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/initrd.h> 22 #include <linux/pagemap.h> 23 #include <linux/bootmem.h> 24 #include <linux/proc_fs.h> 25 #include <linux/pci.h> 26 #include <linux/pfn.h> 27 #include <linux/poison.h> 28 #include <linux/dma-mapping.h> 29 #include <linux/module.h> 30 #include <linux/memory_hotplug.h> 31 #include <linux/nmi.h> 32 33 #include <asm/processor.h> 34 #include <asm/system.h> 35 #include <asm/uaccess.h> 36 #include <asm/pgtable.h> 37 #include <asm/pgalloc.h> 38 #include <asm/dma.h> 39 #include <asm/fixmap.h> 40 #include <asm/e820.h> 41 #include <asm/apic.h> 42 #include <asm/tlb.h> 43 #include <asm/mmu_context.h> 44 #include <asm/proto.h> 45 #include <asm/smp.h> 46 #include <asm/sections.h> 47 #include <asm/kdebug.h> 48 #include <asm/numa.h> 49 #include <asm/cacheflush.h> 50 51 /* 52 * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries. 53 * The direct mapping extends to max_pfn_mapped, so that we can directly access 54 * apertures, ACPI and other tables without having to play with fixmaps. 55 */ 56 unsigned long max_low_pfn_mapped; 57 unsigned long max_pfn_mapped; 58 59 static unsigned long dma_reserve __initdata; 60 61 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); 62 63 int direct_gbpages __meminitdata 64 #ifdef CONFIG_DIRECT_GBPAGES 65 = 1 66 #endif 67 ; 68 69 static int __init parse_direct_gbpages_off(char *arg) 70 { 71 direct_gbpages = 0; 72 return 0; 73 } 74 early_param("nogbpages", parse_direct_gbpages_off); 75 76 static int __init parse_direct_gbpages_on(char *arg) 77 { 78 direct_gbpages = 1; 79 return 0; 80 } 81 early_param("gbpages", parse_direct_gbpages_on); 82 83 /* 84 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the 85 * physical space so we can cache the place of the first one and move 86 * around without checking the pgd every time. 87 */ 88 89 int after_bootmem; 90 91 static __init void *spp_getpage(void) 92 { 93 void *ptr; 94 95 if (after_bootmem) 96 ptr = (void *) get_zeroed_page(GFP_ATOMIC); 97 else 98 ptr = alloc_bootmem_pages(PAGE_SIZE); 99 100 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) { 101 panic("set_pte_phys: cannot allocate page data %s\n", 102 after_bootmem ? "after bootmem" : ""); 103 } 104 105 pr_debug("spp_getpage %p\n", ptr); 106 107 return ptr; 108 } 109 110 void 111 set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte) 112 { 113 pud_t *pud; 114 pmd_t *pmd; 115 pte_t *pte; 116 117 pud = pud_page + pud_index(vaddr); 118 if (pud_none(*pud)) { 119 pmd = (pmd_t *) spp_getpage(); 120 pud_populate(&init_mm, pud, pmd); 121 if (pmd != pmd_offset(pud, 0)) { 122 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n", 123 pmd, pmd_offset(pud, 0)); 124 return; 125 } 126 } 127 pmd = pmd_offset(pud, vaddr); 128 if (pmd_none(*pmd)) { 129 pte = (pte_t *) spp_getpage(); 130 pmd_populate_kernel(&init_mm, pmd, pte); 131 if (pte != pte_offset_kernel(pmd, 0)) { 132 printk(KERN_ERR "PAGETABLE BUG #02!\n"); 133 return; 134 } 135 } 136 137 pte = pte_offset_kernel(pmd, vaddr); 138 if (!pte_none(*pte) && pte_val(new_pte) && 139 pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask)) 140 pte_ERROR(*pte); 141 set_pte(pte, new_pte); 142 143 /* 144 * It's enough to flush this one mapping. 145 * (PGE mappings get flushed as well) 146 */ 147 __flush_tlb_one(vaddr); 148 } 149 150 void 151 set_pte_vaddr(unsigned long vaddr, pte_t pteval) 152 { 153 pgd_t *pgd; 154 pud_t *pud_page; 155 156 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval)); 157 158 pgd = pgd_offset_k(vaddr); 159 if (pgd_none(*pgd)) { 160 printk(KERN_ERR 161 "PGD FIXMAP MISSING, it should be setup in head.S!\n"); 162 return; 163 } 164 pud_page = (pud_t*)pgd_page_vaddr(*pgd); 165 set_pte_vaddr_pud(pud_page, vaddr, pteval); 166 } 167 168 /* 169 * Create large page table mappings for a range of physical addresses. 170 */ 171 static void __init __init_extra_mapping(unsigned long phys, unsigned long size, 172 pgprot_t prot) 173 { 174 pgd_t *pgd; 175 pud_t *pud; 176 pmd_t *pmd; 177 178 BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK)); 179 for (; size; phys += PMD_SIZE, size -= PMD_SIZE) { 180 pgd = pgd_offset_k((unsigned long)__va(phys)); 181 if (pgd_none(*pgd)) { 182 pud = (pud_t *) spp_getpage(); 183 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE | 184 _PAGE_USER)); 185 } 186 pud = pud_offset(pgd, (unsigned long)__va(phys)); 187 if (pud_none(*pud)) { 188 pmd = (pmd_t *) spp_getpage(); 189 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | 190 _PAGE_USER)); 191 } 192 pmd = pmd_offset(pud, phys); 193 BUG_ON(!pmd_none(*pmd)); 194 set_pmd(pmd, __pmd(phys | pgprot_val(prot))); 195 } 196 } 197 198 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size) 199 { 200 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE); 201 } 202 203 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size) 204 { 205 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE); 206 } 207 208 /* 209 * The head.S code sets up the kernel high mapping: 210 * 211 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text) 212 * 213 * phys_addr holds the negative offset to the kernel, which is added 214 * to the compile time generated pmds. This results in invalid pmds up 215 * to the point where we hit the physaddr 0 mapping. 216 * 217 * We limit the mappings to the region from _text to _end. _end is 218 * rounded up to the 2MB boundary. This catches the invalid pmds as 219 * well, as they are located before _text: 220 */ 221 void __init cleanup_highmap(void) 222 { 223 unsigned long vaddr = __START_KERNEL_map; 224 unsigned long end = round_up((unsigned long)_end, PMD_SIZE) - 1; 225 pmd_t *pmd = level2_kernel_pgt; 226 pmd_t *last_pmd = pmd + PTRS_PER_PMD; 227 228 for (; pmd < last_pmd; pmd++, vaddr += PMD_SIZE) { 229 if (pmd_none(*pmd)) 230 continue; 231 if (vaddr < (unsigned long) _text || vaddr > end) 232 set_pmd(pmd, __pmd(0)); 233 } 234 } 235 236 static unsigned long __initdata table_start; 237 static unsigned long __meminitdata table_end; 238 static unsigned long __meminitdata table_top; 239 240 static __meminit void *alloc_low_page(unsigned long *phys) 241 { 242 unsigned long pfn = table_end++; 243 void *adr; 244 245 if (after_bootmem) { 246 adr = (void *)get_zeroed_page(GFP_ATOMIC); 247 *phys = __pa(adr); 248 249 return adr; 250 } 251 252 if (pfn >= table_top) 253 panic("alloc_low_page: ran out of memory"); 254 255 adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE); 256 memset(adr, 0, PAGE_SIZE); 257 *phys = pfn * PAGE_SIZE; 258 return adr; 259 } 260 261 static __meminit void unmap_low_page(void *adr) 262 { 263 if (after_bootmem) 264 return; 265 266 early_iounmap(adr, PAGE_SIZE); 267 } 268 269 static unsigned long __meminit 270 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end) 271 { 272 unsigned pages = 0; 273 unsigned long last_map_addr = end; 274 int i; 275 276 pte_t *pte = pte_page + pte_index(addr); 277 278 for(i = pte_index(addr); i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) { 279 280 if (addr >= end) { 281 if (!after_bootmem) { 282 for(; i < PTRS_PER_PTE; i++, pte++) 283 set_pte(pte, __pte(0)); 284 } 285 break; 286 } 287 288 if (pte_val(*pte)) 289 continue; 290 291 if (0) 292 printk(" pte=%p addr=%lx pte=%016lx\n", 293 pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte); 294 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL)); 295 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE; 296 pages++; 297 } 298 update_page_count(PG_LEVEL_4K, pages); 299 300 return last_map_addr; 301 } 302 303 static unsigned long __meminit 304 phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end) 305 { 306 pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd); 307 308 return phys_pte_init(pte, address, end); 309 } 310 311 static unsigned long __meminit 312 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end, 313 unsigned long page_size_mask) 314 { 315 unsigned long pages = 0; 316 unsigned long last_map_addr = end; 317 318 int i = pmd_index(address); 319 320 for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) { 321 unsigned long pte_phys; 322 pmd_t *pmd = pmd_page + pmd_index(address); 323 pte_t *pte; 324 325 if (address >= end) { 326 if (!after_bootmem) { 327 for (; i < PTRS_PER_PMD; i++, pmd++) 328 set_pmd(pmd, __pmd(0)); 329 } 330 break; 331 } 332 333 if (pmd_val(*pmd)) { 334 if (!pmd_large(*pmd)) 335 last_map_addr = phys_pte_update(pmd, address, 336 end); 337 continue; 338 } 339 340 if (page_size_mask & (1<<PG_LEVEL_2M)) { 341 pages++; 342 set_pte((pte_t *)pmd, 343 pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); 344 last_map_addr = (address & PMD_MASK) + PMD_SIZE; 345 continue; 346 } 347 348 pte = alloc_low_page(&pte_phys); 349 last_map_addr = phys_pte_init(pte, address, end); 350 unmap_low_page(pte); 351 352 pmd_populate_kernel(&init_mm, pmd, __va(pte_phys)); 353 } 354 update_page_count(PG_LEVEL_2M, pages); 355 return last_map_addr; 356 } 357 358 static unsigned long __meminit 359 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end, 360 unsigned long page_size_mask) 361 { 362 pmd_t *pmd = pmd_offset(pud, 0); 363 unsigned long last_map_addr; 364 365 spin_lock(&init_mm.page_table_lock); 366 last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask); 367 spin_unlock(&init_mm.page_table_lock); 368 __flush_tlb_all(); 369 return last_map_addr; 370 } 371 372 static unsigned long __meminit 373 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end, 374 unsigned long page_size_mask) 375 { 376 unsigned long pages = 0; 377 unsigned long last_map_addr = end; 378 int i = pud_index(addr); 379 380 for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) { 381 unsigned long pmd_phys; 382 pud_t *pud = pud_page + pud_index(addr); 383 pmd_t *pmd; 384 385 if (addr >= end) 386 break; 387 388 if (!after_bootmem && 389 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) { 390 set_pud(pud, __pud(0)); 391 continue; 392 } 393 394 if (pud_val(*pud)) { 395 if (!pud_large(*pud)) 396 last_map_addr = phys_pmd_update(pud, addr, end, 397 page_size_mask); 398 continue; 399 } 400 401 if (page_size_mask & (1<<PG_LEVEL_1G)) { 402 pages++; 403 set_pte((pte_t *)pud, 404 pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL_LARGE)); 405 last_map_addr = (addr & PUD_MASK) + PUD_SIZE; 406 continue; 407 } 408 409 pmd = alloc_low_page(&pmd_phys); 410 411 spin_lock(&init_mm.page_table_lock); 412 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask); 413 unmap_low_page(pmd); 414 pud_populate(&init_mm, pud, __va(pmd_phys)); 415 spin_unlock(&init_mm.page_table_lock); 416 417 } 418 __flush_tlb_all(); 419 update_page_count(PG_LEVEL_1G, pages); 420 421 return last_map_addr; 422 } 423 424 static unsigned long __meminit 425 phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end, 426 unsigned long page_size_mask) 427 { 428 pud_t *pud; 429 430 pud = (pud_t *)pgd_page_vaddr(*pgd); 431 432 return phys_pud_init(pud, addr, end, page_size_mask); 433 } 434 435 static void __init find_early_table_space(unsigned long end) 436 { 437 unsigned long puds, pmds, ptes, tables, start; 438 439 puds = (end + PUD_SIZE - 1) >> PUD_SHIFT; 440 tables = round_up(puds * sizeof(pud_t), PAGE_SIZE); 441 if (direct_gbpages) { 442 unsigned long extra; 443 extra = end - ((end>>PUD_SHIFT) << PUD_SHIFT); 444 pmds = (extra + PMD_SIZE - 1) >> PMD_SHIFT; 445 } else 446 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT; 447 tables += round_up(pmds * sizeof(pmd_t), PAGE_SIZE); 448 449 if (cpu_has_pse) { 450 unsigned long extra; 451 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT); 452 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT; 453 } else 454 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT; 455 tables += round_up(ptes * sizeof(pte_t), PAGE_SIZE); 456 457 /* 458 * RED-PEN putting page tables only on node 0 could 459 * cause a hotspot and fill up ZONE_DMA. The page tables 460 * need roughly 0.5KB per GB. 461 */ 462 start = 0x8000; 463 table_start = find_e820_area(start, end, tables, PAGE_SIZE); 464 if (table_start == -1UL) 465 panic("Cannot find space for the kernel page tables"); 466 467 table_start >>= PAGE_SHIFT; 468 table_end = table_start; 469 table_top = table_start + (tables >> PAGE_SHIFT); 470 471 printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n", 472 end, table_start << PAGE_SHIFT, table_top << PAGE_SHIFT); 473 } 474 475 static void __init init_gbpages(void) 476 { 477 if (direct_gbpages && cpu_has_gbpages) 478 printk(KERN_INFO "Using GB pages for direct mapping\n"); 479 else 480 direct_gbpages = 0; 481 } 482 483 static unsigned long __init kernel_physical_mapping_init(unsigned long start, 484 unsigned long end, 485 unsigned long page_size_mask) 486 { 487 488 unsigned long next, last_map_addr = end; 489 490 start = (unsigned long)__va(start); 491 end = (unsigned long)__va(end); 492 493 for (; start < end; start = next) { 494 pgd_t *pgd = pgd_offset_k(start); 495 unsigned long pud_phys; 496 pud_t *pud; 497 498 next = (start + PGDIR_SIZE) & PGDIR_MASK; 499 if (next > end) 500 next = end; 501 502 if (pgd_val(*pgd)) { 503 last_map_addr = phys_pud_update(pgd, __pa(start), 504 __pa(end), page_size_mask); 505 continue; 506 } 507 508 if (after_bootmem) 509 pud = pud_offset(pgd, start & PGDIR_MASK); 510 else 511 pud = alloc_low_page(&pud_phys); 512 513 last_map_addr = phys_pud_init(pud, __pa(start), __pa(next), 514 page_size_mask); 515 unmap_low_page(pud); 516 pgd_populate(&init_mm, pgd_offset_k(start), 517 __va(pud_phys)); 518 } 519 520 return last_map_addr; 521 } 522 523 struct map_range { 524 unsigned long start; 525 unsigned long end; 526 unsigned page_size_mask; 527 }; 528 529 #define NR_RANGE_MR 5 530 531 static int save_mr(struct map_range *mr, int nr_range, 532 unsigned long start_pfn, unsigned long end_pfn, 533 unsigned long page_size_mask) 534 { 535 536 if (start_pfn < end_pfn) { 537 if (nr_range >= NR_RANGE_MR) 538 panic("run out of range for init_memory_mapping\n"); 539 mr[nr_range].start = start_pfn<<PAGE_SHIFT; 540 mr[nr_range].end = end_pfn<<PAGE_SHIFT; 541 mr[nr_range].page_size_mask = page_size_mask; 542 nr_range++; 543 } 544 545 return nr_range; 546 } 547 548 /* 549 * Setup the direct mapping of the physical memory at PAGE_OFFSET. 550 * This runs before bootmem is initialized and gets pages directly from 551 * the physical memory. To access them they are temporarily mapped. 552 */ 553 unsigned long __init_refok init_memory_mapping(unsigned long start, 554 unsigned long end) 555 { 556 unsigned long last_map_addr = 0; 557 unsigned long page_size_mask = 0; 558 unsigned long start_pfn, end_pfn; 559 560 struct map_range mr[NR_RANGE_MR]; 561 int nr_range, i; 562 563 printk(KERN_INFO "init_memory_mapping\n"); 564 565 /* 566 * Find space for the kernel direct mapping tables. 567 * 568 * Later we should allocate these tables in the local node of the 569 * memory mapped. Unfortunately this is done currently before the 570 * nodes are discovered. 571 */ 572 if (!after_bootmem) 573 init_gbpages(); 574 575 if (direct_gbpages) 576 page_size_mask |= 1 << PG_LEVEL_1G; 577 if (cpu_has_pse) 578 page_size_mask |= 1 << PG_LEVEL_2M; 579 580 memset(mr, 0, sizeof(mr)); 581 nr_range = 0; 582 583 /* head if not big page alignment ?*/ 584 start_pfn = start >> PAGE_SHIFT; 585 end_pfn = ((start + (PMD_SIZE - 1)) >> PMD_SHIFT) 586 << (PMD_SHIFT - PAGE_SHIFT); 587 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); 588 589 /* big page (2M) range*/ 590 start_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT) 591 << (PMD_SHIFT - PAGE_SHIFT); 592 end_pfn = ((start + (PUD_SIZE - 1))>>PUD_SHIFT) 593 << (PUD_SHIFT - PAGE_SHIFT); 594 if (end_pfn > ((end>>PUD_SHIFT)<<(PUD_SHIFT - PAGE_SHIFT))) 595 end_pfn = ((end>>PUD_SHIFT)<<(PUD_SHIFT - PAGE_SHIFT)); 596 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 597 page_size_mask & (1<<PG_LEVEL_2M)); 598 599 /* big page (1G) range */ 600 start_pfn = end_pfn; 601 end_pfn = (end>>PUD_SHIFT) << (PUD_SHIFT - PAGE_SHIFT); 602 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 603 page_size_mask & 604 ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G))); 605 606 /* tail is not big page (1G) alignment */ 607 start_pfn = end_pfn; 608 end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT); 609 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 610 page_size_mask & (1<<PG_LEVEL_2M)); 611 612 /* tail is not big page (2M) alignment */ 613 start_pfn = end_pfn; 614 end_pfn = end>>PAGE_SHIFT; 615 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); 616 617 /* try to merge same page size and continuous */ 618 for (i = 0; nr_range > 1 && i < nr_range - 1; i++) { 619 unsigned long old_start; 620 if (mr[i].end != mr[i+1].start || 621 mr[i].page_size_mask != mr[i+1].page_size_mask) 622 continue; 623 /* move it */ 624 old_start = mr[i].start; 625 memmove(&mr[i], &mr[i+1], 626 (nr_range - 1 - i) * sizeof (struct map_range)); 627 mr[i].start = old_start; 628 nr_range--; 629 } 630 631 for (i = 0; i < nr_range; i++) 632 printk(KERN_DEBUG " %010lx - %010lx page %s\n", 633 mr[i].start, mr[i].end, 634 (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":( 635 (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k")); 636 637 if (!after_bootmem) 638 find_early_table_space(end); 639 640 for (i = 0; i < nr_range; i++) 641 last_map_addr = kernel_physical_mapping_init( 642 mr[i].start, mr[i].end, 643 mr[i].page_size_mask); 644 645 if (!after_bootmem) 646 mmu_cr4_features = read_cr4(); 647 __flush_tlb_all(); 648 649 if (!after_bootmem && table_end > table_start) 650 reserve_early(table_start << PAGE_SHIFT, 651 table_end << PAGE_SHIFT, "PGTABLE"); 652 653 printk(KERN_INFO "last_map_addr: %lx end: %lx\n", 654 last_map_addr, end); 655 656 if (!after_bootmem) 657 early_memtest(start, end); 658 659 return last_map_addr >> PAGE_SHIFT; 660 } 661 662 #ifndef CONFIG_NUMA 663 void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn) 664 { 665 unsigned long bootmap_size, bootmap; 666 667 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT; 668 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size, 669 PAGE_SIZE); 670 if (bootmap == -1L) 671 panic("Cannot find bootmem map of size %ld\n", bootmap_size); 672 /* don't touch min_low_pfn */ 673 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT, 674 0, end_pfn); 675 e820_register_active_regions(0, start_pfn, end_pfn); 676 free_bootmem_with_active_regions(0, end_pfn); 677 early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT); 678 reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT); 679 } 680 681 void __init paging_init(void) 682 { 683 unsigned long max_zone_pfns[MAX_NR_ZONES]; 684 685 memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); 686 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN; 687 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN; 688 max_zone_pfns[ZONE_NORMAL] = max_pfn; 689 690 memory_present(0, 0, max_pfn); 691 sparse_init(); 692 free_area_init_nodes(max_zone_pfns); 693 } 694 #endif 695 696 /* 697 * Memory hotplug specific functions 698 */ 699 #ifdef CONFIG_MEMORY_HOTPLUG 700 /* 701 * Memory is added always to NORMAL zone. This means you will never get 702 * additional DMA/DMA32 memory. 703 */ 704 int arch_add_memory(int nid, u64 start, u64 size) 705 { 706 struct pglist_data *pgdat = NODE_DATA(nid); 707 struct zone *zone = pgdat->node_zones + ZONE_NORMAL; 708 unsigned long last_mapped_pfn, start_pfn = start >> PAGE_SHIFT; 709 unsigned long nr_pages = size >> PAGE_SHIFT; 710 int ret; 711 712 last_mapped_pfn = init_memory_mapping(start, start + size-1); 713 if (last_mapped_pfn > max_pfn_mapped) 714 max_pfn_mapped = last_mapped_pfn; 715 716 ret = __add_pages(zone, start_pfn, nr_pages); 717 WARN_ON(1); 718 719 return ret; 720 } 721 EXPORT_SYMBOL_GPL(arch_add_memory); 722 723 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA) 724 int memory_add_physaddr_to_nid(u64 start) 725 { 726 return 0; 727 } 728 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid); 729 #endif 730 731 #endif /* CONFIG_MEMORY_HOTPLUG */ 732 733 /* 734 * devmem_is_allowed() checks to see if /dev/mem access to a certain address 735 * is valid. The argument is a physical page number. 736 * 737 * 738 * On x86, access has to be given to the first megabyte of ram because that area 739 * contains bios code and data regions used by X and dosemu and similar apps. 740 * Access has to be given to non-kernel-ram areas as well, these contain the PCI 741 * mmio resources as well as potential bios/acpi data regions. 742 */ 743 int devmem_is_allowed(unsigned long pagenr) 744 { 745 if (pagenr <= 256) 746 return 1; 747 if (!page_is_ram(pagenr)) 748 return 1; 749 return 0; 750 } 751 752 753 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel, 754 kcore_modules, kcore_vsyscall; 755 756 void __init mem_init(void) 757 { 758 long codesize, reservedpages, datasize, initsize; 759 760 pci_iommu_alloc(); 761 762 /* clear_bss() already clear the empty_zero_page */ 763 764 reservedpages = 0; 765 766 /* this will put all low memory onto the freelists */ 767 #ifdef CONFIG_NUMA 768 totalram_pages = numa_free_all_bootmem(); 769 #else 770 totalram_pages = free_all_bootmem(); 771 #endif 772 reservedpages = max_pfn - totalram_pages - 773 absent_pages_in_range(0, max_pfn); 774 after_bootmem = 1; 775 776 codesize = (unsigned long) &_etext - (unsigned long) &_text; 777 datasize = (unsigned long) &_edata - (unsigned long) &_etext; 778 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin; 779 780 /* Register memory areas for /proc/kcore */ 781 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT); 782 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START, 783 VMALLOC_END-VMALLOC_START); 784 kclist_add(&kcore_kernel, &_stext, _end - _stext); 785 kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN); 786 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START, 787 VSYSCALL_END - VSYSCALL_START); 788 789 printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, " 790 "%ldk reserved, %ldk data, %ldk init)\n", 791 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10), 792 max_pfn << (PAGE_SHIFT-10), 793 codesize >> 10, 794 reservedpages << (PAGE_SHIFT-10), 795 datasize >> 10, 796 initsize >> 10); 797 798 cpa_init(); 799 } 800 801 void free_init_pages(char *what, unsigned long begin, unsigned long end) 802 { 803 unsigned long addr = begin; 804 805 if (addr >= end) 806 return; 807 808 /* 809 * If debugging page accesses then do not free this memory but 810 * mark them not present - any buggy init-section access will 811 * create a kernel page fault: 812 */ 813 #ifdef CONFIG_DEBUG_PAGEALLOC 814 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n", 815 begin, PAGE_ALIGN(end)); 816 set_memory_np(begin, (end - begin) >> PAGE_SHIFT); 817 #else 818 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10); 819 820 for (; addr < end; addr += PAGE_SIZE) { 821 ClearPageReserved(virt_to_page(addr)); 822 init_page_count(virt_to_page(addr)); 823 memset((void *)(addr & ~(PAGE_SIZE-1)), 824 POISON_FREE_INITMEM, PAGE_SIZE); 825 free_page(addr); 826 totalram_pages++; 827 } 828 #endif 829 } 830 831 void free_initmem(void) 832 { 833 free_init_pages("unused kernel memory", 834 (unsigned long)(&__init_begin), 835 (unsigned long)(&__init_end)); 836 } 837 838 #ifdef CONFIG_DEBUG_RODATA 839 const int rodata_test_data = 0xC3; 840 EXPORT_SYMBOL_GPL(rodata_test_data); 841 842 void mark_rodata_ro(void) 843 { 844 unsigned long start = PFN_ALIGN(_stext), end = PFN_ALIGN(__end_rodata); 845 unsigned long rodata_start = 846 ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK; 847 848 #ifdef CONFIG_DYNAMIC_FTRACE 849 /* Dynamic tracing modifies the kernel text section */ 850 start = rodata_start; 851 #endif 852 853 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", 854 (end - start) >> 10); 855 set_memory_ro(start, (end - start) >> PAGE_SHIFT); 856 857 /* 858 * The rodata section (but not the kernel text!) should also be 859 * not-executable. 860 */ 861 set_memory_nx(rodata_start, (end - rodata_start) >> PAGE_SHIFT); 862 863 rodata_test(); 864 865 #ifdef CONFIG_CPA_DEBUG 866 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end); 867 set_memory_rw(start, (end-start) >> PAGE_SHIFT); 868 869 printk(KERN_INFO "Testing CPA: again\n"); 870 set_memory_ro(start, (end-start) >> PAGE_SHIFT); 871 #endif 872 } 873 874 #endif 875 876 #ifdef CONFIG_BLK_DEV_INITRD 877 void free_initrd_mem(unsigned long start, unsigned long end) 878 { 879 free_init_pages("initrd memory", start, end); 880 } 881 #endif 882 883 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len, 884 int flags) 885 { 886 #ifdef CONFIG_NUMA 887 int nid, next_nid; 888 int ret; 889 #endif 890 unsigned long pfn = phys >> PAGE_SHIFT; 891 892 if (pfn >= max_pfn) { 893 /* 894 * This can happen with kdump kernels when accessing 895 * firmware tables: 896 */ 897 if (pfn < max_pfn_mapped) 898 return -EFAULT; 899 900 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %lu\n", 901 phys, len); 902 return -EFAULT; 903 } 904 905 /* Should check here against the e820 map to avoid double free */ 906 #ifdef CONFIG_NUMA 907 nid = phys_to_nid(phys); 908 next_nid = phys_to_nid(phys + len - 1); 909 if (nid == next_nid) 910 ret = reserve_bootmem_node(NODE_DATA(nid), phys, len, flags); 911 else 912 ret = reserve_bootmem(phys, len, flags); 913 914 if (ret != 0) 915 return ret; 916 917 #else 918 reserve_bootmem(phys, len, BOOTMEM_DEFAULT); 919 #endif 920 921 if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) { 922 dma_reserve += len / PAGE_SIZE; 923 set_dma_reserve(dma_reserve); 924 } 925 926 return 0; 927 } 928 929 int kern_addr_valid(unsigned long addr) 930 { 931 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT; 932 pgd_t *pgd; 933 pud_t *pud; 934 pmd_t *pmd; 935 pte_t *pte; 936 937 if (above != 0 && above != -1UL) 938 return 0; 939 940 pgd = pgd_offset_k(addr); 941 if (pgd_none(*pgd)) 942 return 0; 943 944 pud = pud_offset(pgd, addr); 945 if (pud_none(*pud)) 946 return 0; 947 948 pmd = pmd_offset(pud, addr); 949 if (pmd_none(*pmd)) 950 return 0; 951 952 if (pmd_large(*pmd)) 953 return pfn_valid(pmd_pfn(*pmd)); 954 955 pte = pte_offset_kernel(pmd, addr); 956 if (pte_none(*pte)) 957 return 0; 958 959 return pfn_valid(pte_pfn(*pte)); 960 } 961 962 /* 963 * A pseudo VMA to allow ptrace access for the vsyscall page. This only 964 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does 965 * not need special handling anymore: 966 */ 967 static struct vm_area_struct gate_vma = { 968 .vm_start = VSYSCALL_START, 969 .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE), 970 .vm_page_prot = PAGE_READONLY_EXEC, 971 .vm_flags = VM_READ | VM_EXEC 972 }; 973 974 struct vm_area_struct *get_gate_vma(struct task_struct *tsk) 975 { 976 #ifdef CONFIG_IA32_EMULATION 977 if (test_tsk_thread_flag(tsk, TIF_IA32)) 978 return NULL; 979 #endif 980 return &gate_vma; 981 } 982 983 int in_gate_area(struct task_struct *task, unsigned long addr) 984 { 985 struct vm_area_struct *vma = get_gate_vma(task); 986 987 if (!vma) 988 return 0; 989 990 return (addr >= vma->vm_start) && (addr < vma->vm_end); 991 } 992 993 /* 994 * Use this when you have no reliable task/vma, typically from interrupt 995 * context. It is less reliable than using the task's vma and may give 996 * false positives: 997 */ 998 int in_gate_area_no_task(unsigned long addr) 999 { 1000 return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END); 1001 } 1002 1003 const char *arch_vma_name(struct vm_area_struct *vma) 1004 { 1005 if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso) 1006 return "[vdso]"; 1007 if (vma == &gate_vma) 1008 return "[vsyscall]"; 1009 return NULL; 1010 } 1011 1012 #ifdef CONFIG_SPARSEMEM_VMEMMAP 1013 /* 1014 * Initialise the sparsemem vmemmap using huge-pages at the PMD level. 1015 */ 1016 static long __meminitdata addr_start, addr_end; 1017 static void __meminitdata *p_start, *p_end; 1018 static int __meminitdata node_start; 1019 1020 int __meminit 1021 vmemmap_populate(struct page *start_page, unsigned long size, int node) 1022 { 1023 unsigned long addr = (unsigned long)start_page; 1024 unsigned long end = (unsigned long)(start_page + size); 1025 unsigned long next; 1026 pgd_t *pgd; 1027 pud_t *pud; 1028 pmd_t *pmd; 1029 1030 for (; addr < end; addr = next) { 1031 void *p = NULL; 1032 1033 pgd = vmemmap_pgd_populate(addr, node); 1034 if (!pgd) 1035 return -ENOMEM; 1036 1037 pud = vmemmap_pud_populate(pgd, addr, node); 1038 if (!pud) 1039 return -ENOMEM; 1040 1041 if (!cpu_has_pse) { 1042 next = (addr + PAGE_SIZE) & PAGE_MASK; 1043 pmd = vmemmap_pmd_populate(pud, addr, node); 1044 1045 if (!pmd) 1046 return -ENOMEM; 1047 1048 p = vmemmap_pte_populate(pmd, addr, node); 1049 1050 if (!p) 1051 return -ENOMEM; 1052 1053 addr_end = addr + PAGE_SIZE; 1054 p_end = p + PAGE_SIZE; 1055 } else { 1056 next = pmd_addr_end(addr, end); 1057 1058 pmd = pmd_offset(pud, addr); 1059 if (pmd_none(*pmd)) { 1060 pte_t entry; 1061 1062 p = vmemmap_alloc_block(PMD_SIZE, node); 1063 if (!p) 1064 return -ENOMEM; 1065 1066 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, 1067 PAGE_KERNEL_LARGE); 1068 set_pmd(pmd, __pmd(pte_val(entry))); 1069 1070 /* check to see if we have contiguous blocks */ 1071 if (p_end != p || node_start != node) { 1072 if (p_start) 1073 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", 1074 addr_start, addr_end-1, p_start, p_end-1, node_start); 1075 addr_start = addr; 1076 node_start = node; 1077 p_start = p; 1078 } 1079 1080 addr_end = addr + PMD_SIZE; 1081 p_end = p + PMD_SIZE; 1082 } else 1083 vmemmap_verify((pte_t *)pmd, node, addr, next); 1084 } 1085 1086 } 1087 return 0; 1088 } 1089 1090 void __meminit vmemmap_populate_print_last(void) 1091 { 1092 if (p_start) { 1093 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n", 1094 addr_start, addr_end-1, p_start, p_end-1, node_start); 1095 p_start = NULL; 1096 p_end = NULL; 1097 node_start = 0; 1098 } 1099 } 1100 #endif 1101