1 /* 2 * Copyright (C) 1995 Linus Torvalds 3 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs. 4 * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar 5 */ 6 #include <linux/sched.h> /* test_thread_flag(), ... */ 7 #include <linux/kdebug.h> /* oops_begin/end, ... */ 8 #include <linux/module.h> /* search_exception_table */ 9 #include <linux/bootmem.h> /* max_low_pfn */ 10 #include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */ 11 #include <linux/mmiotrace.h> /* kmmio_handler, ... */ 12 #include <linux/perf_event.h> /* perf_sw_event */ 13 #include <linux/hugetlb.h> /* hstate_index_to_shift */ 14 #include <linux/prefetch.h> /* prefetchw */ 15 #include <linux/context_tracking.h> /* exception_enter(), ... */ 16 #include <linux/uaccess.h> /* faulthandler_disabled() */ 17 18 #include <asm/cpufeature.h> /* boot_cpu_has, ... */ 19 #include <asm/traps.h> /* dotraplinkage, ... */ 20 #include <asm/pgalloc.h> /* pgd_*(), ... */ 21 #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */ 22 #include <asm/fixmap.h> /* VSYSCALL_ADDR */ 23 #include <asm/vsyscall.h> /* emulate_vsyscall */ 24 #include <asm/vm86.h> /* struct vm86 */ 25 #include <asm/mmu_context.h> /* vma_pkey() */ 26 27 #define CREATE_TRACE_POINTS 28 #include <asm/trace/exceptions.h> 29 30 /* 31 * Page fault error code bits: 32 * 33 * bit 0 == 0: no page found 1: protection fault 34 * bit 1 == 0: read access 1: write access 35 * bit 2 == 0: kernel-mode access 1: user-mode access 36 * bit 3 == 1: use of reserved bit detected 37 * bit 4 == 1: fault was an instruction fetch 38 * bit 5 == 1: protection keys block access 39 */ 40 enum x86_pf_error_code { 41 42 PF_PROT = 1 << 0, 43 PF_WRITE = 1 << 1, 44 PF_USER = 1 << 2, 45 PF_RSVD = 1 << 3, 46 PF_INSTR = 1 << 4, 47 PF_PK = 1 << 5, 48 }; 49 50 /* 51 * Returns 0 if mmiotrace is disabled, or if the fault is not 52 * handled by mmiotrace: 53 */ 54 static nokprobe_inline int 55 kmmio_fault(struct pt_regs *regs, unsigned long addr) 56 { 57 if (unlikely(is_kmmio_active())) 58 if (kmmio_handler(regs, addr) == 1) 59 return -1; 60 return 0; 61 } 62 63 static nokprobe_inline int kprobes_fault(struct pt_regs *regs) 64 { 65 int ret = 0; 66 67 /* kprobe_running() needs smp_processor_id() */ 68 if (kprobes_built_in() && !user_mode(regs)) { 69 preempt_disable(); 70 if (kprobe_running() && kprobe_fault_handler(regs, 14)) 71 ret = 1; 72 preempt_enable(); 73 } 74 75 return ret; 76 } 77 78 /* 79 * Prefetch quirks: 80 * 81 * 32-bit mode: 82 * 83 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. 84 * Check that here and ignore it. 85 * 86 * 64-bit mode: 87 * 88 * Sometimes the CPU reports invalid exceptions on prefetch. 89 * Check that here and ignore it. 90 * 91 * Opcode checker based on code by Richard Brunner. 92 */ 93 static inline int 94 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, 95 unsigned char opcode, int *prefetch) 96 { 97 unsigned char instr_hi = opcode & 0xf0; 98 unsigned char instr_lo = opcode & 0x0f; 99 100 switch (instr_hi) { 101 case 0x20: 102 case 0x30: 103 /* 104 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. 105 * In X86_64 long mode, the CPU will signal invalid 106 * opcode if some of these prefixes are present so 107 * X86_64 will never get here anyway 108 */ 109 return ((instr_lo & 7) == 0x6); 110 #ifdef CONFIG_X86_64 111 case 0x40: 112 /* 113 * In AMD64 long mode 0x40..0x4F are valid REX prefixes 114 * Need to figure out under what instruction mode the 115 * instruction was issued. Could check the LDT for lm, 116 * but for now it's good enough to assume that long 117 * mode only uses well known segments or kernel. 118 */ 119 return (!user_mode(regs) || user_64bit_mode(regs)); 120 #endif 121 case 0x60: 122 /* 0x64 thru 0x67 are valid prefixes in all modes. */ 123 return (instr_lo & 0xC) == 0x4; 124 case 0xF0: 125 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ 126 return !instr_lo || (instr_lo>>1) == 1; 127 case 0x00: 128 /* Prefetch instruction is 0x0F0D or 0x0F18 */ 129 if (probe_kernel_address(instr, opcode)) 130 return 0; 131 132 *prefetch = (instr_lo == 0xF) && 133 (opcode == 0x0D || opcode == 0x18); 134 return 0; 135 default: 136 return 0; 137 } 138 } 139 140 static int 141 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) 142 { 143 unsigned char *max_instr; 144 unsigned char *instr; 145 int prefetch = 0; 146 147 /* 148 * If it was a exec (instruction fetch) fault on NX page, then 149 * do not ignore the fault: 150 */ 151 if (error_code & PF_INSTR) 152 return 0; 153 154 instr = (void *)convert_ip_to_linear(current, regs); 155 max_instr = instr + 15; 156 157 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX) 158 return 0; 159 160 while (instr < max_instr) { 161 unsigned char opcode; 162 163 if (probe_kernel_address(instr, opcode)) 164 break; 165 166 instr++; 167 168 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) 169 break; 170 } 171 return prefetch; 172 } 173 174 /* 175 * A protection key fault means that the PKRU value did not allow 176 * access to some PTE. Userspace can figure out what PKRU was 177 * from the XSAVE state, and this function fills out a field in 178 * siginfo so userspace can discover which protection key was set 179 * on the PTE. 180 * 181 * If we get here, we know that the hardware signaled a PF_PK 182 * fault and that there was a VMA once we got in the fault 183 * handler. It does *not* guarantee that the VMA we find here 184 * was the one that we faulted on. 185 * 186 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4); 187 * 2. T1 : set PKRU to deny access to pkey=4, touches page 188 * 3. T1 : faults... 189 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5); 190 * 5. T1 : enters fault handler, takes mmap_sem, etc... 191 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really 192 * faulted on a pte with its pkey=4. 193 */ 194 static void fill_sig_info_pkey(int si_code, siginfo_t *info, 195 struct vm_area_struct *vma) 196 { 197 /* This is effectively an #ifdef */ 198 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 199 return; 200 201 /* Fault not from Protection Keys: nothing to do */ 202 if (si_code != SEGV_PKUERR) 203 return; 204 /* 205 * force_sig_info_fault() is called from a number of 206 * contexts, some of which have a VMA and some of which 207 * do not. The PF_PK handing happens after we have a 208 * valid VMA, so we should never reach this without a 209 * valid VMA. 210 */ 211 if (!vma) { 212 WARN_ONCE(1, "PKU fault with no VMA passed in"); 213 info->si_pkey = 0; 214 return; 215 } 216 /* 217 * si_pkey should be thought of as a strong hint, but not 218 * absolutely guranteed to be 100% accurate because of 219 * the race explained above. 220 */ 221 info->si_pkey = vma_pkey(vma); 222 } 223 224 static void 225 force_sig_info_fault(int si_signo, int si_code, unsigned long address, 226 struct task_struct *tsk, struct vm_area_struct *vma, 227 int fault) 228 { 229 unsigned lsb = 0; 230 siginfo_t info; 231 232 info.si_signo = si_signo; 233 info.si_errno = 0; 234 info.si_code = si_code; 235 info.si_addr = (void __user *)address; 236 if (fault & VM_FAULT_HWPOISON_LARGE) 237 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); 238 if (fault & VM_FAULT_HWPOISON) 239 lsb = PAGE_SHIFT; 240 info.si_addr_lsb = lsb; 241 242 fill_sig_info_pkey(si_code, &info, vma); 243 244 force_sig_info(si_signo, &info, tsk); 245 } 246 247 DEFINE_SPINLOCK(pgd_lock); 248 LIST_HEAD(pgd_list); 249 250 #ifdef CONFIG_X86_32 251 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) 252 { 253 unsigned index = pgd_index(address); 254 pgd_t *pgd_k; 255 pud_t *pud, *pud_k; 256 pmd_t *pmd, *pmd_k; 257 258 pgd += index; 259 pgd_k = init_mm.pgd + index; 260 261 if (!pgd_present(*pgd_k)) 262 return NULL; 263 264 /* 265 * set_pgd(pgd, *pgd_k); here would be useless on PAE 266 * and redundant with the set_pmd() on non-PAE. As would 267 * set_pud. 268 */ 269 pud = pud_offset(pgd, address); 270 pud_k = pud_offset(pgd_k, address); 271 if (!pud_present(*pud_k)) 272 return NULL; 273 274 pmd = pmd_offset(pud, address); 275 pmd_k = pmd_offset(pud_k, address); 276 if (!pmd_present(*pmd_k)) 277 return NULL; 278 279 if (!pmd_present(*pmd)) 280 set_pmd(pmd, *pmd_k); 281 else 282 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); 283 284 return pmd_k; 285 } 286 287 void vmalloc_sync_all(void) 288 { 289 unsigned long address; 290 291 if (SHARED_KERNEL_PMD) 292 return; 293 294 for (address = VMALLOC_START & PMD_MASK; 295 address >= TASK_SIZE_MAX && address < FIXADDR_TOP; 296 address += PMD_SIZE) { 297 struct page *page; 298 299 spin_lock(&pgd_lock); 300 list_for_each_entry(page, &pgd_list, lru) { 301 spinlock_t *pgt_lock; 302 pmd_t *ret; 303 304 /* the pgt_lock only for Xen */ 305 pgt_lock = &pgd_page_get_mm(page)->page_table_lock; 306 307 spin_lock(pgt_lock); 308 ret = vmalloc_sync_one(page_address(page), address); 309 spin_unlock(pgt_lock); 310 311 if (!ret) 312 break; 313 } 314 spin_unlock(&pgd_lock); 315 } 316 } 317 318 /* 319 * 32-bit: 320 * 321 * Handle a fault on the vmalloc or module mapping area 322 */ 323 static noinline int vmalloc_fault(unsigned long address) 324 { 325 unsigned long pgd_paddr; 326 pmd_t *pmd_k; 327 pte_t *pte_k; 328 329 /* Make sure we are in vmalloc area: */ 330 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 331 return -1; 332 333 WARN_ON_ONCE(in_nmi()); 334 335 /* 336 * Synchronize this task's top level page-table 337 * with the 'reference' page table. 338 * 339 * Do _not_ use "current" here. We might be inside 340 * an interrupt in the middle of a task switch.. 341 */ 342 pgd_paddr = read_cr3(); 343 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); 344 if (!pmd_k) 345 return -1; 346 347 if (pmd_huge(*pmd_k)) 348 return 0; 349 350 pte_k = pte_offset_kernel(pmd_k, address); 351 if (!pte_present(*pte_k)) 352 return -1; 353 354 return 0; 355 } 356 NOKPROBE_SYMBOL(vmalloc_fault); 357 358 /* 359 * Did it hit the DOS screen memory VA from vm86 mode? 360 */ 361 static inline void 362 check_v8086_mode(struct pt_regs *regs, unsigned long address, 363 struct task_struct *tsk) 364 { 365 #ifdef CONFIG_VM86 366 unsigned long bit; 367 368 if (!v8086_mode(regs) || !tsk->thread.vm86) 369 return; 370 371 bit = (address - 0xA0000) >> PAGE_SHIFT; 372 if (bit < 32) 373 tsk->thread.vm86->screen_bitmap |= 1 << bit; 374 #endif 375 } 376 377 static bool low_pfn(unsigned long pfn) 378 { 379 return pfn < max_low_pfn; 380 } 381 382 static void dump_pagetable(unsigned long address) 383 { 384 pgd_t *base = __va(read_cr3()); 385 pgd_t *pgd = &base[pgd_index(address)]; 386 pmd_t *pmd; 387 pte_t *pte; 388 389 #ifdef CONFIG_X86_PAE 390 printk("*pdpt = %016Lx ", pgd_val(*pgd)); 391 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) 392 goto out; 393 #endif 394 pmd = pmd_offset(pud_offset(pgd, address), address); 395 printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); 396 397 /* 398 * We must not directly access the pte in the highpte 399 * case if the page table is located in highmem. 400 * And let's rather not kmap-atomic the pte, just in case 401 * it's allocated already: 402 */ 403 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) 404 goto out; 405 406 pte = pte_offset_kernel(pmd, address); 407 printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); 408 out: 409 printk("\n"); 410 } 411 412 #else /* CONFIG_X86_64: */ 413 414 void vmalloc_sync_all(void) 415 { 416 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END, 0); 417 } 418 419 /* 420 * 64-bit: 421 * 422 * Handle a fault on the vmalloc area 423 */ 424 static noinline int vmalloc_fault(unsigned long address) 425 { 426 pgd_t *pgd, *pgd_ref; 427 pud_t *pud, *pud_ref; 428 pmd_t *pmd, *pmd_ref; 429 pte_t *pte, *pte_ref; 430 431 /* Make sure we are in vmalloc area: */ 432 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 433 return -1; 434 435 WARN_ON_ONCE(in_nmi()); 436 437 /* 438 * Copy kernel mappings over when needed. This can also 439 * happen within a race in page table update. In the later 440 * case just flush: 441 */ 442 pgd = (pgd_t *)__va(read_cr3()) + pgd_index(address); 443 pgd_ref = pgd_offset_k(address); 444 if (pgd_none(*pgd_ref)) 445 return -1; 446 447 if (pgd_none(*pgd)) { 448 set_pgd(pgd, *pgd_ref); 449 arch_flush_lazy_mmu_mode(); 450 } else { 451 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); 452 } 453 454 /* 455 * Below here mismatches are bugs because these lower tables 456 * are shared: 457 */ 458 459 pud = pud_offset(pgd, address); 460 pud_ref = pud_offset(pgd_ref, address); 461 if (pud_none(*pud_ref)) 462 return -1; 463 464 if (pud_none(*pud) || pud_pfn(*pud) != pud_pfn(*pud_ref)) 465 BUG(); 466 467 if (pud_huge(*pud)) 468 return 0; 469 470 pmd = pmd_offset(pud, address); 471 pmd_ref = pmd_offset(pud_ref, address); 472 if (pmd_none(*pmd_ref)) 473 return -1; 474 475 if (pmd_none(*pmd) || pmd_pfn(*pmd) != pmd_pfn(*pmd_ref)) 476 BUG(); 477 478 if (pmd_huge(*pmd)) 479 return 0; 480 481 pte_ref = pte_offset_kernel(pmd_ref, address); 482 if (!pte_present(*pte_ref)) 483 return -1; 484 485 pte = pte_offset_kernel(pmd, address); 486 487 /* 488 * Don't use pte_page here, because the mappings can point 489 * outside mem_map, and the NUMA hash lookup cannot handle 490 * that: 491 */ 492 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) 493 BUG(); 494 495 return 0; 496 } 497 NOKPROBE_SYMBOL(vmalloc_fault); 498 499 #ifdef CONFIG_CPU_SUP_AMD 500 static const char errata93_warning[] = 501 KERN_ERR 502 "******* Your BIOS seems to not contain a fix for K8 errata #93\n" 503 "******* Working around it, but it may cause SEGVs or burn power.\n" 504 "******* Please consider a BIOS update.\n" 505 "******* Disabling USB legacy in the BIOS may also help.\n"; 506 #endif 507 508 /* 509 * No vm86 mode in 64-bit mode: 510 */ 511 static inline void 512 check_v8086_mode(struct pt_regs *regs, unsigned long address, 513 struct task_struct *tsk) 514 { 515 } 516 517 static int bad_address(void *p) 518 { 519 unsigned long dummy; 520 521 return probe_kernel_address((unsigned long *)p, dummy); 522 } 523 524 static void dump_pagetable(unsigned long address) 525 { 526 pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK); 527 pgd_t *pgd = base + pgd_index(address); 528 pud_t *pud; 529 pmd_t *pmd; 530 pte_t *pte; 531 532 if (bad_address(pgd)) 533 goto bad; 534 535 printk("PGD %lx ", pgd_val(*pgd)); 536 537 if (!pgd_present(*pgd)) 538 goto out; 539 540 pud = pud_offset(pgd, address); 541 if (bad_address(pud)) 542 goto bad; 543 544 printk("PUD %lx ", pud_val(*pud)); 545 if (!pud_present(*pud) || pud_large(*pud)) 546 goto out; 547 548 pmd = pmd_offset(pud, address); 549 if (bad_address(pmd)) 550 goto bad; 551 552 printk("PMD %lx ", pmd_val(*pmd)); 553 if (!pmd_present(*pmd) || pmd_large(*pmd)) 554 goto out; 555 556 pte = pte_offset_kernel(pmd, address); 557 if (bad_address(pte)) 558 goto bad; 559 560 printk("PTE %lx", pte_val(*pte)); 561 out: 562 printk("\n"); 563 return; 564 bad: 565 printk("BAD\n"); 566 } 567 568 #endif /* CONFIG_X86_64 */ 569 570 /* 571 * Workaround for K8 erratum #93 & buggy BIOS. 572 * 573 * BIOS SMM functions are required to use a specific workaround 574 * to avoid corruption of the 64bit RIP register on C stepping K8. 575 * 576 * A lot of BIOS that didn't get tested properly miss this. 577 * 578 * The OS sees this as a page fault with the upper 32bits of RIP cleared. 579 * Try to work around it here. 580 * 581 * Note we only handle faults in kernel here. 582 * Does nothing on 32-bit. 583 */ 584 static int is_errata93(struct pt_regs *regs, unsigned long address) 585 { 586 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) 587 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD 588 || boot_cpu_data.x86 != 0xf) 589 return 0; 590 591 if (address != regs->ip) 592 return 0; 593 594 if ((address >> 32) != 0) 595 return 0; 596 597 address |= 0xffffffffUL << 32; 598 if ((address >= (u64)_stext && address <= (u64)_etext) || 599 (address >= MODULES_VADDR && address <= MODULES_END)) { 600 printk_once(errata93_warning); 601 regs->ip = address; 602 return 1; 603 } 604 #endif 605 return 0; 606 } 607 608 /* 609 * Work around K8 erratum #100 K8 in compat mode occasionally jumps 610 * to illegal addresses >4GB. 611 * 612 * We catch this in the page fault handler because these addresses 613 * are not reachable. Just detect this case and return. Any code 614 * segment in LDT is compatibility mode. 615 */ 616 static int is_errata100(struct pt_regs *regs, unsigned long address) 617 { 618 #ifdef CONFIG_X86_64 619 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) 620 return 1; 621 #endif 622 return 0; 623 } 624 625 static int is_f00f_bug(struct pt_regs *regs, unsigned long address) 626 { 627 #ifdef CONFIG_X86_F00F_BUG 628 unsigned long nr; 629 630 /* 631 * Pentium F0 0F C7 C8 bug workaround: 632 */ 633 if (boot_cpu_has_bug(X86_BUG_F00F)) { 634 nr = (address - idt_descr.address) >> 3; 635 636 if (nr == 6) { 637 do_invalid_op(regs, 0); 638 return 1; 639 } 640 } 641 #endif 642 return 0; 643 } 644 645 static const char nx_warning[] = KERN_CRIT 646 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; 647 static const char smep_warning[] = KERN_CRIT 648 "unable to execute userspace code (SMEP?) (uid: %d)\n"; 649 650 static void 651 show_fault_oops(struct pt_regs *regs, unsigned long error_code, 652 unsigned long address) 653 { 654 if (!oops_may_print()) 655 return; 656 657 if (error_code & PF_INSTR) { 658 unsigned int level; 659 pgd_t *pgd; 660 pte_t *pte; 661 662 pgd = __va(read_cr3() & PHYSICAL_PAGE_MASK); 663 pgd += pgd_index(address); 664 665 pte = lookup_address_in_pgd(pgd, address, &level); 666 667 if (pte && pte_present(*pte) && !pte_exec(*pte)) 668 printk(nx_warning, from_kuid(&init_user_ns, current_uid())); 669 if (pte && pte_present(*pte) && pte_exec(*pte) && 670 (pgd_flags(*pgd) & _PAGE_USER) && 671 (__read_cr4() & X86_CR4_SMEP)) 672 printk(smep_warning, from_kuid(&init_user_ns, current_uid())); 673 } 674 675 printk(KERN_ALERT "BUG: unable to handle kernel "); 676 if (address < PAGE_SIZE) 677 printk(KERN_CONT "NULL pointer dereference"); 678 else 679 printk(KERN_CONT "paging request"); 680 681 printk(KERN_CONT " at %p\n", (void *) address); 682 printk(KERN_ALERT "IP:"); 683 printk_address(regs->ip); 684 685 dump_pagetable(address); 686 } 687 688 static noinline void 689 pgtable_bad(struct pt_regs *regs, unsigned long error_code, 690 unsigned long address) 691 { 692 struct task_struct *tsk; 693 unsigned long flags; 694 int sig; 695 696 flags = oops_begin(); 697 tsk = current; 698 sig = SIGKILL; 699 700 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", 701 tsk->comm, address); 702 dump_pagetable(address); 703 704 tsk->thread.cr2 = address; 705 tsk->thread.trap_nr = X86_TRAP_PF; 706 tsk->thread.error_code = error_code; 707 708 if (__die("Bad pagetable", regs, error_code)) 709 sig = 0; 710 711 oops_end(flags, regs, sig); 712 } 713 714 static noinline void 715 no_context(struct pt_regs *regs, unsigned long error_code, 716 unsigned long address, int signal, int si_code) 717 { 718 struct task_struct *tsk = current; 719 unsigned long flags; 720 int sig; 721 /* No context means no VMA to pass down */ 722 struct vm_area_struct *vma = NULL; 723 724 /* Are we prepared to handle this kernel fault? */ 725 if (fixup_exception(regs, X86_TRAP_PF)) { 726 /* 727 * Any interrupt that takes a fault gets the fixup. This makes 728 * the below recursive fault logic only apply to a faults from 729 * task context. 730 */ 731 if (in_interrupt()) 732 return; 733 734 /* 735 * Per the above we're !in_interrupt(), aka. task context. 736 * 737 * In this case we need to make sure we're not recursively 738 * faulting through the emulate_vsyscall() logic. 739 */ 740 if (current->thread.sig_on_uaccess_err && signal) { 741 tsk->thread.trap_nr = X86_TRAP_PF; 742 tsk->thread.error_code = error_code | PF_USER; 743 tsk->thread.cr2 = address; 744 745 /* XXX: hwpoison faults will set the wrong code. */ 746 force_sig_info_fault(signal, si_code, address, 747 tsk, vma, 0); 748 } 749 750 /* 751 * Barring that, we can do the fixup and be happy. 752 */ 753 return; 754 } 755 756 /* 757 * 32-bit: 758 * 759 * Valid to do another page fault here, because if this fault 760 * had been triggered by is_prefetch fixup_exception would have 761 * handled it. 762 * 763 * 64-bit: 764 * 765 * Hall of shame of CPU/BIOS bugs. 766 */ 767 if (is_prefetch(regs, error_code, address)) 768 return; 769 770 if (is_errata93(regs, address)) 771 return; 772 773 /* 774 * Oops. The kernel tried to access some bad page. We'll have to 775 * terminate things with extreme prejudice: 776 */ 777 flags = oops_begin(); 778 779 show_fault_oops(regs, error_code, address); 780 781 if (task_stack_end_corrupted(tsk)) 782 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); 783 784 tsk->thread.cr2 = address; 785 tsk->thread.trap_nr = X86_TRAP_PF; 786 tsk->thread.error_code = error_code; 787 788 sig = SIGKILL; 789 if (__die("Oops", regs, error_code)) 790 sig = 0; 791 792 /* Executive summary in case the body of the oops scrolled away */ 793 printk(KERN_DEFAULT "CR2: %016lx\n", address); 794 795 oops_end(flags, regs, sig); 796 } 797 798 /* 799 * Print out info about fatal segfaults, if the show_unhandled_signals 800 * sysctl is set: 801 */ 802 static inline void 803 show_signal_msg(struct pt_regs *regs, unsigned long error_code, 804 unsigned long address, struct task_struct *tsk) 805 { 806 if (!unhandled_signal(tsk, SIGSEGV)) 807 return; 808 809 if (!printk_ratelimit()) 810 return; 811 812 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx", 813 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, 814 tsk->comm, task_pid_nr(tsk), address, 815 (void *)regs->ip, (void *)regs->sp, error_code); 816 817 print_vma_addr(KERN_CONT " in ", regs->ip); 818 819 printk(KERN_CONT "\n"); 820 } 821 822 static void 823 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 824 unsigned long address, struct vm_area_struct *vma, 825 int si_code) 826 { 827 struct task_struct *tsk = current; 828 829 /* User mode accesses just cause a SIGSEGV */ 830 if (error_code & PF_USER) { 831 /* 832 * It's possible to have interrupts off here: 833 */ 834 local_irq_enable(); 835 836 /* 837 * Valid to do another page fault here because this one came 838 * from user space: 839 */ 840 if (is_prefetch(regs, error_code, address)) 841 return; 842 843 if (is_errata100(regs, address)) 844 return; 845 846 #ifdef CONFIG_X86_64 847 /* 848 * Instruction fetch faults in the vsyscall page might need 849 * emulation. 850 */ 851 if (unlikely((error_code & PF_INSTR) && 852 ((address & ~0xfff) == VSYSCALL_ADDR))) { 853 if (emulate_vsyscall(regs, address)) 854 return; 855 } 856 #endif 857 858 /* 859 * To avoid leaking information about the kernel page table 860 * layout, pretend that user-mode accesses to kernel addresses 861 * are always protection faults. 862 */ 863 if (address >= TASK_SIZE_MAX) 864 error_code |= PF_PROT; 865 866 if (likely(show_unhandled_signals)) 867 show_signal_msg(regs, error_code, address, tsk); 868 869 tsk->thread.cr2 = address; 870 tsk->thread.error_code = error_code; 871 tsk->thread.trap_nr = X86_TRAP_PF; 872 873 force_sig_info_fault(SIGSEGV, si_code, address, tsk, vma, 0); 874 875 return; 876 } 877 878 if (is_f00f_bug(regs, address)) 879 return; 880 881 no_context(regs, error_code, address, SIGSEGV, si_code); 882 } 883 884 static noinline void 885 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 886 unsigned long address, struct vm_area_struct *vma) 887 { 888 __bad_area_nosemaphore(regs, error_code, address, vma, SEGV_MAPERR); 889 } 890 891 static void 892 __bad_area(struct pt_regs *regs, unsigned long error_code, 893 unsigned long address, struct vm_area_struct *vma, int si_code) 894 { 895 struct mm_struct *mm = current->mm; 896 897 /* 898 * Something tried to access memory that isn't in our memory map.. 899 * Fix it, but check if it's kernel or user first.. 900 */ 901 up_read(&mm->mmap_sem); 902 903 __bad_area_nosemaphore(regs, error_code, address, vma, si_code); 904 } 905 906 static noinline void 907 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) 908 { 909 __bad_area(regs, error_code, address, NULL, SEGV_MAPERR); 910 } 911 912 static inline bool bad_area_access_from_pkeys(unsigned long error_code, 913 struct vm_area_struct *vma) 914 { 915 /* This code is always called on the current mm */ 916 bool foreign = false; 917 918 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 919 return false; 920 if (error_code & PF_PK) 921 return true; 922 /* this checks permission keys on the VMA: */ 923 if (!arch_vma_access_permitted(vma, (error_code & PF_WRITE), 924 (error_code & PF_INSTR), foreign)) 925 return true; 926 return false; 927 } 928 929 static noinline void 930 bad_area_access_error(struct pt_regs *regs, unsigned long error_code, 931 unsigned long address, struct vm_area_struct *vma) 932 { 933 /* 934 * This OSPKE check is not strictly necessary at runtime. 935 * But, doing it this way allows compiler optimizations 936 * if pkeys are compiled out. 937 */ 938 if (bad_area_access_from_pkeys(error_code, vma)) 939 __bad_area(regs, error_code, address, vma, SEGV_PKUERR); 940 else 941 __bad_area(regs, error_code, address, vma, SEGV_ACCERR); 942 } 943 944 static void 945 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, 946 struct vm_area_struct *vma, unsigned int fault) 947 { 948 struct task_struct *tsk = current; 949 int code = BUS_ADRERR; 950 951 /* Kernel mode? Handle exceptions or die: */ 952 if (!(error_code & PF_USER)) { 953 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); 954 return; 955 } 956 957 /* User-space => ok to do another page fault: */ 958 if (is_prefetch(regs, error_code, address)) 959 return; 960 961 tsk->thread.cr2 = address; 962 tsk->thread.error_code = error_code; 963 tsk->thread.trap_nr = X86_TRAP_PF; 964 965 #ifdef CONFIG_MEMORY_FAILURE 966 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { 967 printk(KERN_ERR 968 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", 969 tsk->comm, tsk->pid, address); 970 code = BUS_MCEERR_AR; 971 } 972 #endif 973 force_sig_info_fault(SIGBUS, code, address, tsk, vma, fault); 974 } 975 976 static noinline void 977 mm_fault_error(struct pt_regs *regs, unsigned long error_code, 978 unsigned long address, struct vm_area_struct *vma, 979 unsigned int fault) 980 { 981 if (fatal_signal_pending(current) && !(error_code & PF_USER)) { 982 no_context(regs, error_code, address, 0, 0); 983 return; 984 } 985 986 if (fault & VM_FAULT_OOM) { 987 /* Kernel mode? Handle exceptions or die: */ 988 if (!(error_code & PF_USER)) { 989 no_context(regs, error_code, address, 990 SIGSEGV, SEGV_MAPERR); 991 return; 992 } 993 994 /* 995 * We ran out of memory, call the OOM killer, and return the 996 * userspace (which will retry the fault, or kill us if we got 997 * oom-killed): 998 */ 999 pagefault_out_of_memory(); 1000 } else { 1001 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| 1002 VM_FAULT_HWPOISON_LARGE)) 1003 do_sigbus(regs, error_code, address, vma, fault); 1004 else if (fault & VM_FAULT_SIGSEGV) 1005 bad_area_nosemaphore(regs, error_code, address, vma); 1006 else 1007 BUG(); 1008 } 1009 } 1010 1011 static int spurious_fault_check(unsigned long error_code, pte_t *pte) 1012 { 1013 if ((error_code & PF_WRITE) && !pte_write(*pte)) 1014 return 0; 1015 1016 if ((error_code & PF_INSTR) && !pte_exec(*pte)) 1017 return 0; 1018 /* 1019 * Note: We do not do lazy flushing on protection key 1020 * changes, so no spurious fault will ever set PF_PK. 1021 */ 1022 if ((error_code & PF_PK)) 1023 return 1; 1024 1025 return 1; 1026 } 1027 1028 /* 1029 * Handle a spurious fault caused by a stale TLB entry. 1030 * 1031 * This allows us to lazily refresh the TLB when increasing the 1032 * permissions of a kernel page (RO -> RW or NX -> X). Doing it 1033 * eagerly is very expensive since that implies doing a full 1034 * cross-processor TLB flush, even if no stale TLB entries exist 1035 * on other processors. 1036 * 1037 * Spurious faults may only occur if the TLB contains an entry with 1038 * fewer permission than the page table entry. Non-present (P = 0) 1039 * and reserved bit (R = 1) faults are never spurious. 1040 * 1041 * There are no security implications to leaving a stale TLB when 1042 * increasing the permissions on a page. 1043 * 1044 * Returns non-zero if a spurious fault was handled, zero otherwise. 1045 * 1046 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 1047 * (Optional Invalidation). 1048 */ 1049 static noinline int 1050 spurious_fault(unsigned long error_code, unsigned long address) 1051 { 1052 pgd_t *pgd; 1053 pud_t *pud; 1054 pmd_t *pmd; 1055 pte_t *pte; 1056 int ret; 1057 1058 /* 1059 * Only writes to RO or instruction fetches from NX may cause 1060 * spurious faults. 1061 * 1062 * These could be from user or supervisor accesses but the TLB 1063 * is only lazily flushed after a kernel mapping protection 1064 * change, so user accesses are not expected to cause spurious 1065 * faults. 1066 */ 1067 if (error_code != (PF_WRITE | PF_PROT) 1068 && error_code != (PF_INSTR | PF_PROT)) 1069 return 0; 1070 1071 pgd = init_mm.pgd + pgd_index(address); 1072 if (!pgd_present(*pgd)) 1073 return 0; 1074 1075 pud = pud_offset(pgd, address); 1076 if (!pud_present(*pud)) 1077 return 0; 1078 1079 if (pud_large(*pud)) 1080 return spurious_fault_check(error_code, (pte_t *) pud); 1081 1082 pmd = pmd_offset(pud, address); 1083 if (!pmd_present(*pmd)) 1084 return 0; 1085 1086 if (pmd_large(*pmd)) 1087 return spurious_fault_check(error_code, (pte_t *) pmd); 1088 1089 pte = pte_offset_kernel(pmd, address); 1090 if (!pte_present(*pte)) 1091 return 0; 1092 1093 ret = spurious_fault_check(error_code, pte); 1094 if (!ret) 1095 return 0; 1096 1097 /* 1098 * Make sure we have permissions in PMD. 1099 * If not, then there's a bug in the page tables: 1100 */ 1101 ret = spurious_fault_check(error_code, (pte_t *) pmd); 1102 WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); 1103 1104 return ret; 1105 } 1106 NOKPROBE_SYMBOL(spurious_fault); 1107 1108 int show_unhandled_signals = 1; 1109 1110 static inline int 1111 access_error(unsigned long error_code, struct vm_area_struct *vma) 1112 { 1113 /* This is only called for the current mm, so: */ 1114 bool foreign = false; 1115 /* 1116 * Make sure to check the VMA so that we do not perform 1117 * faults just to hit a PF_PK as soon as we fill in a 1118 * page. 1119 */ 1120 if (!arch_vma_access_permitted(vma, (error_code & PF_WRITE), 1121 (error_code & PF_INSTR), foreign)) 1122 return 1; 1123 1124 if (error_code & PF_WRITE) { 1125 /* write, present and write, not present: */ 1126 if (unlikely(!(vma->vm_flags & VM_WRITE))) 1127 return 1; 1128 return 0; 1129 } 1130 1131 /* read, present: */ 1132 if (unlikely(error_code & PF_PROT)) 1133 return 1; 1134 1135 /* read, not present: */ 1136 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) 1137 return 1; 1138 1139 return 0; 1140 } 1141 1142 static int fault_in_kernel_space(unsigned long address) 1143 { 1144 return address >= TASK_SIZE_MAX; 1145 } 1146 1147 static inline bool smap_violation(int error_code, struct pt_regs *regs) 1148 { 1149 if (!IS_ENABLED(CONFIG_X86_SMAP)) 1150 return false; 1151 1152 if (!static_cpu_has(X86_FEATURE_SMAP)) 1153 return false; 1154 1155 if (error_code & PF_USER) 1156 return false; 1157 1158 if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC)) 1159 return false; 1160 1161 return true; 1162 } 1163 1164 /* 1165 * This routine handles page faults. It determines the address, 1166 * and the problem, and then passes it off to one of the appropriate 1167 * routines. 1168 * 1169 * This function must have noinline because both callers 1170 * {,trace_}do_page_fault() have notrace on. Having this an actual function 1171 * guarantees there's a function trace entry. 1172 */ 1173 static noinline void 1174 __do_page_fault(struct pt_regs *regs, unsigned long error_code, 1175 unsigned long address) 1176 { 1177 struct vm_area_struct *vma; 1178 struct task_struct *tsk; 1179 struct mm_struct *mm; 1180 int fault, major = 0; 1181 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; 1182 1183 tsk = current; 1184 mm = tsk->mm; 1185 1186 /* 1187 * Detect and handle instructions that would cause a page fault for 1188 * both a tracked kernel page and a userspace page. 1189 */ 1190 if (kmemcheck_active(regs)) 1191 kmemcheck_hide(regs); 1192 prefetchw(&mm->mmap_sem); 1193 1194 if (unlikely(kmmio_fault(regs, address))) 1195 return; 1196 1197 /* 1198 * We fault-in kernel-space virtual memory on-demand. The 1199 * 'reference' page table is init_mm.pgd. 1200 * 1201 * NOTE! We MUST NOT take any locks for this case. We may 1202 * be in an interrupt or a critical region, and should 1203 * only copy the information from the master page table, 1204 * nothing more. 1205 * 1206 * This verifies that the fault happens in kernel space 1207 * (error_code & 4) == 0, and that the fault was not a 1208 * protection error (error_code & 9) == 0. 1209 */ 1210 if (unlikely(fault_in_kernel_space(address))) { 1211 if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) { 1212 if (vmalloc_fault(address) >= 0) 1213 return; 1214 1215 if (kmemcheck_fault(regs, address, error_code)) 1216 return; 1217 } 1218 1219 /* Can handle a stale RO->RW TLB: */ 1220 if (spurious_fault(error_code, address)) 1221 return; 1222 1223 /* kprobes don't want to hook the spurious faults: */ 1224 if (kprobes_fault(regs)) 1225 return; 1226 /* 1227 * Don't take the mm semaphore here. If we fixup a prefetch 1228 * fault we could otherwise deadlock: 1229 */ 1230 bad_area_nosemaphore(regs, error_code, address, NULL); 1231 1232 return; 1233 } 1234 1235 /* kprobes don't want to hook the spurious faults: */ 1236 if (unlikely(kprobes_fault(regs))) 1237 return; 1238 1239 if (unlikely(error_code & PF_RSVD)) 1240 pgtable_bad(regs, error_code, address); 1241 1242 if (unlikely(smap_violation(error_code, regs))) { 1243 bad_area_nosemaphore(regs, error_code, address, NULL); 1244 return; 1245 } 1246 1247 /* 1248 * If we're in an interrupt, have no user context or are running 1249 * in a region with pagefaults disabled then we must not take the fault 1250 */ 1251 if (unlikely(faulthandler_disabled() || !mm)) { 1252 bad_area_nosemaphore(regs, error_code, address, NULL); 1253 return; 1254 } 1255 1256 /* 1257 * It's safe to allow irq's after cr2 has been saved and the 1258 * vmalloc fault has been handled. 1259 * 1260 * User-mode registers count as a user access even for any 1261 * potential system fault or CPU buglet: 1262 */ 1263 if (user_mode(regs)) { 1264 local_irq_enable(); 1265 error_code |= PF_USER; 1266 flags |= FAULT_FLAG_USER; 1267 } else { 1268 if (regs->flags & X86_EFLAGS_IF) 1269 local_irq_enable(); 1270 } 1271 1272 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 1273 1274 if (error_code & PF_WRITE) 1275 flags |= FAULT_FLAG_WRITE; 1276 if (error_code & PF_INSTR) 1277 flags |= FAULT_FLAG_INSTRUCTION; 1278 1279 /* 1280 * When running in the kernel we expect faults to occur only to 1281 * addresses in user space. All other faults represent errors in 1282 * the kernel and should generate an OOPS. Unfortunately, in the 1283 * case of an erroneous fault occurring in a code path which already 1284 * holds mmap_sem we will deadlock attempting to validate the fault 1285 * against the address space. Luckily the kernel only validly 1286 * references user space from well defined areas of code, which are 1287 * listed in the exceptions table. 1288 * 1289 * As the vast majority of faults will be valid we will only perform 1290 * the source reference check when there is a possibility of a 1291 * deadlock. Attempt to lock the address space, if we cannot we then 1292 * validate the source. If this is invalid we can skip the address 1293 * space check, thus avoiding the deadlock: 1294 */ 1295 if (unlikely(!down_read_trylock(&mm->mmap_sem))) { 1296 if ((error_code & PF_USER) == 0 && 1297 !search_exception_tables(regs->ip)) { 1298 bad_area_nosemaphore(regs, error_code, address, NULL); 1299 return; 1300 } 1301 retry: 1302 down_read(&mm->mmap_sem); 1303 } else { 1304 /* 1305 * The above down_read_trylock() might have succeeded in 1306 * which case we'll have missed the might_sleep() from 1307 * down_read(): 1308 */ 1309 might_sleep(); 1310 } 1311 1312 vma = find_vma(mm, address); 1313 if (unlikely(!vma)) { 1314 bad_area(regs, error_code, address); 1315 return; 1316 } 1317 if (likely(vma->vm_start <= address)) 1318 goto good_area; 1319 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { 1320 bad_area(regs, error_code, address); 1321 return; 1322 } 1323 if (error_code & PF_USER) { 1324 /* 1325 * Accessing the stack below %sp is always a bug. 1326 * The large cushion allows instructions like enter 1327 * and pusha to work. ("enter $65535, $31" pushes 1328 * 32 pointers and then decrements %sp by 65535.) 1329 */ 1330 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { 1331 bad_area(regs, error_code, address); 1332 return; 1333 } 1334 } 1335 if (unlikely(expand_stack(vma, address))) { 1336 bad_area(regs, error_code, address); 1337 return; 1338 } 1339 1340 /* 1341 * Ok, we have a good vm_area for this memory access, so 1342 * we can handle it.. 1343 */ 1344 good_area: 1345 if (unlikely(access_error(error_code, vma))) { 1346 bad_area_access_error(regs, error_code, address, vma); 1347 return; 1348 } 1349 1350 /* 1351 * If for any reason at all we couldn't handle the fault, 1352 * make sure we exit gracefully rather than endlessly redo 1353 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if 1354 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked. 1355 */ 1356 fault = handle_mm_fault(vma, address, flags); 1357 major |= fault & VM_FAULT_MAJOR; 1358 1359 /* 1360 * If we need to retry the mmap_sem has already been released, 1361 * and if there is a fatal signal pending there is no guarantee 1362 * that we made any progress. Handle this case first. 1363 */ 1364 if (unlikely(fault & VM_FAULT_RETRY)) { 1365 /* Retry at most once */ 1366 if (flags & FAULT_FLAG_ALLOW_RETRY) { 1367 flags &= ~FAULT_FLAG_ALLOW_RETRY; 1368 flags |= FAULT_FLAG_TRIED; 1369 if (!fatal_signal_pending(tsk)) 1370 goto retry; 1371 } 1372 1373 /* User mode? Just return to handle the fatal exception */ 1374 if (flags & FAULT_FLAG_USER) 1375 return; 1376 1377 /* Not returning to user mode? Handle exceptions or die: */ 1378 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); 1379 return; 1380 } 1381 1382 up_read(&mm->mmap_sem); 1383 if (unlikely(fault & VM_FAULT_ERROR)) { 1384 mm_fault_error(regs, error_code, address, vma, fault); 1385 return; 1386 } 1387 1388 /* 1389 * Major/minor page fault accounting. If any of the events 1390 * returned VM_FAULT_MAJOR, we account it as a major fault. 1391 */ 1392 if (major) { 1393 tsk->maj_flt++; 1394 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); 1395 } else { 1396 tsk->min_flt++; 1397 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); 1398 } 1399 1400 check_v8086_mode(regs, address, tsk); 1401 } 1402 NOKPROBE_SYMBOL(__do_page_fault); 1403 1404 dotraplinkage void notrace 1405 do_page_fault(struct pt_regs *regs, unsigned long error_code) 1406 { 1407 unsigned long address = read_cr2(); /* Get the faulting address */ 1408 enum ctx_state prev_state; 1409 1410 /* 1411 * We must have this function tagged with __kprobes, notrace and call 1412 * read_cr2() before calling anything else. To avoid calling any kind 1413 * of tracing machinery before we've observed the CR2 value. 1414 * 1415 * exception_{enter,exit}() contain all sorts of tracepoints. 1416 */ 1417 1418 prev_state = exception_enter(); 1419 __do_page_fault(regs, error_code, address); 1420 exception_exit(prev_state); 1421 } 1422 NOKPROBE_SYMBOL(do_page_fault); 1423 1424 #ifdef CONFIG_TRACING 1425 static nokprobe_inline void 1426 trace_page_fault_entries(unsigned long address, struct pt_regs *regs, 1427 unsigned long error_code) 1428 { 1429 if (user_mode(regs)) 1430 trace_page_fault_user(address, regs, error_code); 1431 else 1432 trace_page_fault_kernel(address, regs, error_code); 1433 } 1434 1435 dotraplinkage void notrace 1436 trace_do_page_fault(struct pt_regs *regs, unsigned long error_code) 1437 { 1438 /* 1439 * The exception_enter and tracepoint processing could 1440 * trigger another page faults (user space callchain 1441 * reading) and destroy the original cr2 value, so read 1442 * the faulting address now. 1443 */ 1444 unsigned long address = read_cr2(); 1445 enum ctx_state prev_state; 1446 1447 prev_state = exception_enter(); 1448 trace_page_fault_entries(address, regs, error_code); 1449 __do_page_fault(regs, error_code, address); 1450 exception_exit(prev_state); 1451 } 1452 NOKPROBE_SYMBOL(trace_do_page_fault); 1453 #endif /* CONFIG_TRACING */ 1454