigned long vaddr;
 266         int i, pmds;
 267
 268         pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 269         vaddr = __START_KERNEL_map;
 270         pmd = level2_kernel_pgt;
 271         last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1;
 272
 273         for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) {
 274                 for (i = 0; i < pmds; i++) {
 275                         if (pmd_present(pmd[i]))
 276                                 goto continue_outer_loop;
 277                 }
 278                 vaddr += addr & ~PMD_MASK;
 279                 addr &= PMD_MASK;
 280
 281                 for (i = 0; i < pmds; i++, addr += PMD_SIZE)
 282                         set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
 283                 __flush_tlb_all();
 284
 285                 return (void *)vaddr;
 286 continue_outer_loop:
 287                 ;
 288         }
 289         printk(KERN_ERR "early_ioremap(0x%lx, %lu) failed\n", addr, size);
 290
 291         return NULL;
 292 }
 293
 294 /*
 295  * To avoid virtual aliases later:
 296  */
 297 __meminit void early_iounmap(void *addr, unsigned long size)
 298 {
 299         unsigned long vaddr;
 300         pmd_t *pmd;
 301         int i, pmds;
 302
 303         vaddr = (unsigned long)addr;
 304         pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
 305         pmd = level2_kernel_pgt + pmd_index(vaddr);
 306
 307         for (i = 0; i < pmds; i++)
 308                 pmd_clear(pmd + i);
 309
 310         __flush_tlb_all();
 311 }
 312
 313 static unsigned long __meminit
 314 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
 315 {
 316         int i = pmd_index(address);
 317
 318         for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
 319                 pmd_t *pmd = pmd_page + pmd_index(address);
 320
 321                 if (address >= end) {
 322                         if (!after_bootmem) {
 323                                 for (; i < PTRS_PER_PMD; i++, pmd++)
 324                                         set_pmd(pmd, __pmd(0));
 325                         }
 326                         break;
 327                 }
 328
 329                 if (pmd_val(*pmd))
 330                         continue;
 331
 332                 set_pte((pte_t *)pmd,
 333                         pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 334         }
 335         return address;
 336 }
 337
 338 static unsigned long __meminit
 339 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
 340 {
 341         pmd_t *pmd = pmd_offset(pud, 0);
 342         unsigned long last_map_addr;
 343
 344         spin_lock(&init_mm.page_table_lock);
 345         last_map_addr = phys_pmd_init(pmd, address, end);
 346         spin_unlock(&init_mm.page_table_lock);
 347         __flush_tlb_all();
 348         return last_map_addr;
 349 }
 350
 351 static unsigned long __meminit
 352 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
 353 {
 354         unsigned long last_map_addr = end;
 355         int i = pud_index(addr);
 356
 357         for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
 358                 unsigned long pmd_phys;
 359                 pud_t *pud = pud_page + pud_index(addr);
 360                 pmd_t *pmd;
 361
 362                 if (addr >= end)
 363                         break;
 364
 365                 if (!after_bootmem &&
 366                                 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
 367                         set_pud(pud, __pud(0));
 368                         continue;
 369                 }
 370
 371                 if (pud_val(*pud)) {
 372                         if (!pud_large(*pud))
 373                                 last_map_addr = phys_pmd_update(pud, addr, end);
 374                         continue;
 375                 }
 376
 377                 if (direct_gbpages) {
 378                         set_pte((pte_t *)pud,
 379                                 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
 380                         last_map_addr = (addr & PUD_MASK) + PUD_SIZE;
 381                         continue;
 382                 }
 383
 384                 pmd = alloc_low_page(&pmd_phys);
 385
 386                 spin_lock(&init_mm.page_table_lock);
 387                 set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE));
 388                 last_map_addr = phys_pmd_init(pmd, addr, end);
 389                 spin_unlock(&init_mm.page_table_lock);
 390
 391                 unmap_low_page(pmd);
 392         }
 393         __flush_tlb_all();
 394
 395         return last_map_addr >> PAGE_SHIFT;
 396 }
 397
 398 static void __init find_early_table_space(unsigned long end)
 399 {
 400         unsigned long puds, pmds, tables, start;
 401
 402         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 403         tables = round_up(puds * sizeof(pud_t), PAGE_SIZE);
 404         if (!direct_gbpages) {
 405                 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 406                 tables += round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
 407         }
 408
 409         /*
 410          * RED-PEN putting page tables only on node 0 could
 411          * cause a hotspot and fill up ZONE_DMA. The page tables
 412          * need roughly 0.5KB per GB.
 413          */
 414         start = 0x8000;
 415         table_start = find_e820_area(start, end, tables, PAGE_SIZE);
 416         if (table_start == -1UL)
 417                 panic("Cannot find space for the kernel page tables");
 418
 419         table_start >>= PAGE_SHIFT;
 420         table_end = table_start;
 421
 422         early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n",
 423                 end, table_start << PAGE_SHIFT,
 424                 (table_start << PAGE_SHIFT) + tables);
 425 }
 426
 427 static void __init init_gbpages(void)
 428 {
 429         if (direct_gbpages && cpu_has_gbpages)
 430                 printk(KERN_INFO "Using GB pages for direct mapping\n");
 431         else
 432                 direct_gbpages = 0;
 433 }
 434
 435 #ifdef CONFIG_MEMTEST_BOOTPARAM
 436
 437 static void __init memtest(unsigned long start_phys, unsigned long size,
 438                                  unsigned pattern)
 439 {
 440         unsigned long i;
 441         unsigned long *start;
 442         unsigned long start_bad;
 443         unsigned long last_bad;
 444         unsigned long val;
 445         unsigned long start_phys_aligned;
 446         unsigned long count;
 447         unsigned long incr;
 448
 449         switch (pattern) {
 450         case 0:
 451                 val = 0UL;
 452                 break;
 453         case 1:
 454                 val = -1UL;
 455                 break;
 456         case 2:
 457                 val = 0x5555555555555555UL;
 458                 break;
 459         case 3:
 460                 val = 0xaaaaaaaaaaaaaaaaUL;
 461                 break;
 462         default:
 463                 return;
 464         }
 465
 466         incr = sizeof(unsigned long);
 467         start_phys_aligned = ALIGN(start_phys, incr);
 468         count = (size - (start_phys_aligned - start_phys))/incr;
 469         start = __va(start_phys_aligned);
 470         start_bad = 0;
 471         last_bad = 0;
 472
 473         for (i = 0; i < count; i++)
 474                 start[i] = val;
 475         for (i = 0; i < count; i++, start++, start_phys_aligned += incr) {
 476                 if (*start != val) {
 477                         if (start_phys_aligned == last_bad + incr) {
 478                                 last_bad += incr;
 479                         } else {
 480                                 if (start_bad) {
 481                                         printk(KERN_CONT "\n  %016lx bad mem addr %016lx - %016lx reserved",
 482                                                 val, start_bad, last_bad + incr);
 483                                         reserve_early(start_bad, last_bad - start_bad, "BAD RAM");
 484                                 }
 485                                 start_bad = last_bad = start_phys_aligned;
 486                         }
 487                 }
 488         }
 489         if (start_bad) {
 490                 printk(KERN_CONT "\n  %016lx bad mem addr %016lx - %016lx reserved",
 491                         val, start_bad, last_bad + incr);
 492                 reserve_early(start_bad, last_bad - start_bad, "BAD RAM");
 493         }
 494
 495 }
 496
 497 static int memtest_pattern __initdata = CONFIG_MEMTEST_BOOTPARAM_VALUE;
 498
 499 static int __init parse_memtest(char *arg)
 500 {
 501         if (arg)
 502                 memtest_pattern = simple_strtoul(arg, NULL, 0);
 503         return 0;
 504 }
 505
 506 early_param("memtest", parse_memtest);
 507
 508 static void __init early_memtest(unsigned long start, unsigned long end)
 509 {
 510         unsigned long t_start, t_size;
 511         unsigned pattern;
 512
 513         if (!memtest_pattern)
 514                 return;
 515
 516         printk(KERN_INFO "early_memtest: pattern num %d", memtest_pattern);
 517         for (pattern = 0; pattern < memtest_pattern; pattern++) {
 518                 t_start = start;
 519                 t_size = 0;
 520                 while (t_start < end) {
 521                         t_start = find_e820_area_size(t_start, &t_size, 1);
 522
 523                         /* done ? */
 524                         if (t_start >= end)
 525                                 break;
 526                         if (t_start + t_size > end)
 527                                 t_size = end - t_start;
 528
 529                         printk(KERN_CONT "\n  %016lx - %016lx pattern %d",
 530                                 t_start, t_start + t_size, pattern);
 531
 532                         memtest(t_start, t_size, pattern);
 533
 534                         t_start += t_size;
 535                 }
 536         }
 537         printk(KERN_CONT "\n");
 538 }
 539 #else
 540 static void __init early_memtest(unsigned long start, unsigned long end)
 541 {
 542 }
 543 #endif
 544
 545 /*
 546  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 547  * This runs before bootmem is initialized and gets pages directly from
 548  * the physical memory. To access them they are temporarily mapped.
 549  */
 550 unsigned long __init_refok init_memory_mapping(unsigned long start, unsigned long end)
 551 {
 552         unsigned long next, last_map_addr = end;
 553         unsigned long start_phys = start, end_phys = end;
 554
 555         printk(KERN_INFO "init_memory_mapping\n");
 556
 557         /*
 558          * Find space for the kernel direct mapping tables.
 559          *
 560          * Later we should allocate these tables in the local node of the
 561          * memory mapped. Unfortunately this is done currently before the
 562          * nodes are discovered.
 563          */
 564         if (!after_bootmem) {
 565                 init_gbpages();
 566                 find_early_table_space(end);
 567         }
 568
 569         start = (unsigned long)__va(start);
 570         end = (unsigned long)__va(end);
 571
 572         for (; start < end; start = next) {
 573                 pgd_t *pgd = pgd_offset_k(start);
 574                 unsigned long pud_phys;
 575                 pud_t *pud;
 576
 577                 if (after_bootmem)
 578                         pud = pud_offset(pgd, start & PGDIR_MASK);
 579                 else
 580                         pud = alloc_low_page(&pud_phys);
 581
 582                 next = start + PGDIR_SIZE;
 583                 if (next > end)
 584                         next = end;
 585                 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next));
 586                 if (!after_bootmem)
 587                         set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys));
 588                 unmap_low_page(pud);
 589         }
 590
 591         if (!after_bootmem)
 592                 mmu_cr4_features = read_cr4();
 593         __flush_tlb_all();
 594
 595         if (!after_bootmem)
 596                 reserve_early(table_start << PAGE_SHIFT,
 597                                  table_end << PAGE_SHIFT, "PGTABLE");
 598
 599         if (!after_bootmem)
 600                 early_memtest(start_phys, end_phys);
 601
 602         return last_map_addr;
 603 }
 604
 605 #ifndef CONFIG_NUMA
 606 void __init paging_init(void)
 607 {
 608         unsigned long max_zone_pfns[MAX_NR_ZONES];
 609
 610         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 611         max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 612         max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
 613         max_zone_pfns[ZONE_NORMAL] = end_pfn;
 614
 615         memory_present(0, 0, end_pfn);
 616         sparse_init();
 617         free_area_init_nodes(max_zone_pfns);
 618 }
 619 #endif
 620
 621 /*
 622  * Memory hotplug specific functions
 623  */
 624 void online_page(struct page *page)
 625 {
 626         ClearPageReserved(page);
 627         init_page_count(page);
 628         __free_page(page);
 629         totalram_pages++;
 630         num_physpages++;
 631 }
 632
 633 #ifdef CONFIG_MEMORY_HOTPLUG
 634 /*
 635  * Memory is added always to NORMAL zone. This means you will never get
 636  * additional DMA/DMA32 memory.
 637  */
 638 int arch_add_memory(int nid, u64 start, u64 size)
 639 {
 640         struct pglist_data *pgdat = NODE_DATA(nid);
 641         struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
 642         unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT;
 643         unsigned long nr_pages = size >> PAGE_SHIFT;
 644         int ret;
 645
 646         last_mapped_pfn = init_memory_mapping(start, start + size-1);
 647         if (last_mapped_pfn > max_pfn_mapped)
 648                 max_pfn_mapped = last_mapped_pfn;
 649
 650         ret = __add_pages(zone, start_pfn, nr_pages);
 651         WARN_ON(1);
 652
 653         return ret;
 654 }
 655 EXPORT_SYMBOL_GPL(arch_add_memory);
 656
 657 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
 658 int memory_add_physaddr_to_nid(u64 start)
 659 {
 660         return 0;
 661 }
 662 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
 663 #endif
 664
 665 #endif /* CONFIG_MEMORY_HOTPLUG */
 666
 667 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
 668                          kcore_modules, kcore_vsyscall;
 669
 670 void __init mem_init(void)
 671 {
 672         long codesize, reservedpages, datasize, initsize;
 673
 674         pci_iommu_alloc();
 675
 676         /* clear_bss() already clear the empty_zero_page */
 677
 678         reservedpages = 0;
 679
 680         /* this will put all low memory onto the freelists */
 681 #ifdef CONFIG_NUMA
 682         totalram_pages = numa_free_all_bootmem();
 683 #else
 684         totalram_pages = free_all_bootmem();
 685 #endif
 686         reservedpages = end_pfn - totalram_pages -
 687                                         absent_pages_in_range(0, end_pfn);
 688         after_bootmem = 1;
 689
 690         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 691         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 692         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 693
 694         /* Register memory areas for /proc/kcore */
 695         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 696         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 697                    VMALLOC_END-VMALLOC_START);
 698         kclist_add(&kcore_kernel, &_stext, _end - _stext);
 699         kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
 700         kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
 701                                  VSYSCALL_END - VSYSCALL_START);
 702
 703         printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
 704                                 "%ldk reserved, %ldk data, %ldk init)\n",
 705                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 706                 end_pfn << (PAGE_SHIFT-10),
 707                 codesize >> 10,
 708                 reservedpages << (PAGE_SHIFT-10),
 709                 datasize >> 10,
 710                 initsize >> 10);
 711
 712         cpa_init();
 713 }
 714
 715 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 716 {
 717         unsigned long addr = begin;
 718
 719         if (addr >= end)
 720                 return;
 721
 722         /*
 723          * If debugging page accesses then do not free this memory but
 724          * mark them not present - any buggy init-section access will
 725          * create a kernel page fault:
 726          */
 727 #ifdef CONFIG_DEBUG_PAGEALLOC
 728         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
 729                 begin, PAGE_ALIGN(end));
 730         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 731 #else
 732         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 733
 734         for (; addr < end; addr += PAGE_SIZE) {
 735                 ClearPageReserved(virt_to_page(addr));
 736                 init_page_count(virt_to_page(addr));
 737                 memset((void *)(addr & ~(PAGE_SIZE-1)),
 738                         POISON_FREE_INITMEM, PAGE_SIZE);
 739                 free_page(addr);
 740                 totalram_pages++;
 741         }
 742 #endif
 743 }
 744
 745 void free_initmem(void)
 746 {
 747         free_init_pages("unused kernel memory",
 748                         (unsigned long)(&__init_begin),
 749                         (unsigned long)(&__init_end));
 750 }
 751
 752 #ifdef CONFIG_DEBUG_RODATA
 753 const int rodata_test_data = 0xC3;
 754 EXPORT_SYMBOL_GPL(rodata_test_data);
 755
 756 void mark_rodata_ro(void)
 757 {
 758         unsigned long start = PFN_ALIGN(_stext), end = PFN_ALIGN(__end_rodata);
 759
 760         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 761                (end - start) >> 10);
 762         set_memory_ro(start, (end - start) >> PAGE_SHIFT);
 763
 764         /*
 765          * The rodata section (but not the kernel text!) should also be
 766          * not-executable.
 767          */
 768         start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
 769         set_memory_nx(start, (end - start) >> PAGE_SHIFT);
 770
 771         rodata_test();
 772
 773 #ifdef CONFIG_CPA_DEBUG
 774         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
 775         set_memory_rw(start, (end-start) >> PAGE_SHIFT);
 776
 777         printk(KERN_INFO "Testing CPA: again\n");
 778         set_memory_ro(start, (end-start) >> PAGE_SHIFT);
 779 #endif
 780 }
 781
 782 #endif
 783
 784 #ifdef CONFIG_BLK_DEV_INITRD
 785 void free_initrd_mem(unsigned long start, unsigned long end)
 786 {
 787         free_init_pages("initrd memory", start, end);
 788 }
 789 #endif
 790
 791 void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
 792 {
 793 #ifdef CONFIG_NUMA
 794         int nid = phys_to_nid(phys);
 795 #endif
 796         unsigned long pfn = phys >> PAGE_SHIFT;
 797
 798         if (pfn >= end_pfn) {
 799                 /*
 800                  * This can happen with kdump kernels when accessing
 801                  * firmware tables:
 802                  */
 803                 if (pfn < max_pfn_mapped)
 804                         return;
 805
 806                 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n",
 807                                 phys, len);
 808                 return;
 809         }
 810
 811         /* Should check here against the e820 map to avoid double free */
 812 #ifdef CONFIG_NUMA
 813         reserve_bootmem_node(NODE_DATA(nid), phys, len, BOOTMEM_DEFAULT);
 814 #else
 815         reserve_bootmem(phys, len, BOOTMEM_DEFAULT);
 816 #endif
 817         if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
 818                 dma_reserve += len / PAGE_SIZE;
 819                 set_dma_reserve(dma_reserve);
 820         }
 821 }
 822
 823 int kern_addr_valid(unsigned long addr)
 824 {
 825         unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
 826         pgd_t *pgd;
 827         pud_t *pud;
 828         pmd_t *pmd;
 829         pte_t *pte;
 830
 831         if (above != 0 && above != -1UL)
 832                 return 0;
 833
 834         pgd = pgd_offset_k(addr);
 835         if (pgd_none(*pgd))
 836                 return 0;
 837
 838         pud = pud_offset(pgd, addr);
 839         if (pud_none(*pud))
 840                 return 0;
 841
 842         pmd = pmd_offset(pud, addr);
 843         if (pmd_none(*pmd))
 844                 return 0;
 845
 846         if (pmd_large(*pmd))
 847                 return pfn_valid(pmd_pfn(*pmd));
 848
 849         pte = pte_offset_kernel(pmd, addr);
 850         if (pte_none(*pte))
 851                 return 0;
 852
 853         return pfn_valid(pte_pfn(*pte));
 854 }
 855
 856 /*
 857  * A pseudo VMA to allow ptrace access for the vsyscall page.  This only
 858  * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
 859  * not need special handling anymore:
 860  */
 861 static struct vm_area_struct gate_vma = {
 862         .vm_start       = VSYSCALL_START,
 863         .vm_end         = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
 864         .vm_page_prot   = PAGE_READONLY_EXEC,
 865         .vm_flags       = VM_READ | VM_EXEC
 866 };
 867
 868 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 869 {
 870 #ifdef CONFIG_IA32_EMULATION
 871         if (test_tsk_thread_flag(tsk, TIF_IA32))
 872                 return NULL;
 873 #endif
 874         return &gate_vma;
 875 }
 876
 877 int in_gate_area(struct task_struct *task, unsigned long addr)
 878 {
 879         struct vm_area_struct *vma = get_gate_vma(task);
 880
 881         if (!vma)
 882                 return 0;
 883
 884         return (addr >= vma->vm_start) && (addr < vma->vm_end);
 885 }
 886
 887 /*
 888  * Use this when you have no reliable task/vma, typically from interrupt
 889  * context. It is less reliable than using the task's vma and may give
 890  * false positives:
 891  */
 892 int in_gate_area_no_task(unsigned long addr)
 893 {
 894         return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 895 }
 896
 897 const char *arch_vma_name(struct vm_area_struct *vma)
 898 {
 899         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 900                 return "[vdso]";
 901         if (vma == &gate_vma)
 902                 return "[vsyscall]";
 903         return NULL;
 904 }
 905
 906 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 907 /*
 908  * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
 909  */
 910 int __meminit
 911 vmemmap_populate(struct page *start_page, unsigned long size, int node)
 912 {
 913         unsigned long addr = (unsigned long)start_page;
 914         unsigned long end = (unsigned long)(start_page + size);
 915         unsigned long next;
 916         pgd_t *pgd;
 917         pud_t *pud;
 918         pmd_t *pmd;
 919
 920         for (; addr < end; addr = next) {
 921                 next = pmd_addr_end(addr, end);
 922
 923                 pgd = vmemmap_pgd_populate(addr, node);
 924                 if (!pgd)
 925                         return -ENOMEM;
 926
 927                 pud = vmemmap_pud_populate(pgd, addr, node);
 928                 if (!pud)
 929                         return -ENOMEM;
 930
 931                 pmd = pmd_offset(pud, addr);
 932                 if (pmd_none(*pmd)) {
 933                         pte_t entry;
 934                         void *p;
 935
 936                         p = vmemmap_alloc_block(PMD_SIZE, node);
 937                         if (!p)
 938                                 return -ENOMEM;
 939
 940                         entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
 941                                                         PAGE_KERNEL_LARGE);
 942                         set_pmd(pmd, __pmd(pte_val(entry)));
 943
 944                         printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n",
 945                                 addr, addr + PMD_SIZE - 1, p, node);
 946                 } else {
 947                         vmemmap_verify((pte_t *)pmd, node, addr, next);
 948                 }
 949         }
 950         return 0;
 951 }
 952 #endif