1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 1995 Linus Torvalds 4 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs. 5 * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar 6 */ 7 #include <linux/sched.h> /* test_thread_flag(), ... */ 8 #include <linux/sched/task_stack.h> /* task_stack_*(), ... */ 9 #include <linux/kdebug.h> /* oops_begin/end, ... */ 10 #include <linux/extable.h> /* search_exception_tables */ 11 #include <linux/memblock.h> /* max_low_pfn */ 12 #include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */ 13 #include <linux/mmiotrace.h> /* kmmio_handler, ... */ 14 #include <linux/perf_event.h> /* perf_sw_event */ 15 #include <linux/hugetlb.h> /* hstate_index_to_shift */ 16 #include <linux/prefetch.h> /* prefetchw */ 17 #include <linux/context_tracking.h> /* exception_enter(), ... */ 18 #include <linux/uaccess.h> /* faulthandler_disabled() */ 19 #include <linux/efi.h> /* efi_recover_from_page_fault()*/ 20 #include <linux/mm_types.h> 21 22 #include <asm/cpufeature.h> /* boot_cpu_has, ... */ 23 #include <asm/traps.h> /* dotraplinkage, ... */ 24 #include <asm/pgalloc.h> /* pgd_*(), ... */ 25 #include <asm/fixmap.h> /* VSYSCALL_ADDR */ 26 #include <asm/vsyscall.h> /* emulate_vsyscall */ 27 #include <asm/vm86.h> /* struct vm86 */ 28 #include <asm/mmu_context.h> /* vma_pkey() */ 29 #include <asm/efi.h> /* efi_recover_from_page_fault()*/ 30 #include <asm/desc.h> /* store_idt(), ... */ 31 #include <asm/cpu_entry_area.h> /* exception stack */ 32 #include <asm/pgtable_areas.h> /* VMALLOC_START, ... */ 33 #include <asm/kvm_para.h> /* kvm_handle_async_pf */ 34 35 #define CREATE_TRACE_POINTS 36 #include <asm/trace/exceptions.h> 37 38 /* 39 * Returns 0 if mmiotrace is disabled, or if the fault is not 40 * handled by mmiotrace: 41 */ 42 static nokprobe_inline int 43 kmmio_fault(struct pt_regs *regs, unsigned long addr) 44 { 45 if (unlikely(is_kmmio_active())) 46 if (kmmio_handler(regs, addr) == 1) 47 return -1; 48 return 0; 49 } 50 51 /* 52 * Prefetch quirks: 53 * 54 * 32-bit mode: 55 * 56 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. 57 * Check that here and ignore it. 58 * 59 * 64-bit mode: 60 * 61 * Sometimes the CPU reports invalid exceptions on prefetch. 62 * Check that here and ignore it. 63 * 64 * Opcode checker based on code by Richard Brunner. 65 */ 66 static inline int 67 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, 68 unsigned char opcode, int *prefetch) 69 { 70 unsigned char instr_hi = opcode & 0xf0; 71 unsigned char instr_lo = opcode & 0x0f; 72 73 switch (instr_hi) { 74 case 0x20: 75 case 0x30: 76 /* 77 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. 78 * In X86_64 long mode, the CPU will signal invalid 79 * opcode if some of these prefixes are present so 80 * X86_64 will never get here anyway 81 */ 82 return ((instr_lo & 7) == 0x6); 83 #ifdef CONFIG_X86_64 84 case 0x40: 85 /* 86 * In AMD64 long mode 0x40..0x4F are valid REX prefixes 87 * Need to figure out under what instruction mode the 88 * instruction was issued. Could check the LDT for lm, 89 * but for now it's good enough to assume that long 90 * mode only uses well known segments or kernel. 91 */ 92 return (!user_mode(regs) || user_64bit_mode(regs)); 93 #endif 94 case 0x60: 95 /* 0x64 thru 0x67 are valid prefixes in all modes. */ 96 return (instr_lo & 0xC) == 0x4; 97 case 0xF0: 98 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ 99 return !instr_lo || (instr_lo>>1) == 1; 100 case 0x00: 101 /* Prefetch instruction is 0x0F0D or 0x0F18 */ 102 if (probe_kernel_address(instr, opcode)) 103 return 0; 104 105 *prefetch = (instr_lo == 0xF) && 106 (opcode == 0x0D || opcode == 0x18); 107 return 0; 108 default: 109 return 0; 110 } 111 } 112 113 static int 114 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) 115 { 116 unsigned char *max_instr; 117 unsigned char *instr; 118 int prefetch = 0; 119 120 /* 121 * If it was a exec (instruction fetch) fault on NX page, then 122 * do not ignore the fault: 123 */ 124 if (error_code & X86_PF_INSTR) 125 return 0; 126 127 instr = (void *)convert_ip_to_linear(current, regs); 128 max_instr = instr + 15; 129 130 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX) 131 return 0; 132 133 while (instr < max_instr) { 134 unsigned char opcode; 135 136 if (probe_kernel_address(instr, opcode)) 137 break; 138 139 instr++; 140 141 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) 142 break; 143 } 144 return prefetch; 145 } 146 147 DEFINE_SPINLOCK(pgd_lock); 148 LIST_HEAD(pgd_list); 149 150 #ifdef CONFIG_X86_32 151 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) 152 { 153 unsigned index = pgd_index(address); 154 pgd_t *pgd_k; 155 p4d_t *p4d, *p4d_k; 156 pud_t *pud, *pud_k; 157 pmd_t *pmd, *pmd_k; 158 159 pgd += index; 160 pgd_k = init_mm.pgd + index; 161 162 if (!pgd_present(*pgd_k)) 163 return NULL; 164 165 /* 166 * set_pgd(pgd, *pgd_k); here would be useless on PAE 167 * and redundant with the set_pmd() on non-PAE. As would 168 * set_p4d/set_pud. 169 */ 170 p4d = p4d_offset(pgd, address); 171 p4d_k = p4d_offset(pgd_k, address); 172 if (!p4d_present(*p4d_k)) 173 return NULL; 174 175 pud = pud_offset(p4d, address); 176 pud_k = pud_offset(p4d_k, address); 177 if (!pud_present(*pud_k)) 178 return NULL; 179 180 pmd = pmd_offset(pud, address); 181 pmd_k = pmd_offset(pud_k, address); 182 183 if (pmd_present(*pmd) != pmd_present(*pmd_k)) 184 set_pmd(pmd, *pmd_k); 185 186 if (!pmd_present(*pmd_k)) 187 return NULL; 188 else 189 BUG_ON(pmd_pfn(*pmd) != pmd_pfn(*pmd_k)); 190 191 return pmd_k; 192 } 193 194 void arch_sync_kernel_mappings(unsigned long start, unsigned long end) 195 { 196 unsigned long addr; 197 198 for (addr = start & PMD_MASK; 199 addr >= TASK_SIZE_MAX && addr < VMALLOC_END; 200 addr += PMD_SIZE) { 201 struct page *page; 202 203 spin_lock(&pgd_lock); 204 list_for_each_entry(page, &pgd_list, lru) { 205 spinlock_t *pgt_lock; 206 207 /* the pgt_lock only for Xen */ 208 pgt_lock = &pgd_page_get_mm(page)->page_table_lock; 209 210 spin_lock(pgt_lock); 211 vmalloc_sync_one(page_address(page), addr); 212 spin_unlock(pgt_lock); 213 } 214 spin_unlock(&pgd_lock); 215 } 216 } 217 218 /* 219 * Did it hit the DOS screen memory VA from vm86 mode? 220 */ 221 static inline void 222 check_v8086_mode(struct pt_regs *regs, unsigned long address, 223 struct task_struct *tsk) 224 { 225 #ifdef CONFIG_VM86 226 unsigned long bit; 227 228 if (!v8086_mode(regs) || !tsk->thread.vm86) 229 return; 230 231 bit = (address - 0xA0000) >> PAGE_SHIFT; 232 if (bit < 32) 233 tsk->thread.vm86->screen_bitmap |= 1 << bit; 234 #endif 235 } 236 237 static bool low_pfn(unsigned long pfn) 238 { 239 return pfn < max_low_pfn; 240 } 241 242 static void dump_pagetable(unsigned long address) 243 { 244 pgd_t *base = __va(read_cr3_pa()); 245 pgd_t *pgd = &base[pgd_index(address)]; 246 p4d_t *p4d; 247 pud_t *pud; 248 pmd_t *pmd; 249 pte_t *pte; 250 251 #ifdef CONFIG_X86_PAE 252 pr_info("*pdpt = %016Lx ", pgd_val(*pgd)); 253 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) 254 goto out; 255 #define pr_pde pr_cont 256 #else 257 #define pr_pde pr_info 258 #endif 259 p4d = p4d_offset(pgd, address); 260 pud = pud_offset(p4d, address); 261 pmd = pmd_offset(pud, address); 262 pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); 263 #undef pr_pde 264 265 /* 266 * We must not directly access the pte in the highpte 267 * case if the page table is located in highmem. 268 * And let's rather not kmap-atomic the pte, just in case 269 * it's allocated already: 270 */ 271 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) 272 goto out; 273 274 pte = pte_offset_kernel(pmd, address); 275 pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); 276 out: 277 pr_cont("\n"); 278 } 279 280 #else /* CONFIG_X86_64: */ 281 282 #ifdef CONFIG_CPU_SUP_AMD 283 static const char errata93_warning[] = 284 KERN_ERR 285 "******* Your BIOS seems to not contain a fix for K8 errata #93\n" 286 "******* Working around it, but it may cause SEGVs or burn power.\n" 287 "******* Please consider a BIOS update.\n" 288 "******* Disabling USB legacy in the BIOS may also help.\n"; 289 #endif 290 291 /* 292 * No vm86 mode in 64-bit mode: 293 */ 294 static inline void 295 check_v8086_mode(struct pt_regs *regs, unsigned long address, 296 struct task_struct *tsk) 297 { 298 } 299 300 static int bad_address(void *p) 301 { 302 unsigned long dummy; 303 304 return probe_kernel_address((unsigned long *)p, dummy); 305 } 306 307 static void dump_pagetable(unsigned long address) 308 { 309 pgd_t *base = __va(read_cr3_pa()); 310 pgd_t *pgd = base + pgd_index(address); 311 p4d_t *p4d; 312 pud_t *pud; 313 pmd_t *pmd; 314 pte_t *pte; 315 316 if (bad_address(pgd)) 317 goto bad; 318 319 pr_info("PGD %lx ", pgd_val(*pgd)); 320 321 if (!pgd_present(*pgd)) 322 goto out; 323 324 p4d = p4d_offset(pgd, address); 325 if (bad_address(p4d)) 326 goto bad; 327 328 pr_cont("P4D %lx ", p4d_val(*p4d)); 329 if (!p4d_present(*p4d) || p4d_large(*p4d)) 330 goto out; 331 332 pud = pud_offset(p4d, address); 333 if (bad_address(pud)) 334 goto bad; 335 336 pr_cont("PUD %lx ", pud_val(*pud)); 337 if (!pud_present(*pud) || pud_large(*pud)) 338 goto out; 339 340 pmd = pmd_offset(pud, address); 341 if (bad_address(pmd)) 342 goto bad; 343 344 pr_cont("PMD %lx ", pmd_val(*pmd)); 345 if (!pmd_present(*pmd) || pmd_large(*pmd)) 346 goto out; 347 348 pte = pte_offset_kernel(pmd, address); 349 if (bad_address(pte)) 350 goto bad; 351 352 pr_cont("PTE %lx", pte_val(*pte)); 353 out: 354 pr_cont("\n"); 355 return; 356 bad: 357 pr_info("BAD\n"); 358 } 359 360 #endif /* CONFIG_X86_64 */ 361 362 /* 363 * Workaround for K8 erratum #93 & buggy BIOS. 364 * 365 * BIOS SMM functions are required to use a specific workaround 366 * to avoid corruption of the 64bit RIP register on C stepping K8. 367 * 368 * A lot of BIOS that didn't get tested properly miss this. 369 * 370 * The OS sees this as a page fault with the upper 32bits of RIP cleared. 371 * Try to work around it here. 372 * 373 * Note we only handle faults in kernel here. 374 * Does nothing on 32-bit. 375 */ 376 static int is_errata93(struct pt_regs *regs, unsigned long address) 377 { 378 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) 379 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD 380 || boot_cpu_data.x86 != 0xf) 381 return 0; 382 383 if (address != regs->ip) 384 return 0; 385 386 if ((address >> 32) != 0) 387 return 0; 388 389 address |= 0xffffffffUL << 32; 390 if ((address >= (u64)_stext && address <= (u64)_etext) || 391 (address >= MODULES_VADDR && address <= MODULES_END)) { 392 printk_once(errata93_warning); 393 regs->ip = address; 394 return 1; 395 } 396 #endif 397 return 0; 398 } 399 400 /* 401 * Work around K8 erratum #100 K8 in compat mode occasionally jumps 402 * to illegal addresses >4GB. 403 * 404 * We catch this in the page fault handler because these addresses 405 * are not reachable. Just detect this case and return. Any code 406 * segment in LDT is compatibility mode. 407 */ 408 static int is_errata100(struct pt_regs *regs, unsigned long address) 409 { 410 #ifdef CONFIG_X86_64 411 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) 412 return 1; 413 #endif 414 return 0; 415 } 416 417 static int is_f00f_bug(struct pt_regs *regs, unsigned long address) 418 { 419 #ifdef CONFIG_X86_F00F_BUG 420 unsigned long nr; 421 422 /* 423 * Pentium F0 0F C7 C8 bug workaround: 424 */ 425 if (boot_cpu_has_bug(X86_BUG_F00F)) { 426 nr = (address - idt_descr.address) >> 3; 427 428 if (nr == 6) { 429 do_invalid_op(regs, 0); 430 return 1; 431 } 432 } 433 #endif 434 return 0; 435 } 436 437 static void show_ldttss(const struct desc_ptr *gdt, const char *name, u16 index) 438 { 439 u32 offset = (index >> 3) * sizeof(struct desc_struct); 440 unsigned long addr; 441 struct ldttss_desc desc; 442 443 if (index == 0) { 444 pr_alert("%s: NULL\n", name); 445 return; 446 } 447 448 if (offset + sizeof(struct ldttss_desc) >= gdt->size) { 449 pr_alert("%s: 0x%hx -- out of bounds\n", name, index); 450 return; 451 } 452 453 if (probe_kernel_read(&desc, (void *)(gdt->address + offset), 454 sizeof(struct ldttss_desc))) { 455 pr_alert("%s: 0x%hx -- GDT entry is not readable\n", 456 name, index); 457 return; 458 } 459 460 addr = desc.base0 | (desc.base1 << 16) | ((unsigned long)desc.base2 << 24); 461 #ifdef CONFIG_X86_64 462 addr |= ((u64)desc.base3 << 32); 463 #endif 464 pr_alert("%s: 0x%hx -- base=0x%lx limit=0x%x\n", 465 name, index, addr, (desc.limit0 | (desc.limit1 << 16))); 466 } 467 468 static void 469 show_fault_oops(struct pt_regs *regs, unsigned long error_code, unsigned long address) 470 { 471 if (!oops_may_print()) 472 return; 473 474 if (error_code & X86_PF_INSTR) { 475 unsigned int level; 476 pgd_t *pgd; 477 pte_t *pte; 478 479 pgd = __va(read_cr3_pa()); 480 pgd += pgd_index(address); 481 482 pte = lookup_address_in_pgd(pgd, address, &level); 483 484 if (pte && pte_present(*pte) && !pte_exec(*pte)) 485 pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n", 486 from_kuid(&init_user_ns, current_uid())); 487 if (pte && pte_present(*pte) && pte_exec(*pte) && 488 (pgd_flags(*pgd) & _PAGE_USER) && 489 (__read_cr4() & X86_CR4_SMEP)) 490 pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n", 491 from_kuid(&init_user_ns, current_uid())); 492 } 493 494 if (address < PAGE_SIZE && !user_mode(regs)) 495 pr_alert("BUG: kernel NULL pointer dereference, address: %px\n", 496 (void *)address); 497 else 498 pr_alert("BUG: unable to handle page fault for address: %px\n", 499 (void *)address); 500 501 pr_alert("#PF: %s %s in %s mode\n", 502 (error_code & X86_PF_USER) ? "user" : "supervisor", 503 (error_code & X86_PF_INSTR) ? "instruction fetch" : 504 (error_code & X86_PF_WRITE) ? "write access" : 505 "read access", 506 user_mode(regs) ? "user" : "kernel"); 507 pr_alert("#PF: error_code(0x%04lx) - %s\n", error_code, 508 !(error_code & X86_PF_PROT) ? "not-present page" : 509 (error_code & X86_PF_RSVD) ? "reserved bit violation" : 510 (error_code & X86_PF_PK) ? "protection keys violation" : 511 "permissions violation"); 512 513 if (!(error_code & X86_PF_USER) && user_mode(regs)) { 514 struct desc_ptr idt, gdt; 515 u16 ldtr, tr; 516 517 /* 518 * This can happen for quite a few reasons. The more obvious 519 * ones are faults accessing the GDT, or LDT. Perhaps 520 * surprisingly, if the CPU tries to deliver a benign or 521 * contributory exception from user code and gets a page fault 522 * during delivery, the page fault can be delivered as though 523 * it originated directly from user code. This could happen 524 * due to wrong permissions on the IDT, GDT, LDT, TSS, or 525 * kernel or IST stack. 526 */ 527 store_idt(&idt); 528 529 /* Usable even on Xen PV -- it's just slow. */ 530 native_store_gdt(&gdt); 531 532 pr_alert("IDT: 0x%lx (limit=0x%hx) GDT: 0x%lx (limit=0x%hx)\n", 533 idt.address, idt.size, gdt.address, gdt.size); 534 535 store_ldt(ldtr); 536 show_ldttss(&gdt, "LDTR", ldtr); 537 538 store_tr(tr); 539 show_ldttss(&gdt, "TR", tr); 540 } 541 542 dump_pagetable(address); 543 } 544 545 static noinline void 546 pgtable_bad(struct pt_regs *regs, unsigned long error_code, 547 unsigned long address) 548 { 549 struct task_struct *tsk; 550 unsigned long flags; 551 int sig; 552 553 flags = oops_begin(); 554 tsk = current; 555 sig = SIGKILL; 556 557 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", 558 tsk->comm, address); 559 dump_pagetable(address); 560 561 if (__die("Bad pagetable", regs, error_code)) 562 sig = 0; 563 564 oops_end(flags, regs, sig); 565 } 566 567 static void set_signal_archinfo(unsigned long address, 568 unsigned long error_code) 569 { 570 struct task_struct *tsk = current; 571 572 /* 573 * To avoid leaking information about the kernel page 574 * table layout, pretend that user-mode accesses to 575 * kernel addresses are always protection faults. 576 * 577 * NB: This means that failed vsyscalls with vsyscall=none 578 * will have the PROT bit. This doesn't leak any 579 * information and does not appear to cause any problems. 580 */ 581 if (address >= TASK_SIZE_MAX) 582 error_code |= X86_PF_PROT; 583 584 tsk->thread.trap_nr = X86_TRAP_PF; 585 tsk->thread.error_code = error_code | X86_PF_USER; 586 tsk->thread.cr2 = address; 587 } 588 589 static noinline void 590 no_context(struct pt_regs *regs, unsigned long error_code, 591 unsigned long address, int signal, int si_code) 592 { 593 struct task_struct *tsk = current; 594 unsigned long flags; 595 int sig; 596 597 if (user_mode(regs)) { 598 /* 599 * This is an implicit supervisor-mode access from user 600 * mode. Bypass all the kernel-mode recovery code and just 601 * OOPS. 602 */ 603 goto oops; 604 } 605 606 /* Are we prepared to handle this kernel fault? */ 607 if (fixup_exception(regs, X86_TRAP_PF, error_code, address)) { 608 /* 609 * Any interrupt that takes a fault gets the fixup. This makes 610 * the below recursive fault logic only apply to a faults from 611 * task context. 612 */ 613 if (in_interrupt()) 614 return; 615 616 /* 617 * Per the above we're !in_interrupt(), aka. task context. 618 * 619 * In this case we need to make sure we're not recursively 620 * faulting through the emulate_vsyscall() logic. 621 */ 622 if (current->thread.sig_on_uaccess_err && signal) { 623 set_signal_archinfo(address, error_code); 624 625 /* XXX: hwpoison faults will set the wrong code. */ 626 force_sig_fault(signal, si_code, (void __user *)address); 627 } 628 629 /* 630 * Barring that, we can do the fixup and be happy. 631 */ 632 return; 633 } 634 635 #ifdef CONFIG_VMAP_STACK 636 /* 637 * Stack overflow? During boot, we can fault near the initial 638 * stack in the direct map, but that's not an overflow -- check 639 * that we're in vmalloc space to avoid this. 640 */ 641 if (is_vmalloc_addr((void *)address) && 642 (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) || 643 address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) { 644 unsigned long stack = __this_cpu_ist_top_va(DF) - sizeof(void *); 645 /* 646 * We're likely to be running with very little stack space 647 * left. It's plausible that we'd hit this condition but 648 * double-fault even before we get this far, in which case 649 * we're fine: the double-fault handler will deal with it. 650 * 651 * We don't want to make it all the way into the oops code 652 * and then double-fault, though, because we're likely to 653 * break the console driver and lose most of the stack dump. 654 */ 655 asm volatile ("movq %[stack], %%rsp\n\t" 656 "call handle_stack_overflow\n\t" 657 "1: jmp 1b" 658 : ASM_CALL_CONSTRAINT 659 : "D" ("kernel stack overflow (page fault)"), 660 "S" (regs), "d" (address), 661 [stack] "rm" (stack)); 662 unreachable(); 663 } 664 #endif 665 666 /* 667 * 32-bit: 668 * 669 * Valid to do another page fault here, because if this fault 670 * had been triggered by is_prefetch fixup_exception would have 671 * handled it. 672 * 673 * 64-bit: 674 * 675 * Hall of shame of CPU/BIOS bugs. 676 */ 677 if (is_prefetch(regs, error_code, address)) 678 return; 679 680 if (is_errata93(regs, address)) 681 return; 682 683 /* 684 * Buggy firmware could access regions which might page fault, try to 685 * recover from such faults. 686 */ 687 if (IS_ENABLED(CONFIG_EFI)) 688 efi_recover_from_page_fault(address); 689 690 oops: 691 /* 692 * Oops. The kernel tried to access some bad page. We'll have to 693 * terminate things with extreme prejudice: 694 */ 695 flags = oops_begin(); 696 697 show_fault_oops(regs, error_code, address); 698 699 if (task_stack_end_corrupted(tsk)) 700 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); 701 702 sig = SIGKILL; 703 if (__die("Oops", regs, error_code)) 704 sig = 0; 705 706 /* Executive summary in case the body of the oops scrolled away */ 707 printk(KERN_DEFAULT "CR2: %016lx\n", address); 708 709 oops_end(flags, regs, sig); 710 } 711 712 /* 713 * Print out info about fatal segfaults, if the show_unhandled_signals 714 * sysctl is set: 715 */ 716 static inline void 717 show_signal_msg(struct pt_regs *regs, unsigned long error_code, 718 unsigned long address, struct task_struct *tsk) 719 { 720 const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG; 721 722 if (!unhandled_signal(tsk, SIGSEGV)) 723 return; 724 725 if (!printk_ratelimit()) 726 return; 727 728 printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx", 729 loglvl, tsk->comm, task_pid_nr(tsk), address, 730 (void *)regs->ip, (void *)regs->sp, error_code); 731 732 print_vma_addr(KERN_CONT " in ", regs->ip); 733 734 printk(KERN_CONT "\n"); 735 736 show_opcodes(regs, loglvl); 737 } 738 739 /* 740 * The (legacy) vsyscall page is the long page in the kernel portion 741 * of the address space that has user-accessible permissions. 742 */ 743 static bool is_vsyscall_vaddr(unsigned long vaddr) 744 { 745 return unlikely((vaddr & PAGE_MASK) == VSYSCALL_ADDR); 746 } 747 748 static void 749 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 750 unsigned long address, u32 pkey, int si_code) 751 { 752 struct task_struct *tsk = current; 753 754 /* User mode accesses just cause a SIGSEGV */ 755 if (user_mode(regs) && (error_code & X86_PF_USER)) { 756 /* 757 * It's possible to have interrupts off here: 758 */ 759 local_irq_enable(); 760 761 /* 762 * Valid to do another page fault here because this one came 763 * from user space: 764 */ 765 if (is_prefetch(regs, error_code, address)) 766 return; 767 768 if (is_errata100(regs, address)) 769 return; 770 771 /* 772 * To avoid leaking information about the kernel page table 773 * layout, pretend that user-mode accesses to kernel addresses 774 * are always protection faults. 775 */ 776 if (address >= TASK_SIZE_MAX) 777 error_code |= X86_PF_PROT; 778 779 if (likely(show_unhandled_signals)) 780 show_signal_msg(regs, error_code, address, tsk); 781 782 set_signal_archinfo(address, error_code); 783 784 if (si_code == SEGV_PKUERR) 785 force_sig_pkuerr((void __user *)address, pkey); 786 787 force_sig_fault(SIGSEGV, si_code, (void __user *)address); 788 789 return; 790 } 791 792 if (is_f00f_bug(regs, address)) 793 return; 794 795 no_context(regs, error_code, address, SIGSEGV, si_code); 796 } 797 798 static noinline void 799 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 800 unsigned long address) 801 { 802 __bad_area_nosemaphore(regs, error_code, address, 0, SEGV_MAPERR); 803 } 804 805 static void 806 __bad_area(struct pt_regs *regs, unsigned long error_code, 807 unsigned long address, u32 pkey, int si_code) 808 { 809 struct mm_struct *mm = current->mm; 810 /* 811 * Something tried to access memory that isn't in our memory map.. 812 * Fix it, but check if it's kernel or user first.. 813 */ 814 mmap_read_unlock(mm); 815 816 __bad_area_nosemaphore(regs, error_code, address, pkey, si_code); 817 } 818 819 static noinline void 820 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) 821 { 822 __bad_area(regs, error_code, address, 0, SEGV_MAPERR); 823 } 824 825 static inline bool bad_area_access_from_pkeys(unsigned long error_code, 826 struct vm_area_struct *vma) 827 { 828 /* This code is always called on the current mm */ 829 bool foreign = false; 830 831 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 832 return false; 833 if (error_code & X86_PF_PK) 834 return true; 835 /* this checks permission keys on the VMA: */ 836 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), 837 (error_code & X86_PF_INSTR), foreign)) 838 return true; 839 return false; 840 } 841 842 static noinline void 843 bad_area_access_error(struct pt_regs *regs, unsigned long error_code, 844 unsigned long address, struct vm_area_struct *vma) 845 { 846 /* 847 * This OSPKE check is not strictly necessary at runtime. 848 * But, doing it this way allows compiler optimizations 849 * if pkeys are compiled out. 850 */ 851 if (bad_area_access_from_pkeys(error_code, vma)) { 852 /* 853 * A protection key fault means that the PKRU value did not allow 854 * access to some PTE. Userspace can figure out what PKRU was 855 * from the XSAVE state. This function captures the pkey from 856 * the vma and passes it to userspace so userspace can discover 857 * which protection key was set on the PTE. 858 * 859 * If we get here, we know that the hardware signaled a X86_PF_PK 860 * fault and that there was a VMA once we got in the fault 861 * handler. It does *not* guarantee that the VMA we find here 862 * was the one that we faulted on. 863 * 864 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4); 865 * 2. T1 : set PKRU to deny access to pkey=4, touches page 866 * 3. T1 : faults... 867 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5); 868 * 5. T1 : enters fault handler, takes mmap_lock, etc... 869 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really 870 * faulted on a pte with its pkey=4. 871 */ 872 u32 pkey = vma_pkey(vma); 873 874 __bad_area(regs, error_code, address, pkey, SEGV_PKUERR); 875 } else { 876 __bad_area(regs, error_code, address, 0, SEGV_ACCERR); 877 } 878 } 879 880 static void 881 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, 882 vm_fault_t fault) 883 { 884 /* Kernel mode? Handle exceptions or die: */ 885 if (!(error_code & X86_PF_USER)) { 886 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); 887 return; 888 } 889 890 /* User-space => ok to do another page fault: */ 891 if (is_prefetch(regs, error_code, address)) 892 return; 893 894 set_signal_archinfo(address, error_code); 895 896 #ifdef CONFIG_MEMORY_FAILURE 897 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { 898 struct task_struct *tsk = current; 899 unsigned lsb = 0; 900 901 pr_err( 902 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", 903 tsk->comm, tsk->pid, address); 904 if (fault & VM_FAULT_HWPOISON_LARGE) 905 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); 906 if (fault & VM_FAULT_HWPOISON) 907 lsb = PAGE_SHIFT; 908 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb); 909 return; 910 } 911 #endif 912 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address); 913 } 914 915 static noinline void 916 mm_fault_error(struct pt_regs *regs, unsigned long error_code, 917 unsigned long address, vm_fault_t fault) 918 { 919 if (fatal_signal_pending(current) && !(error_code & X86_PF_USER)) { 920 no_context(regs, error_code, address, 0, 0); 921 return; 922 } 923 924 if (fault & VM_FAULT_OOM) { 925 /* Kernel mode? Handle exceptions or die: */ 926 if (!(error_code & X86_PF_USER)) { 927 no_context(regs, error_code, address, 928 SIGSEGV, SEGV_MAPERR); 929 return; 930 } 931 932 /* 933 * We ran out of memory, call the OOM killer, and return the 934 * userspace (which will retry the fault, or kill us if we got 935 * oom-killed): 936 */ 937 pagefault_out_of_memory(); 938 } else { 939 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| 940 VM_FAULT_HWPOISON_LARGE)) 941 do_sigbus(regs, error_code, address, fault); 942 else if (fault & VM_FAULT_SIGSEGV) 943 bad_area_nosemaphore(regs, error_code, address); 944 else 945 BUG(); 946 } 947 } 948 949 static int spurious_kernel_fault_check(unsigned long error_code, pte_t *pte) 950 { 951 if ((error_code & X86_PF_WRITE) && !pte_write(*pte)) 952 return 0; 953 954 if ((error_code & X86_PF_INSTR) && !pte_exec(*pte)) 955 return 0; 956 957 return 1; 958 } 959 960 /* 961 * Handle a spurious fault caused by a stale TLB entry. 962 * 963 * This allows us to lazily refresh the TLB when increasing the 964 * permissions of a kernel page (RO -> RW or NX -> X). Doing it 965 * eagerly is very expensive since that implies doing a full 966 * cross-processor TLB flush, even if no stale TLB entries exist 967 * on other processors. 968 * 969 * Spurious faults may only occur if the TLB contains an entry with 970 * fewer permission than the page table entry. Non-present (P = 0) 971 * and reserved bit (R = 1) faults are never spurious. 972 * 973 * There are no security implications to leaving a stale TLB when 974 * increasing the permissions on a page. 975 * 976 * Returns non-zero if a spurious fault was handled, zero otherwise. 977 * 978 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 979 * (Optional Invalidation). 980 */ 981 static noinline int 982 spurious_kernel_fault(unsigned long error_code, unsigned long address) 983 { 984 pgd_t *pgd; 985 p4d_t *p4d; 986 pud_t *pud; 987 pmd_t *pmd; 988 pte_t *pte; 989 int ret; 990 991 /* 992 * Only writes to RO or instruction fetches from NX may cause 993 * spurious faults. 994 * 995 * These could be from user or supervisor accesses but the TLB 996 * is only lazily flushed after a kernel mapping protection 997 * change, so user accesses are not expected to cause spurious 998 * faults. 999 */ 1000 if (error_code != (X86_PF_WRITE | X86_PF_PROT) && 1001 error_code != (X86_PF_INSTR | X86_PF_PROT)) 1002 return 0; 1003 1004 pgd = init_mm.pgd + pgd_index(address); 1005 if (!pgd_present(*pgd)) 1006 return 0; 1007 1008 p4d = p4d_offset(pgd, address); 1009 if (!p4d_present(*p4d)) 1010 return 0; 1011 1012 if (p4d_large(*p4d)) 1013 return spurious_kernel_fault_check(error_code, (pte_t *) p4d); 1014 1015 pud = pud_offset(p4d, address); 1016 if (!pud_present(*pud)) 1017 return 0; 1018 1019 if (pud_large(*pud)) 1020 return spurious_kernel_fault_check(error_code, (pte_t *) pud); 1021 1022 pmd = pmd_offset(pud, address); 1023 if (!pmd_present(*pmd)) 1024 return 0; 1025 1026 if (pmd_large(*pmd)) 1027 return spurious_kernel_fault_check(error_code, (pte_t *) pmd); 1028 1029 pte = pte_offset_kernel(pmd, address); 1030 if (!pte_present(*pte)) 1031 return 0; 1032 1033 ret = spurious_kernel_fault_check(error_code, pte); 1034 if (!ret) 1035 return 0; 1036 1037 /* 1038 * Make sure we have permissions in PMD. 1039 * If not, then there's a bug in the page tables: 1040 */ 1041 ret = spurious_kernel_fault_check(error_code, (pte_t *) pmd); 1042 WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); 1043 1044 return ret; 1045 } 1046 NOKPROBE_SYMBOL(spurious_kernel_fault); 1047 1048 int show_unhandled_signals = 1; 1049 1050 static inline int 1051 access_error(unsigned long error_code, struct vm_area_struct *vma) 1052 { 1053 /* This is only called for the current mm, so: */ 1054 bool foreign = false; 1055 1056 /* 1057 * Read or write was blocked by protection keys. This is 1058 * always an unconditional error and can never result in 1059 * a follow-up action to resolve the fault, like a COW. 1060 */ 1061 if (error_code & X86_PF_PK) 1062 return 1; 1063 1064 /* 1065 * Make sure to check the VMA so that we do not perform 1066 * faults just to hit a X86_PF_PK as soon as we fill in a 1067 * page. 1068 */ 1069 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), 1070 (error_code & X86_PF_INSTR), foreign)) 1071 return 1; 1072 1073 if (error_code & X86_PF_WRITE) { 1074 /* write, present and write, not present: */ 1075 if (unlikely(!(vma->vm_flags & VM_WRITE))) 1076 return 1; 1077 return 0; 1078 } 1079 1080 /* read, present: */ 1081 if (unlikely(error_code & X86_PF_PROT)) 1082 return 1; 1083 1084 /* read, not present: */ 1085 if (unlikely(!vma_is_accessible(vma))) 1086 return 1; 1087 1088 return 0; 1089 } 1090 1091 static int fault_in_kernel_space(unsigned long address) 1092 { 1093 /* 1094 * On 64-bit systems, the vsyscall page is at an address above 1095 * TASK_SIZE_MAX, but is not considered part of the kernel 1096 * address space. 1097 */ 1098 if (IS_ENABLED(CONFIG_X86_64) && is_vsyscall_vaddr(address)) 1099 return false; 1100 1101 return address >= TASK_SIZE_MAX; 1102 } 1103 1104 /* 1105 * Called for all faults where 'address' is part of the kernel address 1106 * space. Might get called for faults that originate from *code* that 1107 * ran in userspace or the kernel. 1108 */ 1109 static void 1110 do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code, 1111 unsigned long address) 1112 { 1113 /* 1114 * Protection keys exceptions only happen on user pages. We 1115 * have no user pages in the kernel portion of the address 1116 * space, so do not expect them here. 1117 */ 1118 WARN_ON_ONCE(hw_error_code & X86_PF_PK); 1119 1120 /* Was the fault spurious, caused by lazy TLB invalidation? */ 1121 if (spurious_kernel_fault(hw_error_code, address)) 1122 return; 1123 1124 /* kprobes don't want to hook the spurious faults: */ 1125 if (kprobe_page_fault(regs, X86_TRAP_PF)) 1126 return; 1127 1128 /* 1129 * Note, despite being a "bad area", there are quite a few 1130 * acceptable reasons to get here, such as erratum fixups 1131 * and handling kernel code that can fault, like get_user(). 1132 * 1133 * Don't take the mm semaphore here. If we fixup a prefetch 1134 * fault we could otherwise deadlock: 1135 */ 1136 bad_area_nosemaphore(regs, hw_error_code, address); 1137 } 1138 NOKPROBE_SYMBOL(do_kern_addr_fault); 1139 1140 /* Handle faults in the user portion of the address space */ 1141 static inline 1142 void do_user_addr_fault(struct pt_regs *regs, 1143 unsigned long hw_error_code, 1144 unsigned long address) 1145 { 1146 struct vm_area_struct *vma; 1147 struct task_struct *tsk; 1148 struct mm_struct *mm; 1149 vm_fault_t fault, major = 0; 1150 unsigned int flags = FAULT_FLAG_DEFAULT; 1151 1152 tsk = current; 1153 mm = tsk->mm; 1154 1155 /* kprobes don't want to hook the spurious faults: */ 1156 if (unlikely(kprobe_page_fault(regs, X86_TRAP_PF))) 1157 return; 1158 1159 /* 1160 * Reserved bits are never expected to be set on 1161 * entries in the user portion of the page tables. 1162 */ 1163 if (unlikely(hw_error_code & X86_PF_RSVD)) 1164 pgtable_bad(regs, hw_error_code, address); 1165 1166 /* 1167 * If SMAP is on, check for invalid kernel (supervisor) access to user 1168 * pages in the user address space. The odd case here is WRUSS, 1169 * which, according to the preliminary documentation, does not respect 1170 * SMAP and will have the USER bit set so, in all cases, SMAP 1171 * enforcement appears to be consistent with the USER bit. 1172 */ 1173 if (unlikely(cpu_feature_enabled(X86_FEATURE_SMAP) && 1174 !(hw_error_code & X86_PF_USER) && 1175 !(regs->flags & X86_EFLAGS_AC))) 1176 { 1177 bad_area_nosemaphore(regs, hw_error_code, address); 1178 return; 1179 } 1180 1181 /* 1182 * If we're in an interrupt, have no user context or are running 1183 * in a region with pagefaults disabled then we must not take the fault 1184 */ 1185 if (unlikely(faulthandler_disabled() || !mm)) { 1186 bad_area_nosemaphore(regs, hw_error_code, address); 1187 return; 1188 } 1189 1190 /* 1191 * It's safe to allow irq's after cr2 has been saved and the 1192 * vmalloc fault has been handled. 1193 * 1194 * User-mode registers count as a user access even for any 1195 * potential system fault or CPU buglet: 1196 */ 1197 if (user_mode(regs)) { 1198 local_irq_enable(); 1199 flags |= FAULT_FLAG_USER; 1200 } else { 1201 if (regs->flags & X86_EFLAGS_IF) 1202 local_irq_enable(); 1203 } 1204 1205 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 1206 1207 if (hw_error_code & X86_PF_WRITE) 1208 flags |= FAULT_FLAG_WRITE; 1209 if (hw_error_code & X86_PF_INSTR) 1210 flags |= FAULT_FLAG_INSTRUCTION; 1211 1212 #ifdef CONFIG_X86_64 1213 /* 1214 * Faults in the vsyscall page might need emulation. The 1215 * vsyscall page is at a high address (>PAGE_OFFSET), but is 1216 * considered to be part of the user address space. 1217 * 1218 * The vsyscall page does not have a "real" VMA, so do this 1219 * emulation before we go searching for VMAs. 1220 * 1221 * PKRU never rejects instruction fetches, so we don't need 1222 * to consider the PF_PK bit. 1223 */ 1224 if (is_vsyscall_vaddr(address)) { 1225 if (emulate_vsyscall(hw_error_code, regs, address)) 1226 return; 1227 } 1228 #endif 1229 1230 /* 1231 * Kernel-mode access to the user address space should only occur 1232 * on well-defined single instructions listed in the exception 1233 * tables. But, an erroneous kernel fault occurring outside one of 1234 * those areas which also holds mmap_lock might deadlock attempting 1235 * to validate the fault against the address space. 1236 * 1237 * Only do the expensive exception table search when we might be at 1238 * risk of a deadlock. This happens if we 1239 * 1. Failed to acquire mmap_lock, and 1240 * 2. The access did not originate in userspace. 1241 */ 1242 if (unlikely(!mmap_read_trylock(mm))) { 1243 if (!user_mode(regs) && !search_exception_tables(regs->ip)) { 1244 /* 1245 * Fault from code in kernel from 1246 * which we do not expect faults. 1247 */ 1248 bad_area_nosemaphore(regs, hw_error_code, address); 1249 return; 1250 } 1251 retry: 1252 mmap_read_lock(mm); 1253 } else { 1254 /* 1255 * The above down_read_trylock() might have succeeded in 1256 * which case we'll have missed the might_sleep() from 1257 * down_read(): 1258 */ 1259 might_sleep(); 1260 } 1261 1262 vma = find_vma(mm, address); 1263 if (unlikely(!vma)) { 1264 bad_area(regs, hw_error_code, address); 1265 return; 1266 } 1267 if (likely(vma->vm_start <= address)) 1268 goto good_area; 1269 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { 1270 bad_area(regs, hw_error_code, address); 1271 return; 1272 } 1273 if (unlikely(expand_stack(vma, address))) { 1274 bad_area(regs, hw_error_code, address); 1275 return; 1276 } 1277 1278 /* 1279 * Ok, we have a good vm_area for this memory access, so 1280 * we can handle it.. 1281 */ 1282 good_area: 1283 if (unlikely(access_error(hw_error_code, vma))) { 1284 bad_area_access_error(regs, hw_error_code, address, vma); 1285 return; 1286 } 1287 1288 /* 1289 * If for any reason at all we couldn't handle the fault, 1290 * make sure we exit gracefully rather than endlessly redo 1291 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if 1292 * we get VM_FAULT_RETRY back, the mmap_lock has been unlocked. 1293 * 1294 * Note that handle_userfault() may also release and reacquire mmap_lock 1295 * (and not return with VM_FAULT_RETRY), when returning to userland to 1296 * repeat the page fault later with a VM_FAULT_NOPAGE retval 1297 * (potentially after handling any pending signal during the return to 1298 * userland). The return to userland is identified whenever 1299 * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags. 1300 */ 1301 fault = handle_mm_fault(vma, address, flags); 1302 major |= fault & VM_FAULT_MAJOR; 1303 1304 /* Quick path to respond to signals */ 1305 if (fault_signal_pending(fault, regs)) { 1306 if (!user_mode(regs)) 1307 no_context(regs, hw_error_code, address, SIGBUS, 1308 BUS_ADRERR); 1309 return; 1310 } 1311 1312 /* 1313 * If we need to retry the mmap_lock has already been released, 1314 * and if there is a fatal signal pending there is no guarantee 1315 * that we made any progress. Handle this case first. 1316 */ 1317 if (unlikely((fault & VM_FAULT_RETRY) && 1318 (flags & FAULT_FLAG_ALLOW_RETRY))) { 1319 flags |= FAULT_FLAG_TRIED; 1320 goto retry; 1321 } 1322 1323 mmap_read_unlock(mm); 1324 if (unlikely(fault & VM_FAULT_ERROR)) { 1325 mm_fault_error(regs, hw_error_code, address, fault); 1326 return; 1327 } 1328 1329 /* 1330 * Major/minor page fault accounting. If any of the events 1331 * returned VM_FAULT_MAJOR, we account it as a major fault. 1332 */ 1333 if (major) { 1334 tsk->maj_flt++; 1335 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); 1336 } else { 1337 tsk->min_flt++; 1338 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); 1339 } 1340 1341 check_v8086_mode(regs, address, tsk); 1342 } 1343 NOKPROBE_SYMBOL(do_user_addr_fault); 1344 1345 static __always_inline void 1346 trace_page_fault_entries(struct pt_regs *regs, unsigned long error_code, 1347 unsigned long address) 1348 { 1349 if (!trace_pagefault_enabled()) 1350 return; 1351 1352 if (user_mode(regs)) 1353 trace_page_fault_user(address, regs, error_code); 1354 else 1355 trace_page_fault_kernel(address, regs, error_code); 1356 } 1357 1358 dotraplinkage void 1359 do_page_fault(struct pt_regs *regs, unsigned long hw_error_code, 1360 unsigned long address) 1361 { 1362 prefetchw(¤t->mm->mmap_lock); 1363 /* 1364 * KVM has two types of events that are, logically, interrupts, but 1365 * are unfortunately delivered using the #PF vector. These events are 1366 * "you just accessed valid memory, but the host doesn't have it right 1367 * now, so I'll put you to sleep if you continue" and "that memory 1368 * you tried to access earlier is available now." 1369 * 1370 * We are relying on the interrupted context being sane (valid RSP, 1371 * relevant locks not held, etc.), which is fine as long as the 1372 * interrupted context had IF=1. We are also relying on the KVM 1373 * async pf type field and CR2 being read consistently instead of 1374 * getting values from real and async page faults mixed up. 1375 * 1376 * Fingers crossed. 1377 */ 1378 if (kvm_handle_async_pf(regs, (u32)address)) 1379 return; 1380 1381 trace_page_fault_entries(regs, hw_error_code, address); 1382 1383 if (unlikely(kmmio_fault(regs, address))) 1384 return; 1385 1386 /* Was the fault on kernel-controlled part of the address space? */ 1387 if (unlikely(fault_in_kernel_space(address))) 1388 do_kern_addr_fault(regs, hw_error_code, address); 1389 else 1390 do_user_addr_fault(regs, hw_error_code, address); 1391 } 1392 NOKPROBE_SYMBOL(do_page_fault); 1393