1 /* 2 * 3 * Copyright (C) 1995 Linus Torvalds 4 * 5 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 6 */ 7 8 #include <linux/module.h> 9 #include <linux/signal.h> 10 #include <linux/sched.h> 11 #include <linux/kernel.h> 12 #include <linux/errno.h> 13 #include <linux/string.h> 14 #include <linux/types.h> 15 #include <linux/ptrace.h> 16 #include <linux/mman.h> 17 #include <linux/mm.h> 18 #include <linux/hugetlb.h> 19 #include <linux/swap.h> 20 #include <linux/smp.h> 21 #include <linux/init.h> 22 #include <linux/highmem.h> 23 #include <linux/pagemap.h> 24 #include <linux/pci.h> 25 #include <linux/pfn.h> 26 #include <linux/poison.h> 27 #include <linux/bootmem.h> 28 #include <linux/memblock.h> 29 #include <linux/proc_fs.h> 30 #include <linux/memory_hotplug.h> 31 #include <linux/initrd.h> 32 #include <linux/cpumask.h> 33 #include <linux/gfp.h> 34 35 #include <asm/asm.h> 36 #include <asm/bios_ebda.h> 37 #include <asm/processor.h> 38 #include <asm/uaccess.h> 39 #include <asm/pgtable.h> 40 #include <asm/dma.h> 41 #include <asm/fixmap.h> 42 #include <asm/e820.h> 43 #include <asm/apic.h> 44 #include <asm/bugs.h> 45 #include <asm/tlb.h> 46 #include <asm/tlbflush.h> 47 #include <asm/olpc_ofw.h> 48 #include <asm/pgalloc.h> 49 #include <asm/sections.h> 50 #include <asm/paravirt.h> 51 #include <asm/setup.h> 52 #include <asm/cacheflush.h> 53 #include <asm/page_types.h> 54 #include <asm/init.h> 55 56 #include "mm_internal.h" 57 58 unsigned long highstart_pfn, highend_pfn; 59 60 static noinline int do_test_wp_bit(void); 61 62 bool __read_mostly __vmalloc_start_set = false; 63 64 /* 65 * Creates a middle page table and puts a pointer to it in the 66 * given global directory entry. This only returns the gd entry 67 * in non-PAE compilation mode, since the middle layer is folded. 68 */ 69 static pmd_t * __init one_md_table_init(pgd_t *pgd) 70 { 71 pud_t *pud; 72 pmd_t *pmd_table; 73 74 #ifdef CONFIG_X86_PAE 75 if (!(pgd_val(*pgd) & _PAGE_PRESENT)) { 76 pmd_table = (pmd_t *)alloc_low_page(); 77 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT); 78 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT)); 79 pud = pud_offset(pgd, 0); 80 BUG_ON(pmd_table != pmd_offset(pud, 0)); 81 82 return pmd_table; 83 } 84 #endif 85 pud = pud_offset(pgd, 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 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT); 101 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE)); 102 BUG_ON(page_table != pte_offset_kernel(pmd, 0)); 103 } 104 105 return pte_offset_kernel(pmd, 0); 106 } 107 108 pmd_t * __init populate_extra_pmd(unsigned long vaddr) 109 { 110 int pgd_idx = pgd_index(vaddr); 111 int pmd_idx = pmd_index(vaddr); 112 113 return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx; 114 } 115 116 pte_t * __init populate_extra_pte(unsigned long vaddr) 117 { 118 int pte_idx = pte_index(vaddr); 119 pmd_t *pmd; 120 121 pmd = populate_extra_pmd(vaddr); 122 return one_page_table_init(pmd) + pte_idx; 123 } 124 125 static unsigned long __init 126 page_table_range_init_count(unsigned long start, unsigned long end) 127 { 128 unsigned long count = 0; 129 #ifdef CONFIG_HIGHMEM 130 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT; 131 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT; 132 int pgd_idx, pmd_idx; 133 unsigned long vaddr; 134 135 if (pmd_idx_kmap_begin == pmd_idx_kmap_end) 136 return 0; 137 138 vaddr = start; 139 pgd_idx = pgd_index(vaddr); 140 pmd_idx = pmd_index(vaddr); 141 142 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) { 143 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); 144 pmd_idx++) { 145 if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin && 146 (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) 147 count++; 148 vaddr += PMD_SIZE; 149 } 150 pmd_idx = 0; 151 } 152 #endif 153 return count; 154 } 155 156 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd, 157 unsigned long vaddr, pte_t *lastpte, 158 void **adr) 159 { 160 #ifdef CONFIG_HIGHMEM 161 /* 162 * Something (early fixmap) may already have put a pte 163 * page here, which causes the page table allocation 164 * to become nonlinear. Attempt to fix it, and if it 165 * is still nonlinear then we have to bug. 166 */ 167 int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT; 168 int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT; 169 170 if (pmd_idx_kmap_begin != pmd_idx_kmap_end 171 && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin 172 && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) { 173 pte_t *newpte; 174 int i; 175 176 BUG_ON(after_bootmem); 177 newpte = *adr; 178 for (i = 0; i < PTRS_PER_PTE; i++) 179 set_pte(newpte + i, pte[i]); 180 *adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE); 181 182 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT); 183 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE)); 184 BUG_ON(newpte != pte_offset_kernel(pmd, 0)); 185 __flush_tlb_all(); 186 187 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT); 188 pte = newpte; 189 } 190 BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1) 191 && vaddr > fix_to_virt(FIX_KMAP_END) 192 && lastpte && lastpte + PTRS_PER_PTE != pte); 193 #endif 194 return pte; 195 } 196 197 /* 198 * This function initializes a certain range of kernel virtual memory 199 * with new bootmem page tables, everywhere page tables are missing in 200 * the given range. 201 * 202 * NOTE: The pagetables are allocated contiguous on the physical space 203 * so we can cache the place of the first one and move around without 204 * checking the pgd every time. 205 */ 206 static void __init 207 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base) 208 { 209 int pgd_idx, pmd_idx; 210 unsigned long vaddr; 211 pgd_t *pgd; 212 pmd_t *pmd; 213 pte_t *pte = NULL; 214 unsigned long count = page_table_range_init_count(start, end); 215 void *adr = NULL; 216 217 if (count) 218 adr = alloc_low_pages(count); 219 220 vaddr = start; 221 pgd_idx = pgd_index(vaddr); 222 pmd_idx = pmd_index(vaddr); 223 pgd = pgd_base + pgd_idx; 224 225 for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) { 226 pmd = one_md_table_init(pgd); 227 pmd = pmd + pmd_index(vaddr); 228 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); 229 pmd++, pmd_idx++) { 230 pte = page_table_kmap_check(one_page_table_init(pmd), 231 pmd, vaddr, pte, &adr); 232 233 vaddr += PMD_SIZE; 234 } 235 pmd_idx = 0; 236 } 237 } 238 239 static inline int is_kernel_text(unsigned long addr) 240 { 241 if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end) 242 return 1; 243 return 0; 244 } 245 246 /* 247 * This maps the physical memory to kernel virtual address space, a total 248 * of max_low_pfn pages, by creating page tables starting from address 249 * PAGE_OFFSET: 250 */ 251 unsigned long __init 252 kernel_physical_mapping_init(unsigned long start, 253 unsigned long end, 254 unsigned long page_size_mask) 255 { 256 int use_pse = page_size_mask == (1<<PG_LEVEL_2M); 257 unsigned long last_map_addr = end; 258 unsigned long start_pfn, end_pfn; 259 pgd_t *pgd_base = swapper_pg_dir; 260 int pgd_idx, pmd_idx, pte_ofs; 261 unsigned long pfn; 262 pgd_t *pgd; 263 pmd_t *pmd; 264 pte_t *pte; 265 unsigned pages_2m, pages_4k; 266 int mapping_iter; 267 268 start_pfn = start >> PAGE_SHIFT; 269 end_pfn = end >> PAGE_SHIFT; 270 271 /* 272 * First iteration will setup identity mapping using large/small pages 273 * based on use_pse, with other attributes same as set by 274 * the early code in head_32.S 275 * 276 * Second iteration will setup the appropriate attributes (NX, GLOBAL..) 277 * as desired for the kernel identity mapping. 278 * 279 * This two pass mechanism conforms to the TLB app note which says: 280 * 281 * "Software should not write to a paging-structure entry in a way 282 * that would change, for any linear address, both the page size 283 * and either the page frame or attributes." 284 */ 285 mapping_iter = 1; 286 287 if (!boot_cpu_has(X86_FEATURE_PSE)) 288 use_pse = 0; 289 290 repeat: 291 pages_2m = pages_4k = 0; 292 pfn = start_pfn; 293 pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 294 pgd = pgd_base + pgd_idx; 295 for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) { 296 pmd = one_md_table_init(pgd); 297 298 if (pfn >= end_pfn) 299 continue; 300 #ifdef CONFIG_X86_PAE 301 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 302 pmd += pmd_idx; 303 #else 304 pmd_idx = 0; 305 #endif 306 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn; 307 pmd++, pmd_idx++) { 308 unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET; 309 310 /* 311 * Map with big pages if possible, otherwise 312 * create normal page tables: 313 */ 314 if (use_pse) { 315 unsigned int addr2; 316 pgprot_t prot = PAGE_KERNEL_LARGE; 317 /* 318 * first pass will use the same initial 319 * identity mapping attribute + _PAGE_PSE. 320 */ 321 pgprot_t init_prot = 322 __pgprot(PTE_IDENT_ATTR | 323 _PAGE_PSE); 324 325 pfn &= PMD_MASK >> PAGE_SHIFT; 326 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE + 327 PAGE_OFFSET + PAGE_SIZE-1; 328 329 if (is_kernel_text(addr) || 330 is_kernel_text(addr2)) 331 prot = PAGE_KERNEL_LARGE_EXEC; 332 333 pages_2m++; 334 if (mapping_iter == 1) 335 set_pmd(pmd, pfn_pmd(pfn, init_prot)); 336 else 337 set_pmd(pmd, pfn_pmd(pfn, prot)); 338 339 pfn += PTRS_PER_PTE; 340 continue; 341 } 342 pte = one_page_table_init(pmd); 343 344 pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET); 345 pte += pte_ofs; 346 for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn; 347 pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) { 348 pgprot_t prot = PAGE_KERNEL; 349 /* 350 * first pass will use the same initial 351 * identity mapping attribute. 352 */ 353 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR); 354 355 if (is_kernel_text(addr)) 356 prot = PAGE_KERNEL_EXEC; 357 358 pages_4k++; 359 if (mapping_iter == 1) { 360 set_pte(pte, pfn_pte(pfn, init_prot)); 361 last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE; 362 } else 363 set_pte(pte, pfn_pte(pfn, prot)); 364 } 365 } 366 } 367 if (mapping_iter == 1) { 368 /* 369 * update direct mapping page count only in the first 370 * iteration. 371 */ 372 update_page_count(PG_LEVEL_2M, pages_2m); 373 update_page_count(PG_LEVEL_4K, pages_4k); 374 375 /* 376 * local global flush tlb, which will flush the previous 377 * mappings present in both small and large page TLB's. 378 */ 379 __flush_tlb_all(); 380 381 /* 382 * Second iteration will set the actual desired PTE attributes. 383 */ 384 mapping_iter = 2; 385 goto repeat; 386 } 387 return last_map_addr; 388 } 389 390 pte_t *kmap_pte; 391 392 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr) 393 { 394 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), 395 vaddr), vaddr), vaddr); 396 } 397 398 static void __init kmap_init(void) 399 { 400 unsigned long kmap_vstart; 401 402 /* 403 * Cache the first kmap pte: 404 */ 405 kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN); 406 kmap_pte = kmap_get_fixmap_pte(kmap_vstart); 407 } 408 409 #ifdef CONFIG_HIGHMEM 410 static void __init permanent_kmaps_init(pgd_t *pgd_base) 411 { 412 unsigned long vaddr; 413 pgd_t *pgd; 414 pud_t *pud; 415 pmd_t *pmd; 416 pte_t *pte; 417 418 vaddr = PKMAP_BASE; 419 page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base); 420 421 pgd = swapper_pg_dir + pgd_index(vaddr); 422 pud = pud_offset(pgd, vaddr); 423 pmd = pmd_offset(pud, vaddr); 424 pte = pte_offset_kernel(pmd, vaddr); 425 pkmap_page_table = pte; 426 } 427 428 void __init add_highpages_with_active_regions(int nid, 429 unsigned long start_pfn, unsigned long end_pfn) 430 { 431 phys_addr_t start, end; 432 u64 i; 433 434 for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) { 435 unsigned long pfn = clamp_t(unsigned long, PFN_UP(start), 436 start_pfn, end_pfn); 437 unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end), 438 start_pfn, end_pfn); 439 for ( ; pfn < e_pfn; pfn++) 440 if (pfn_valid(pfn)) 441 free_highmem_page(pfn_to_page(pfn)); 442 } 443 } 444 #else 445 static inline void permanent_kmaps_init(pgd_t *pgd_base) 446 { 447 } 448 #endif /* CONFIG_HIGHMEM */ 449 450 void __init native_pagetable_init(void) 451 { 452 unsigned long pfn, va; 453 pgd_t *pgd, *base = swapper_pg_dir; 454 pud_t *pud; 455 pmd_t *pmd; 456 pte_t *pte; 457 458 /* 459 * Remove any mappings which extend past the end of physical 460 * memory from the boot time page table. 461 * In virtual address space, we should have at least two pages 462 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END 463 * definition. And max_low_pfn is set to VMALLOC_END physical 464 * address. If initial memory mapping is doing right job, we 465 * should have pte used near max_low_pfn or one pmd is not present. 466 */ 467 for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) { 468 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT); 469 pgd = base + pgd_index(va); 470 if (!pgd_present(*pgd)) 471 break; 472 473 pud = pud_offset(pgd, va); 474 pmd = pmd_offset(pud, va); 475 if (!pmd_present(*pmd)) 476 break; 477 478 /* should not be large page here */ 479 if (pmd_large(*pmd)) { 480 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", 481 pfn, pmd, __pa(pmd)); 482 BUG_ON(1); 483 } 484 485 pte = pte_offset_kernel(pmd, va); 486 if (!pte_present(*pte)) 487 break; 488 489 printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n", 490 pfn, pmd, __pa(pmd), pte, __pa(pte)); 491 pte_clear(NULL, va, pte); 492 } 493 paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT); 494 paging_init(); 495 } 496 497 /* 498 * Build a proper pagetable for the kernel mappings. Up until this 499 * point, we've been running on some set of pagetables constructed by 500 * the boot process. 501 * 502 * If we're booting on native hardware, this will be a pagetable 503 * constructed in arch/x86/kernel/head_32.S. The root of the 504 * pagetable will be swapper_pg_dir. 505 * 506 * If we're booting paravirtualized under a hypervisor, then there are 507 * more options: we may already be running PAE, and the pagetable may 508 * or may not be based in swapper_pg_dir. In any case, 509 * paravirt_pagetable_init() will set up swapper_pg_dir 510 * appropriately for the rest of the initialization to work. 511 * 512 * In general, pagetable_init() assumes that the pagetable may already 513 * be partially populated, and so it avoids stomping on any existing 514 * mappings. 515 */ 516 void __init early_ioremap_page_table_range_init(void) 517 { 518 pgd_t *pgd_base = swapper_pg_dir; 519 unsigned long vaddr, end; 520 521 /* 522 * Fixed mappings, only the page table structure has to be 523 * created - mappings will be set by set_fixmap(): 524 */ 525 vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK; 526 end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK; 527 page_table_range_init(vaddr, end, pgd_base); 528 early_ioremap_reset(); 529 } 530 531 static void __init pagetable_init(void) 532 { 533 pgd_t *pgd_base = swapper_pg_dir; 534 535 permanent_kmaps_init(pgd_base); 536 } 537 538 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL); 539 EXPORT_SYMBOL_GPL(__supported_pte_mask); 540 541 /* user-defined highmem size */ 542 static unsigned int highmem_pages = -1; 543 544 /* 545 * highmem=size forces highmem to be exactly 'size' bytes. 546 * This works even on boxes that have no highmem otherwise. 547 * This also works to reduce highmem size on bigger boxes. 548 */ 549 static int __init parse_highmem(char *arg) 550 { 551 if (!arg) 552 return -EINVAL; 553 554 highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT; 555 return 0; 556 } 557 early_param("highmem", parse_highmem); 558 559 #define MSG_HIGHMEM_TOO_BIG \ 560 "highmem size (%luMB) is bigger than pages available (%luMB)!\n" 561 562 #define MSG_LOWMEM_TOO_SMALL \ 563 "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n" 564 /* 565 * All of RAM fits into lowmem - but if user wants highmem 566 * artificially via the highmem=x boot parameter then create 567 * it: 568 */ 569 static void __init lowmem_pfn_init(void) 570 { 571 /* max_low_pfn is 0, we already have early_res support */ 572 max_low_pfn = max_pfn; 573 574 if (highmem_pages == -1) 575 highmem_pages = 0; 576 #ifdef CONFIG_HIGHMEM 577 if (highmem_pages >= max_pfn) { 578 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG, 579 pages_to_mb(highmem_pages), pages_to_mb(max_pfn)); 580 highmem_pages = 0; 581 } 582 if (highmem_pages) { 583 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) { 584 printk(KERN_ERR MSG_LOWMEM_TOO_SMALL, 585 pages_to_mb(highmem_pages)); 586 highmem_pages = 0; 587 } 588 max_low_pfn -= highmem_pages; 589 } 590 #else 591 if (highmem_pages) 592 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n"); 593 #endif 594 } 595 596 #define MSG_HIGHMEM_TOO_SMALL \ 597 "only %luMB highmem pages available, ignoring highmem size of %luMB!\n" 598 599 #define MSG_HIGHMEM_TRIMMED \ 600 "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n" 601 /* 602 * We have more RAM than fits into lowmem - we try to put it into 603 * highmem, also taking the highmem=x boot parameter into account: 604 */ 605 static void __init highmem_pfn_init(void) 606 { 607 max_low_pfn = MAXMEM_PFN; 608 609 if (highmem_pages == -1) 610 highmem_pages = max_pfn - MAXMEM_PFN; 611 612 if (highmem_pages + MAXMEM_PFN < max_pfn) 613 max_pfn = MAXMEM_PFN + highmem_pages; 614 615 if (highmem_pages + MAXMEM_PFN > max_pfn) { 616 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL, 617 pages_to_mb(max_pfn - MAXMEM_PFN), 618 pages_to_mb(highmem_pages)); 619 highmem_pages = 0; 620 } 621 #ifndef CONFIG_HIGHMEM 622 /* Maximum memory usable is what is directly addressable */ 623 printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20); 624 if (max_pfn > MAX_NONPAE_PFN) 625 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n"); 626 else 627 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n"); 628 max_pfn = MAXMEM_PFN; 629 #else /* !CONFIG_HIGHMEM */ 630 #ifndef CONFIG_HIGHMEM64G 631 if (max_pfn > MAX_NONPAE_PFN) { 632 max_pfn = MAX_NONPAE_PFN; 633 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED); 634 } 635 #endif /* !CONFIG_HIGHMEM64G */ 636 #endif /* !CONFIG_HIGHMEM */ 637 } 638 639 /* 640 * Determine low and high memory ranges: 641 */ 642 void __init find_low_pfn_range(void) 643 { 644 /* it could update max_pfn */ 645 646 if (max_pfn <= MAXMEM_PFN) 647 lowmem_pfn_init(); 648 else 649 highmem_pfn_init(); 650 } 651 652 #ifndef CONFIG_NEED_MULTIPLE_NODES 653 void __init initmem_init(void) 654 { 655 #ifdef CONFIG_HIGHMEM 656 highstart_pfn = highend_pfn = max_pfn; 657 if (max_pfn > max_low_pfn) 658 highstart_pfn = max_low_pfn; 659 printk(KERN_NOTICE "%ldMB HIGHMEM available.\n", 660 pages_to_mb(highend_pfn - highstart_pfn)); 661 high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1; 662 #else 663 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1; 664 #endif 665 666 memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0); 667 sparse_memory_present_with_active_regions(0); 668 669 #ifdef CONFIG_FLATMEM 670 max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn; 671 #endif 672 __vmalloc_start_set = true; 673 674 printk(KERN_NOTICE "%ldMB LOWMEM available.\n", 675 pages_to_mb(max_low_pfn)); 676 677 setup_bootmem_allocator(); 678 } 679 #endif /* !CONFIG_NEED_MULTIPLE_NODES */ 680 681 void __init setup_bootmem_allocator(void) 682 { 683 printk(KERN_INFO " mapped low ram: 0 - %08lx\n", 684 max_pfn_mapped<<PAGE_SHIFT); 685 printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT); 686 } 687 688 /* 689 * paging_init() sets up the page tables - note that the first 8MB are 690 * already mapped by head.S. 691 * 692 * This routines also unmaps the page at virtual kernel address 0, so 693 * that we can trap those pesky NULL-reference errors in the kernel. 694 */ 695 void __init paging_init(void) 696 { 697 pagetable_init(); 698 699 __flush_tlb_all(); 700 701 kmap_init(); 702 703 /* 704 * NOTE: at this point the bootmem allocator is fully available. 705 */ 706 olpc_dt_build_devicetree(); 707 sparse_memory_present_with_active_regions(MAX_NUMNODES); 708 sparse_init(); 709 zone_sizes_init(); 710 } 711 712 /* 713 * Test if the WP bit works in supervisor mode. It isn't supported on 386's 714 * and also on some strange 486's. All 586+'s are OK. This used to involve 715 * black magic jumps to work around some nasty CPU bugs, but fortunately the 716 * switch to using exceptions got rid of all that. 717 */ 718 static void __init test_wp_bit(void) 719 { 720 printk(KERN_INFO 721 "Checking if this processor honours the WP bit even in supervisor mode..."); 722 723 /* Any page-aligned address will do, the test is non-destructive */ 724 __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_KERNEL_RO); 725 boot_cpu_data.wp_works_ok = do_test_wp_bit(); 726 clear_fixmap(FIX_WP_TEST); 727 728 if (!boot_cpu_data.wp_works_ok) { 729 printk(KERN_CONT "No.\n"); 730 panic("Linux doesn't support CPUs with broken WP."); 731 } else { 732 printk(KERN_CONT "Ok.\n"); 733 } 734 } 735 736 void __init mem_init(void) 737 { 738 pci_iommu_alloc(); 739 740 #ifdef CONFIG_FLATMEM 741 BUG_ON(!mem_map); 742 #endif 743 /* 744 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to 745 * be done before free_all_bootmem(). Memblock use free low memory for 746 * temporary data (see find_range_array()) and for this purpose can use 747 * pages that was already passed to the buddy allocator, hence marked as 748 * not accessible in the page tables when compiled with 749 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not 750 * important here. 751 */ 752 set_highmem_pages_init(); 753 754 /* this will put all low memory onto the freelists */ 755 free_all_bootmem(); 756 757 after_bootmem = 1; 758 759 mem_init_print_info(NULL); 760 printk(KERN_INFO "virtual kernel memory layout:\n" 761 " fixmap : 0x%08lx - 0x%08lx (%4ld kB)\n" 762 #ifdef CONFIG_HIGHMEM 763 " pkmap : 0x%08lx - 0x%08lx (%4ld kB)\n" 764 #endif 765 " vmalloc : 0x%08lx - 0x%08lx (%4ld MB)\n" 766 " lowmem : 0x%08lx - 0x%08lx (%4ld MB)\n" 767 " .init : 0x%08lx - 0x%08lx (%4ld kB)\n" 768 " .data : 0x%08lx - 0x%08lx (%4ld kB)\n" 769 " .text : 0x%08lx - 0x%08lx (%4ld kB)\n", 770 FIXADDR_START, FIXADDR_TOP, 771 (FIXADDR_TOP - FIXADDR_START) >> 10, 772 773 #ifdef CONFIG_HIGHMEM 774 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, 775 (LAST_PKMAP*PAGE_SIZE) >> 10, 776 #endif 777 778 VMALLOC_START, VMALLOC_END, 779 (VMALLOC_END - VMALLOC_START) >> 20, 780 781 (unsigned long)__va(0), (unsigned long)high_memory, 782 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20, 783 784 (unsigned long)&__init_begin, (unsigned long)&__init_end, 785 ((unsigned long)&__init_end - 786 (unsigned long)&__init_begin) >> 10, 787 788 (unsigned long)&_etext, (unsigned long)&_edata, 789 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10, 790 791 (unsigned long)&_text, (unsigned long)&_etext, 792 ((unsigned long)&_etext - (unsigned long)&_text) >> 10); 793 794 /* 795 * Check boundaries twice: Some fundamental inconsistencies can 796 * be detected at build time already. 797 */ 798 #define __FIXADDR_TOP (-PAGE_SIZE) 799 #ifdef CONFIG_HIGHMEM 800 BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); 801 BUILD_BUG_ON(VMALLOC_END > PKMAP_BASE); 802 #endif 803 #define high_memory (-128UL << 20) 804 BUILD_BUG_ON(VMALLOC_START >= VMALLOC_END); 805 #undef high_memory 806 #undef __FIXADDR_TOP 807 808 #ifdef CONFIG_HIGHMEM 809 BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE > FIXADDR_START); 810 BUG_ON(VMALLOC_END > PKMAP_BASE); 811 #endif 812 BUG_ON(VMALLOC_START >= VMALLOC_END); 813 BUG_ON((unsigned long)high_memory > VMALLOC_START); 814 815 if (boot_cpu_data.wp_works_ok < 0) 816 test_wp_bit(); 817 } 818 819 #ifdef CONFIG_MEMORY_HOTPLUG 820 int arch_add_memory(int nid, u64 start, u64 size, bool for_device) 821 { 822 struct pglist_data *pgdata = NODE_DATA(nid); 823 struct zone *zone = pgdata->node_zones + 824 zone_for_memory(nid, start, size, ZONE_HIGHMEM, for_device); 825 unsigned long start_pfn = start >> PAGE_SHIFT; 826 unsigned long nr_pages = size >> PAGE_SHIFT; 827 828 return __add_pages(nid, zone, start_pfn, nr_pages); 829 } 830 831 #ifdef CONFIG_MEMORY_HOTREMOVE 832 int arch_remove_memory(u64 start, u64 size) 833 { 834 unsigned long start_pfn = start >> PAGE_SHIFT; 835 unsigned long nr_pages = size >> PAGE_SHIFT; 836 struct zone *zone; 837 838 zone = page_zone(pfn_to_page(start_pfn)); 839 return __remove_pages(zone, start_pfn, nr_pages); 840 } 841 #endif 842 #endif 843 844 /* 845 * This function cannot be __init, since exceptions don't work in that 846 * section. Put this after the callers, so that it cannot be inlined. 847 */ 848 static noinline int do_test_wp_bit(void) 849 { 850 char tmp_reg; 851 int flag; 852 853 __asm__ __volatile__( 854 " movb %0, %1 \n" 855 "1: movb %1, %0 \n" 856 " xorl %2, %2 \n" 857 "2: \n" 858 _ASM_EXTABLE(1b,2b) 859 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)), 860 "=q" (tmp_reg), 861 "=r" (flag) 862 :"2" (1) 863 :"memory"); 864 865 return flag; 866 } 867 868 const int rodata_test_data = 0xC3; 869 EXPORT_SYMBOL_GPL(rodata_test_data); 870 871 int kernel_set_to_readonly __read_mostly; 872 873 void set_kernel_text_rw(void) 874 { 875 unsigned long start = PFN_ALIGN(_text); 876 unsigned long size = PFN_ALIGN(_etext) - start; 877 878 if (!kernel_set_to_readonly) 879 return; 880 881 pr_debug("Set kernel text: %lx - %lx for read write\n", 882 start, start+size); 883 884 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT); 885 } 886 887 void set_kernel_text_ro(void) 888 { 889 unsigned long start = PFN_ALIGN(_text); 890 unsigned long size = PFN_ALIGN(_etext) - start; 891 892 if (!kernel_set_to_readonly) 893 return; 894 895 pr_debug("Set kernel text: %lx - %lx for read only\n", 896 start, start+size); 897 898 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 899 } 900 901 static void mark_nxdata_nx(void) 902 { 903 /* 904 * When this called, init has already been executed and released, 905 * so everything past _etext should be NX. 906 */ 907 unsigned long start = PFN_ALIGN(_etext); 908 /* 909 * This comes from is_kernel_text upper limit. Also HPAGE where used: 910 */ 911 unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start; 912 913 if (__supported_pte_mask & _PAGE_NX) 914 printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10); 915 set_pages_nx(virt_to_page(start), size >> PAGE_SHIFT); 916 } 917 918 void mark_rodata_ro(void) 919 { 920 unsigned long start = PFN_ALIGN(_text); 921 unsigned long size = PFN_ALIGN(_etext) - start; 922 923 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 924 printk(KERN_INFO "Write protecting the kernel text: %luk\n", 925 size >> 10); 926 927 kernel_set_to_readonly = 1; 928 929 #ifdef CONFIG_CPA_DEBUG 930 printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n", 931 start, start+size); 932 set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT); 933 934 printk(KERN_INFO "Testing CPA: write protecting again\n"); 935 set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT); 936 #endif 937 938 start += size; 939 size = (unsigned long)__end_rodata - start; 940 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 941 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n", 942 size >> 10); 943 rodata_test(); 944 945 #ifdef CONFIG_CPA_DEBUG 946 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size); 947 set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT); 948 949 printk(KERN_INFO "Testing CPA: write protecting again\n"); 950 set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT); 951 #endif 952 mark_nxdata_nx(); 953 if (__supported_pte_mask & _PAGE_NX) 954 debug_checkwx(); 955 } 956