1 #include <linux/gfp.h> 2 #include <linux/initrd.h> 3 #include <linux/ioport.h> 4 #include <linux/swap.h> 5 #include <linux/memblock.h> 6 #include <linux/bootmem.h> /* for max_low_pfn */ 7 8 #include <asm/cacheflush.h> 9 #include <asm/e820.h> 10 #include <asm/init.h> 11 #include <asm/page.h> 12 #include <asm/page_types.h> 13 #include <asm/sections.h> 14 #include <asm/setup.h> 15 #include <asm/tlbflush.h> 16 #include <asm/tlb.h> 17 #include <asm/proto.h> 18 #include <asm/dma.h> /* for MAX_DMA_PFN */ 19 #include <asm/microcode.h> 20 21 /* 22 * We need to define the tracepoints somewhere, and tlb.c 23 * is only compied when SMP=y. 24 */ 25 #define CREATE_TRACE_POINTS 26 #include <trace/events/tlb.h> 27 28 #include "mm_internal.h" 29 30 /* 31 * Tables translating between page_cache_type_t and pte encoding. 32 * Minimal supported modes are defined statically, modified if more supported 33 * cache modes are available. 34 * Index into __cachemode2pte_tbl is the cachemode. 35 * Index into __pte2cachemode_tbl are the caching attribute bits of the pte 36 * (_PAGE_PWT, _PAGE_PCD, _PAGE_PAT) at index bit positions 0, 1, 2. 37 */ 38 uint16_t __cachemode2pte_tbl[_PAGE_CACHE_MODE_NUM] = { 39 [_PAGE_CACHE_MODE_WB] = 0, 40 [_PAGE_CACHE_MODE_WC] = _PAGE_PWT, 41 [_PAGE_CACHE_MODE_UC_MINUS] = _PAGE_PCD, 42 [_PAGE_CACHE_MODE_UC] = _PAGE_PCD | _PAGE_PWT, 43 [_PAGE_CACHE_MODE_WT] = _PAGE_PCD, 44 [_PAGE_CACHE_MODE_WP] = _PAGE_PCD, 45 }; 46 EXPORT_SYMBOL(__cachemode2pte_tbl); 47 uint8_t __pte2cachemode_tbl[8] = { 48 [__pte2cm_idx(0)] = _PAGE_CACHE_MODE_WB, 49 [__pte2cm_idx(_PAGE_PWT)] = _PAGE_CACHE_MODE_WC, 50 [__pte2cm_idx(_PAGE_PCD)] = _PAGE_CACHE_MODE_UC_MINUS, 51 [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD)] = _PAGE_CACHE_MODE_UC, 52 [__pte2cm_idx(_PAGE_PAT)] = _PAGE_CACHE_MODE_WB, 53 [__pte2cm_idx(_PAGE_PWT | _PAGE_PAT)] = _PAGE_CACHE_MODE_WC, 54 [__pte2cm_idx(_PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC_MINUS, 55 [__pte2cm_idx(_PAGE_PWT | _PAGE_PCD | _PAGE_PAT)] = _PAGE_CACHE_MODE_UC, 56 }; 57 EXPORT_SYMBOL(__pte2cachemode_tbl); 58 59 static unsigned long __initdata pgt_buf_start; 60 static unsigned long __initdata pgt_buf_end; 61 static unsigned long __initdata pgt_buf_top; 62 63 static unsigned long min_pfn_mapped; 64 65 static bool __initdata can_use_brk_pgt = true; 66 67 /* 68 * Pages returned are already directly mapped. 69 * 70 * Changing that is likely to break Xen, see commit: 71 * 72 * 279b706 x86,xen: introduce x86_init.mapping.pagetable_reserve 73 * 74 * for detailed information. 75 */ 76 __ref void *alloc_low_pages(unsigned int num) 77 { 78 unsigned long pfn; 79 int i; 80 81 if (after_bootmem) { 82 unsigned int order; 83 84 order = get_order((unsigned long)num << PAGE_SHIFT); 85 return (void *)__get_free_pages(GFP_ATOMIC | __GFP_NOTRACK | 86 __GFP_ZERO, order); 87 } 88 89 if ((pgt_buf_end + num) > pgt_buf_top || !can_use_brk_pgt) { 90 unsigned long ret; 91 if (min_pfn_mapped >= max_pfn_mapped) 92 panic("alloc_low_pages: ran out of memory"); 93 ret = memblock_find_in_range(min_pfn_mapped << PAGE_SHIFT, 94 max_pfn_mapped << PAGE_SHIFT, 95 PAGE_SIZE * num , PAGE_SIZE); 96 if (!ret) 97 panic("alloc_low_pages: can not alloc memory"); 98 memblock_reserve(ret, PAGE_SIZE * num); 99 pfn = ret >> PAGE_SHIFT; 100 } else { 101 pfn = pgt_buf_end; 102 pgt_buf_end += num; 103 printk(KERN_DEBUG "BRK [%#010lx, %#010lx] PGTABLE\n", 104 pfn << PAGE_SHIFT, (pgt_buf_end << PAGE_SHIFT) - 1); 105 } 106 107 for (i = 0; i < num; i++) { 108 void *adr; 109 110 adr = __va((pfn + i) << PAGE_SHIFT); 111 clear_page(adr); 112 } 113 114 return __va(pfn << PAGE_SHIFT); 115 } 116 117 /* need 3 4k for initial PMD_SIZE, 3 4k for 0-ISA_END_ADDRESS */ 118 #define INIT_PGT_BUF_SIZE (6 * PAGE_SIZE) 119 RESERVE_BRK(early_pgt_alloc, INIT_PGT_BUF_SIZE); 120 void __init early_alloc_pgt_buf(void) 121 { 122 unsigned long tables = INIT_PGT_BUF_SIZE; 123 phys_addr_t base; 124 125 base = __pa(extend_brk(tables, PAGE_SIZE)); 126 127 pgt_buf_start = base >> PAGE_SHIFT; 128 pgt_buf_end = pgt_buf_start; 129 pgt_buf_top = pgt_buf_start + (tables >> PAGE_SHIFT); 130 } 131 132 int after_bootmem; 133 134 int direct_gbpages 135 #ifdef CONFIG_DIRECT_GBPAGES 136 = 1 137 #endif 138 ; 139 140 static void __init init_gbpages(void) 141 { 142 #ifdef CONFIG_X86_64 143 if (direct_gbpages && cpu_has_gbpages) 144 printk(KERN_INFO "Using GB pages for direct mapping\n"); 145 else 146 direct_gbpages = 0; 147 #endif 148 } 149 150 struct map_range { 151 unsigned long start; 152 unsigned long end; 153 unsigned page_size_mask; 154 }; 155 156 static int page_size_mask; 157 158 static void __init probe_page_size_mask(void) 159 { 160 init_gbpages(); 161 162 #if !defined(CONFIG_DEBUG_PAGEALLOC) && !defined(CONFIG_KMEMCHECK) 163 /* 164 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages. 165 * This will simplify cpa(), which otherwise needs to support splitting 166 * large pages into small in interrupt context, etc. 167 */ 168 if (direct_gbpages) 169 page_size_mask |= 1 << PG_LEVEL_1G; 170 if (cpu_has_pse) 171 page_size_mask |= 1 << PG_LEVEL_2M; 172 #endif 173 174 /* Enable PSE if available */ 175 if (cpu_has_pse) 176 cr4_set_bits_and_update_boot(X86_CR4_PSE); 177 178 /* Enable PGE if available */ 179 if (cpu_has_pge) { 180 cr4_set_bits_and_update_boot(X86_CR4_PGE); 181 __supported_pte_mask |= _PAGE_GLOBAL; 182 } 183 } 184 185 #ifdef CONFIG_X86_32 186 #define NR_RANGE_MR 3 187 #else /* CONFIG_X86_64 */ 188 #define NR_RANGE_MR 5 189 #endif 190 191 static int __meminit save_mr(struct map_range *mr, int nr_range, 192 unsigned long start_pfn, unsigned long end_pfn, 193 unsigned long page_size_mask) 194 { 195 if (start_pfn < end_pfn) { 196 if (nr_range >= NR_RANGE_MR) 197 panic("run out of range for init_memory_mapping\n"); 198 mr[nr_range].start = start_pfn<<PAGE_SHIFT; 199 mr[nr_range].end = end_pfn<<PAGE_SHIFT; 200 mr[nr_range].page_size_mask = page_size_mask; 201 nr_range++; 202 } 203 204 return nr_range; 205 } 206 207 /* 208 * adjust the page_size_mask for small range to go with 209 * big page size instead small one if nearby are ram too. 210 */ 211 static void __init_refok adjust_range_page_size_mask(struct map_range *mr, 212 int nr_range) 213 { 214 int i; 215 216 for (i = 0; i < nr_range; i++) { 217 if ((page_size_mask & (1<<PG_LEVEL_2M)) && 218 !(mr[i].page_size_mask & (1<<PG_LEVEL_2M))) { 219 unsigned long start = round_down(mr[i].start, PMD_SIZE); 220 unsigned long end = round_up(mr[i].end, PMD_SIZE); 221 222 #ifdef CONFIG_X86_32 223 if ((end >> PAGE_SHIFT) > max_low_pfn) 224 continue; 225 #endif 226 227 if (memblock_is_region_memory(start, end - start)) 228 mr[i].page_size_mask |= 1<<PG_LEVEL_2M; 229 } 230 if ((page_size_mask & (1<<PG_LEVEL_1G)) && 231 !(mr[i].page_size_mask & (1<<PG_LEVEL_1G))) { 232 unsigned long start = round_down(mr[i].start, PUD_SIZE); 233 unsigned long end = round_up(mr[i].end, PUD_SIZE); 234 235 if (memblock_is_region_memory(start, end - start)) 236 mr[i].page_size_mask |= 1<<PG_LEVEL_1G; 237 } 238 } 239 } 240 241 static const char *page_size_string(struct map_range *mr) 242 { 243 static const char str_1g[] = "1G"; 244 static const char str_2m[] = "2M"; 245 static const char str_4m[] = "4M"; 246 static const char str_4k[] = "4k"; 247 248 if (mr->page_size_mask & (1<<PG_LEVEL_1G)) 249 return str_1g; 250 /* 251 * 32-bit without PAE has a 4M large page size. 252 * PG_LEVEL_2M is misnamed, but we can at least 253 * print out the right size in the string. 254 */ 255 if (IS_ENABLED(CONFIG_X86_32) && 256 !IS_ENABLED(CONFIG_X86_PAE) && 257 mr->page_size_mask & (1<<PG_LEVEL_2M)) 258 return str_4m; 259 260 if (mr->page_size_mask & (1<<PG_LEVEL_2M)) 261 return str_2m; 262 263 return str_4k; 264 } 265 266 static int __meminit split_mem_range(struct map_range *mr, int nr_range, 267 unsigned long start, 268 unsigned long end) 269 { 270 unsigned long start_pfn, end_pfn, limit_pfn; 271 unsigned long pfn; 272 int i; 273 274 limit_pfn = PFN_DOWN(end); 275 276 /* head if not big page alignment ? */ 277 pfn = start_pfn = PFN_DOWN(start); 278 #ifdef CONFIG_X86_32 279 /* 280 * Don't use a large page for the first 2/4MB of memory 281 * because there are often fixed size MTRRs in there 282 * and overlapping MTRRs into large pages can cause 283 * slowdowns. 284 */ 285 if (pfn == 0) 286 end_pfn = PFN_DOWN(PMD_SIZE); 287 else 288 end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); 289 #else /* CONFIG_X86_64 */ 290 end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); 291 #endif 292 if (end_pfn > limit_pfn) 293 end_pfn = limit_pfn; 294 if (start_pfn < end_pfn) { 295 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); 296 pfn = end_pfn; 297 } 298 299 /* big page (2M) range */ 300 start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); 301 #ifdef CONFIG_X86_32 302 end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); 303 #else /* CONFIG_X86_64 */ 304 end_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); 305 if (end_pfn > round_down(limit_pfn, PFN_DOWN(PMD_SIZE))) 306 end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); 307 #endif 308 309 if (start_pfn < end_pfn) { 310 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 311 page_size_mask & (1<<PG_LEVEL_2M)); 312 pfn = end_pfn; 313 } 314 315 #ifdef CONFIG_X86_64 316 /* big page (1G) range */ 317 start_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); 318 end_pfn = round_down(limit_pfn, PFN_DOWN(PUD_SIZE)); 319 if (start_pfn < end_pfn) { 320 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 321 page_size_mask & 322 ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G))); 323 pfn = end_pfn; 324 } 325 326 /* tail is not big page (1G) alignment */ 327 start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); 328 end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); 329 if (start_pfn < end_pfn) { 330 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 331 page_size_mask & (1<<PG_LEVEL_2M)); 332 pfn = end_pfn; 333 } 334 #endif 335 336 /* tail is not big page (2M) alignment */ 337 start_pfn = pfn; 338 end_pfn = limit_pfn; 339 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); 340 341 if (!after_bootmem) 342 adjust_range_page_size_mask(mr, nr_range); 343 344 /* try to merge same page size and continuous */ 345 for (i = 0; nr_range > 1 && i < nr_range - 1; i++) { 346 unsigned long old_start; 347 if (mr[i].end != mr[i+1].start || 348 mr[i].page_size_mask != mr[i+1].page_size_mask) 349 continue; 350 /* move it */ 351 old_start = mr[i].start; 352 memmove(&mr[i], &mr[i+1], 353 (nr_range - 1 - i) * sizeof(struct map_range)); 354 mr[i--].start = old_start; 355 nr_range--; 356 } 357 358 for (i = 0; i < nr_range; i++) 359 printk(KERN_DEBUG " [mem %#010lx-%#010lx] page %s\n", 360 mr[i].start, mr[i].end - 1, 361 page_size_string(&mr[i])); 362 363 return nr_range; 364 } 365 366 struct range pfn_mapped[E820_X_MAX]; 367 int nr_pfn_mapped; 368 369 static void add_pfn_range_mapped(unsigned long start_pfn, unsigned long end_pfn) 370 { 371 nr_pfn_mapped = add_range_with_merge(pfn_mapped, E820_X_MAX, 372 nr_pfn_mapped, start_pfn, end_pfn); 373 nr_pfn_mapped = clean_sort_range(pfn_mapped, E820_X_MAX); 374 375 max_pfn_mapped = max(max_pfn_mapped, end_pfn); 376 377 if (start_pfn < (1UL<<(32-PAGE_SHIFT))) 378 max_low_pfn_mapped = max(max_low_pfn_mapped, 379 min(end_pfn, 1UL<<(32-PAGE_SHIFT))); 380 } 381 382 bool pfn_range_is_mapped(unsigned long start_pfn, unsigned long end_pfn) 383 { 384 int i; 385 386 for (i = 0; i < nr_pfn_mapped; i++) 387 if ((start_pfn >= pfn_mapped[i].start) && 388 (end_pfn <= pfn_mapped[i].end)) 389 return true; 390 391 return false; 392 } 393 394 /* 395 * Setup the direct mapping of the physical memory at PAGE_OFFSET. 396 * This runs before bootmem is initialized and gets pages directly from 397 * the physical memory. To access them they are temporarily mapped. 398 */ 399 unsigned long __init_refok init_memory_mapping(unsigned long start, 400 unsigned long end) 401 { 402 struct map_range mr[NR_RANGE_MR]; 403 unsigned long ret = 0; 404 int nr_range, i; 405 406 pr_info("init_memory_mapping: [mem %#010lx-%#010lx]\n", 407 start, end - 1); 408 409 memset(mr, 0, sizeof(mr)); 410 nr_range = split_mem_range(mr, 0, start, end); 411 412 for (i = 0; i < nr_range; i++) 413 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end, 414 mr[i].page_size_mask); 415 416 add_pfn_range_mapped(start >> PAGE_SHIFT, ret >> PAGE_SHIFT); 417 418 return ret >> PAGE_SHIFT; 419 } 420 421 /* 422 * We need to iterate through the E820 memory map and create direct mappings 423 * for only E820_RAM and E820_KERN_RESERVED regions. We cannot simply 424 * create direct mappings for all pfns from [0 to max_low_pfn) and 425 * [4GB to max_pfn) because of possible memory holes in high addresses 426 * that cannot be marked as UC by fixed/variable range MTRRs. 427 * Depending on the alignment of E820 ranges, this may possibly result 428 * in using smaller size (i.e. 4K instead of 2M or 1G) page tables. 429 * 430 * init_mem_mapping() calls init_range_memory_mapping() with big range. 431 * That range would have hole in the middle or ends, and only ram parts 432 * will be mapped in init_range_memory_mapping(). 433 */ 434 static unsigned long __init init_range_memory_mapping( 435 unsigned long r_start, 436 unsigned long r_end) 437 { 438 unsigned long start_pfn, end_pfn; 439 unsigned long mapped_ram_size = 0; 440 int i; 441 442 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) { 443 u64 start = clamp_val(PFN_PHYS(start_pfn), r_start, r_end); 444 u64 end = clamp_val(PFN_PHYS(end_pfn), r_start, r_end); 445 if (start >= end) 446 continue; 447 448 /* 449 * if it is overlapping with brk pgt, we need to 450 * alloc pgt buf from memblock instead. 451 */ 452 can_use_brk_pgt = max(start, (u64)pgt_buf_end<<PAGE_SHIFT) >= 453 min(end, (u64)pgt_buf_top<<PAGE_SHIFT); 454 init_memory_mapping(start, end); 455 mapped_ram_size += end - start; 456 can_use_brk_pgt = true; 457 } 458 459 return mapped_ram_size; 460 } 461 462 static unsigned long __init get_new_step_size(unsigned long step_size) 463 { 464 /* 465 * Initial mapped size is PMD_SIZE (2M). 466 * We can not set step_size to be PUD_SIZE (1G) yet. 467 * In worse case, when we cross the 1G boundary, and 468 * PG_LEVEL_2M is not set, we will need 1+1+512 pages (2M + 8k) 469 * to map 1G range with PTE. Hence we use one less than the 470 * difference of page table level shifts. 471 * 472 * Don't need to worry about overflow in the top-down case, on 32bit, 473 * when step_size is 0, round_down() returns 0 for start, and that 474 * turns it into 0x100000000ULL. 475 * In the bottom-up case, round_up(x, 0) returns 0 though too, which 476 * needs to be taken into consideration by the code below. 477 */ 478 return step_size << (PMD_SHIFT - PAGE_SHIFT - 1); 479 } 480 481 /** 482 * memory_map_top_down - Map [map_start, map_end) top down 483 * @map_start: start address of the target memory range 484 * @map_end: end address of the target memory range 485 * 486 * This function will setup direct mapping for memory range 487 * [map_start, map_end) in top-down. That said, the page tables 488 * will be allocated at the end of the memory, and we map the 489 * memory in top-down. 490 */ 491 static void __init memory_map_top_down(unsigned long map_start, 492 unsigned long map_end) 493 { 494 unsigned long real_end, start, last_start; 495 unsigned long step_size; 496 unsigned long addr; 497 unsigned long mapped_ram_size = 0; 498 499 /* xen has big range in reserved near end of ram, skip it at first.*/ 500 addr = memblock_find_in_range(map_start, map_end, PMD_SIZE, PMD_SIZE); 501 real_end = addr + PMD_SIZE; 502 503 /* step_size need to be small so pgt_buf from BRK could cover it */ 504 step_size = PMD_SIZE; 505 max_pfn_mapped = 0; /* will get exact value next */ 506 min_pfn_mapped = real_end >> PAGE_SHIFT; 507 last_start = start = real_end; 508 509 /* 510 * We start from the top (end of memory) and go to the bottom. 511 * The memblock_find_in_range() gets us a block of RAM from the 512 * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages 513 * for page table. 514 */ 515 while (last_start > map_start) { 516 if (last_start > step_size) { 517 start = round_down(last_start - 1, step_size); 518 if (start < map_start) 519 start = map_start; 520 } else 521 start = map_start; 522 mapped_ram_size += init_range_memory_mapping(start, 523 last_start); 524 last_start = start; 525 min_pfn_mapped = last_start >> PAGE_SHIFT; 526 if (mapped_ram_size >= step_size) 527 step_size = get_new_step_size(step_size); 528 } 529 530 if (real_end < map_end) 531 init_range_memory_mapping(real_end, map_end); 532 } 533 534 /** 535 * memory_map_bottom_up - Map [map_start, map_end) bottom up 536 * @map_start: start address of the target memory range 537 * @map_end: end address of the target memory range 538 * 539 * This function will setup direct mapping for memory range 540 * [map_start, map_end) in bottom-up. Since we have limited the 541 * bottom-up allocation above the kernel, the page tables will 542 * be allocated just above the kernel and we map the memory 543 * in [map_start, map_end) in bottom-up. 544 */ 545 static void __init memory_map_bottom_up(unsigned long map_start, 546 unsigned long map_end) 547 { 548 unsigned long next, start; 549 unsigned long mapped_ram_size = 0; 550 /* step_size need to be small so pgt_buf from BRK could cover it */ 551 unsigned long step_size = PMD_SIZE; 552 553 start = map_start; 554 min_pfn_mapped = start >> PAGE_SHIFT; 555 556 /* 557 * We start from the bottom (@map_start) and go to the top (@map_end). 558 * The memblock_find_in_range() gets us a block of RAM from the 559 * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages 560 * for page table. 561 */ 562 while (start < map_end) { 563 if (step_size && map_end - start > step_size) { 564 next = round_up(start + 1, step_size); 565 if (next > map_end) 566 next = map_end; 567 } else { 568 next = map_end; 569 } 570 571 mapped_ram_size += init_range_memory_mapping(start, next); 572 start = next; 573 574 if (mapped_ram_size >= step_size) 575 step_size = get_new_step_size(step_size); 576 } 577 } 578 579 void __init init_mem_mapping(void) 580 { 581 unsigned long end; 582 583 probe_page_size_mask(); 584 585 #ifdef CONFIG_X86_64 586 end = max_pfn << PAGE_SHIFT; 587 #else 588 end = max_low_pfn << PAGE_SHIFT; 589 #endif 590 591 /* the ISA range is always mapped regardless of memory holes */ 592 init_memory_mapping(0, ISA_END_ADDRESS); 593 594 /* 595 * If the allocation is in bottom-up direction, we setup direct mapping 596 * in bottom-up, otherwise we setup direct mapping in top-down. 597 */ 598 if (memblock_bottom_up()) { 599 unsigned long kernel_end = __pa_symbol(_end); 600 601 /* 602 * we need two separate calls here. This is because we want to 603 * allocate page tables above the kernel. So we first map 604 * [kernel_end, end) to make memory above the kernel be mapped 605 * as soon as possible. And then use page tables allocated above 606 * the kernel to map [ISA_END_ADDRESS, kernel_end). 607 */ 608 memory_map_bottom_up(kernel_end, end); 609 memory_map_bottom_up(ISA_END_ADDRESS, kernel_end); 610 } else { 611 memory_map_top_down(ISA_END_ADDRESS, end); 612 } 613 614 #ifdef CONFIG_X86_64 615 if (max_pfn > max_low_pfn) { 616 /* can we preseve max_low_pfn ?*/ 617 max_low_pfn = max_pfn; 618 } 619 #else 620 early_ioremap_page_table_range_init(); 621 #endif 622 623 load_cr3(swapper_pg_dir); 624 __flush_tlb_all(); 625 626 early_memtest(0, max_pfn_mapped << PAGE_SHIFT); 627 } 628 629 /* 630 * devmem_is_allowed() checks to see if /dev/mem access to a certain address 631 * is valid. The argument is a physical page number. 632 * 633 * 634 * On x86, access has to be given to the first megabyte of ram because that area 635 * contains BIOS code and data regions used by X and dosemu and similar apps. 636 * Access has to be given to non-kernel-ram areas as well, these contain the PCI 637 * mmio resources as well as potential bios/acpi data regions. 638 */ 639 int devmem_is_allowed(unsigned long pagenr) 640 { 641 if (pagenr < 256) 642 return 1; 643 if (iomem_is_exclusive(pagenr << PAGE_SHIFT)) 644 return 0; 645 if (!page_is_ram(pagenr)) 646 return 1; 647 return 0; 648 } 649 650 void free_init_pages(char *what, unsigned long begin, unsigned long end) 651 { 652 unsigned long begin_aligned, end_aligned; 653 654 /* Make sure boundaries are page aligned */ 655 begin_aligned = PAGE_ALIGN(begin); 656 end_aligned = end & PAGE_MASK; 657 658 if (WARN_ON(begin_aligned != begin || end_aligned != end)) { 659 begin = begin_aligned; 660 end = end_aligned; 661 } 662 663 if (begin >= end) 664 return; 665 666 /* 667 * If debugging page accesses then do not free this memory but 668 * mark them not present - any buggy init-section access will 669 * create a kernel page fault: 670 */ 671 #ifdef CONFIG_DEBUG_PAGEALLOC 672 printk(KERN_INFO "debug: unmapping init [mem %#010lx-%#010lx]\n", 673 begin, end - 1); 674 set_memory_np(begin, (end - begin) >> PAGE_SHIFT); 675 #else 676 /* 677 * We just marked the kernel text read only above, now that 678 * we are going to free part of that, we need to make that 679 * writeable and non-executable first. 680 */ 681 set_memory_nx(begin, (end - begin) >> PAGE_SHIFT); 682 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT); 683 684 free_reserved_area((void *)begin, (void *)end, POISON_FREE_INITMEM, what); 685 #endif 686 } 687 688 void free_initmem(void) 689 { 690 free_init_pages("unused kernel", 691 (unsigned long)(&__init_begin), 692 (unsigned long)(&__init_end)); 693 } 694 695 #ifdef CONFIG_BLK_DEV_INITRD 696 void __init free_initrd_mem(unsigned long start, unsigned long end) 697 { 698 #ifdef CONFIG_MICROCODE_EARLY 699 /* 700 * Remember, initrd memory may contain microcode or other useful things. 701 * Before we lose initrd mem, we need to find a place to hold them 702 * now that normal virtual memory is enabled. 703 */ 704 save_microcode_in_initrd(); 705 #endif 706 707 /* 708 * end could be not aligned, and We can not align that, 709 * decompresser could be confused by aligned initrd_end 710 * We already reserve the end partial page before in 711 * - i386_start_kernel() 712 * - x86_64_start_kernel() 713 * - relocate_initrd() 714 * So here We can do PAGE_ALIGN() safely to get partial page to be freed 715 */ 716 free_init_pages("initrd", start, PAGE_ALIGN(end)); 717 } 718 #endif 719 720 void __init zone_sizes_init(void) 721 { 722 unsigned long max_zone_pfns[MAX_NR_ZONES]; 723 724 memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); 725 726 #ifdef CONFIG_ZONE_DMA 727 max_zone_pfns[ZONE_DMA] = min(MAX_DMA_PFN, max_low_pfn); 728 #endif 729 #ifdef CONFIG_ZONE_DMA32 730 max_zone_pfns[ZONE_DMA32] = min(MAX_DMA32_PFN, max_low_pfn); 731 #endif 732 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 733 #ifdef CONFIG_HIGHMEM 734 max_zone_pfns[ZONE_HIGHMEM] = max_pfn; 735 #endif 736 737 free_area_init_nodes(max_zone_pfns); 738 } 739 740 DEFINE_PER_CPU_SHARED_ALIGNED(struct tlb_state, cpu_tlbstate) = { 741 #ifdef CONFIG_SMP 742 .active_mm = &init_mm, 743 .state = 0, 744 #endif 745 .cr4 = ~0UL, /* fail hard if we screw up cr4 shadow initialization */ 746 }; 747 EXPORT_SYMBOL_GPL(cpu_tlbstate); 748 749 void update_cache_mode_entry(unsigned entry, enum page_cache_mode cache) 750 { 751 /* entry 0 MUST be WB (hardwired to speed up translations) */ 752 BUG_ON(!entry && cache != _PAGE_CACHE_MODE_WB); 753 754 __cachemode2pte_tbl[cache] = __cm_idx2pte(entry); 755 __pte2cachemode_tbl[entry] = cache; 756 } 757