1 /* 2 * 3 * Copyright (C) 1995 Linus Torvalds 4 * 5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 6 */ 7 8 #include <linux/module.h> 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/hugetlb.h> 19 #include <linux/swap.h> 20 #include <linux/smp.h> 21 #include <linux/init.h> 22 #include <linux/highmem.h> 23 #include <linux/pagemap.h> 24 #include <linux/pfn.h> 25 #include <linux/poison.h> 26 #include <linux/bootmem.h> 27 #include <linux/slab.h> 28 #include <linux/proc_fs.h> 29 #include <linux/memory_hotplug.h> 30 #include <linux/initrd.h> 31 #include <linux/cpumask.h> 32 33 #include <asm/asm.h> 34 #include <asm/processor.h> 35 #include <asm/system.h> 36 #include <asm/uaccess.h> 37 #include <asm/pgtable.h> 38 #include <asm/dma.h> 39 #include <asm/fixmap.h> 40 #include <asm/e820.h> 41 #include <asm/apic.h> 42 #include <asm/bugs.h> 43 #include <asm/tlb.h> 44 #include <asm/tlbflush.h> 45 #include <asm/pgalloc.h> 46 #include <asm/sections.h> 47 #include <asm/paravirt.h> 48 #include <asm/setup.h> 49 #include <asm/cacheflush.h> 50 51 unsigned int __VMALLOC_RESERVE = 128 << 20; 52 53 unsigned long max_pfn_mapped; 54 55 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); 56 unsigned long highstart_pfn, highend_pfn; 57 58 static noinline int do_test_wp_bit(void); 59 60 /* 61 * Creates a middle page table and puts a pointer to it in the 62 * given global directory entry. This only returns the gd entry 63 * in non-PAE compilation mode, since the middle layer is folded. 64 */ 65 static pmd_t * __init one_md_table_init(pgd_t *pgd) 66 { 67 pud_t *pud; 68 pmd_t *pmd_table; 69 70 #ifdef CONFIG_X86_PAE 71 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) { 72 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE); 73 74 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT); 75 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT)); 76 pud = pud_offset(pgd, 0); 77 BUG_ON(pmd_table != pmd_offset(pud, 0)); 78 } 79 #endif 80 pud = pud_offset(pgd, 0); 81 pmd_table = pmd_offset(pud, 0); 82 83 return pmd_table; 84 } 85 86 /* 87 * Create a page table and place a pointer to it in a middle page 88 * directory entry: 89 */ 90 static pte_t * __init one_page_table_init(pmd_t *pmd) 91 { 92 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) { 93 pte_t *page_table = NULL; 94 95 #ifdef CONFIG_DEBUG_PAGEALLOC 96 page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE); 97 #endif 98 if (!page_table) { 99 page_table = 100 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE); 101 } 102 103 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT); 104 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE)); 105 BUG_ON(page_table != pte_offset_kernel(pmd, 0)); 106 } 107 108 return pte_offset_kernel(pmd, 0); 109 } 110 111 /* 112 * This function initializes a certain range of kernel virtual memory 113 * with new bootmem page tables, everywhere page tables are missing in 114 * the given range. 115 * 116 * NOTE: The pagetables are allocated contiguous on the physical space 117 * so we can cache the place of the first one and move around without 118 * checking the pgd every time. 119 */ 120 static void __init 121 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base) 122 { 123 int pgd_idx, pmd_idx; 124 unsigned long vaddr; 125 pgd_t *pgd; 126 pmd_t *pmd; 127 128 vaddr = start; 129 pgd_idx = pgd_index(vaddr); 130 pmd_idx = pmd_index(vaddr); 131 pgd = pgd_base + pgd_idx; 132 133 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) { 134 pmd = one_md_table_init(pgd); 135 pmd = pmd + pmd_index(vaddr); 136 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); 137 pmd++, pmd_idx++) { 138 one_page_table_init(pmd); 139 140 vaddr += PMD_SIZE; 141 } 142 pmd_idx = 0; 143 } 144 } 145 146 static inline int is_kernel_text(unsigned long addr) 147 { 148 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end) 149 return 1; 150 return 0; 151 } 152 153 /* 154 * This maps the physical memory to kernel virtual address space, a total 155 * of max_low_pfn pages, by creating page tables starting from address 156 * PAGE_OFFSET: 157 */ 158 static void __init kernel_physical_mapping_init(pgd_t *pgd_base) 159 { 160 int pgd_idx, pmd_idx, pte_ofs; 161 unsigned long pfn; 162 pgd_t *pgd; 163 pmd_t *pmd; 164 pte_t *pte; 165 166 pgd_idx = pgd_index(PAGE_OFFSET); 167 pgd = pgd_base + pgd_idx; 168 pfn = 0; 169 170 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) { 171 pmd = one_md_table_init(pgd); 172 if (pfn >= max_low_pfn) 173 continue; 174 175 for (pmd_idx = 0; 176 pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; 177 pmd++, pmd_idx++) { 178 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET; 179 180 /* 181 * Map with big pages if possible, otherwise 182 * create normal page tables: 183 * 184 * Don't use a large page for the first 2/4MB of memory 185 * because there are often fixed size MTRRs in there 186 * and overlapping MTRRs into large pages can cause 187 * slowdowns. 188 */ 189 if (cpu_has_pse && !(pgd_idx == 0 && pmd_idx == 0)) { 190 unsigned int addr2; 191 pgprot_t prot = PAGE_KERNEL_LARGE; 192 193 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE + 194 PAGE_OFFSET + PAGE_SIZE-1; 195 196 if (is_kernel_text(addr) || 197 is_kernel_text(addr2)) 198 prot = PAGE_KERNEL_LARGE_EXEC; 199 200 set_pmd(pmd, pfn_pmd(pfn, prot)); 201 202 pfn += PTRS_PER_PTE; 203 max_pfn_mapped = pfn; 204 continue; 205 } 206 pte = one_page_table_init(pmd); 207 208 for (pte_ofs = 0; 209 pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; 210 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) { 211 pgprot_t prot = PAGE_KERNEL; 212 213 if (is_kernel_text(addr)) 214 prot = PAGE_KERNEL_EXEC; 215 216 set_pte(pte, pfn_pte(pfn, prot)); 217 } 218 max_pfn_mapped = pfn; 219 } 220 } 221 } 222 223 static inline int page_kills_ppro(unsigned long pagenr) 224 { 225 if (pagenr >= 0x70000 && pagenr <= 0x7003F) 226 return 1; 227 return 0; 228 } 229 230 /* 231 * devmem_is_allowed() checks to see if /dev/mem access to a certain address 232 * is valid. The argument is a physical page number. 233 * 234 * 235 * On x86, access has to be given to the first megabyte of ram because that area 236 * contains bios code and data regions used by X and dosemu and similar apps. 237 * Access has to be given to non-kernel-ram areas as well, these contain the PCI 238 * mmio resources as well as potential bios/acpi data regions. 239 */ 240 int devmem_is_allowed(unsigned long pagenr) 241 { 242 if (pagenr <= 256) 243 return 1; 244 if (!page_is_ram(pagenr)) 245 return 1; 246 return 0; 247 } 248 249 #ifdef CONFIG_HIGHMEM 250 pte_t *kmap_pte; 251 pgprot_t kmap_prot; 252 253 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr) 254 { 255 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), 256 vaddr), vaddr), vaddr); 257 } 258 259 static void __init kmap_init(void) 260 { 261 unsigned long kmap_vstart; 262 263 /* 264 * Cache the first kmap pte: 265 */ 266 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN); 267 kmap_pte = kmap_get_fixmap_pte(kmap_vstart); 268 269 kmap_prot = PAGE_KERNEL; 270 } 271 272 static void __init permanent_kmaps_init(pgd_t *pgd_base) 273 { 274 unsigned long vaddr; 275 pgd_t *pgd; 276 pud_t *pud; 277 pmd_t *pmd; 278 pte_t *pte; 279 280 vaddr = PKMAP_BASE; 281 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base); 282 283 pgd = swapper_pg_dir + pgd_index(vaddr); 284 pud = pud_offset(pgd, vaddr); 285 pmd = pmd_offset(pud, vaddr); 286 pte = pte_offset_kernel(pmd, vaddr); 287 pkmap_page_table = pte; 288 } 289 290 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro) 291 { 292 if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) { 293 ClearPageReserved(page); 294 init_page_count(page); 295 __free_page(page); 296 totalhigh_pages++; 297 } else 298 SetPageReserved(page); 299 } 300 301 #ifndef CONFIG_NUMA 302 static void __init set_highmem_pages_init(int bad_ppro) 303 { 304 int pfn; 305 306 for (pfn = highstart_pfn; pfn < highend_pfn; pfn++) { 307 /* 308 * Holes under sparsemem might not have no mem_map[]: 309 */ 310 if (pfn_valid(pfn)) 311 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro); 312 } 313 totalram_pages += totalhigh_pages; 314 } 315 #endif /* !CONFIG_NUMA */ 316 317 #else 318 # define kmap_init() do { } while (0) 319 # define permanent_kmaps_init(pgd_base) do { } while (0) 320 # define set_highmem_pages_init(bad_ppro) do { } while (0) 321 #endif /* CONFIG_HIGHMEM */ 322 323 pteval_t __PAGE_KERNEL = _PAGE_KERNEL; 324 EXPORT_SYMBOL(__PAGE_KERNEL); 325 326 pteval_t __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC; 327 328 void __init native_pagetable_setup_start(pgd_t *base) 329 { 330 unsigned long pfn, va; 331 pgd_t *pgd; 332 pud_t *pud; 333 pmd_t *pmd; 334 pte_t *pte; 335 336 /* 337 * Remove any mappings which extend past the end of physical 338 * memory from the boot time page table: 339 */ 340 for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) { 341 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT); 342 pgd = base + pgd_index(va); 343 if (!pgd_present(*pgd)) 344 break; 345 346 pud = pud_offset(pgd, va); 347 pmd = pmd_offset(pud, va); 348 if (!pmd_present(*pmd)) 349 break; 350 351 pte = pte_offset_kernel(pmd, va); 352 if (!pte_present(*pte)) 353 break; 354 355 pte_clear(NULL, va, pte); 356 } 357 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT); 358 } 359 360 void __init native_pagetable_setup_done(pgd_t *base) 361 { 362 } 363 364 /* 365 * Build a proper pagetable for the kernel mappings. Up until this 366 * point, we've been running on some set of pagetables constructed by 367 * the boot process. 368 * 369 * If we're booting on native hardware, this will be a pagetable 370 * constructed in arch/x86/kernel/head_32.S. The root of the 371 * pagetable will be swapper_pg_dir. 372 * 373 * If we're booting paravirtualized under a hypervisor, then there are 374 * more options: we may already be running PAE, and the pagetable may 375 * or may not be based in swapper_pg_dir. In any case, 376 * paravirt_pagetable_setup_start() will set up swapper_pg_dir 377 * appropriately for the rest of the initialization to work. 378 * 379 * In general, pagetable_init() assumes that the pagetable may already 380 * be partially populated, and so it avoids stomping on any existing 381 * mappings. 382 */ 383 static void __init pagetable_init(void) 384 { 385 pgd_t *pgd_base = swapper_pg_dir; 386 unsigned long vaddr, end; 387 388 paravirt_pagetable_setup_start(pgd_base); 389 390 /* Enable PSE if available */ 391 if (cpu_has_pse) 392 set_in_cr4(X86_CR4_PSE); 393 394 /* Enable PGE if available */ 395 if (cpu_has_pge) { 396 set_in_cr4(X86_CR4_PGE); 397 __PAGE_KERNEL |= _PAGE_GLOBAL; 398 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL; 399 } 400 401 kernel_physical_mapping_init(pgd_base); 402 remap_numa_kva(); 403 404 /* 405 * Fixed mappings, only the page table structure has to be 406 * created - mappings will be set by set_fixmap(): 407 */ 408 early_ioremap_clear(); 409 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK; 410 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK; 411 page_table_range_init(vaddr, end, pgd_base); 412 early_ioremap_reset(); 413 414 permanent_kmaps_init(pgd_base); 415 416 paravirt_pagetable_setup_done(pgd_base); 417 } 418 419 #ifdef CONFIG_ACPI_SLEEP 420 /* 421 * ACPI suspend needs this for resume, because things like the intel-agp 422 * driver might have split up a kernel 4MB mapping. 423 */ 424 char swsusp_pg_dir[PAGE_SIZE] 425 __attribute__ ((aligned(PAGE_SIZE))); 426 427 static inline void save_pg_dir(void) 428 { 429 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE); 430 } 431 #else /* !CONFIG_ACPI_SLEEP */ 432 static inline void save_pg_dir(void) 433 { 434 } 435 #endif /* !CONFIG_ACPI_SLEEP */ 436 437 void zap_low_mappings(void) 438 { 439 int i; 440 441 /* 442 * Zap initial low-memory mappings. 443 * 444 * Note that "pgd_clear()" doesn't do it for 445 * us, because pgd_clear() is a no-op on i386. 446 */ 447 for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) { 448 #ifdef CONFIG_X86_PAE 449 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page))); 450 #else 451 set_pgd(swapper_pg_dir+i, __pgd(0)); 452 #endif 453 } 454 flush_tlb_all(); 455 } 456 457 int nx_enabled; 458 459 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_NX; 460 EXPORT_SYMBOL_GPL(__supported_pte_mask); 461 462 #ifdef CONFIG_X86_PAE 463 464 static int disable_nx __initdata; 465 466 /* 467 * noexec = on|off 468 * 469 * Control non executable mappings. 470 * 471 * on Enable 472 * off Disable 473 */ 474 static int __init noexec_setup(char *str) 475 { 476 if (!str || !strcmp(str, "on")) { 477 if (cpu_has_nx) { 478 __supported_pte_mask |= _PAGE_NX; 479 disable_nx = 0; 480 } 481 } else { 482 if (!strcmp(str, "off")) { 483 disable_nx = 1; 484 __supported_pte_mask &= ~_PAGE_NX; 485 } else { 486 return -EINVAL; 487 } 488 } 489 490 return 0; 491 } 492 early_param("noexec", noexec_setup); 493 494 static void __init set_nx(void) 495 { 496 unsigned int v[4], l, h; 497 498 if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) { 499 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]); 500 501 if ((v[3] & (1 << 20)) && !disable_nx) { 502 rdmsr(MSR_EFER, l, h); 503 l |= EFER_NX; 504 wrmsr(MSR_EFER, l, h); 505 nx_enabled = 1; 506 __supported_pte_mask |= _PAGE_NX; 507 } 508 } 509 } 510 #endif 511 512 /* 513 * paging_init() sets up the page tables - note that the first 8MB are 514 * already mapped by head.S. 515 * 516 * This routines also unmaps the page at virtual kernel address 0, so 517 * that we can trap those pesky NULL-reference errors in the kernel. 518 */ 519 void __init paging_init(void) 520 { 521 #ifdef CONFIG_X86_PAE 522 set_nx(); 523 if (nx_enabled) 524 printk(KERN_INFO "NX (Execute Disable) protection: active\n"); 525 #endif 526 pagetable_init(); 527 528 load_cr3(swapper_pg_dir); 529 530 __flush_tlb_all(); 531 532 kmap_init(); 533 } 534 535 /* 536 * Test if the WP bit works in supervisor mode. It isn't supported on 386's 537 * and also on some strange 486's. All 586+'s are OK. This used to involve 538 * black magic jumps to work around some nasty CPU bugs, but fortunately the 539 * switch to using exceptions got rid of all that. 540 */ 541 static void __init test_wp_bit(void) 542 { 543 printk(KERN_INFO 544 "Checking if this processor honours the WP bit even in supervisor mode..."); 545 546 /* Any page-aligned address will do, the test is non-destructive */ 547 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY); 548 boot_cpu_data.wp_works_ok = do_test_wp_bit(); 549 clear_fixmap(FIX_WP_TEST); 550 551 if (!boot_cpu_data.wp_works_ok) { 552 printk(KERN_CONT "No.\n"); 553 #ifdef CONFIG_X86_WP_WORKS_OK 554 panic( 555 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!"); 556 #endif 557 } else { 558 printk(KERN_CONT "Ok.\n"); 559 } 560 } 561 562 static struct kcore_list kcore_mem, kcore_vmalloc; 563 564 void __init mem_init(void) 565 { 566 int codesize, reservedpages, datasize, initsize; 567 int tmp, bad_ppro; 568 569 #ifdef CONFIG_FLATMEM 570 BUG_ON(!mem_map); 571 #endif 572 bad_ppro = ppro_with_ram_bug(); 573 574 #ifdef CONFIG_HIGHMEM 575 /* check that fixmap and pkmap do not overlap */ 576 if (PKMAP_BASE + LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) { 577 printk(KERN_ERR 578 "fixmap and kmap areas overlap - this will crash\n"); 579 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n", 580 PKMAP_BASE, PKMAP_BASE + LAST_PKMAP*PAGE_SIZE, 581 FIXADDR_START); 582 BUG(); 583 } 584 #endif 585 /* this will put all low memory onto the freelists */ 586 totalram_pages += free_all_bootmem(); 587 588 reservedpages = 0; 589 for (tmp = 0; tmp < max_low_pfn; tmp++) 590 /* 591 * Only count reserved RAM pages: 592 */ 593 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp))) 594 reservedpages++; 595 596 set_highmem_pages_init(bad_ppro); 597 598 codesize = (unsigned long) &_etext - (unsigned long) &_text; 599 datasize = (unsigned long) &_edata - (unsigned long) &_etext; 600 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; 601 602 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT); 603 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, 604 VMALLOC_END-VMALLOC_START); 605 606 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, " 607 "%dk reserved, %dk data, %dk init, %ldk highmem)\n", 608 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), 609 num_physpages << (PAGE_SHIFT-10), 610 codesize >> 10, 611 reservedpages << (PAGE_SHIFT-10), 612 datasize >> 10, 613 initsize >> 10, 614 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10)) 615 ); 616 617 #if 1 /* double-sanity-check paranoia */ 618 printk(KERN_INFO "virtual kernel memory layout:\n" 619 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n" 620 #ifdef CONFIG_HIGHMEM 621 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n" 622 #endif 623 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n" 624 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n" 625 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n" 626 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n" 627 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n", 628 FIXADDR_START, FIXADDR_TOP, 629 (FIXADDR_TOP - FIXADDR_START) >> 10, 630 631 #ifdef CONFIG_HIGHMEM 632 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, 633 (LAST_PKMAP*PAGE_SIZE) >> 10, 634 #endif 635 636 VMALLOC_START, VMALLOC_END, 637 (VMALLOC_END - VMALLOC_START) >> 20, 638 639 (unsigned long)__va(0), (unsigned long)high_memory, 640 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20, 641 642 (unsigned long)&__init_begin, (unsigned long)&__init_end, 643 ((unsigned long)&__init_end - 644 (unsigned long)&__init_begin) >> 10, 645 646 (unsigned long)&_etext, (unsigned long)&_edata, 647 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10, 648 649 (unsigned long)&_text, (unsigned long)&_etext, 650 ((unsigned long)&_etext - (unsigned long)&_text) >> 10); 651 652 #ifdef CONFIG_HIGHMEM 653 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); 654 BUG_ON(VMALLOC_END > PKMAP_BASE); 655 #endif 656 BUG_ON(VMALLOC_START > VMALLOC_END); 657 BUG_ON((unsigned long)high_memory > VMALLOC_START); 658 #endif /* double-sanity-check paranoia */ 659 660 if (boot_cpu_data.wp_works_ok < 0) 661 test_wp_bit(); 662 663 cpa_init(); 664 save_pg_dir(); 665 zap_low_mappings(); 666 } 667 668 #ifdef CONFIG_MEMORY_HOTPLUG 669 int arch_add_memory(int nid, u64 start, u64 size) 670 { 671 struct pglist_data *pgdata = NODE_DATA(nid); 672 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM; 673 unsigned long start_pfn = start >> PAGE_SHIFT; 674 unsigned long nr_pages = size >> PAGE_SHIFT; 675 676 return __add_pages(zone, start_pfn, nr_pages); 677 } 678 #endif 679 680 /* 681 * This function cannot be __init, since exceptions don't work in that 682 * section. Put this after the callers, so that it cannot be inlined. 683 */ 684 static noinline int do_test_wp_bit(void) 685 { 686 char tmp_reg; 687 int flag; 688 689 __asm__ __volatile__( 690 " movb %0, %1 \n" 691 "1: movb %1, %0 \n" 692 " xorl %2, %2 \n" 693 "2: \n" 694 _ASM_EXTABLE(1b,2b) 695 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)), 696 "=q" (tmp_reg), 697 "=r" (flag) 698 :"2" (1) 699 :"memory"); 700 701 return flag; 702 } 703 704 #ifdef CONFIG_DEBUG_RODATA 705 const int rodata_test_data = 0xC3; 706 EXPORT_SYMBOL_GPL(rodata_test_data); 707 708 void mark_rodata_ro(void) 709 { 710 unsigned long start = PFN_ALIGN(_text); 711 unsigned long size = PFN_ALIGN(_etext) - start; 712 713 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 714 printk(KERN_INFO "Write protecting the kernel text: %luk\n", 715 size >> 10); 716 717 #ifdef CONFIG_CPA_DEBUG 718 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n", 719 start, start+size); 720 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT); 721 722 printk(KERN_INFO "Testing CPA: write protecting again\n"); 723 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT); 724 #endif 725 start += size; 726 size = (unsigned long)__end_rodata - start; 727 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 728 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", 729 size >> 10); 730 rodata_test(); 731 732 #ifdef CONFIG_CPA_DEBUG 733 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size); 734 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT); 735 736 printk(KERN_INFO "Testing CPA: write protecting again\n"); 737 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 738 #endif 739 } 740 #endif 741 742 void free_init_pages(char *what, unsigned long begin, unsigned long end) 743 { 744 #ifdef CONFIG_DEBUG_PAGEALLOC 745 /* 746 * If debugging page accesses then do not free this memory but 747 * mark them not present - any buggy init-section access will 748 * create a kernel page fault: 749 */ 750 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n", 751 begin, PAGE_ALIGN(end)); 752 set_memory_np(begin, (end - begin) >> PAGE_SHIFT); 753 #else 754 unsigned long addr; 755 756 /* 757 * We just marked the kernel text read only above, now that 758 * we are going to free part of that, we need to make that 759 * writeable first. 760 */ 761 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT); 762 763 for (addr = begin; addr < end; addr += PAGE_SIZE) { 764 ClearPageReserved(virt_to_page(addr)); 765 init_page_count(virt_to_page(addr)); 766 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE); 767 free_page(addr); 768 totalram_pages++; 769 } 770 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10); 771 #endif 772 } 773 774 void free_initmem(void) 775 { 776 free_init_pages("unused kernel memory", 777 (unsigned long)(&__init_begin), 778 (unsigned long)(&__init_end)); 779 } 780 781 #ifdef CONFIG_BLK_DEV_INITRD 782 void free_initrd_mem(unsigned long start, unsigned long end) 783 { 784 free_init_pages("initrd memory", start, end); 785 } 786 #endif 787