init_64.c revision f89112502805c1f6a6955f90ad158e538edb319d
1/* 2 * linux/arch/x86_64/mm/init.c 3 * 4 * Copyright (C) 1995 Linus Torvalds 5 * Copyright (C) 2000 Pavel Machek <pavel@ucw.cz> 6 * Copyright (C) 2002,2003 Andi Kleen <ak@suse.de> 7 */ 8 9#include <linux/signal.h> 10#include <linux/sched.h> 11#include <linux/kernel.h> 12#include <linux/errno.h> 13#include <linux/string.h> 14#include <linux/types.h> 15#include <linux/ptrace.h> 16#include <linux/mman.h> 17#include <linux/mm.h> 18#include <linux/swap.h> 19#include <linux/smp.h> 20#include <linux/init.h> 21#include <linux/initrd.h> 22#include <linux/pagemap.h> 23#include <linux/bootmem.h> 24#include <linux/memblock.h> 25#include <linux/proc_fs.h> 26#include <linux/pci.h> 27#include <linux/pfn.h> 28#include <linux/poison.h> 29#include <linux/dma-mapping.h> 30#include <linux/module.h> 31#include <linux/memory_hotplug.h> 32#include <linux/nmi.h> 33#include <linux/gfp.h> 34 35#include <asm/processor.h> 36#include <asm/bios_ebda.h> 37#include <asm/system.h> 38#include <asm/uaccess.h> 39#include <asm/pgtable.h> 40#include <asm/pgalloc.h> 41#include <asm/dma.h> 42#include <asm/fixmap.h> 43#include <asm/e820.h> 44#include <asm/apic.h> 45#include <asm/tlb.h> 46#include <asm/mmu_context.h> 47#include <asm/proto.h> 48#include <asm/smp.h> 49#include <asm/sections.h> 50#include <asm/kdebug.h> 51#include <asm/numa.h> 52#include <asm/cacheflush.h> 53#include <asm/init.h> 54 55static int __init parse_direct_gbpages_off(char *arg) 56{ 57 direct_gbpages = 0; 58 return 0; 59} 60early_param("nogbpages", parse_direct_gbpages_off); 61 62static int __init parse_direct_gbpages_on(char *arg) 63{ 64 direct_gbpages = 1; 65 return 0; 66} 67early_param("gbpages", parse_direct_gbpages_on); 68 69/* 70 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the 71 * physical space so we can cache the place of the first one and move 72 * around without checking the pgd every time. 73 */ 74 75pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP; 76EXPORT_SYMBOL_GPL(__supported_pte_mask); 77 78int force_personality32; 79 80/* 81 * noexec32=on|off 82 * Control non executable heap for 32bit processes. 83 * To control the stack too use noexec=off 84 * 85 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default) 86 * off PROT_READ implies PROT_EXEC 87 */ 88static int __init nonx32_setup(char *str) 89{ 90 if (!strcmp(str, "on")) 91 force_personality32 &= ~READ_IMPLIES_EXEC; 92 else if (!strcmp(str, "off")) 93 force_personality32 |= READ_IMPLIES_EXEC; 94 return 1; 95} 96__setup("noexec32=", nonx32_setup); 97 98/* 99 * When memory was added/removed make sure all the processes MM have 100 * suitable PGD entries in the local PGD level page. 101 */ 102void sync_global_pgds(unsigned long start, unsigned long end) 103{ 104 unsigned long address; 105 106 for (address = start; address <= end; address += PGDIR_SIZE) { 107 const pgd_t *pgd_ref = pgd_offset_k(address); 108 unsigned long flags; 109 struct page *page; 110 111 if (pgd_none(*pgd_ref)) 112 continue; 113 114 spin_lock_irqsave(&pgd_lock, flags); 115 list_for_each_entry(page, &pgd_list, lru) { 116 pgd_t *pgd; 117 spinlock_t *pgt_lock; 118 119 pgd = (pgd_t *)page_address(page) + pgd_index(address); 120 pgt_lock = &pgd_page_get_mm(page)->page_table_lock; 121 spin_lock(pgt_lock); 122 123 if (pgd_none(*pgd)) 124 set_pgd(pgd, *pgd_ref); 125 else 126 BUG_ON(pgd_page_vaddr(*pgd) 127 != pgd_page_vaddr(*pgd_ref)); 128 129 spin_unlock(pgt_lock); 130 } 131 spin_unlock_irqrestore(&pgd_lock, flags); 132 } 133} 134 135/* 136 * NOTE: This function is marked __ref because it calls __init function 137 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0. 138 */ 139static __ref void *spp_getpage(void) 140{ 141 void *ptr; 142 143 if (after_bootmem) 144 ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK); 145 else 146 ptr = alloc_bootmem_pages(PAGE_SIZE); 147 148 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) { 149 panic("set_pte_phys: cannot allocate page data %s\n", 150 after_bootmem ? "after bootmem" : ""); 151 } 152 153 pr_debug("spp_getpage %p\n", ptr); 154 155 return ptr; 156} 157 158static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr) 159{ 160 if (pgd_none(*pgd)) { 161 pud_t *pud = (pud_t *)spp_getpage(); 162 pgd_populate(&init_mm, pgd, pud); 163 if (pud != pud_offset(pgd, 0)) 164 printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n", 165 pud, pud_offset(pgd, 0)); 166 } 167 return pud_offset(pgd, vaddr); 168} 169 170static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr) 171{ 172 if (pud_none(*pud)) { 173 pmd_t *pmd = (pmd_t *) spp_getpage(); 174 pud_populate(&init_mm, pud, pmd); 175 if (pmd != pmd_offset(pud, 0)) 176 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n", 177 pmd, pmd_offset(pud, 0)); 178 } 179 return pmd_offset(pud, vaddr); 180} 181 182static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr) 183{ 184 if (pmd_none(*pmd)) { 185 pte_t *pte = (pte_t *) spp_getpage(); 186 pmd_populate_kernel(&init_mm, pmd, pte); 187 if (pte != pte_offset_kernel(pmd, 0)) 188 printk(KERN_ERR "PAGETABLE BUG #02!\n"); 189 } 190 return pte_offset_kernel(pmd, vaddr); 191} 192 193void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte) 194{ 195 pud_t *pud; 196 pmd_t *pmd; 197 pte_t *pte; 198 199 pud = pud_page + pud_index(vaddr); 200 pmd = fill_pmd(pud, vaddr); 201 pte = fill_pte(pmd, vaddr); 202 203 set_pte(pte, new_pte); 204 205 /* 206 * It's enough to flush this one mapping. 207 * (PGE mappings get flushed as well) 208 */ 209 __flush_tlb_one(vaddr); 210} 211 212void set_pte_vaddr(unsigned long vaddr, pte_t pteval) 213{ 214 pgd_t *pgd; 215 pud_t *pud_page; 216 217 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval)); 218 219 pgd = pgd_offset_k(vaddr); 220 if (pgd_none(*pgd)) { 221 printk(KERN_ERR 222 "PGD FIXMAP MISSING, it should be setup in head.S!\n"); 223 return; 224 } 225 pud_page = (pud_t*)pgd_page_vaddr(*pgd); 226 set_pte_vaddr_pud(pud_page, vaddr, pteval); 227} 228 229pmd_t * __init populate_extra_pmd(unsigned long vaddr) 230{ 231 pgd_t *pgd; 232 pud_t *pud; 233 234 pgd = pgd_offset_k(vaddr); 235 pud = fill_pud(pgd, vaddr); 236 return fill_pmd(pud, vaddr); 237} 238 239pte_t * __init populate_extra_pte(unsigned long vaddr) 240{ 241 pmd_t *pmd; 242 243 pmd = populate_extra_pmd(vaddr); 244 return fill_pte(pmd, vaddr); 245} 246 247/* 248 * Create large page table mappings for a range of physical addresses. 249 */ 250static void __init __init_extra_mapping(unsigned long phys, unsigned long size, 251 pgprot_t prot) 252{ 253 pgd_t *pgd; 254 pud_t *pud; 255 pmd_t *pmd; 256 257 BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK)); 258 for (; size; phys += PMD_SIZE, size -= PMD_SIZE) { 259 pgd = pgd_offset_k((unsigned long)__va(phys)); 260 if (pgd_none(*pgd)) { 261 pud = (pud_t *) spp_getpage(); 262 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE | 263 _PAGE_USER)); 264 } 265 pud = pud_offset(pgd, (unsigned long)__va(phys)); 266 if (pud_none(*pud)) { 267 pmd = (pmd_t *) spp_getpage(); 268 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | 269 _PAGE_USER)); 270 } 271 pmd = pmd_offset(pud, phys); 272 BUG_ON(!pmd_none(*pmd)); 273 set_pmd(pmd, __pmd(phys | pgprot_val(prot))); 274 } 275} 276 277void __init init_extra_mapping_wb(unsigned long phys, unsigned long size) 278{ 279 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE); 280} 281 282void __init init_extra_mapping_uc(unsigned long phys, unsigned long size) 283{ 284 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE); 285} 286 287/* 288 * The head.S code sets up the kernel high mapping: 289 * 290 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text) 291 * 292 * phys_addr holds the negative offset to the kernel, which is added 293 * to the compile time generated pmds. This results in invalid pmds up 294 * to the point where we hit the physaddr 0 mapping. 295 * 296 * We limit the mappings to the region from _text to _end. _end is 297 * rounded up to the 2MB boundary. This catches the invalid pmds as 298 * well, as they are located before _text: 299 */ 300void __init cleanup_highmap(void) 301{ 302 unsigned long vaddr = __START_KERNEL_map; 303 unsigned long end = roundup((unsigned long)_end, PMD_SIZE) - 1; 304 pmd_t *pmd = level2_kernel_pgt; 305 pmd_t *last_pmd = pmd + PTRS_PER_PMD; 306 307 for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) { 308 if (pmd_none(*pmd)) 309 continue; 310 if (vaddr < (unsigned long) _text || vaddr > end) 311 set_pmd(pmd, __pmd(0)); 312 } 313} 314 315static __ref void *alloc_low_page(unsigned long *phys) 316{ 317 unsigned long pfn = pgt_buf_end++; 318 void *adr; 319 320 if (after_bootmem) { 321 adr = (void *)get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK); 322 *phys = __pa(adr); 323 324 return adr; 325 } 326 327 if (pfn >= pgt_buf_top) 328 panic("alloc_low_page: ran out of memory"); 329 330 adr = early_memremap(pfn * PAGE_SIZE, PAGE_SIZE); 331 clear_page(adr); 332 *phys = pfn * PAGE_SIZE; 333 return adr; 334} 335 336static __ref void *map_low_page(void *virt) 337{ 338 void *adr; 339 unsigned long phys, left; 340 341 if (after_bootmem) 342 return virt; 343 344 phys = __pa(virt); 345 left = phys & (PAGE_SIZE - 1); 346 adr = early_memremap(phys & PAGE_MASK, PAGE_SIZE); 347 adr = (void *)(((unsigned long)adr) | left); 348 349 return adr; 350} 351 352static __ref void unmap_low_page(void *adr) 353{ 354 if (after_bootmem) 355 return; 356 357 early_iounmap((void *)((unsigned long)adr & PAGE_MASK), PAGE_SIZE); 358} 359 360static unsigned long __meminit 361phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end, 362 pgprot_t prot) 363{ 364 unsigned pages = 0; 365 unsigned long last_map_addr = end; 366 int i; 367 368 pte_t *pte = pte_page + pte_index(addr); 369 370 for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) { 371 372 if (addr >= end) { 373 if (!after_bootmem) { 374 for(; i < PTRS_PER_PTE; i++, pte++) 375 set_pte(pte, __pte(0)); 376 } 377 break; 378 } 379 380 /* 381 * We will re-use the existing mapping. 382 * Xen for example has some special requirements, like mapping 383 * pagetable pages as RO. So assume someone who pre-setup 384 * these mappings are more intelligent. 385 */ 386 if (pte_val(*pte)) { 387 pages++; 388 continue; 389 } 390 391 if (0) 392 printk(" pte=%p addr=%lx pte=%016lx\n", 393 pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte); 394 pages++; 395 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot)); 396 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE; 397 } 398 399 update_page_count(PG_LEVEL_4K, pages); 400 401 return last_map_addr; 402} 403 404static unsigned long __meminit 405phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end, 406 unsigned long page_size_mask, pgprot_t prot) 407{ 408 unsigned long pages = 0; 409 unsigned long last_map_addr = end; 410 411 int i = pmd_index(address); 412 413 for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) { 414 unsigned long pte_phys; 415 pmd_t *pmd = pmd_page + pmd_index(address); 416 pte_t *pte; 417 pgprot_t new_prot = prot; 418 419 if (address >= end) { 420 if (!after_bootmem) { 421 for (; i < PTRS_PER_PMD; i++, pmd++) 422 set_pmd(pmd, __pmd(0)); 423 } 424 break; 425 } 426 427 if (pmd_val(*pmd)) { 428 if (!pmd_large(*pmd)) { 429 spin_lock(&init_mm.page_table_lock); 430 pte = map_low_page((pte_t *)pmd_page_vaddr(*pmd)); 431 last_map_addr = phys_pte_init(pte, address, 432 end, prot); 433 unmap_low_page(pte); 434 spin_unlock(&init_mm.page_table_lock); 435 continue; 436 } 437 /* 438 * If we are ok with PG_LEVEL_2M mapping, then we will 439 * use the existing mapping, 440 * 441 * Otherwise, we will split the large page mapping but 442 * use the same existing protection bits except for 443 * large page, so that we don't violate Intel's TLB 444 * Application note (317080) which says, while changing 445 * the page sizes, new and old translations should 446 * not differ with respect to page frame and 447 * attributes. 448 */ 449 if (page_size_mask & (1 << PG_LEVEL_2M)) { 450 pages++; 451 continue; 452 } 453 new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd)); 454 } 455 456 if (page_size_mask & (1<<PG_LEVEL_2M)) { 457 pages++; 458 spin_lock(&init_mm.page_table_lock); 459 set_pte((pte_t *)pmd, 460 pfn_pte(address >> PAGE_SHIFT, 461 __pgprot(pgprot_val(prot) | _PAGE_PSE))); 462 spin_unlock(&init_mm.page_table_lock); 463 last_map_addr = (address & PMD_MASK) + PMD_SIZE; 464 continue; 465 } 466 467 pte = alloc_low_page(&pte_phys); 468 last_map_addr = phys_pte_init(pte, address, end, new_prot); 469 unmap_low_page(pte); 470 471 spin_lock(&init_mm.page_table_lock); 472 pmd_populate_kernel(&init_mm, pmd, __va(pte_phys)); 473 spin_unlock(&init_mm.page_table_lock); 474 } 475 update_page_count(PG_LEVEL_2M, pages); 476 return last_map_addr; 477} 478 479static unsigned long __meminit 480phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end, 481 unsigned long page_size_mask) 482{ 483 unsigned long pages = 0; 484 unsigned long last_map_addr = end; 485 int i = pud_index(addr); 486 487 for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) { 488 unsigned long pmd_phys; 489 pud_t *pud = pud_page + pud_index(addr); 490 pmd_t *pmd; 491 pgprot_t prot = PAGE_KERNEL; 492 493 if (addr >= end) 494 break; 495 496 if (!after_bootmem && 497 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) { 498 set_pud(pud, __pud(0)); 499 continue; 500 } 501 502 if (pud_val(*pud)) { 503 if (!pud_large(*pud)) { 504 pmd = map_low_page(pmd_offset(pud, 0)); 505 last_map_addr = phys_pmd_init(pmd, addr, end, 506 page_size_mask, prot); 507 unmap_low_page(pmd); 508 __flush_tlb_all(); 509 continue; 510 } 511 /* 512 * If we are ok with PG_LEVEL_1G mapping, then we will 513 * use the existing mapping. 514 * 515 * Otherwise, we will split the gbpage mapping but use 516 * the same existing protection bits except for large 517 * page, so that we don't violate Intel's TLB 518 * Application note (317080) which says, while changing 519 * the page sizes, new and old translations should 520 * not differ with respect to page frame and 521 * attributes. 522 */ 523 if (page_size_mask & (1 << PG_LEVEL_1G)) { 524 pages++; 525 continue; 526 } 527 prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud)); 528 } 529 530 if (page_size_mask & (1<<PG_LEVEL_1G)) { 531 pages++; 532 spin_lock(&init_mm.page_table_lock); 533 set_pte((pte_t *)pud, 534 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); 535 spin_unlock(&init_mm.page_table_lock); 536 last_map_addr = (addr & PUD_MASK) + PUD_SIZE; 537 continue; 538 } 539 540 pmd = alloc_low_page(&pmd_phys); 541 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask, 542 prot); 543 unmap_low_page(pmd); 544 545 spin_lock(&init_mm.page_table_lock); 546 pud_populate(&init_mm, pud, __va(pmd_phys)); 547 spin_unlock(&init_mm.page_table_lock); 548 } 549 __flush_tlb_all(); 550 551 update_page_count(PG_LEVEL_1G, pages); 552 553 return last_map_addr; 554} 555 556unsigned long __meminit 557kernel_physical_mapping_init(unsigned long start, 558 unsigned long end, 559 unsigned long page_size_mask) 560{ 561 bool pgd_changed = false; 562 unsigned long next, last_map_addr = end; 563 unsigned long addr; 564 565 start = (unsigned long)__va(start); 566 end = (unsigned long)__va(end); 567 addr = start; 568 569 for (; start < end; start = next) { 570 pgd_t *pgd = pgd_offset_k(start); 571 unsigned long pud_phys; 572 pud_t *pud; 573 574 next = (start + PGDIR_SIZE) & PGDIR_MASK; 575 if (next > end) 576 next = end; 577 578 if (pgd_val(*pgd)) { 579 pud = map_low_page((pud_t *)pgd_page_vaddr(*pgd)); 580 last_map_addr = phys_pud_init(pud, __pa(start), 581 __pa(end), page_size_mask); 582 unmap_low_page(pud); 583 continue; 584 } 585 586 pud = alloc_low_page(&pud_phys); 587 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next), 588 page_size_mask); 589 unmap_low_page(pud); 590 591 spin_lock(&init_mm.page_table_lock); 592 pgd_populate(&init_mm, pgd, __va(pud_phys)); 593 spin_unlock(&init_mm.page_table_lock); 594 pgd_changed = true; 595 } 596 597 if (pgd_changed) 598 sync_global_pgds(addr, end); 599 600 __flush_tlb_all(); 601 602 return last_map_addr; 603} 604 605#ifndef CONFIG_NUMA 606void __init initmem_init(void) 607{ 608 memblock_x86_register_active_regions(0, 0, max_pfn); 609} 610#endif 611 612void __init paging_init(void) 613{ 614 unsigned long max_zone_pfns[MAX_NR_ZONES]; 615 616 memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); 617 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; 618 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; 619 max_zone_pfns[ZONE_NORMAL] = max_pfn; 620 621 sparse_memory_present_with_active_regions(MAX_NUMNODES); 622 sparse_init(); 623 624 /* 625 * clear the default setting with node 0 626 * note: don't use nodes_clear here, that is really clearing when 627 * numa support is not compiled in, and later node_set_state 628 * will not set it back. 629 */ 630 node_clear_state(0, N_NORMAL_MEMORY); 631 632 free_area_init_nodes(max_zone_pfns); 633} 634 635/* 636 * Memory hotplug specific functions 637 */ 638#ifdef CONFIG_MEMORY_HOTPLUG 639/* 640 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need 641 * updating. 642 */ 643static void update_end_of_memory_vars(u64 start, u64 size) 644{ 645 unsigned long end_pfn = PFN_UP(start + size); 646 647 if (end_pfn > max_pfn) { 648 max_pfn = end_pfn; 649 max_low_pfn = end_pfn; 650 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 651 } 652} 653 654/* 655 * Memory is added always to NORMAL zone. This means you will never get 656 * additional DMA/DMA32 memory. 657 */ 658int arch_add_memory(int nid, u64 start, u64 size) 659{ 660 struct pglist_data *pgdat = NODE_DATA(nid); 661 struct zone *zone = pgdat->node_zones + ZONE_NORMAL; 662 unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT; 663 unsigned long nr_pages = size >> PAGE_SHIFT; 664 int ret; 665 666 last_mapped_pfn = init_memory_mapping(start, start + size); 667 if (last_mapped_pfn > max_pfn_mapped) 668 max_pfn_mapped = last_mapped_pfn; 669 670 ret = __add_pages(nid, zone, start_pfn, nr_pages); 671 WARN_ON_ONCE(ret); 672 673 /* update max_pfn, max_low_pfn and high_memory */ 674 update_end_of_memory_vars(start, size); 675 676 return ret; 677} 678EXPORT_SYMBOL_GPL(arch_add_memory); 679 680#if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA) 681int memory_add_physaddr_to_nid(u64 start) 682{ 683 return 0; 684} 685EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid); 686#endif 687 688#endif /* CONFIG_MEMORY_HOTPLUG */ 689 690static struct kcore_list kcore_vsyscall; 691 692void __init mem_init(void) 693{ 694 long codesize, reservedpages, datasize, initsize; 695 unsigned long absent_pages; 696 697 pci_iommu_alloc(); 698 699 /* clear_bss() already clear the empty_zero_page */ 700 701 reservedpages = 0; 702 703 /* this will put all low memory onto the freelists */ 704#ifdef CONFIG_NUMA 705 totalram_pages = numa_free_all_bootmem(); 706#else 707 totalram_pages = free_all_bootmem(); 708#endif 709 710 absent_pages = absent_pages_in_range(0, max_pfn); 711 reservedpages = max_pfn - totalram_pages - absent_pages; 712 after_bootmem = 1; 713 714 codesize = (unsigned long) &_etext - (unsigned long) &_text; 715 datasize = (unsigned long) &_edata - (unsigned long) &_etext; 716 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; 717 718 /* Register memory areas for /proc/kcore */ 719 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START, 720 VSYSCALL_END - VSYSCALL_START, KCORE_OTHER); 721 722 printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, " 723 "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n", 724 nr_free_pages() << (PAGE_SHIFT-10), 725 max_pfn << (PAGE_SHIFT-10), 726 codesize >> 10, 727 absent_pages << (PAGE_SHIFT-10), 728 reservedpages << (PAGE_SHIFT-10), 729 datasize >> 10, 730 initsize >> 10); 731} 732 733#ifdef CONFIG_DEBUG_RODATA 734const int rodata_test_data = 0xC3; 735EXPORT_SYMBOL_GPL(rodata_test_data); 736 737int kernel_set_to_readonly; 738 739void set_kernel_text_rw(void) 740{ 741 unsigned long start = PFN_ALIGN(_text); 742 unsigned long end = PFN_ALIGN(__stop___ex_table); 743 744 if (!kernel_set_to_readonly) 745 return; 746 747 pr_debug("Set kernel text: %lx - %lx for read write\n", 748 start, end); 749 750 /* 751 * Make the kernel identity mapping for text RW. Kernel text 752 * mapping will always be RO. Refer to the comment in 753 * static_protections() in pageattr.c 754 */ 755 set_memory_rw(start, (end - start) >> PAGE_SHIFT); 756} 757 758void set_kernel_text_ro(void) 759{ 760 unsigned long start = PFN_ALIGN(_text); 761 unsigned long end = PFN_ALIGN(__stop___ex_table); 762 763 if (!kernel_set_to_readonly) 764 return; 765 766 pr_debug("Set kernel text: %lx - %lx for read only\n", 767 start, end); 768 769 /* 770 * Set the kernel identity mapping for text RO. 771 */ 772 set_memory_ro(start, (end - start) >> PAGE_SHIFT); 773} 774 775void mark_rodata_ro(void) 776{ 777 unsigned long start = PFN_ALIGN(_text); 778 unsigned long rodata_start = 779 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK; 780 unsigned long end = (unsigned long) &__end_rodata_hpage_align; 781 unsigned long text_end = PAGE_ALIGN((unsigned long) &__stop___ex_table); 782 unsigned long rodata_end = PAGE_ALIGN((unsigned long) &__end_rodata); 783 unsigned long data_start = (unsigned long) &_sdata; 784 785 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", 786 (end - start) >> 10); 787 set_memory_ro(start, (end - start) >> PAGE_SHIFT); 788 789 kernel_set_to_readonly = 1; 790 791 /* 792 * The rodata section (but not the kernel text!) should also be 793 * not-executable. 794 */ 795 set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT); 796 797 rodata_test(); 798 799#ifdef CONFIG_CPA_DEBUG 800 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end); 801 set_memory_rw(start, (end-start) >> PAGE_SHIFT); 802 803 printk(KERN_INFO "Testing CPA: again\n"); 804 set_memory_ro(start, (end-start) >> PAGE_SHIFT); 805#endif 806 807 free_init_pages("unused kernel memory", 808 (unsigned long) page_address(virt_to_page(text_end)), 809 (unsigned long) 810 page_address(virt_to_page(rodata_start))); 811 free_init_pages("unused kernel memory", 812 (unsigned long) page_address(virt_to_page(rodata_end)), 813 (unsigned long) page_address(virt_to_page(data_start))); 814} 815 816#endif 817 818int kern_addr_valid(unsigned long addr) 819{ 820 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT; 821 pgd_t *pgd; 822 pud_t *pud; 823 pmd_t *pmd; 824 pte_t *pte; 825 826 if (above != 0 && above != -1UL) 827 return 0; 828 829 pgd = pgd_offset_k(addr); 830 if (pgd_none(*pgd)) 831 return 0; 832 833 pud = pud_offset(pgd, addr); 834 if (pud_none(*pud)) 835 return 0; 836 837 pmd = pmd_offset(pud, addr); 838 if (pmd_none(*pmd)) 839 return 0; 840 841 if (pmd_large(*pmd)) 842 return pfn_valid(pmd_pfn(*pmd)); 843 844 pte = pte_offset_kernel(pmd, addr); 845 if (pte_none(*pte)) 846 return 0; 847 848 return pfn_valid(pte_pfn(*pte)); 849} 850 851/* 852 * A pseudo VMA to allow ptrace access for the vsyscall page. This only 853 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does 854 * not need special handling anymore: 855 */ 856static struct vm_area_struct gate_vma = { 857 .vm_start = VSYSCALL_START, 858 .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE), 859 .vm_page_prot = PAGE_READONLY_EXEC, 860 .vm_flags = VM_READ | VM_EXEC 861}; 862 863struct vm_area_struct *get_gate_vma(struct task_struct *tsk) 864{ 865#ifdef CONFIG_IA32_EMULATION 866 if (test_tsk_thread_flag(tsk, TIF_IA32)) 867 return NULL; 868#endif 869 return &gate_vma; 870} 871 872int in_gate_area(struct task_struct *task, unsigned long addr) 873{ 874 struct vm_area_struct *vma = get_gate_vma(task); 875 876 if (!vma) 877 return 0; 878 879 return (addr >= vma->vm_start) && (addr < vma->vm_end); 880} 881 882/* 883 * Use this when you have no reliable task/vma, typically from interrupt 884 * context. It is less reliable than using the task's vma and may give 885 * false positives: 886 */ 887int in_gate_area_no_task(unsigned long addr) 888{ 889 return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END); 890} 891 892const char *arch_vma_name(struct vm_area_struct *vma) 893{ 894 if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso) 895 return "[vdso]"; 896 if (vma == &gate_vma) 897 return "[vsyscall]"; 898 return NULL; 899} 900 901#ifdef CONFIG_SPARSEMEM_VMEMMAP 902/* 903 * Initialise the sparsemem vmemmap using huge-pages at the PMD level. 904 */ 905static long __meminitdata addr_start, addr_end; 906static void __meminitdata *p_start, *p_end; 907static int __meminitdata node_start; 908 909int __meminit 910vmemmap_populate(struct page *start_page, unsigned long size, int node) 911{ 912 unsigned long addr = (unsigned long)start_page; 913 unsigned long end = (unsigned long)(start_page + size); 914 unsigned long next; 915 pgd_t *pgd; 916 pud_t *pud; 917 pmd_t *pmd; 918 919 for (; addr < end; addr = next) { 920 void *p = NULL; 921 922 pgd = vmemmap_pgd_populate(addr, node); 923 if (!pgd) 924 return -ENOMEM; 925 926 pud = vmemmap_pud_populate(pgd, addr, node); 927 if (!pud) 928 return -ENOMEM; 929 930 if (!cpu_has_pse) { 931 next = (addr + PAGE_SIZE) & PAGE_MASK; 932 pmd = vmemmap_pmd_populate(pud, addr, node); 933 934 if (!pmd) 935 return -ENOMEM; 936 937 p = vmemmap_pte_populate(pmd, addr, node); 938 939 if (!p) 940 return -ENOMEM; 941 942 addr_end = addr + PAGE_SIZE; 943 p_end = p + PAGE_SIZE; 944 } else { 945 next = pmd_addr_end(addr, end); 946 947 pmd = pmd_offset(pud, addr); 948 if (pmd_none(*pmd)) { 949 pte_t entry; 950 951 p = vmemmap_alloc_block_buf(PMD_SIZE, node); 952 if (!p) 953 return -ENOMEM; 954 955 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, 956 PAGE_KERNEL_LARGE); 957 set_pmd(pmd, __pmd(pte_val(entry))); 958 959 /* check to see if we have contiguous blocks */ 960 if (p_end != p || node_start != node) { 961 if (p_start) 962 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", 963 addr_start, addr_end-1, p_start, p_end-1, node_start); 964 addr_start = addr; 965 node_start = node; 966 p_start = p; 967 } 968 969 addr_end = addr + PMD_SIZE; 970 p_end = p + PMD_SIZE; 971 } else 972 vmemmap_verify((pte_t *)pmd, node, addr, next); 973 } 974 975 } 976 sync_global_pgds((unsigned long)start_page, end); 977 return 0; 978} 979 980void __meminit vmemmap_populate_print_last(void) 981{ 982 if (p_start) { 983 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", 984 addr_start, addr_end-1, p_start, p_end-1, node_start); 985 p_start = NULL; 986 p_end = NULL; 987 node_start = 0; 988 } 989} 990#endif 991