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/pci.h> 25 #include <linux/pfn.h> 26 #include <linux/poison.h> 27 #include <linux/bootmem.h> 28 #include <linux/proc_fs.h> 29 #include <linux/memory_hotplug.h> 30 #include <linux/initrd.h> 31 #include <linux/cpumask.h> 32 #include <linux/gfp.h> 33 34 #include <asm/asm.h> 35 #include <asm/bios_ebda.h> 36 #include <asm/processor.h> 37 #include <asm/system.h> 38 #include <asm/uaccess.h> 39 #include <asm/pgtable.h> 40 #include <asm/dma.h> 41 #include <asm/fixmap.h> 42 #include <asm/e820.h> 43 #include <asm/apic.h> 44 #include <asm/bugs.h> 45 #include <asm/tlb.h> 46 #include <asm/tlbflush.h> 47 #include <asm/pgalloc.h> 48 #include <asm/sections.h> 49 #include <asm/paravirt.h> 50 #include <asm/setup.h> 51 #include <asm/cacheflush.h> 52 #include <asm/page_types.h> 53 #include <asm/init.h> 54 55 unsigned long highstart_pfn, highend_pfn; 56 57 static noinline int do_test_wp_bit(void); 58 59 bool __read_mostly __vmalloc_start_set = false; 60 61 static __init void *alloc_low_page(void) 62 { 63 unsigned long pfn = e820_table_end++; 64 void *adr; 65 66 if (pfn >= e820_table_top) 67 panic("alloc_low_page: ran out of memory"); 68 69 adr = __va(pfn * PAGE_SIZE); 70 memset(adr, 0, PAGE_SIZE); 71 return adr; 72 } 73 74 /* 75 * Creates a middle page table and puts a pointer to it in the 76 * given global directory entry. This only returns the gd entry 77 * in non-PAE compilation mode, since the middle layer is folded. 78 */ 79 static pmd_t * __init one_md_table_init(pgd_t *pgd) 80 { 81 pud_t *pud; 82 pmd_t *pmd_table; 83 84 #ifdef CONFIG_X86_PAE 85 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) { 86 if (after_bootmem) 87 pmd_table = (pmd_t *)alloc_bootmem_pages(PAGE_SIZE); 88 else 89 pmd_table = (pmd_t *)alloc_low_page(); 90 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT); 91 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT)); 92 pud = pud_offset(pgd, 0); 93 BUG_ON(pmd_table != pmd_offset(pud, 0)); 94 95 return pmd_table; 96 } 97 #endif 98 pud = pud_offset(pgd, 0); 99 pmd_table = pmd_offset(pud, 0); 100 101 return pmd_table; 102 } 103 104 /* 105 * Create a page table and place a pointer to it in a middle page 106 * directory entry: 107 */ 108 static pte_t * __init one_page_table_init(pmd_t *pmd) 109 { 110 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) { 111 pte_t *page_table = NULL; 112 113 if (after_bootmem) { 114 #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK) 115 page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE); 116 #endif 117 if (!page_table) 118 page_table = 119 (pte_t *)alloc_bootmem_pages(PAGE_SIZE); 120 } else 121 page_table = (pte_t *)alloc_low_page(); 122 123 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT); 124 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE)); 125 BUG_ON(page_table != pte_offset_kernel(pmd, 0)); 126 } 127 128 return pte_offset_kernel(pmd, 0); 129 } 130 131 pmd_t * __init populate_extra_pmd(unsigned long vaddr) 132 { 133 int pgd_idx = pgd_index(vaddr); 134 int pmd_idx = pmd_index(vaddr); 135 136 return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx; 137 } 138 139 pte_t * __init populate_extra_pte(unsigned long vaddr) 140 { 141 int pte_idx = pte_index(vaddr); 142 pmd_t *pmd; 143 144 pmd = populate_extra_pmd(vaddr); 145 return one_page_table_init(pmd) + pte_idx; 146 } 147 148 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd, 149 unsigned long vaddr, pte_t *lastpte) 150 { 151 #ifdef CONFIG_HIGHMEM 152 /* 153 * Something (early fixmap) may already have put a pte 154 * page here, which causes the page table allocation 155 * to become nonlinear. Attempt to fix it, and if it 156 * is still nonlinear then we have to bug. 157 */ 158 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT; 159 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT; 160 161 if (pmd_idx_kmap_begin != pmd_idx_kmap_end 162 && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin 163 && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end 164 && ((__pa(pte) >> PAGE_SHIFT) < e820_table_start 165 || (__pa(pte) >> PAGE_SHIFT) >= e820_table_end)) { 166 pte_t *newpte; 167 int i; 168 169 BUG_ON(after_bootmem); 170 newpte = alloc_low_page(); 171 for (i = 0; i < PTRS_PER_PTE; i++) 172 set_pte(newpte + i, pte[i]); 173 174 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT); 175 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE)); 176 BUG_ON(newpte != pte_offset_kernel(pmd, 0)); 177 __flush_tlb_all(); 178 179 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT); 180 pte = newpte; 181 } 182 BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1) 183 && vaddr > fix_to_virt(FIX_KMAP_END) 184 && lastpte && lastpte + PTRS_PER_PTE != pte); 185 #endif 186 return pte; 187 } 188 189 /* 190 * This function initializes a certain range of kernel virtual memory 191 * with new bootmem page tables, everywhere page tables are missing in 192 * the given range. 193 * 194 * NOTE: The pagetables are allocated contiguous on the physical space 195 * so we can cache the place of the first one and move around without 196 * checking the pgd every time. 197 */ 198 static void __init 199 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base) 200 { 201 int pgd_idx, pmd_idx; 202 unsigned long vaddr; 203 pgd_t *pgd; 204 pmd_t *pmd; 205 pte_t *pte = NULL; 206 207 vaddr = start; 208 pgd_idx = pgd_index(vaddr); 209 pmd_idx = pmd_index(vaddr); 210 pgd = pgd_base + pgd_idx; 211 212 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) { 213 pmd = one_md_table_init(pgd); 214 pmd = pmd + pmd_index(vaddr); 215 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); 216 pmd++, pmd_idx++) { 217 pte = page_table_kmap_check(one_page_table_init(pmd), 218 pmd, vaddr, pte); 219 220 vaddr += PMD_SIZE; 221 } 222 pmd_idx = 0; 223 } 224 } 225 226 static inline int is_kernel_text(unsigned long addr) 227 { 228 if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end) 229 return 1; 230 return 0; 231 } 232 233 /* 234 * This maps the physical memory to kernel virtual address space, a total 235 * of max_low_pfn pages, by creating page tables starting from address 236 * PAGE_OFFSET: 237 */ 238 unsigned long __init 239 kernel_physical_mapping_init(unsigned long start, 240 unsigned long end, 241 unsigned long page_size_mask) 242 { 243 int use_pse = page_size_mask == (1<<PG_LEVEL_2M); 244 unsigned long last_map_addr = end; 245 unsigned long start_pfn, end_pfn; 246 pgd_t *pgd_base = swapper_pg_dir; 247 int pgd_idx, pmd_idx, pte_ofs; 248 unsigned long pfn; 249 pgd_t *pgd; 250 pmd_t *pmd; 251 pte_t *pte; 252 unsigned pages_2m, pages_4k; 253 int mapping_iter; 254 255 start_pfn = start >> PAGE_SHIFT; 256 end_pfn = end >> PAGE_SHIFT; 257 258 /* 259 * First iteration will setup identity mapping using large/small pages 260 * based on use_pse, with other attributes same as set by 261 * the early code in head_32.S 262 * 263 * Second iteration will setup the appropriate attributes (NX, GLOBAL..) 264 * as desired for the kernel identity mapping. 265 * 266 * This two pass mechanism conforms to the TLB app note which says: 267 * 268 * "Software should not write to a paging-structure entry in a way 269 * that would change, for any linear address, both the page size 270 * and either the page frame or attributes." 271 */ 272 mapping_iter = 1; 273 274 if (!cpu_has_pse) 275 use_pse = 0; 276 277 repeat: 278 pages_2m = pages_4k = 0; 279 pfn = start_pfn; 280 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 281 pgd = pgd_base + pgd_idx; 282 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) { 283 pmd = one_md_table_init(pgd); 284 285 if (pfn >= end_pfn) 286 continue; 287 #ifdef CONFIG_X86_PAE 288 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 289 pmd += pmd_idx; 290 #else 291 pmd_idx = 0; 292 #endif 293 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn; 294 pmd++, pmd_idx++) { 295 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET; 296 297 /* 298 * Map with big pages if possible, otherwise 299 * create normal page tables: 300 */ 301 if (use_pse) { 302 unsigned int addr2; 303 pgprot_t prot = PAGE_KERNEL_LARGE; 304 /* 305 * first pass will use the same initial 306 * identity mapping attribute + _PAGE_PSE. 307 */ 308 pgprot_t init_prot = 309 __pgprot(PTE_IDENT_ATTR | 310 _PAGE_PSE); 311 312 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE + 313 PAGE_OFFSET + PAGE_SIZE-1; 314 315 if (is_kernel_text(addr) || 316 is_kernel_text(addr2)) 317 prot = PAGE_KERNEL_LARGE_EXEC; 318 319 pages_2m++; 320 if (mapping_iter == 1) 321 set_pmd(pmd, pfn_pmd(pfn, init_prot)); 322 else 323 set_pmd(pmd, pfn_pmd(pfn, prot)); 324 325 pfn += PTRS_PER_PTE; 326 continue; 327 } 328 pte = one_page_table_init(pmd); 329 330 pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 331 pte += pte_ofs; 332 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn; 333 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) { 334 pgprot_t prot = PAGE_KERNEL; 335 /* 336 * first pass will use the same initial 337 * identity mapping attribute. 338 */ 339 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR); 340 341 if (is_kernel_text(addr)) 342 prot = PAGE_KERNEL_EXEC; 343 344 pages_4k++; 345 if (mapping_iter == 1) { 346 set_pte(pte, pfn_pte(pfn, init_prot)); 347 last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE; 348 } else 349 set_pte(pte, pfn_pte(pfn, prot)); 350 } 351 } 352 } 353 if (mapping_iter == 1) { 354 /* 355 * update direct mapping page count only in the first 356 * iteration. 357 */ 358 update_page_count(PG_LEVEL_2M, pages_2m); 359 update_page_count(PG_LEVEL_4K, pages_4k); 360 361 /* 362 * local global flush tlb, which will flush the previous 363 * mappings present in both small and large page TLB's. 364 */ 365 __flush_tlb_all(); 366 367 /* 368 * Second iteration will set the actual desired PTE attributes. 369 */ 370 mapping_iter = 2; 371 goto repeat; 372 } 373 return last_map_addr; 374 } 375 376 pte_t *kmap_pte; 377 pgprot_t kmap_prot; 378 379 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr) 380 { 381 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), 382 vaddr), vaddr), vaddr); 383 } 384 385 static void __init kmap_init(void) 386 { 387 unsigned long kmap_vstart; 388 389 /* 390 * Cache the first kmap pte: 391 */ 392 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN); 393 kmap_pte = kmap_get_fixmap_pte(kmap_vstart); 394 395 kmap_prot = PAGE_KERNEL; 396 } 397 398 #ifdef CONFIG_HIGHMEM 399 static void __init permanent_kmaps_init(pgd_t *pgd_base) 400 { 401 unsigned long vaddr; 402 pgd_t *pgd; 403 pud_t *pud; 404 pmd_t *pmd; 405 pte_t *pte; 406 407 vaddr = PKMAP_BASE; 408 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base); 409 410 pgd = swapper_pg_dir + pgd_index(vaddr); 411 pud = pud_offset(pgd, vaddr); 412 pmd = pmd_offset(pud, vaddr); 413 pte = pte_offset_kernel(pmd, vaddr); 414 pkmap_page_table = pte; 415 } 416 417 static void __init add_one_highpage_init(struct page *page) 418 { 419 ClearPageReserved(page); 420 init_page_count(page); 421 __free_page(page); 422 totalhigh_pages++; 423 } 424 425 struct add_highpages_data { 426 unsigned long start_pfn; 427 unsigned long end_pfn; 428 }; 429 430 static int __init add_highpages_work_fn(unsigned long start_pfn, 431 unsigned long end_pfn, void *datax) 432 { 433 int node_pfn; 434 struct page *page; 435 unsigned long final_start_pfn, final_end_pfn; 436 struct add_highpages_data *data; 437 438 data = (struct add_highpages_data *)datax; 439 440 final_start_pfn = max(start_pfn, data->start_pfn); 441 final_end_pfn = min(end_pfn, data->end_pfn); 442 if (final_start_pfn >= final_end_pfn) 443 return 0; 444 445 for (node_pfn = final_start_pfn; node_pfn < final_end_pfn; 446 node_pfn++) { 447 if (!pfn_valid(node_pfn)) 448 continue; 449 page = pfn_to_page(node_pfn); 450 add_one_highpage_init(page); 451 } 452 453 return 0; 454 455 } 456 457 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn, 458 unsigned long end_pfn) 459 { 460 struct add_highpages_data data; 461 462 data.start_pfn = start_pfn; 463 data.end_pfn = end_pfn; 464 465 work_with_active_regions(nid, add_highpages_work_fn, &data); 466 } 467 468 #else 469 static inline void permanent_kmaps_init(pgd_t *pgd_base) 470 { 471 } 472 #endif /* CONFIG_HIGHMEM */ 473 474 void __init native_pagetable_setup_start(pgd_t *base) 475 { 476 unsigned long pfn, va; 477 pgd_t *pgd; 478 pud_t *pud; 479 pmd_t *pmd; 480 pte_t *pte; 481 482 /* 483 * Remove any mappings which extend past the end of physical 484 * memory from the boot time page table: 485 */ 486 for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) { 487 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT); 488 pgd = base + pgd_index(va); 489 if (!pgd_present(*pgd)) 490 break; 491 492 pud = pud_offset(pgd, va); 493 pmd = pmd_offset(pud, va); 494 if (!pmd_present(*pmd)) 495 break; 496 497 pte = pte_offset_kernel(pmd, va); 498 if (!pte_present(*pte)) 499 break; 500 501 pte_clear(NULL, va, pte); 502 } 503 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT); 504 } 505 506 void __init native_pagetable_setup_done(pgd_t *base) 507 { 508 } 509 510 /* 511 * Build a proper pagetable for the kernel mappings. Up until this 512 * point, we've been running on some set of pagetables constructed by 513 * the boot process. 514 * 515 * If we're booting on native hardware, this will be a pagetable 516 * constructed in arch/x86/kernel/head_32.S. The root of the 517 * pagetable will be swapper_pg_dir. 518 * 519 * If we're booting paravirtualized under a hypervisor, then there are 520 * more options: we may already be running PAE, and the pagetable may 521 * or may not be based in swapper_pg_dir. In any case, 522 * paravirt_pagetable_setup_start() will set up swapper_pg_dir 523 * appropriately for the rest of the initialization to work. 524 * 525 * In general, pagetable_init() assumes that the pagetable may already 526 * be partially populated, and so it avoids stomping on any existing 527 * mappings. 528 */ 529 void __init early_ioremap_page_table_range_init(void) 530 { 531 pgd_t *pgd_base = swapper_pg_dir; 532 unsigned long vaddr, end; 533 534 /* 535 * Fixed mappings, only the page table structure has to be 536 * created - mappings will be set by set_fixmap(): 537 */ 538 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK; 539 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK; 540 page_table_range_init(vaddr, end, pgd_base); 541 early_ioremap_reset(); 542 } 543 544 static void __init pagetable_init(void) 545 { 546 pgd_t *pgd_base = swapper_pg_dir; 547 548 permanent_kmaps_init(pgd_base); 549 } 550 551 #ifdef CONFIG_ACPI_SLEEP 552 /* 553 * ACPI suspend needs this for resume, because things like the intel-agp 554 * driver might have split up a kernel 4MB mapping. 555 */ 556 char swsusp_pg_dir[PAGE_SIZE] 557 __attribute__ ((aligned(PAGE_SIZE))); 558 559 static inline void save_pg_dir(void) 560 { 561 memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE); 562 } 563 #else /* !CONFIG_ACPI_SLEEP */ 564 static inline void save_pg_dir(void) 565 { 566 } 567 #endif /* !CONFIG_ACPI_SLEEP */ 568 569 void zap_low_mappings(bool early) 570 { 571 int i; 572 573 /* 574 * Zap initial low-memory mappings. 575 * 576 * Note that "pgd_clear()" doesn't do it for 577 * us, because pgd_clear() is a no-op on i386. 578 */ 579 for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) { 580 #ifdef CONFIG_X86_PAE 581 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page))); 582 #else 583 set_pgd(swapper_pg_dir+i, __pgd(0)); 584 #endif 585 } 586 587 if (early) 588 __flush_tlb(); 589 else 590 flush_tlb_all(); 591 } 592 593 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP); 594 EXPORT_SYMBOL_GPL(__supported_pte_mask); 595 596 /* user-defined highmem size */ 597 static unsigned int highmem_pages = -1; 598 599 /* 600 * highmem=size forces highmem to be exactly 'size' bytes. 601 * This works even on boxes that have no highmem otherwise. 602 * This also works to reduce highmem size on bigger boxes. 603 */ 604 static int __init parse_highmem(char *arg) 605 { 606 if (!arg) 607 return -EINVAL; 608 609 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT; 610 return 0; 611 } 612 early_param("highmem", parse_highmem); 613 614 #define MSG_HIGHMEM_TOO_BIG \ 615 "highmem size (%luMB) is bigger than pages available (%luMB)!\n" 616 617 #define MSG_LOWMEM_TOO_SMALL \ 618 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n" 619 /* 620 * All of RAM fits into lowmem - but if user wants highmem 621 * artificially via the highmem=x boot parameter then create 622 * it: 623 */ 624 void __init lowmem_pfn_init(void) 625 { 626 /* max_low_pfn is 0, we already have early_res support */ 627 max_low_pfn = max_pfn; 628 629 if (highmem_pages == -1) 630 highmem_pages = 0; 631 #ifdef CONFIG_HIGHMEM 632 if (highmem_pages >= max_pfn) { 633 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG, 634 pages_to_mb(highmem_pages), pages_to_mb(max_pfn)); 635 highmem_pages = 0; 636 } 637 if (highmem_pages) { 638 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) { 639 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL, 640 pages_to_mb(highmem_pages)); 641 highmem_pages = 0; 642 } 643 max_low_pfn -= highmem_pages; 644 } 645 #else 646 if (highmem_pages) 647 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n"); 648 #endif 649 } 650 651 #define MSG_HIGHMEM_TOO_SMALL \ 652 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n" 653 654 #define MSG_HIGHMEM_TRIMMED \ 655 "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n" 656 /* 657 * We have more RAM than fits into lowmem - we try to put it into 658 * highmem, also taking the highmem=x boot parameter into account: 659 */ 660 void __init highmem_pfn_init(void) 661 { 662 max_low_pfn = MAXMEM_PFN; 663 664 if (highmem_pages == -1) 665 highmem_pages = max_pfn - MAXMEM_PFN; 666 667 if (highmem_pages + MAXMEM_PFN < max_pfn) 668 max_pfn = MAXMEM_PFN + highmem_pages; 669 670 if (highmem_pages + MAXMEM_PFN > max_pfn) { 671 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL, 672 pages_to_mb(max_pfn - MAXMEM_PFN), 673 pages_to_mb(highmem_pages)); 674 highmem_pages = 0; 675 } 676 #ifndef CONFIG_HIGHMEM 677 /* Maximum memory usable is what is directly addressable */ 678 printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20); 679 if (max_pfn > MAX_NONPAE_PFN) 680 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n"); 681 else 682 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n"); 683 max_pfn = MAXMEM_PFN; 684 #else /* !CONFIG_HIGHMEM */ 685 #ifndef CONFIG_HIGHMEM64G 686 if (max_pfn > MAX_NONPAE_PFN) { 687 max_pfn = MAX_NONPAE_PFN; 688 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED); 689 } 690 #endif /* !CONFIG_HIGHMEM64G */ 691 #endif /* !CONFIG_HIGHMEM */ 692 } 693 694 /* 695 * Determine low and high memory ranges: 696 */ 697 void __init find_low_pfn_range(void) 698 { 699 /* it could update max_pfn */ 700 701 if (max_pfn <= MAXMEM_PFN) 702 lowmem_pfn_init(); 703 else 704 highmem_pfn_init(); 705 } 706 707 #ifndef CONFIG_NEED_MULTIPLE_NODES 708 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn, 709 int acpi, int k8) 710 { 711 #ifdef CONFIG_HIGHMEM 712 highstart_pfn = highend_pfn = max_pfn; 713 if (max_pfn > max_low_pfn) 714 highstart_pfn = max_low_pfn; 715 e820_register_active_regions(0, 0, highend_pfn); 716 sparse_memory_present_with_active_regions(0); 717 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n", 718 pages_to_mb(highend_pfn - highstart_pfn)); 719 num_physpages = highend_pfn; 720 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1; 721 #else 722 e820_register_active_regions(0, 0, max_low_pfn); 723 sparse_memory_present_with_active_regions(0); 724 num_physpages = max_low_pfn; 725 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1; 726 #endif 727 #ifdef CONFIG_FLATMEM 728 max_mapnr = num_physpages; 729 #endif 730 __vmalloc_start_set = true; 731 732 printk(KERN_NOTICE "%ldMB LOWMEM available.\n", 733 pages_to_mb(max_low_pfn)); 734 735 setup_bootmem_allocator(); 736 } 737 #endif /* !CONFIG_NEED_MULTIPLE_NODES */ 738 739 static void __init zone_sizes_init(void) 740 { 741 unsigned long max_zone_pfns[MAX_NR_ZONES]; 742 memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); 743 max_zone_pfns[ZONE_DMA] = 744 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT; 745 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 746 #ifdef CONFIG_HIGHMEM 747 max_zone_pfns[ZONE_HIGHMEM] = highend_pfn; 748 #endif 749 750 free_area_init_nodes(max_zone_pfns); 751 } 752 753 #ifndef CONFIG_NO_BOOTMEM 754 static unsigned long __init setup_node_bootmem(int nodeid, 755 unsigned long start_pfn, 756 unsigned long end_pfn, 757 unsigned long bootmap) 758 { 759 unsigned long bootmap_size; 760 761 /* don't touch min_low_pfn */ 762 bootmap_size = init_bootmem_node(NODE_DATA(nodeid), 763 bootmap >> PAGE_SHIFT, 764 start_pfn, end_pfn); 765 printk(KERN_INFO " node %d low ram: %08lx - %08lx\n", 766 nodeid, start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT); 767 printk(KERN_INFO " node %d bootmap %08lx - %08lx\n", 768 nodeid, bootmap, bootmap + bootmap_size); 769 free_bootmem_with_active_regions(nodeid, end_pfn); 770 771 return bootmap + bootmap_size; 772 } 773 #endif 774 775 void __init setup_bootmem_allocator(void) 776 { 777 #ifndef CONFIG_NO_BOOTMEM 778 int nodeid; 779 unsigned long bootmap_size, bootmap; 780 /* 781 * Initialize the boot-time allocator (with low memory only): 782 */ 783 bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT; 784 bootmap = find_e820_area(0, max_pfn_mapped<<PAGE_SHIFT, bootmap_size, 785 PAGE_SIZE); 786 if (bootmap == -1L) 787 panic("Cannot find bootmem map of size %ld\n", bootmap_size); 788 reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP"); 789 #endif 790 791 printk(KERN_INFO " mapped low ram: 0 - %08lx\n", 792 max_pfn_mapped<<PAGE_SHIFT); 793 printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT); 794 795 #ifndef CONFIG_NO_BOOTMEM 796 for_each_online_node(nodeid) { 797 unsigned long start_pfn, end_pfn; 798 799 #ifdef CONFIG_NEED_MULTIPLE_NODES 800 start_pfn = node_start_pfn[nodeid]; 801 end_pfn = node_end_pfn[nodeid]; 802 if (start_pfn > max_low_pfn) 803 continue; 804 if (end_pfn > max_low_pfn) 805 end_pfn = max_low_pfn; 806 #else 807 start_pfn = 0; 808 end_pfn = max_low_pfn; 809 #endif 810 bootmap = setup_node_bootmem(nodeid, start_pfn, end_pfn, 811 bootmap); 812 } 813 #endif 814 815 after_bootmem = 1; 816 } 817 818 /* 819 * paging_init() sets up the page tables - note that the first 8MB are 820 * already mapped by head.S. 821 * 822 * This routines also unmaps the page at virtual kernel address 0, so 823 * that we can trap those pesky NULL-reference errors in the kernel. 824 */ 825 void __init paging_init(void) 826 { 827 pagetable_init(); 828 829 __flush_tlb_all(); 830 831 kmap_init(); 832 833 /* 834 * NOTE: at this point the bootmem allocator is fully available. 835 */ 836 sparse_init(); 837 zone_sizes_init(); 838 } 839 840 /* 841 * Test if the WP bit works in supervisor mode. It isn't supported on 386's 842 * and also on some strange 486's. All 586+'s are OK. This used to involve 843 * black magic jumps to work around some nasty CPU bugs, but fortunately the 844 * switch to using exceptions got rid of all that. 845 */ 846 static void __init test_wp_bit(void) 847 { 848 printk(KERN_INFO 849 "Checking if this processor honours the WP bit even in supervisor mode..."); 850 851 /* Any page-aligned address will do, the test is non-destructive */ 852 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY); 853 boot_cpu_data.wp_works_ok = do_test_wp_bit(); 854 clear_fixmap(FIX_WP_TEST); 855 856 if (!boot_cpu_data.wp_works_ok) { 857 printk(KERN_CONT "No.\n"); 858 #ifdef CONFIG_X86_WP_WORKS_OK 859 panic( 860 "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!"); 861 #endif 862 } else { 863 printk(KERN_CONT "Ok.\n"); 864 } 865 } 866 867 void __init mem_init(void) 868 { 869 int codesize, reservedpages, datasize, initsize; 870 int tmp; 871 872 pci_iommu_alloc(); 873 874 #ifdef CONFIG_FLATMEM 875 BUG_ON(!mem_map); 876 #endif 877 /* this will put all low memory onto the freelists */ 878 totalram_pages += free_all_bootmem(); 879 880 reservedpages = 0; 881 for (tmp = 0; tmp < max_low_pfn; tmp++) 882 /* 883 * Only count reserved RAM pages: 884 */ 885 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp))) 886 reservedpages++; 887 888 set_highmem_pages_init(); 889 890 codesize = (unsigned long) &_etext - (unsigned long) &_text; 891 datasize = (unsigned long) &_edata - (unsigned long) &_etext; 892 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; 893 894 printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, " 895 "%dk reserved, %dk data, %dk init, %ldk highmem)\n", 896 nr_free_pages() << (PAGE_SHIFT-10), 897 num_physpages << (PAGE_SHIFT-10), 898 codesize >> 10, 899 reservedpages << (PAGE_SHIFT-10), 900 datasize >> 10, 901 initsize >> 10, 902 totalhigh_pages << (PAGE_SHIFT-10)); 903 904 printk(KERN_INFO "virtual kernel memory layout:\n" 905 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n" 906 #ifdef CONFIG_HIGHMEM 907 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n" 908 #endif 909 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n" 910 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n" 911 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n" 912 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n" 913 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n", 914 FIXADDR_START, FIXADDR_TOP, 915 (FIXADDR_TOP - FIXADDR_START) >> 10, 916 917 #ifdef CONFIG_HIGHMEM 918 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, 919 (LAST_PKMAP*PAGE_SIZE) >> 10, 920 #endif 921 922 VMALLOC_START, VMALLOC_END, 923 (VMALLOC_END - VMALLOC_START) >> 20, 924 925 (unsigned long)__va(0), (unsigned long)high_memory, 926 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20, 927 928 (unsigned long)&__init_begin, (unsigned long)&__init_end, 929 ((unsigned long)&__init_end - 930 (unsigned long)&__init_begin) >> 10, 931 932 (unsigned long)&_etext, (unsigned long)&_edata, 933 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10, 934 935 (unsigned long)&_text, (unsigned long)&_etext, 936 ((unsigned long)&_etext - (unsigned long)&_text) >> 10); 937 938 /* 939 * Check boundaries twice: Some fundamental inconsistencies can 940 * be detected at build time already. 941 */ 942 #define __FIXADDR_TOP (-PAGE_SIZE) 943 #ifdef CONFIG_HIGHMEM 944 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); 945 BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE); 946 #endif 947 #define high_memory (-128UL << 20) 948 BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END); 949 #undef high_memory 950 #undef __FIXADDR_TOP 951 952 #ifdef CONFIG_HIGHMEM 953 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); 954 BUG_ON(VMALLOC_END > PKMAP_BASE); 955 #endif 956 BUG_ON(VMALLOC_START >= VMALLOC_END); 957 BUG_ON((unsigned long)high_memory > VMALLOC_START); 958 959 if (boot_cpu_data.wp_works_ok < 0) 960 test_wp_bit(); 961 962 save_pg_dir(); 963 zap_low_mappings(true); 964 } 965 966 #ifdef CONFIG_MEMORY_HOTPLUG 967 int arch_add_memory(int nid, u64 start, u64 size) 968 { 969 struct pglist_data *pgdata = NODE_DATA(nid); 970 struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM; 971 unsigned long start_pfn = start >> PAGE_SHIFT; 972 unsigned long nr_pages = size >> PAGE_SHIFT; 973 974 return __add_pages(nid, zone, start_pfn, nr_pages); 975 } 976 #endif 977 978 /* 979 * This function cannot be __init, since exceptions don't work in that 980 * section. Put this after the callers, so that it cannot be inlined. 981 */ 982 static noinline int do_test_wp_bit(void) 983 { 984 char tmp_reg; 985 int flag; 986 987 __asm__ __volatile__( 988 " movb %0, %1 \n" 989 "1: movb %1, %0 \n" 990 " xorl %2, %2 \n" 991 "2: \n" 992 _ASM_EXTABLE(1b,2b) 993 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)), 994 "=q" (tmp_reg), 995 "=r" (flag) 996 :"2" (1) 997 :"memory"); 998 999 return flag; 1000 } 1001 1002 #ifdef CONFIG_DEBUG_RODATA 1003 const int rodata_test_data = 0xC3; 1004 EXPORT_SYMBOL_GPL(rodata_test_data); 1005 1006 int kernel_set_to_readonly __read_mostly; 1007 1008 void set_kernel_text_rw(void) 1009 { 1010 unsigned long start = PFN_ALIGN(_text); 1011 unsigned long size = PFN_ALIGN(_etext) - start; 1012 1013 if (!kernel_set_to_readonly) 1014 return; 1015 1016 pr_debug("Set kernel text: %lx - %lx for read write\n", 1017 start, start+size); 1018 1019 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT); 1020 } 1021 1022 void set_kernel_text_ro(void) 1023 { 1024 unsigned long start = PFN_ALIGN(_text); 1025 unsigned long size = PFN_ALIGN(_etext) - start; 1026 1027 if (!kernel_set_to_readonly) 1028 return; 1029 1030 pr_debug("Set kernel text: %lx - %lx for read only\n", 1031 start, start+size); 1032 1033 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 1034 } 1035 1036 void mark_rodata_ro(void) 1037 { 1038 unsigned long start = PFN_ALIGN(_text); 1039 unsigned long size = PFN_ALIGN(_etext) - start; 1040 1041 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 1042 printk(KERN_INFO "Write protecting the kernel text: %luk\n", 1043 size >> 10); 1044 1045 kernel_set_to_readonly = 1; 1046 1047 #ifdef CONFIG_CPA_DEBUG 1048 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n", 1049 start, start+size); 1050 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT); 1051 1052 printk(KERN_INFO "Testing CPA: write protecting again\n"); 1053 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT); 1054 #endif 1055 1056 start += size; 1057 size = (unsigned long)__end_rodata - start; 1058 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 1059 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", 1060 size >> 10); 1061 rodata_test(); 1062 1063 #ifdef CONFIG_CPA_DEBUG 1064 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size); 1065 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT); 1066 1067 printk(KERN_INFO "Testing CPA: write protecting again\n"); 1068 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 1069 #endif 1070 } 1071 #endif 1072 1073 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len, 1074 int flags) 1075 { 1076 return reserve_bootmem(phys, len, flags); 1077 } 1078