1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * 4 * Copyright (C) 1995 Linus Torvalds 5 * 6 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 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/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/memblock.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 <linux/uaccess.h> 38 #include <asm/dma.h> 39 #include <asm/fixmap.h> 40 #include <asm/e820/api.h> 41 #include <asm/apic.h> 42 #include <asm/bugs.h> 43 #include <asm/tlb.h> 44 #include <asm/tlbflush.h> 45 #include <asm/olpc_ofw.h> 46 #include <asm/pgalloc.h> 47 #include <asm/sections.h> 48 #include <asm/setup.h> 49 #include <asm/set_memory.h> 50 #include <asm/page_types.h> 51 #include <asm/cpu_entry_area.h> 52 #include <asm/init.h> 53 #include <asm/pgtable_areas.h> 54 #include <asm/numa.h> 55 56 #include "mm_internal.h" 57 58 unsigned long highstart_pfn, highend_pfn; 59 60 bool __read_mostly __vmalloc_start_set = false; 61 62 /* 63 * Creates a middle page table and puts a pointer to it in the 64 * given global directory entry. This only returns the gd entry 65 * in non-PAE compilation mode, since the middle layer is folded. 66 */ 67 static pmd_t * __init one_md_table_init(pgd_t *pgd) 68 { 69 p4d_t *p4d; 70 pud_t *pud; 71 pmd_t *pmd_table; 72 73 #ifdef CONFIG_X86_PAE 74 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) { 75 pmd_table = (pmd_t *)alloc_low_page(); 76 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT)); 77 p4d = p4d_offset(pgd, 0); 78 pud = pud_offset(p4d, 0); 79 BUG_ON(pmd_table != pmd_offset(pud, 0)); 80 81 return pmd_table; 82 } 83 #endif 84 p4d = p4d_offset(pgd, 0); 85 pud = pud_offset(p4d, 0); 86 pmd_table = pmd_offset(pud, 0); 87 88 return pmd_table; 89 } 90 91 /* 92 * Create a page table and place a pointer to it in a middle page 93 * directory entry: 94 */ 95 static pte_t * __init one_page_table_init(pmd_t *pmd) 96 { 97 if (!(pmd_val(*pmd) & _PAGE_PRESENT)) { 98 pte_t *page_table = (pte_t *)alloc_low_page(); 99 100 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE)); 101 BUG_ON(page_table != pte_offset_kernel(pmd, 0)); 102 } 103 104 return pte_offset_kernel(pmd, 0); 105 } 106 107 pmd_t * __init populate_extra_pmd(unsigned long vaddr) 108 { 109 int pgd_idx = pgd_index(vaddr); 110 int pmd_idx = pmd_index(vaddr); 111 112 return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx; 113 } 114 115 pte_t * __init populate_extra_pte(unsigned long vaddr) 116 { 117 int pte_idx = pte_index(vaddr); 118 pmd_t *pmd; 119 120 pmd = populate_extra_pmd(vaddr); 121 return one_page_table_init(pmd) + pte_idx; 122 } 123 124 static unsigned long __init 125 page_table_range_init_count(unsigned long start, unsigned long end) 126 { 127 unsigned long count = 0; 128 #ifdef CONFIG_HIGHMEM 129 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT; 130 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT; 131 int pgd_idx, pmd_idx; 132 unsigned long vaddr; 133 134 if (pmd_idx_kmap_begin == pmd_idx_kmap_end) 135 return 0; 136 137 vaddr = start; 138 pgd_idx = pgd_index(vaddr); 139 pmd_idx = pmd_index(vaddr); 140 141 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) { 142 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); 143 pmd_idx++) { 144 if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin && 145 (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) 146 count++; 147 vaddr += PMD_SIZE; 148 } 149 pmd_idx = 0; 150 } 151 #endif 152 return count; 153 } 154 155 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd, 156 unsigned long vaddr, pte_t *lastpte, 157 void **adr) 158 { 159 #ifdef CONFIG_HIGHMEM 160 /* 161 * Something (early fixmap) may already have put a pte 162 * page here, which causes the page table allocation 163 * to become nonlinear. Attempt to fix it, and if it 164 * is still nonlinear then we have to bug. 165 */ 166 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT; 167 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT; 168 169 if (pmd_idx_kmap_begin != pmd_idx_kmap_end 170 && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin 171 && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) { 172 pte_t *newpte; 173 int i; 174 175 BUG_ON(after_bootmem); 176 newpte = *adr; 177 for (i = 0; i < PTRS_PER_PTE; i++) 178 set_pte(newpte + i, pte[i]); 179 *adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE); 180 181 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE)); 182 BUG_ON(newpte != pte_offset_kernel(pmd, 0)); 183 __flush_tlb_all(); 184 185 pte = newpte; 186 } 187 BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1) 188 && vaddr > fix_to_virt(FIX_KMAP_END) 189 && lastpte && lastpte + PTRS_PER_PTE != pte); 190 #endif 191 return pte; 192 } 193 194 /* 195 * This function initializes a certain range of kernel virtual memory 196 * with new bootmem page tables, everywhere page tables are missing in 197 * the given range. 198 * 199 * NOTE: The pagetables are allocated contiguous on the physical space 200 * so we can cache the place of the first one and move around without 201 * checking the pgd every time. 202 */ 203 static void __init 204 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base) 205 { 206 int pgd_idx, pmd_idx; 207 unsigned long vaddr; 208 pgd_t *pgd; 209 pmd_t *pmd; 210 pte_t *pte = NULL; 211 unsigned long count = page_table_range_init_count(start, end); 212 void *adr = NULL; 213 214 if (count) 215 adr = alloc_low_pages(count); 216 217 vaddr = start; 218 pgd_idx = pgd_index(vaddr); 219 pmd_idx = pmd_index(vaddr); 220 pgd = pgd_base + pgd_idx; 221 222 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) { 223 pmd = one_md_table_init(pgd); 224 pmd = pmd + pmd_index(vaddr); 225 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); 226 pmd++, pmd_idx++) { 227 pte = page_table_kmap_check(one_page_table_init(pmd), 228 pmd, vaddr, pte, &adr); 229 230 vaddr += PMD_SIZE; 231 } 232 pmd_idx = 0; 233 } 234 } 235 236 static inline int is_x86_32_kernel_text(unsigned long addr) 237 { 238 if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end) 239 return 1; 240 return 0; 241 } 242 243 /* 244 * This maps the physical memory to kernel virtual address space, a total 245 * of max_low_pfn pages, by creating page tables starting from address 246 * PAGE_OFFSET: 247 */ 248 unsigned long __init 249 kernel_physical_mapping_init(unsigned long start, 250 unsigned long end, 251 unsigned long page_size_mask, 252 pgprot_t prot) 253 { 254 int use_pse = page_size_mask == (1<<PG_LEVEL_2M); 255 unsigned long last_map_addr = end; 256 unsigned long start_pfn, end_pfn; 257 pgd_t *pgd_base = swapper_pg_dir; 258 int pgd_idx, pmd_idx, pte_ofs; 259 unsigned long pfn; 260 pgd_t *pgd; 261 pmd_t *pmd; 262 pte_t *pte; 263 unsigned pages_2m, pages_4k; 264 int mapping_iter; 265 266 start_pfn = start >> PAGE_SHIFT; 267 end_pfn = end >> PAGE_SHIFT; 268 269 /* 270 * First iteration will setup identity mapping using large/small pages 271 * based on use_pse, with other attributes same as set by 272 * the early code in head_32.S 273 * 274 * Second iteration will setup the appropriate attributes (NX, GLOBAL..) 275 * as desired for the kernel identity mapping. 276 * 277 * This two pass mechanism conforms to the TLB app note which says: 278 * 279 * "Software should not write to a paging-structure entry in a way 280 * that would change, for any linear address, both the page size 281 * and either the page frame or attributes." 282 */ 283 mapping_iter = 1; 284 285 if (!boot_cpu_has(X86_FEATURE_PSE)) 286 use_pse = 0; 287 288 repeat: 289 pages_2m = pages_4k = 0; 290 pfn = start_pfn; 291 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 292 pgd = pgd_base + pgd_idx; 293 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) { 294 pmd = one_md_table_init(pgd); 295 296 if (pfn >= end_pfn) 297 continue; 298 #ifdef CONFIG_X86_PAE 299 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 300 pmd += pmd_idx; 301 #else 302 pmd_idx = 0; 303 #endif 304 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn; 305 pmd++, pmd_idx++) { 306 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET; 307 308 /* 309 * Map with big pages if possible, otherwise 310 * create normal page tables: 311 */ 312 if (use_pse) { 313 unsigned int addr2; 314 pgprot_t prot = PAGE_KERNEL_LARGE; 315 /* 316 * first pass will use the same initial 317 * identity mapping attribute + _PAGE_PSE. 318 */ 319 pgprot_t init_prot = 320 __pgprot(PTE_IDENT_ATTR | 321 _PAGE_PSE); 322 323 pfn &= PMD_MASK >> PAGE_SHIFT; 324 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE + 325 PAGE_OFFSET + PAGE_SIZE-1; 326 327 if (is_x86_32_kernel_text(addr) || 328 is_x86_32_kernel_text(addr2)) 329 prot = PAGE_KERNEL_LARGE_EXEC; 330 331 pages_2m++; 332 if (mapping_iter == 1) 333 set_pmd(pmd, pfn_pmd(pfn, init_prot)); 334 else 335 set_pmd(pmd, pfn_pmd(pfn, prot)); 336 337 pfn += PTRS_PER_PTE; 338 continue; 339 } 340 pte = one_page_table_init(pmd); 341 342 pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 343 pte += pte_ofs; 344 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn; 345 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) { 346 pgprot_t prot = PAGE_KERNEL; 347 /* 348 * first pass will use the same initial 349 * identity mapping attribute. 350 */ 351 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR); 352 353 if (is_x86_32_kernel_text(addr)) 354 prot = PAGE_KERNEL_EXEC; 355 356 pages_4k++; 357 if (mapping_iter == 1) { 358 set_pte(pte, pfn_pte(pfn, init_prot)); 359 last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE; 360 } else 361 set_pte(pte, pfn_pte(pfn, prot)); 362 } 363 } 364 } 365 if (mapping_iter == 1) { 366 /* 367 * update direct mapping page count only in the first 368 * iteration. 369 */ 370 update_page_count(PG_LEVEL_2M, pages_2m); 371 update_page_count(PG_LEVEL_4K, pages_4k); 372 373 /* 374 * local global flush tlb, which will flush the previous 375 * mappings present in both small and large page TLB's. 376 */ 377 __flush_tlb_all(); 378 379 /* 380 * Second iteration will set the actual desired PTE attributes. 381 */ 382 mapping_iter = 2; 383 goto repeat; 384 } 385 return last_map_addr; 386 } 387 388 #ifdef CONFIG_HIGHMEM 389 static void __init permanent_kmaps_init(pgd_t *pgd_base) 390 { 391 unsigned long vaddr = PKMAP_BASE; 392 393 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base); 394 395 pkmap_page_table = virt_to_kpte(vaddr); 396 } 397 #else 398 static inline void permanent_kmaps_init(pgd_t *pgd_base) 399 { 400 } 401 #endif /* CONFIG_HIGHMEM */ 402 403 void __init sync_initial_page_table(void) 404 { 405 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY, 406 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 407 KERNEL_PGD_PTRS); 408 409 /* 410 * sync back low identity map too. It is used for example 411 * in the 32-bit EFI stub. 412 */ 413 clone_pgd_range(initial_page_table, 414 swapper_pg_dir + KERNEL_PGD_BOUNDARY, 415 min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY)); 416 } 417 418 void __init native_pagetable_init(void) 419 { 420 unsigned long pfn, va; 421 pgd_t *pgd, *base = swapper_pg_dir; 422 p4d_t *p4d; 423 pud_t *pud; 424 pmd_t *pmd; 425 pte_t *pte; 426 427 /* 428 * Remove any mappings which extend past the end of physical 429 * memory from the boot time page table. 430 * In virtual address space, we should have at least two pages 431 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END 432 * definition. And max_low_pfn is set to VMALLOC_END physical 433 * address. If initial memory mapping is doing right job, we 434 * should have pte used near max_low_pfn or one pmd is not present. 435 */ 436 for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) { 437 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT); 438 pgd = base + pgd_index(va); 439 if (!pgd_present(*pgd)) 440 break; 441 442 p4d = p4d_offset(pgd, va); 443 pud = pud_offset(p4d, va); 444 pmd = pmd_offset(pud, va); 445 if (!pmd_present(*pmd)) 446 break; 447 448 /* should not be large page here */ 449 if (pmd_leaf(*pmd)) { 450 pr_warn("try to clear pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx, but pmd is big page and is not using pte !\n", 451 pfn, pmd, __pa(pmd)); 452 BUG_ON(1); 453 } 454 455 pte = pte_offset_kernel(pmd, va); 456 if (!pte_present(*pte)) 457 break; 458 459 printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n", 460 pfn, pmd, __pa(pmd), pte, __pa(pte)); 461 pte_clear(NULL, va, pte); 462 } 463 paging_init(); 464 } 465 466 /* 467 * Build a proper pagetable for the kernel mappings. Up until this 468 * point, we've been running on some set of pagetables constructed by 469 * the boot process. 470 * 471 * This will be a pagetable constructed in arch/x86/kernel/head_32.S. 472 * The root of the pagetable will be swapper_pg_dir. 473 * 474 * In general, pagetable_init() assumes that the pagetable may already 475 * be partially populated, and so it avoids stomping on any existing 476 * mappings. 477 */ 478 void __init early_ioremap_page_table_range_init(void) 479 { 480 pgd_t *pgd_base = swapper_pg_dir; 481 unsigned long vaddr, end; 482 483 /* 484 * Fixed mappings, only the page table structure has to be 485 * created - mappings will be set by set_fixmap(): 486 */ 487 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK; 488 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK; 489 page_table_range_init(vaddr, end, pgd_base); 490 early_ioremap_reset(); 491 } 492 493 static void __init pagetable_init(void) 494 { 495 pgd_t *pgd_base = swapper_pg_dir; 496 497 permanent_kmaps_init(pgd_base); 498 } 499 500 #define DEFAULT_PTE_MASK ~(_PAGE_NX | _PAGE_GLOBAL) 501 /* Bits supported by the hardware: */ 502 pteval_t __supported_pte_mask __read_mostly = DEFAULT_PTE_MASK; 503 /* Bits allowed in normal kernel mappings: */ 504 pteval_t __default_kernel_pte_mask __read_mostly = DEFAULT_PTE_MASK; 505 EXPORT_SYMBOL_GPL(__supported_pte_mask); 506 /* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */ 507 EXPORT_SYMBOL(__default_kernel_pte_mask); 508 509 /* user-defined highmem size */ 510 static unsigned int highmem_pages = -1; 511 512 /* 513 * highmem=size forces highmem to be exactly 'size' bytes. 514 * This works even on boxes that have no highmem otherwise. 515 * This also works to reduce highmem size on bigger boxes. 516 */ 517 static int __init parse_highmem(char *arg) 518 { 519 if (!arg) 520 return -EINVAL; 521 522 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT; 523 return 0; 524 } 525 early_param("highmem", parse_highmem); 526 527 #define MSG_HIGHMEM_TOO_BIG \ 528 "highmem size (%luMB) is bigger than pages available (%luMB)!\n" 529 530 #define MSG_LOWMEM_TOO_SMALL \ 531 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n" 532 /* 533 * All of RAM fits into lowmem - but if user wants highmem 534 * artificially via the highmem=x boot parameter then create 535 * it: 536 */ 537 static void __init lowmem_pfn_init(void) 538 { 539 /* max_low_pfn is 0, we already have early_res support */ 540 max_low_pfn = max_pfn; 541 542 if (highmem_pages == -1) 543 highmem_pages = 0; 544 #ifdef CONFIG_HIGHMEM 545 if (highmem_pages >= max_pfn) { 546 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG, 547 pages_to_mb(highmem_pages), pages_to_mb(max_pfn)); 548 highmem_pages = 0; 549 } 550 if (highmem_pages) { 551 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) { 552 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL, 553 pages_to_mb(highmem_pages)); 554 highmem_pages = 0; 555 } 556 max_low_pfn -= highmem_pages; 557 } 558 #else 559 if (highmem_pages) 560 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n"); 561 #endif 562 } 563 564 #define MSG_HIGHMEM_TOO_SMALL \ 565 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n" 566 567 #define MSG_HIGHMEM_TRIMMED \ 568 "Warning: only 4GB will be used. Support for for CONFIG_HIGHMEM64G was removed!\n" 569 /* 570 * We have more RAM than fits into lowmem - we try to put it into 571 * highmem, also taking the highmem=x boot parameter into account: 572 */ 573 static void __init highmem_pfn_init(void) 574 { 575 max_low_pfn = MAXMEM_PFN; 576 577 if (highmem_pages == -1) 578 highmem_pages = max_pfn - MAXMEM_PFN; 579 580 if (highmem_pages + MAXMEM_PFN < max_pfn) 581 max_pfn = MAXMEM_PFN + highmem_pages; 582 583 if (highmem_pages + MAXMEM_PFN > max_pfn) { 584 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL, 585 pages_to_mb(max_pfn - MAXMEM_PFN), 586 pages_to_mb(highmem_pages)); 587 highmem_pages = 0; 588 } 589 #ifndef CONFIG_HIGHMEM 590 /* Maximum memory usable is what is directly addressable */ 591 printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20); 592 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n"); 593 max_pfn = MAXMEM_PFN; 594 #else /* !CONFIG_HIGHMEM */ 595 if (max_pfn > MAX_NONPAE_PFN) { 596 max_pfn = MAX_NONPAE_PFN; 597 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED); 598 } 599 #endif /* !CONFIG_HIGHMEM */ 600 } 601 602 /* 603 * Determine low and high memory ranges: 604 */ 605 void __init find_low_pfn_range(void) 606 { 607 /* it could update max_pfn */ 608 609 if (max_pfn <= MAXMEM_PFN) 610 lowmem_pfn_init(); 611 else 612 highmem_pfn_init(); 613 } 614 615 #ifndef CONFIG_NUMA 616 void __init initmem_init(void) 617 { 618 #ifdef CONFIG_HIGHMEM 619 highstart_pfn = highend_pfn = max_pfn; 620 if (max_pfn > max_low_pfn) 621 highstart_pfn = max_low_pfn; 622 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n", 623 pages_to_mb(highend_pfn - highstart_pfn)); 624 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1; 625 #else 626 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1; 627 #endif 628 629 memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0); 630 631 __vmalloc_start_set = true; 632 633 printk(KERN_NOTICE "%ldMB LOWMEM available.\n", 634 pages_to_mb(max_low_pfn)); 635 636 setup_bootmem_allocator(); 637 } 638 #endif /* !CONFIG_NUMA */ 639 640 void __init setup_bootmem_allocator(void) 641 { 642 printk(KERN_INFO " mapped low ram: 0 - %08lx\n", 643 max_pfn_mapped<<PAGE_SHIFT); 644 printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT); 645 } 646 647 /* 648 * paging_init() sets up the page tables - note that the first 8MB are 649 * already mapped by head.S. 650 * 651 * This routines also unmaps the page at virtual kernel address 0, so 652 * that we can trap those pesky NULL-reference errors in the kernel. 653 */ 654 void __init paging_init(void) 655 { 656 pagetable_init(); 657 658 __flush_tlb_all(); 659 660 /* 661 * NOTE: at this point the bootmem allocator is fully available. 662 */ 663 olpc_dt_build_devicetree(); 664 sparse_init(); 665 zone_sizes_init(); 666 } 667 668 /* 669 * Test if the WP bit works in supervisor mode. It isn't supported on 386's 670 * and also on some strange 486's. All 586+'s are OK. This used to involve 671 * black magic jumps to work around some nasty CPU bugs, but fortunately the 672 * switch to using exceptions got rid of all that. 673 */ 674 static void __init test_wp_bit(void) 675 { 676 char z = 0; 677 678 printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode..."); 679 680 __set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO); 681 682 if (copy_to_kernel_nofault((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) { 683 clear_fixmap(FIX_WP_TEST); 684 printk(KERN_CONT "Ok.\n"); 685 return; 686 } 687 688 printk(KERN_CONT "No.\n"); 689 panic("Linux doesn't support CPUs with broken WP."); 690 } 691 692 void __init arch_mm_preinit(void) 693 { 694 pci_iommu_alloc(); 695 696 #ifdef CONFIG_FLATMEM 697 BUG_ON(!mem_map); 698 #endif 699 } 700 701 void __init mem_init(void) 702 { 703 after_bootmem = 1; 704 x86_init.hyper.init_after_bootmem(); 705 706 /* 707 * Check boundaries twice: Some fundamental inconsistencies can 708 * be detected at build time already. 709 */ 710 #define __FIXADDR_TOP (-PAGE_SIZE) 711 #ifdef CONFIG_HIGHMEM 712 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); 713 BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE); 714 #endif 715 #define high_memory (-128UL << 20) 716 BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END); 717 #undef high_memory 718 #undef __FIXADDR_TOP 719 720 #ifdef CONFIG_HIGHMEM 721 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); 722 BUG_ON(VMALLOC_END > PKMAP_BASE); 723 #endif 724 BUG_ON(VMALLOC_START >= VMALLOC_END); 725 BUG_ON((unsigned long)high_memory > VMALLOC_START); 726 727 test_wp_bit(); 728 } 729 730 int kernel_set_to_readonly __read_mostly; 731 732 static void mark_nxdata_nx(void) 733 { 734 /* 735 * When this called, init has already been executed and released, 736 * so everything past _etext should be NX. 737 */ 738 unsigned long start = PFN_ALIGN(_etext); 739 /* 740 * This comes from is_x86_32_kernel_text upper limit. Also HPAGE where used: 741 */ 742 unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start; 743 744 if (__supported_pte_mask & _PAGE_NX) 745 printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10); 746 set_memory_nx(start, size >> PAGE_SHIFT); 747 } 748 749 void mark_rodata_ro(void) 750 { 751 unsigned long start = PFN_ALIGN(_text); 752 unsigned long size = (unsigned long)__end_rodata - start; 753 754 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 755 pr_info("Write protecting kernel text and read-only data: %luk\n", 756 size >> 10); 757 758 kernel_set_to_readonly = 1; 759 760 #ifdef CONFIG_CPA_DEBUG 761 pr_info("Testing CPA: Reverting %lx-%lx\n", start, start + size); 762 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT); 763 764 pr_info("Testing CPA: write protecting again\n"); 765 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 766 #endif 767 mark_nxdata_nx(); 768 } 769