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_GPL(__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_GPL(__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 set_in_cr4(X86_CR4_PSE); 177 178 /* Enable PGE if available */ 179 if (cpu_has_pge) { 180 set_in_cr4(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 int __meminit split_mem_range(struct map_range *mr, int nr_range, 242 unsigned long start, 243 unsigned long end) 244 { 245 unsigned long start_pfn, end_pfn, limit_pfn; 246 unsigned long pfn; 247 int i; 248 249 limit_pfn = PFN_DOWN(end); 250 251 /* head if not big page alignment ? */ 252 pfn = start_pfn = PFN_DOWN(start); 253 #ifdef CONFIG_X86_32 254 /* 255 * Don't use a large page for the first 2/4MB of memory 256 * because there are often fixed size MTRRs in there 257 * and overlapping MTRRs into large pages can cause 258 * slowdowns. 259 */ 260 if (pfn == 0) 261 end_pfn = PFN_DOWN(PMD_SIZE); 262 else 263 end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); 264 #else /* CONFIG_X86_64 */ 265 end_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); 266 #endif 267 if (end_pfn > limit_pfn) 268 end_pfn = limit_pfn; 269 if (start_pfn < end_pfn) { 270 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); 271 pfn = end_pfn; 272 } 273 274 /* big page (2M) range */ 275 start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); 276 #ifdef CONFIG_X86_32 277 end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); 278 #else /* CONFIG_X86_64 */ 279 end_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); 280 if (end_pfn > round_down(limit_pfn, PFN_DOWN(PMD_SIZE))) 281 end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); 282 #endif 283 284 if (start_pfn < end_pfn) { 285 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 286 page_size_mask & (1<<PG_LEVEL_2M)); 287 pfn = end_pfn; 288 } 289 290 #ifdef CONFIG_X86_64 291 /* big page (1G) range */ 292 start_pfn = round_up(pfn, PFN_DOWN(PUD_SIZE)); 293 end_pfn = round_down(limit_pfn, PFN_DOWN(PUD_SIZE)); 294 if (start_pfn < end_pfn) { 295 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 296 page_size_mask & 297 ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G))); 298 pfn = end_pfn; 299 } 300 301 /* tail is not big page (1G) alignment */ 302 start_pfn = round_up(pfn, PFN_DOWN(PMD_SIZE)); 303 end_pfn = round_down(limit_pfn, PFN_DOWN(PMD_SIZE)); 304 if (start_pfn < end_pfn) { 305 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 306 page_size_mask & (1<<PG_LEVEL_2M)); 307 pfn = end_pfn; 308 } 309 #endif 310 311 /* tail is not big page (2M) alignment */ 312 start_pfn = pfn; 313 end_pfn = limit_pfn; 314 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0); 315 316 if (!after_bootmem) 317 adjust_range_page_size_mask(mr, nr_range); 318 319 /* try to merge same page size and continuous */ 320 for (i = 0; nr_range > 1 && i < nr_range - 1; i++) { 321 unsigned long old_start; 322 if (mr[i].end != mr[i+1].start || 323 mr[i].page_size_mask != mr[i+1].page_size_mask) 324 continue; 325 /* move it */ 326 old_start = mr[i].start; 327 memmove(&mr[i], &mr[i+1], 328 (nr_range - 1 - i) * sizeof(struct map_range)); 329 mr[i--].start = old_start; 330 nr_range--; 331 } 332 333 for (i = 0; i < nr_range; i++) 334 printk(KERN_DEBUG " [mem %#010lx-%#010lx] page %s\n", 335 mr[i].start, mr[i].end - 1, 336 (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":( 337 (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k")); 338 339 return nr_range; 340 } 341 342 struct range pfn_mapped[E820_X_MAX]; 343 int nr_pfn_mapped; 344 345 static void add_pfn_range_mapped(unsigned long start_pfn, unsigned long end_pfn) 346 { 347 nr_pfn_mapped = add_range_with_merge(pfn_mapped, E820_X_MAX, 348 nr_pfn_mapped, start_pfn, end_pfn); 349 nr_pfn_mapped = clean_sort_range(pfn_mapped, E820_X_MAX); 350 351 max_pfn_mapped = max(max_pfn_mapped, end_pfn); 352 353 if (start_pfn < (1UL<<(32-PAGE_SHIFT))) 354 max_low_pfn_mapped = max(max_low_pfn_mapped, 355 min(end_pfn, 1UL<<(32-PAGE_SHIFT))); 356 } 357 358 bool pfn_range_is_mapped(unsigned long start_pfn, unsigned long end_pfn) 359 { 360 int i; 361 362 for (i = 0; i < nr_pfn_mapped; i++) 363 if ((start_pfn >= pfn_mapped[i].start) && 364 (end_pfn <= pfn_mapped[i].end)) 365 return true; 366 367 return false; 368 } 369 370 /* 371 * Setup the direct mapping of the physical memory at PAGE_OFFSET. 372 * This runs before bootmem is initialized and gets pages directly from 373 * the physical memory. To access them they are temporarily mapped. 374 */ 375 unsigned long __init_refok init_memory_mapping(unsigned long start, 376 unsigned long end) 377 { 378 struct map_range mr[NR_RANGE_MR]; 379 unsigned long ret = 0; 380 int nr_range, i; 381 382 pr_info("init_memory_mapping: [mem %#010lx-%#010lx]\n", 383 start, end - 1); 384 385 memset(mr, 0, sizeof(mr)); 386 nr_range = split_mem_range(mr, 0, start, end); 387 388 for (i = 0; i < nr_range; i++) 389 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end, 390 mr[i].page_size_mask); 391 392 add_pfn_range_mapped(start >> PAGE_SHIFT, ret >> PAGE_SHIFT); 393 394 return ret >> PAGE_SHIFT; 395 } 396 397 /* 398 * We need to iterate through the E820 memory map and create direct mappings 399 * for only E820_RAM and E820_KERN_RESERVED regions. We cannot simply 400 * create direct mappings for all pfns from [0 to max_low_pfn) and 401 * [4GB to max_pfn) because of possible memory holes in high addresses 402 * that cannot be marked as UC by fixed/variable range MTRRs. 403 * Depending on the alignment of E820 ranges, this may possibly result 404 * in using smaller size (i.e. 4K instead of 2M or 1G) page tables. 405 * 406 * init_mem_mapping() calls init_range_memory_mapping() with big range. 407 * That range would have hole in the middle or ends, and only ram parts 408 * will be mapped in init_range_memory_mapping(). 409 */ 410 static unsigned long __init init_range_memory_mapping( 411 unsigned long r_start, 412 unsigned long r_end) 413 { 414 unsigned long start_pfn, end_pfn; 415 unsigned long mapped_ram_size = 0; 416 int i; 417 418 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) { 419 u64 start = clamp_val(PFN_PHYS(start_pfn), r_start, r_end); 420 u64 end = clamp_val(PFN_PHYS(end_pfn), r_start, r_end); 421 if (start >= end) 422 continue; 423 424 /* 425 * if it is overlapping with brk pgt, we need to 426 * alloc pgt buf from memblock instead. 427 */ 428 can_use_brk_pgt = max(start, (u64)pgt_buf_end<<PAGE_SHIFT) >= 429 min(end, (u64)pgt_buf_top<<PAGE_SHIFT); 430 init_memory_mapping(start, end); 431 mapped_ram_size += end - start; 432 can_use_brk_pgt = true; 433 } 434 435 return mapped_ram_size; 436 } 437 438 static unsigned long __init get_new_step_size(unsigned long step_size) 439 { 440 /* 441 * Explain why we shift by 5 and why we don't have to worry about 442 * 'step_size << 5' overflowing: 443 * 444 * initial mapped size is PMD_SIZE (2M). 445 * We can not set step_size to be PUD_SIZE (1G) yet. 446 * In worse case, when we cross the 1G boundary, and 447 * PG_LEVEL_2M is not set, we will need 1+1+512 pages (2M + 8k) 448 * to map 1G range with PTE. Use 5 as shift for now. 449 * 450 * Don't need to worry about overflow, on 32bit, when step_size 451 * is 0, round_down() returns 0 for start, and that turns it 452 * into 0x100000000ULL. 453 */ 454 return step_size << 5; 455 } 456 457 /** 458 * memory_map_top_down - Map [map_start, map_end) top down 459 * @map_start: start address of the target memory range 460 * @map_end: end address of the target memory range 461 * 462 * This function will setup direct mapping for memory range 463 * [map_start, map_end) in top-down. That said, the page tables 464 * will be allocated at the end of the memory, and we map the 465 * memory in top-down. 466 */ 467 static void __init memory_map_top_down(unsigned long map_start, 468 unsigned long map_end) 469 { 470 unsigned long real_end, start, last_start; 471 unsigned long step_size; 472 unsigned long addr; 473 unsigned long mapped_ram_size = 0; 474 unsigned long new_mapped_ram_size; 475 476 /* xen has big range in reserved near end of ram, skip it at first.*/ 477 addr = memblock_find_in_range(map_start, map_end, PMD_SIZE, PMD_SIZE); 478 real_end = addr + PMD_SIZE; 479 480 /* step_size need to be small so pgt_buf from BRK could cover it */ 481 step_size = PMD_SIZE; 482 max_pfn_mapped = 0; /* will get exact value next */ 483 min_pfn_mapped = real_end >> PAGE_SHIFT; 484 last_start = start = real_end; 485 486 /* 487 * We start from the top (end of memory) and go to the bottom. 488 * The memblock_find_in_range() gets us a block of RAM from the 489 * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages 490 * for page table. 491 */ 492 while (last_start > map_start) { 493 if (last_start > step_size) { 494 start = round_down(last_start - 1, step_size); 495 if (start < map_start) 496 start = map_start; 497 } else 498 start = map_start; 499 new_mapped_ram_size = init_range_memory_mapping(start, 500 last_start); 501 last_start = start; 502 min_pfn_mapped = last_start >> PAGE_SHIFT; 503 /* only increase step_size after big range get mapped */ 504 if (new_mapped_ram_size > mapped_ram_size) 505 step_size = get_new_step_size(step_size); 506 mapped_ram_size += new_mapped_ram_size; 507 } 508 509 if (real_end < map_end) 510 init_range_memory_mapping(real_end, map_end); 511 } 512 513 /** 514 * memory_map_bottom_up - Map [map_start, map_end) bottom up 515 * @map_start: start address of the target memory range 516 * @map_end: end address of the target memory range 517 * 518 * This function will setup direct mapping for memory range 519 * [map_start, map_end) in bottom-up. Since we have limited the 520 * bottom-up allocation above the kernel, the page tables will 521 * be allocated just above the kernel and we map the memory 522 * in [map_start, map_end) in bottom-up. 523 */ 524 static void __init memory_map_bottom_up(unsigned long map_start, 525 unsigned long map_end) 526 { 527 unsigned long next, new_mapped_ram_size, start; 528 unsigned long mapped_ram_size = 0; 529 /* step_size need to be small so pgt_buf from BRK could cover it */ 530 unsigned long step_size = PMD_SIZE; 531 532 start = map_start; 533 min_pfn_mapped = start >> PAGE_SHIFT; 534 535 /* 536 * We start from the bottom (@map_start) and go to the top (@map_end). 537 * The memblock_find_in_range() gets us a block of RAM from the 538 * end of RAM in [min_pfn_mapped, max_pfn_mapped) used as new pages 539 * for page table. 540 */ 541 while (start < map_end) { 542 if (map_end - start > step_size) { 543 next = round_up(start + 1, step_size); 544 if (next > map_end) 545 next = map_end; 546 } else 547 next = map_end; 548 549 new_mapped_ram_size = init_range_memory_mapping(start, next); 550 start = next; 551 552 if (new_mapped_ram_size > mapped_ram_size) 553 step_size = get_new_step_size(step_size); 554 mapped_ram_size += new_mapped_ram_size; 555 } 556 } 557 558 void __init init_mem_mapping(void) 559 { 560 unsigned long end; 561 562 probe_page_size_mask(); 563 564 #ifdef CONFIG_X86_64 565 end = max_pfn << PAGE_SHIFT; 566 #else 567 end = max_low_pfn << PAGE_SHIFT; 568 #endif 569 570 /* the ISA range is always mapped regardless of memory holes */ 571 init_memory_mapping(0, ISA_END_ADDRESS); 572 573 /* 574 * If the allocation is in bottom-up direction, we setup direct mapping 575 * in bottom-up, otherwise we setup direct mapping in top-down. 576 */ 577 if (memblock_bottom_up()) { 578 unsigned long kernel_end = __pa_symbol(_end); 579 580 /* 581 * we need two separate calls here. This is because we want to 582 * allocate page tables above the kernel. So we first map 583 * [kernel_end, end) to make memory above the kernel be mapped 584 * as soon as possible. And then use page tables allocated above 585 * the kernel to map [ISA_END_ADDRESS, kernel_end). 586 */ 587 memory_map_bottom_up(kernel_end, end); 588 memory_map_bottom_up(ISA_END_ADDRESS, kernel_end); 589 } else { 590 memory_map_top_down(ISA_END_ADDRESS, end); 591 } 592 593 #ifdef CONFIG_X86_64 594 if (max_pfn > max_low_pfn) { 595 /* can we preseve max_low_pfn ?*/ 596 max_low_pfn = max_pfn; 597 } 598 #else 599 early_ioremap_page_table_range_init(); 600 #endif 601 602 load_cr3(swapper_pg_dir); 603 __flush_tlb_all(); 604 605 early_memtest(0, max_pfn_mapped << PAGE_SHIFT); 606 } 607 608 /* 609 * devmem_is_allowed() checks to see if /dev/mem access to a certain address 610 * is valid. The argument is a physical page number. 611 * 612 * 613 * On x86, access has to be given to the first megabyte of ram because that area 614 * contains bios code and data regions used by X and dosemu and similar apps. 615 * Access has to be given to non-kernel-ram areas as well, these contain the PCI 616 * mmio resources as well as potential bios/acpi data regions. 617 */ 618 int devmem_is_allowed(unsigned long pagenr) 619 { 620 if (pagenr < 256) 621 return 1; 622 if (iomem_is_exclusive(pagenr << PAGE_SHIFT)) 623 return 0; 624 if (!page_is_ram(pagenr)) 625 return 1; 626 return 0; 627 } 628 629 void free_init_pages(char *what, unsigned long begin, unsigned long end) 630 { 631 unsigned long begin_aligned, end_aligned; 632 633 /* Make sure boundaries are page aligned */ 634 begin_aligned = PAGE_ALIGN(begin); 635 end_aligned = end & PAGE_MASK; 636 637 if (WARN_ON(begin_aligned != begin || end_aligned != end)) { 638 begin = begin_aligned; 639 end = end_aligned; 640 } 641 642 if (begin >= end) 643 return; 644 645 /* 646 * If debugging page accesses then do not free this memory but 647 * mark them not present - any buggy init-section access will 648 * create a kernel page fault: 649 */ 650 #ifdef CONFIG_DEBUG_PAGEALLOC 651 printk(KERN_INFO "debug: unmapping init [mem %#010lx-%#010lx]\n", 652 begin, end - 1); 653 set_memory_np(begin, (end - begin) >> PAGE_SHIFT); 654 #else 655 /* 656 * We just marked the kernel text read only above, now that 657 * we are going to free part of that, we need to make that 658 * writeable and non-executable first. 659 */ 660 set_memory_nx(begin, (end - begin) >> PAGE_SHIFT); 661 set_memory_rw(begin, (end - begin) >> PAGE_SHIFT); 662 663 free_reserved_area((void *)begin, (void *)end, POISON_FREE_INITMEM, what); 664 #endif 665 } 666 667 void free_initmem(void) 668 { 669 free_init_pages("unused kernel", 670 (unsigned long)(&__init_begin), 671 (unsigned long)(&__init_end)); 672 } 673 674 #ifdef CONFIG_BLK_DEV_INITRD 675 void __init free_initrd_mem(unsigned long start, unsigned long end) 676 { 677 #ifdef CONFIG_MICROCODE_EARLY 678 /* 679 * Remember, initrd memory may contain microcode or other useful things. 680 * Before we lose initrd mem, we need to find a place to hold them 681 * now that normal virtual memory is enabled. 682 */ 683 save_microcode_in_initrd(); 684 #endif 685 686 /* 687 * end could be not aligned, and We can not align that, 688 * decompresser could be confused by aligned initrd_end 689 * We already reserve the end partial page before in 690 * - i386_start_kernel() 691 * - x86_64_start_kernel() 692 * - relocate_initrd() 693 * So here We can do PAGE_ALIGN() safely to get partial page to be freed 694 */ 695 free_init_pages("initrd", start, PAGE_ALIGN(end)); 696 } 697 #endif 698 699 void __init zone_sizes_init(void) 700 { 701 unsigned long max_zone_pfns[MAX_NR_ZONES]; 702 703 memset(max_zone_pfns, 0, sizeof(max_zone_pfns)); 704 705 #ifdef CONFIG_ZONE_DMA 706 max_zone_pfns[ZONE_DMA] = min(MAX_DMA_PFN, max_low_pfn); 707 #endif 708 #ifdef CONFIG_ZONE_DMA32 709 max_zone_pfns[ZONE_DMA32] = min(MAX_DMA32_PFN, max_low_pfn); 710 #endif 711 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 712 #ifdef CONFIG_HIGHMEM 713 max_zone_pfns[ZONE_HIGHMEM] = max_pfn; 714 #endif 715 716 free_area_init_nodes(max_zone_pfns); 717 } 718 719 void update_cache_mode_entry(unsigned entry, enum page_cache_mode cache) 720 { 721 /* entry 0 MUST be WB (hardwired to speed up translations) */ 722 BUG_ON(!entry && cache != _PAGE_CACHE_MODE_WB); 723 724 __cachemode2pte_tbl[cache] = __cm_idx2pte(entry); 725 __pte2cachemode_tbl[entry] = cache; 726 } 727