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 31 #define CREATE_TRACE_POINTS 32 #include <asm/trace/exceptions.h> 33 34 /* 35 * Returns 0 if mmiotrace is disabled, or if the fault is not 36 * handled by mmiotrace: 37 */ 38 static nokprobe_inline int 39 kmmio_fault(struct pt_regs *regs, unsigned long addr) 40 { 41 if (unlikely(is_kmmio_active())) 42 if (kmmio_handler(regs, addr) == 1) 43 return -1; 44 return 0; 45 } 46 47 static nokprobe_inline int kprobes_fault(struct pt_regs *regs) 48 { 49 if (!kprobes_built_in()) 50 return 0; 51 if (user_mode(regs)) 52 return 0; 53 /* 54 * To be potentially processing a kprobe fault and to be allowed to call 55 * kprobe_running(), we have to be non-preemptible. 56 */ 57 if (preemptible()) 58 return 0; 59 if (!kprobe_running()) 60 return 0; 61 return kprobe_fault_handler(regs, X86_TRAP_PF); 62 } 63 64 /* 65 * Prefetch quirks: 66 * 67 * 32-bit mode: 68 * 69 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. 70 * Check that here and ignore it. 71 * 72 * 64-bit mode: 73 * 74 * Sometimes the CPU reports invalid exceptions on prefetch. 75 * Check that here and ignore it. 76 * 77 * Opcode checker based on code by Richard Brunner. 78 */ 79 static inline int 80 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, 81 unsigned char opcode, int *prefetch) 82 { 83 unsigned char instr_hi = opcode & 0xf0; 84 unsigned char instr_lo = opcode & 0x0f; 85 86 switch (instr_hi) { 87 case 0x20: 88 case 0x30: 89 /* 90 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. 91 * In X86_64 long mode, the CPU will signal invalid 92 * opcode if some of these prefixes are present so 93 * X86_64 will never get here anyway 94 */ 95 return ((instr_lo & 7) == 0x6); 96 #ifdef CONFIG_X86_64 97 case 0x40: 98 /* 99 * In AMD64 long mode 0x40..0x4F are valid REX prefixes 100 * Need to figure out under what instruction mode the 101 * instruction was issued. Could check the LDT for lm, 102 * but for now it's good enough to assume that long 103 * mode only uses well known segments or kernel. 104 */ 105 return (!user_mode(regs) || user_64bit_mode(regs)); 106 #endif 107 case 0x60: 108 /* 0x64 thru 0x67 are valid prefixes in all modes. */ 109 return (instr_lo & 0xC) == 0x4; 110 case 0xF0: 111 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ 112 return !instr_lo || (instr_lo>>1) == 1; 113 case 0x00: 114 /* Prefetch instruction is 0x0F0D or 0x0F18 */ 115 if (probe_kernel_address(instr, opcode)) 116 return 0; 117 118 *prefetch = (instr_lo == 0xF) && 119 (opcode == 0x0D || opcode == 0x18); 120 return 0; 121 default: 122 return 0; 123 } 124 } 125 126 static int 127 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) 128 { 129 unsigned char *max_instr; 130 unsigned char *instr; 131 int prefetch = 0; 132 133 /* 134 * If it was a exec (instruction fetch) fault on NX page, then 135 * do not ignore the fault: 136 */ 137 if (error_code & X86_PF_INSTR) 138 return 0; 139 140 instr = (void *)convert_ip_to_linear(current, regs); 141 max_instr = instr + 15; 142 143 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX) 144 return 0; 145 146 while (instr < max_instr) { 147 unsigned char opcode; 148 149 if (probe_kernel_address(instr, opcode)) 150 break; 151 152 instr++; 153 154 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) 155 break; 156 } 157 return prefetch; 158 } 159 160 DEFINE_SPINLOCK(pgd_lock); 161 LIST_HEAD(pgd_list); 162 163 #ifdef CONFIG_X86_32 164 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) 165 { 166 unsigned index = pgd_index(address); 167 pgd_t *pgd_k; 168 p4d_t *p4d, *p4d_k; 169 pud_t *pud, *pud_k; 170 pmd_t *pmd, *pmd_k; 171 172 pgd += index; 173 pgd_k = init_mm.pgd + index; 174 175 if (!pgd_present(*pgd_k)) 176 return NULL; 177 178 /* 179 * set_pgd(pgd, *pgd_k); here would be useless on PAE 180 * and redundant with the set_pmd() on non-PAE. As would 181 * set_p4d/set_pud. 182 */ 183 p4d = p4d_offset(pgd, address); 184 p4d_k = p4d_offset(pgd_k, address); 185 if (!p4d_present(*p4d_k)) 186 return NULL; 187 188 pud = pud_offset(p4d, address); 189 pud_k = pud_offset(p4d_k, address); 190 if (!pud_present(*pud_k)) 191 return NULL; 192 193 pmd = pmd_offset(pud, address); 194 pmd_k = pmd_offset(pud_k, address); 195 if (!pmd_present(*pmd_k)) 196 return NULL; 197 198 if (!pmd_present(*pmd)) 199 set_pmd(pmd, *pmd_k); 200 else 201 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); 202 203 return pmd_k; 204 } 205 206 void vmalloc_sync_all(void) 207 { 208 unsigned long address; 209 210 if (SHARED_KERNEL_PMD) 211 return; 212 213 for (address = VMALLOC_START & PMD_MASK; 214 address >= TASK_SIZE_MAX && address < FIXADDR_TOP; 215 address += PMD_SIZE) { 216 struct page *page; 217 218 spin_lock(&pgd_lock); 219 list_for_each_entry(page, &pgd_list, lru) { 220 spinlock_t *pgt_lock; 221 pmd_t *ret; 222 223 /* the pgt_lock only for Xen */ 224 pgt_lock = &pgd_page_get_mm(page)->page_table_lock; 225 226 spin_lock(pgt_lock); 227 ret = vmalloc_sync_one(page_address(page), address); 228 spin_unlock(pgt_lock); 229 230 if (!ret) 231 break; 232 } 233 spin_unlock(&pgd_lock); 234 } 235 } 236 237 /* 238 * 32-bit: 239 * 240 * Handle a fault on the vmalloc or module mapping area 241 */ 242 static noinline int vmalloc_fault(unsigned long address) 243 { 244 unsigned long pgd_paddr; 245 pmd_t *pmd_k; 246 pte_t *pte_k; 247 248 /* Make sure we are in vmalloc area: */ 249 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 250 return -1; 251 252 /* 253 * Synchronize this task's top level page-table 254 * with the 'reference' page table. 255 * 256 * Do _not_ use "current" here. We might be inside 257 * an interrupt in the middle of a task switch.. 258 */ 259 pgd_paddr = read_cr3_pa(); 260 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); 261 if (!pmd_k) 262 return -1; 263 264 if (pmd_large(*pmd_k)) 265 return 0; 266 267 pte_k = pte_offset_kernel(pmd_k, address); 268 if (!pte_present(*pte_k)) 269 return -1; 270 271 return 0; 272 } 273 NOKPROBE_SYMBOL(vmalloc_fault); 274 275 /* 276 * Did it hit the DOS screen memory VA from vm86 mode? 277 */ 278 static inline void 279 check_v8086_mode(struct pt_regs *regs, unsigned long address, 280 struct task_struct *tsk) 281 { 282 #ifdef CONFIG_VM86 283 unsigned long bit; 284 285 if (!v8086_mode(regs) || !tsk->thread.vm86) 286 return; 287 288 bit = (address - 0xA0000) >> PAGE_SHIFT; 289 if (bit < 32) 290 tsk->thread.vm86->screen_bitmap |= 1 << bit; 291 #endif 292 } 293 294 static bool low_pfn(unsigned long pfn) 295 { 296 return pfn < max_low_pfn; 297 } 298 299 static void dump_pagetable(unsigned long address) 300 { 301 pgd_t *base = __va(read_cr3_pa()); 302 pgd_t *pgd = &base[pgd_index(address)]; 303 p4d_t *p4d; 304 pud_t *pud; 305 pmd_t *pmd; 306 pte_t *pte; 307 308 #ifdef CONFIG_X86_PAE 309 pr_info("*pdpt = %016Lx ", pgd_val(*pgd)); 310 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) 311 goto out; 312 #define pr_pde pr_cont 313 #else 314 #define pr_pde pr_info 315 #endif 316 p4d = p4d_offset(pgd, address); 317 pud = pud_offset(p4d, address); 318 pmd = pmd_offset(pud, address); 319 pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); 320 #undef pr_pde 321 322 /* 323 * We must not directly access the pte in the highpte 324 * case if the page table is located in highmem. 325 * And let's rather not kmap-atomic the pte, just in case 326 * it's allocated already: 327 */ 328 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) 329 goto out; 330 331 pte = pte_offset_kernel(pmd, address); 332 pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); 333 out: 334 pr_cont("\n"); 335 } 336 337 #else /* CONFIG_X86_64: */ 338 339 void vmalloc_sync_all(void) 340 { 341 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END); 342 } 343 344 /* 345 * 64-bit: 346 * 347 * Handle a fault on the vmalloc area 348 */ 349 static noinline int vmalloc_fault(unsigned long address) 350 { 351 pgd_t *pgd, *pgd_k; 352 p4d_t *p4d, *p4d_k; 353 pud_t *pud; 354 pmd_t *pmd; 355 pte_t *pte; 356 357 /* Make sure we are in vmalloc area: */ 358 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 359 return -1; 360 361 WARN_ON_ONCE(in_nmi()); 362 363 /* 364 * Copy kernel mappings over when needed. This can also 365 * happen within a race in page table update. In the later 366 * case just flush: 367 */ 368 pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address); 369 pgd_k = pgd_offset_k(address); 370 if (pgd_none(*pgd_k)) 371 return -1; 372 373 if (pgtable_l5_enabled()) { 374 if (pgd_none(*pgd)) { 375 set_pgd(pgd, *pgd_k); 376 arch_flush_lazy_mmu_mode(); 377 } else { 378 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_k)); 379 } 380 } 381 382 /* With 4-level paging, copying happens on the p4d level. */ 383 p4d = p4d_offset(pgd, address); 384 p4d_k = p4d_offset(pgd_k, address); 385 if (p4d_none(*p4d_k)) 386 return -1; 387 388 if (p4d_none(*p4d) && !pgtable_l5_enabled()) { 389 set_p4d(p4d, *p4d_k); 390 arch_flush_lazy_mmu_mode(); 391 } else { 392 BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_k)); 393 } 394 395 BUILD_BUG_ON(CONFIG_PGTABLE_LEVELS < 4); 396 397 pud = pud_offset(p4d, address); 398 if (pud_none(*pud)) 399 return -1; 400 401 if (pud_large(*pud)) 402 return 0; 403 404 pmd = pmd_offset(pud, address); 405 if (pmd_none(*pmd)) 406 return -1; 407 408 if (pmd_large(*pmd)) 409 return 0; 410 411 pte = pte_offset_kernel(pmd, address); 412 if (!pte_present(*pte)) 413 return -1; 414 415 return 0; 416 } 417 NOKPROBE_SYMBOL(vmalloc_fault); 418 419 #ifdef CONFIG_CPU_SUP_AMD 420 static const char errata93_warning[] = 421 KERN_ERR 422 "******* Your BIOS seems to not contain a fix for K8 errata #93\n" 423 "******* Working around it, but it may cause SEGVs or burn power.\n" 424 "******* Please consider a BIOS update.\n" 425 "******* Disabling USB legacy in the BIOS may also help.\n"; 426 #endif 427 428 /* 429 * No vm86 mode in 64-bit mode: 430 */ 431 static inline void 432 check_v8086_mode(struct pt_regs *regs, unsigned long address, 433 struct task_struct *tsk) 434 { 435 } 436 437 static int bad_address(void *p) 438 { 439 unsigned long dummy; 440 441 return probe_kernel_address((unsigned long *)p, dummy); 442 } 443 444 static void dump_pagetable(unsigned long address) 445 { 446 pgd_t *base = __va(read_cr3_pa()); 447 pgd_t *pgd = base + pgd_index(address); 448 p4d_t *p4d; 449 pud_t *pud; 450 pmd_t *pmd; 451 pte_t *pte; 452 453 if (bad_address(pgd)) 454 goto bad; 455 456 pr_info("PGD %lx ", pgd_val(*pgd)); 457 458 if (!pgd_present(*pgd)) 459 goto out; 460 461 p4d = p4d_offset(pgd, address); 462 if (bad_address(p4d)) 463 goto bad; 464 465 pr_cont("P4D %lx ", p4d_val(*p4d)); 466 if (!p4d_present(*p4d) || p4d_large(*p4d)) 467 goto out; 468 469 pud = pud_offset(p4d, address); 470 if (bad_address(pud)) 471 goto bad; 472 473 pr_cont("PUD %lx ", pud_val(*pud)); 474 if (!pud_present(*pud) || pud_large(*pud)) 475 goto out; 476 477 pmd = pmd_offset(pud, address); 478 if (bad_address(pmd)) 479 goto bad; 480 481 pr_cont("PMD %lx ", pmd_val(*pmd)); 482 if (!pmd_present(*pmd) || pmd_large(*pmd)) 483 goto out; 484 485 pte = pte_offset_kernel(pmd, address); 486 if (bad_address(pte)) 487 goto bad; 488 489 pr_cont("PTE %lx", pte_val(*pte)); 490 out: 491 pr_cont("\n"); 492 return; 493 bad: 494 pr_info("BAD\n"); 495 } 496 497 #endif /* CONFIG_X86_64 */ 498 499 /* 500 * Workaround for K8 erratum #93 & buggy BIOS. 501 * 502 * BIOS SMM functions are required to use a specific workaround 503 * to avoid corruption of the 64bit RIP register on C stepping K8. 504 * 505 * A lot of BIOS that didn't get tested properly miss this. 506 * 507 * The OS sees this as a page fault with the upper 32bits of RIP cleared. 508 * Try to work around it here. 509 * 510 * Note we only handle faults in kernel here. 511 * Does nothing on 32-bit. 512 */ 513 static int is_errata93(struct pt_regs *regs, unsigned long address) 514 { 515 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) 516 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD 517 || boot_cpu_data.x86 != 0xf) 518 return 0; 519 520 if (address != regs->ip) 521 return 0; 522 523 if ((address >> 32) != 0) 524 return 0; 525 526 address |= 0xffffffffUL << 32; 527 if ((address >= (u64)_stext && address <= (u64)_etext) || 528 (address >= MODULES_VADDR && address <= MODULES_END)) { 529 printk_once(errata93_warning); 530 regs->ip = address; 531 return 1; 532 } 533 #endif 534 return 0; 535 } 536 537 /* 538 * Work around K8 erratum #100 K8 in compat mode occasionally jumps 539 * to illegal addresses >4GB. 540 * 541 * We catch this in the page fault handler because these addresses 542 * are not reachable. Just detect this case and return. Any code 543 * segment in LDT is compatibility mode. 544 */ 545 static int is_errata100(struct pt_regs *regs, unsigned long address) 546 { 547 #ifdef CONFIG_X86_64 548 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) 549 return 1; 550 #endif 551 return 0; 552 } 553 554 static int is_f00f_bug(struct pt_regs *regs, unsigned long address) 555 { 556 #ifdef CONFIG_X86_F00F_BUG 557 unsigned long nr; 558 559 /* 560 * Pentium F0 0F C7 C8 bug workaround: 561 */ 562 if (boot_cpu_has_bug(X86_BUG_F00F)) { 563 nr = (address - idt_descr.address) >> 3; 564 565 if (nr == 6) { 566 do_invalid_op(regs, 0); 567 return 1; 568 } 569 } 570 #endif 571 return 0; 572 } 573 574 static void 575 show_fault_oops(struct pt_regs *regs, unsigned long error_code, 576 unsigned long address) 577 { 578 if (!oops_may_print()) 579 return; 580 581 if (error_code & X86_PF_INSTR) { 582 unsigned int level; 583 pgd_t *pgd; 584 pte_t *pte; 585 586 pgd = __va(read_cr3_pa()); 587 pgd += pgd_index(address); 588 589 pte = lookup_address_in_pgd(pgd, address, &level); 590 591 if (pte && pte_present(*pte) && !pte_exec(*pte)) 592 pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n", 593 from_kuid(&init_user_ns, current_uid())); 594 if (pte && pte_present(*pte) && pte_exec(*pte) && 595 (pgd_flags(*pgd) & _PAGE_USER) && 596 (__read_cr4() & X86_CR4_SMEP)) 597 pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n", 598 from_kuid(&init_user_ns, current_uid())); 599 } 600 601 pr_alert("BUG: unable to handle kernel %s at %px\n", 602 address < PAGE_SIZE ? "NULL pointer dereference" : "paging request", 603 (void *)address); 604 605 dump_pagetable(address); 606 } 607 608 static noinline void 609 pgtable_bad(struct pt_regs *regs, unsigned long error_code, 610 unsigned long address) 611 { 612 struct task_struct *tsk; 613 unsigned long flags; 614 int sig; 615 616 flags = oops_begin(); 617 tsk = current; 618 sig = SIGKILL; 619 620 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", 621 tsk->comm, address); 622 dump_pagetable(address); 623 624 tsk->thread.cr2 = address; 625 tsk->thread.trap_nr = X86_TRAP_PF; 626 tsk->thread.error_code = error_code; 627 628 if (__die("Bad pagetable", regs, error_code)) 629 sig = 0; 630 631 oops_end(flags, regs, sig); 632 } 633 634 static noinline void 635 no_context(struct pt_regs *regs, unsigned long error_code, 636 unsigned long address, int signal, int si_code) 637 { 638 struct task_struct *tsk = current; 639 unsigned long flags; 640 int sig; 641 642 /* Are we prepared to handle this kernel fault? */ 643 if (fixup_exception(regs, X86_TRAP_PF, error_code, address)) { 644 /* 645 * Any interrupt that takes a fault gets the fixup. This makes 646 * the below recursive fault logic only apply to a faults from 647 * task context. 648 */ 649 if (in_interrupt()) 650 return; 651 652 /* 653 * Per the above we're !in_interrupt(), aka. task context. 654 * 655 * In this case we need to make sure we're not recursively 656 * faulting through the emulate_vsyscall() logic. 657 */ 658 if (current->thread.sig_on_uaccess_err && signal) { 659 tsk->thread.trap_nr = X86_TRAP_PF; 660 tsk->thread.error_code = error_code | X86_PF_USER; 661 tsk->thread.cr2 = address; 662 663 /* XXX: hwpoison faults will set the wrong code. */ 664 force_sig_fault(signal, si_code, (void __user *)address, 665 tsk); 666 } 667 668 /* 669 * Barring that, we can do the fixup and be happy. 670 */ 671 return; 672 } 673 674 #ifdef CONFIG_VMAP_STACK 675 /* 676 * Stack overflow? During boot, we can fault near the initial 677 * stack in the direct map, but that's not an overflow -- check 678 * that we're in vmalloc space to avoid this. 679 */ 680 if (is_vmalloc_addr((void *)address) && 681 (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) || 682 address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) { 683 unsigned long stack = this_cpu_read(orig_ist.ist[DOUBLEFAULT_STACK]) - sizeof(void *); 684 /* 685 * We're likely to be running with very little stack space 686 * left. It's plausible that we'd hit this condition but 687 * double-fault even before we get this far, in which case 688 * we're fine: the double-fault handler will deal with it. 689 * 690 * We don't want to make it all the way into the oops code 691 * and then double-fault, though, because we're likely to 692 * break the console driver and lose most of the stack dump. 693 */ 694 asm volatile ("movq %[stack], %%rsp\n\t" 695 "call handle_stack_overflow\n\t" 696 "1: jmp 1b" 697 : ASM_CALL_CONSTRAINT 698 : "D" ("kernel stack overflow (page fault)"), 699 "S" (regs), "d" (address), 700 [stack] "rm" (stack)); 701 unreachable(); 702 } 703 #endif 704 705 /* 706 * 32-bit: 707 * 708 * Valid to do another page fault here, because if this fault 709 * had been triggered by is_prefetch fixup_exception would have 710 * handled it. 711 * 712 * 64-bit: 713 * 714 * Hall of shame of CPU/BIOS bugs. 715 */ 716 if (is_prefetch(regs, error_code, address)) 717 return; 718 719 if (is_errata93(regs, address)) 720 return; 721 722 /* 723 * Buggy firmware could access regions which might page fault, try to 724 * recover from such faults. 725 */ 726 if (IS_ENABLED(CONFIG_EFI)) 727 efi_recover_from_page_fault(address); 728 729 /* 730 * Oops. The kernel tried to access some bad page. We'll have to 731 * terminate things with extreme prejudice: 732 */ 733 flags = oops_begin(); 734 735 show_fault_oops(regs, error_code, address); 736 737 if (task_stack_end_corrupted(tsk)) 738 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); 739 740 tsk->thread.cr2 = address; 741 tsk->thread.trap_nr = X86_TRAP_PF; 742 tsk->thread.error_code = error_code; 743 744 sig = SIGKILL; 745 if (__die("Oops", regs, error_code)) 746 sig = 0; 747 748 /* Executive summary in case the body of the oops scrolled away */ 749 printk(KERN_DEFAULT "CR2: %016lx\n", address); 750 751 oops_end(flags, regs, sig); 752 } 753 754 /* 755 * Print out info about fatal segfaults, if the show_unhandled_signals 756 * sysctl is set: 757 */ 758 static inline void 759 show_signal_msg(struct pt_regs *regs, unsigned long error_code, 760 unsigned long address, struct task_struct *tsk) 761 { 762 const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG; 763 764 if (!unhandled_signal(tsk, SIGSEGV)) 765 return; 766 767 if (!printk_ratelimit()) 768 return; 769 770 printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx", 771 loglvl, tsk->comm, task_pid_nr(tsk), address, 772 (void *)regs->ip, (void *)regs->sp, error_code); 773 774 print_vma_addr(KERN_CONT " in ", regs->ip); 775 776 printk(KERN_CONT "\n"); 777 778 show_opcodes(regs, loglvl); 779 } 780 781 /* 782 * The (legacy) vsyscall page is the long page in the kernel portion 783 * of the address space that has user-accessible permissions. 784 */ 785 static bool is_vsyscall_vaddr(unsigned long vaddr) 786 { 787 return unlikely((vaddr & PAGE_MASK) == VSYSCALL_ADDR); 788 } 789 790 static void 791 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 792 unsigned long address, u32 pkey, int si_code) 793 { 794 struct task_struct *tsk = current; 795 796 /* User mode accesses just cause a SIGSEGV */ 797 if (error_code & X86_PF_USER) { 798 /* 799 * It's possible to have interrupts off here: 800 */ 801 local_irq_enable(); 802 803 /* 804 * Valid to do another page fault here because this one came 805 * from user space: 806 */ 807 if (is_prefetch(regs, error_code, address)) 808 return; 809 810 if (is_errata100(regs, address)) 811 return; 812 813 /* 814 * To avoid leaking information about the kernel page table 815 * layout, pretend that user-mode accesses to kernel addresses 816 * are always protection faults. 817 */ 818 if (address >= TASK_SIZE_MAX) 819 error_code |= X86_PF_PROT; 820 821 if (likely(show_unhandled_signals)) 822 show_signal_msg(regs, error_code, address, tsk); 823 824 tsk->thread.cr2 = address; 825 tsk->thread.error_code = error_code; 826 tsk->thread.trap_nr = X86_TRAP_PF; 827 828 if (si_code == SEGV_PKUERR) 829 force_sig_pkuerr((void __user *)address, pkey); 830 831 force_sig_fault(SIGSEGV, si_code, (void __user *)address, tsk); 832 833 return; 834 } 835 836 if (is_f00f_bug(regs, address)) 837 return; 838 839 no_context(regs, error_code, address, SIGSEGV, si_code); 840 } 841 842 static noinline void 843 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 844 unsigned long address) 845 { 846 __bad_area_nosemaphore(regs, error_code, address, 0, SEGV_MAPERR); 847 } 848 849 static void 850 __bad_area(struct pt_regs *regs, unsigned long error_code, 851 unsigned long address, u32 pkey, int si_code) 852 { 853 struct mm_struct *mm = current->mm; 854 /* 855 * Something tried to access memory that isn't in our memory map.. 856 * Fix it, but check if it's kernel or user first.. 857 */ 858 up_read(&mm->mmap_sem); 859 860 __bad_area_nosemaphore(regs, error_code, address, pkey, si_code); 861 } 862 863 static noinline void 864 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) 865 { 866 __bad_area(regs, error_code, address, 0, SEGV_MAPERR); 867 } 868 869 static inline bool bad_area_access_from_pkeys(unsigned long error_code, 870 struct vm_area_struct *vma) 871 { 872 /* This code is always called on the current mm */ 873 bool foreign = false; 874 875 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 876 return false; 877 if (error_code & X86_PF_PK) 878 return true; 879 /* this checks permission keys on the VMA: */ 880 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), 881 (error_code & X86_PF_INSTR), foreign)) 882 return true; 883 return false; 884 } 885 886 static noinline void 887 bad_area_access_error(struct pt_regs *regs, unsigned long error_code, 888 unsigned long address, struct vm_area_struct *vma) 889 { 890 /* 891 * This OSPKE check is not strictly necessary at runtime. 892 * But, doing it this way allows compiler optimizations 893 * if pkeys are compiled out. 894 */ 895 if (bad_area_access_from_pkeys(error_code, vma)) { 896 /* 897 * A protection key fault means that the PKRU value did not allow 898 * access to some PTE. Userspace can figure out what PKRU was 899 * from the XSAVE state. This function captures the pkey from 900 * the vma and passes it to userspace so userspace can discover 901 * which protection key was set on the PTE. 902 * 903 * If we get here, we know that the hardware signaled a X86_PF_PK 904 * fault and that there was a VMA once we got in the fault 905 * handler. It does *not* guarantee that the VMA we find here 906 * was the one that we faulted on. 907 * 908 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4); 909 * 2. T1 : set PKRU to deny access to pkey=4, touches page 910 * 3. T1 : faults... 911 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5); 912 * 5. T1 : enters fault handler, takes mmap_sem, etc... 913 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really 914 * faulted on a pte with its pkey=4. 915 */ 916 u32 pkey = vma_pkey(vma); 917 918 __bad_area(regs, error_code, address, pkey, SEGV_PKUERR); 919 } else { 920 __bad_area(regs, error_code, address, 0, SEGV_ACCERR); 921 } 922 } 923 924 static void 925 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, 926 unsigned int fault) 927 { 928 struct task_struct *tsk = current; 929 930 /* Kernel mode? Handle exceptions or die: */ 931 if (!(error_code & X86_PF_USER)) { 932 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); 933 return; 934 } 935 936 /* User-space => ok to do another page fault: */ 937 if (is_prefetch(regs, error_code, address)) 938 return; 939 940 tsk->thread.cr2 = address; 941 tsk->thread.error_code = error_code; 942 tsk->thread.trap_nr = X86_TRAP_PF; 943 944 #ifdef CONFIG_MEMORY_FAILURE 945 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { 946 unsigned lsb = 0; 947 948 pr_err( 949 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", 950 tsk->comm, tsk->pid, address); 951 if (fault & VM_FAULT_HWPOISON_LARGE) 952 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); 953 if (fault & VM_FAULT_HWPOISON) 954 lsb = PAGE_SHIFT; 955 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb, tsk); 956 return; 957 } 958 #endif 959 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address, tsk); 960 } 961 962 static noinline void 963 mm_fault_error(struct pt_regs *regs, unsigned long error_code, 964 unsigned long address, vm_fault_t fault) 965 { 966 if (fatal_signal_pending(current) && !(error_code & X86_PF_USER)) { 967 no_context(regs, error_code, address, 0, 0); 968 return; 969 } 970 971 if (fault & VM_FAULT_OOM) { 972 /* Kernel mode? Handle exceptions or die: */ 973 if (!(error_code & X86_PF_USER)) { 974 no_context(regs, error_code, address, 975 SIGSEGV, SEGV_MAPERR); 976 return; 977 } 978 979 /* 980 * We ran out of memory, call the OOM killer, and return the 981 * userspace (which will retry the fault, or kill us if we got 982 * oom-killed): 983 */ 984 pagefault_out_of_memory(); 985 } else { 986 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| 987 VM_FAULT_HWPOISON_LARGE)) 988 do_sigbus(regs, error_code, address, fault); 989 else if (fault & VM_FAULT_SIGSEGV) 990 bad_area_nosemaphore(regs, error_code, address); 991 else 992 BUG(); 993 } 994 } 995 996 static int spurious_kernel_fault_check(unsigned long error_code, pte_t *pte) 997 { 998 if ((error_code & X86_PF_WRITE) && !pte_write(*pte)) 999 return 0; 1000 1001 if ((error_code & X86_PF_INSTR) && !pte_exec(*pte)) 1002 return 0; 1003 1004 return 1; 1005 } 1006 1007 /* 1008 * Handle a spurious fault caused by a stale TLB entry. 1009 * 1010 * This allows us to lazily refresh the TLB when increasing the 1011 * permissions of a kernel page (RO -> RW or NX -> X). Doing it 1012 * eagerly is very expensive since that implies doing a full 1013 * cross-processor TLB flush, even if no stale TLB entries exist 1014 * on other processors. 1015 * 1016 * Spurious faults may only occur if the TLB contains an entry with 1017 * fewer permission than the page table entry. Non-present (P = 0) 1018 * and reserved bit (R = 1) faults are never spurious. 1019 * 1020 * There are no security implications to leaving a stale TLB when 1021 * increasing the permissions on a page. 1022 * 1023 * Returns non-zero if a spurious fault was handled, zero otherwise. 1024 * 1025 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 1026 * (Optional Invalidation). 1027 */ 1028 static noinline int 1029 spurious_kernel_fault(unsigned long error_code, unsigned long address) 1030 { 1031 pgd_t *pgd; 1032 p4d_t *p4d; 1033 pud_t *pud; 1034 pmd_t *pmd; 1035 pte_t *pte; 1036 int ret; 1037 1038 /* 1039 * Only writes to RO or instruction fetches from NX may cause 1040 * spurious faults. 1041 * 1042 * These could be from user or supervisor accesses but the TLB 1043 * is only lazily flushed after a kernel mapping protection 1044 * change, so user accesses are not expected to cause spurious 1045 * faults. 1046 */ 1047 if (error_code != (X86_PF_WRITE | X86_PF_PROT) && 1048 error_code != (X86_PF_INSTR | X86_PF_PROT)) 1049 return 0; 1050 1051 pgd = init_mm.pgd + pgd_index(address); 1052 if (!pgd_present(*pgd)) 1053 return 0; 1054 1055 p4d = p4d_offset(pgd, address); 1056 if (!p4d_present(*p4d)) 1057 return 0; 1058 1059 if (p4d_large(*p4d)) 1060 return spurious_kernel_fault_check(error_code, (pte_t *) p4d); 1061 1062 pud = pud_offset(p4d, address); 1063 if (!pud_present(*pud)) 1064 return 0; 1065 1066 if (pud_large(*pud)) 1067 return spurious_kernel_fault_check(error_code, (pte_t *) pud); 1068 1069 pmd = pmd_offset(pud, address); 1070 if (!pmd_present(*pmd)) 1071 return 0; 1072 1073 if (pmd_large(*pmd)) 1074 return spurious_kernel_fault_check(error_code, (pte_t *) pmd); 1075 1076 pte = pte_offset_kernel(pmd, address); 1077 if (!pte_present(*pte)) 1078 return 0; 1079 1080 ret = spurious_kernel_fault_check(error_code, pte); 1081 if (!ret) 1082 return 0; 1083 1084 /* 1085 * Make sure we have permissions in PMD. 1086 * If not, then there's a bug in the page tables: 1087 */ 1088 ret = spurious_kernel_fault_check(error_code, (pte_t *) pmd); 1089 WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); 1090 1091 return ret; 1092 } 1093 NOKPROBE_SYMBOL(spurious_kernel_fault); 1094 1095 int show_unhandled_signals = 1; 1096 1097 static inline int 1098 access_error(unsigned long error_code, struct vm_area_struct *vma) 1099 { 1100 /* This is only called for the current mm, so: */ 1101 bool foreign = false; 1102 1103 /* 1104 * Read or write was blocked by protection keys. This is 1105 * always an unconditional error and can never result in 1106 * a follow-up action to resolve the fault, like a COW. 1107 */ 1108 if (error_code & X86_PF_PK) 1109 return 1; 1110 1111 /* 1112 * Make sure to check the VMA so that we do not perform 1113 * faults just to hit a X86_PF_PK as soon as we fill in a 1114 * page. 1115 */ 1116 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), 1117 (error_code & X86_PF_INSTR), foreign)) 1118 return 1; 1119 1120 if (error_code & X86_PF_WRITE) { 1121 /* write, present and write, not present: */ 1122 if (unlikely(!(vma->vm_flags & VM_WRITE))) 1123 return 1; 1124 return 0; 1125 } 1126 1127 /* read, present: */ 1128 if (unlikely(error_code & X86_PF_PROT)) 1129 return 1; 1130 1131 /* read, not present: */ 1132 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) 1133 return 1; 1134 1135 return 0; 1136 } 1137 1138 static int fault_in_kernel_space(unsigned long address) 1139 { 1140 /* 1141 * On 64-bit systems, the vsyscall page is at an address above 1142 * TASK_SIZE_MAX, but is not considered part of the kernel 1143 * address space. 1144 */ 1145 if (IS_ENABLED(CONFIG_X86_64) && is_vsyscall_vaddr(address)) 1146 return false; 1147 1148 return address >= TASK_SIZE_MAX; 1149 } 1150 1151 static inline bool smap_violation(int error_code, struct pt_regs *regs) 1152 { 1153 if (!IS_ENABLED(CONFIG_X86_SMAP)) 1154 return false; 1155 1156 if (!static_cpu_has(X86_FEATURE_SMAP)) 1157 return false; 1158 1159 if (error_code & X86_PF_USER) 1160 return false; 1161 1162 if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC)) 1163 return false; 1164 1165 return true; 1166 } 1167 1168 /* 1169 * Called for all faults where 'address' is part of the kernel address 1170 * space. Might get called for faults that originate from *code* that 1171 * ran in userspace or the kernel. 1172 */ 1173 static void 1174 do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code, 1175 unsigned long address) 1176 { 1177 /* 1178 * Protection keys exceptions only happen on user pages. We 1179 * have no user pages in the kernel portion of the address 1180 * space, so do not expect them here. 1181 */ 1182 WARN_ON_ONCE(hw_error_code & X86_PF_PK); 1183 1184 /* 1185 * We can fault-in kernel-space virtual memory on-demand. The 1186 * 'reference' page table is init_mm.pgd. 1187 * 1188 * NOTE! We MUST NOT take any locks for this case. We may 1189 * be in an interrupt or a critical region, and should 1190 * only copy the information from the master page table, 1191 * nothing more. 1192 * 1193 * Before doing this on-demand faulting, ensure that the 1194 * fault is not any of the following: 1195 * 1. A fault on a PTE with a reserved bit set. 1196 * 2. A fault caused by a user-mode access. (Do not demand- 1197 * fault kernel memory due to user-mode accesses). 1198 * 3. A fault caused by a page-level protection violation. 1199 * (A demand fault would be on a non-present page which 1200 * would have X86_PF_PROT==0). 1201 */ 1202 if (!(hw_error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) { 1203 if (vmalloc_fault(address) >= 0) 1204 return; 1205 } 1206 1207 /* Was the fault spurious, caused by lazy TLB invalidation? */ 1208 if (spurious_kernel_fault(hw_error_code, address)) 1209 return; 1210 1211 /* kprobes don't want to hook the spurious faults: */ 1212 if (kprobes_fault(regs)) 1213 return; 1214 1215 /* 1216 * Note, despite being a "bad area", there are quite a few 1217 * acceptable reasons to get here, such as erratum fixups 1218 * and handling kernel code that can fault, like get_user(). 1219 * 1220 * Don't take the mm semaphore here. If we fixup a prefetch 1221 * fault we could otherwise deadlock: 1222 */ 1223 bad_area_nosemaphore(regs, hw_error_code, address); 1224 } 1225 NOKPROBE_SYMBOL(do_kern_addr_fault); 1226 1227 /* Handle faults in the user portion of the address space */ 1228 static inline 1229 void do_user_addr_fault(struct pt_regs *regs, 1230 unsigned long hw_error_code, 1231 unsigned long address) 1232 { 1233 unsigned long sw_error_code; 1234 struct vm_area_struct *vma; 1235 struct task_struct *tsk; 1236 struct mm_struct *mm; 1237 vm_fault_t fault, major = 0; 1238 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; 1239 1240 tsk = current; 1241 mm = tsk->mm; 1242 1243 /* kprobes don't want to hook the spurious faults: */ 1244 if (unlikely(kprobes_fault(regs))) 1245 return; 1246 1247 /* 1248 * Reserved bits are never expected to be set on 1249 * entries in the user portion of the page tables. 1250 */ 1251 if (unlikely(hw_error_code & X86_PF_RSVD)) 1252 pgtable_bad(regs, hw_error_code, address); 1253 1254 /* 1255 * Check for invalid kernel (supervisor) access to user 1256 * pages in the user address space. 1257 */ 1258 if (unlikely(smap_violation(hw_error_code, regs))) { 1259 bad_area_nosemaphore(regs, hw_error_code, address); 1260 return; 1261 } 1262 1263 /* 1264 * If we're in an interrupt, have no user context or are running 1265 * in a region with pagefaults disabled then we must not take the fault 1266 */ 1267 if (unlikely(faulthandler_disabled() || !mm)) { 1268 bad_area_nosemaphore(regs, hw_error_code, address); 1269 return; 1270 } 1271 1272 /* 1273 * hw_error_code is literally the "page fault error code" passed to 1274 * the kernel directly from the hardware. But, we will shortly be 1275 * modifying it in software, so give it a new name. 1276 */ 1277 sw_error_code = hw_error_code; 1278 1279 /* 1280 * It's safe to allow irq's after cr2 has been saved and the 1281 * vmalloc fault has been handled. 1282 * 1283 * User-mode registers count as a user access even for any 1284 * potential system fault or CPU buglet: 1285 */ 1286 if (user_mode(regs)) { 1287 local_irq_enable(); 1288 /* 1289 * Up to this point, X86_PF_USER set in hw_error_code 1290 * indicated a user-mode access. But, after this, 1291 * X86_PF_USER in sw_error_code will indicate either 1292 * that, *or* an implicit kernel(supervisor)-mode access 1293 * which originated from user mode. 1294 */ 1295 if (!(hw_error_code & X86_PF_USER)) { 1296 /* 1297 * The CPU was in user mode, but the CPU says 1298 * the fault was not a user-mode access. 1299 * Must be an implicit kernel-mode access, 1300 * which we do not expect to happen in the 1301 * user address space. 1302 */ 1303 pr_warn_once("kernel-mode error from user-mode: %lx\n", 1304 hw_error_code); 1305 1306 sw_error_code |= X86_PF_USER; 1307 } 1308 flags |= FAULT_FLAG_USER; 1309 } else { 1310 if (regs->flags & X86_EFLAGS_IF) 1311 local_irq_enable(); 1312 } 1313 1314 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 1315 1316 if (sw_error_code & X86_PF_WRITE) 1317 flags |= FAULT_FLAG_WRITE; 1318 if (sw_error_code & X86_PF_INSTR) 1319 flags |= FAULT_FLAG_INSTRUCTION; 1320 1321 #ifdef CONFIG_X86_64 1322 /* 1323 * Instruction fetch faults in the vsyscall page might need 1324 * emulation. The vsyscall page is at a high address 1325 * (>PAGE_OFFSET), but is considered to be part of the user 1326 * address space. 1327 * 1328 * The vsyscall page does not have a "real" VMA, so do this 1329 * emulation before we go searching for VMAs. 1330 */ 1331 if ((sw_error_code & X86_PF_INSTR) && is_vsyscall_vaddr(address)) { 1332 if (emulate_vsyscall(regs, address)) 1333 return; 1334 } 1335 #endif 1336 1337 /* 1338 * Kernel-mode access to the user address space should only occur 1339 * on well-defined single instructions listed in the exception 1340 * tables. But, an erroneous kernel fault occurring outside one of 1341 * those areas which also holds mmap_sem might deadlock attempting 1342 * to validate the fault against the address space. 1343 * 1344 * Only do the expensive exception table search when we might be at 1345 * risk of a deadlock. This happens if we 1346 * 1. Failed to acquire mmap_sem, and 1347 * 2. The access did not originate in userspace. Note: either the 1348 * hardware or earlier page fault code may set X86_PF_USER 1349 * in sw_error_code. 1350 */ 1351 if (unlikely(!down_read_trylock(&mm->mmap_sem))) { 1352 if (!(sw_error_code & X86_PF_USER) && 1353 !search_exception_tables(regs->ip)) { 1354 /* 1355 * Fault from code in kernel from 1356 * which we do not expect faults. 1357 */ 1358 bad_area_nosemaphore(regs, sw_error_code, address); 1359 return; 1360 } 1361 retry: 1362 down_read(&mm->mmap_sem); 1363 } else { 1364 /* 1365 * The above down_read_trylock() might have succeeded in 1366 * which case we'll have missed the might_sleep() from 1367 * down_read(): 1368 */ 1369 might_sleep(); 1370 } 1371 1372 vma = find_vma(mm, address); 1373 if (unlikely(!vma)) { 1374 bad_area(regs, sw_error_code, address); 1375 return; 1376 } 1377 if (likely(vma->vm_start <= address)) 1378 goto good_area; 1379 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { 1380 bad_area(regs, sw_error_code, address); 1381 return; 1382 } 1383 if (sw_error_code & X86_PF_USER) { 1384 /* 1385 * Accessing the stack below %sp is always a bug. 1386 * The large cushion allows instructions like enter 1387 * and pusha to work. ("enter $65535, $31" pushes 1388 * 32 pointers and then decrements %sp by 65535.) 1389 */ 1390 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { 1391 bad_area(regs, sw_error_code, address); 1392 return; 1393 } 1394 } 1395 if (unlikely(expand_stack(vma, address))) { 1396 bad_area(regs, sw_error_code, address); 1397 return; 1398 } 1399 1400 /* 1401 * Ok, we have a good vm_area for this memory access, so 1402 * we can handle it.. 1403 */ 1404 good_area: 1405 if (unlikely(access_error(sw_error_code, vma))) { 1406 bad_area_access_error(regs, sw_error_code, address, vma); 1407 return; 1408 } 1409 1410 /* 1411 * If for any reason at all we couldn't handle the fault, 1412 * make sure we exit gracefully rather than endlessly redo 1413 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if 1414 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked. 1415 * 1416 * Note that handle_userfault() may also release and reacquire mmap_sem 1417 * (and not return with VM_FAULT_RETRY), when returning to userland to 1418 * repeat the page fault later with a VM_FAULT_NOPAGE retval 1419 * (potentially after handling any pending signal during the return to 1420 * userland). The return to userland is identified whenever 1421 * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags. 1422 */ 1423 fault = handle_mm_fault(vma, address, flags); 1424 major |= fault & VM_FAULT_MAJOR; 1425 1426 /* 1427 * If we need to retry the mmap_sem has already been released, 1428 * and if there is a fatal signal pending there is no guarantee 1429 * that we made any progress. Handle this case first. 1430 */ 1431 if (unlikely(fault & VM_FAULT_RETRY)) { 1432 /* Retry at most once */ 1433 if (flags & FAULT_FLAG_ALLOW_RETRY) { 1434 flags &= ~FAULT_FLAG_ALLOW_RETRY; 1435 flags |= FAULT_FLAG_TRIED; 1436 if (!fatal_signal_pending(tsk)) 1437 goto retry; 1438 } 1439 1440 /* User mode? Just return to handle the fatal exception */ 1441 if (flags & FAULT_FLAG_USER) 1442 return; 1443 1444 /* Not returning to user mode? Handle exceptions or die: */ 1445 no_context(regs, sw_error_code, address, SIGBUS, BUS_ADRERR); 1446 return; 1447 } 1448 1449 up_read(&mm->mmap_sem); 1450 if (unlikely(fault & VM_FAULT_ERROR)) { 1451 mm_fault_error(regs, sw_error_code, address, fault); 1452 return; 1453 } 1454 1455 /* 1456 * Major/minor page fault accounting. If any of the events 1457 * returned VM_FAULT_MAJOR, we account it as a major fault. 1458 */ 1459 if (major) { 1460 tsk->maj_flt++; 1461 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); 1462 } else { 1463 tsk->min_flt++; 1464 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); 1465 } 1466 1467 check_v8086_mode(regs, address, tsk); 1468 } 1469 NOKPROBE_SYMBOL(do_user_addr_fault); 1470 1471 /* 1472 * This routine handles page faults. It determines the address, 1473 * and the problem, and then passes it off to one of the appropriate 1474 * routines. 1475 */ 1476 static noinline void 1477 __do_page_fault(struct pt_regs *regs, unsigned long hw_error_code, 1478 unsigned long address) 1479 { 1480 prefetchw(¤t->mm->mmap_sem); 1481 1482 if (unlikely(kmmio_fault(regs, address))) 1483 return; 1484 1485 /* Was the fault on kernel-controlled part of the address space? */ 1486 if (unlikely(fault_in_kernel_space(address))) 1487 do_kern_addr_fault(regs, hw_error_code, address); 1488 else 1489 do_user_addr_fault(regs, hw_error_code, address); 1490 } 1491 NOKPROBE_SYMBOL(__do_page_fault); 1492 1493 static nokprobe_inline void 1494 trace_page_fault_entries(unsigned long address, struct pt_regs *regs, 1495 unsigned long error_code) 1496 { 1497 if (user_mode(regs)) 1498 trace_page_fault_user(address, regs, error_code); 1499 else 1500 trace_page_fault_kernel(address, regs, error_code); 1501 } 1502 1503 /* 1504 * We must have this function blacklisted from kprobes, tagged with notrace 1505 * and call read_cr2() before calling anything else. To avoid calling any 1506 * kind of tracing machinery before we've observed the CR2 value. 1507 * 1508 * exception_{enter,exit}() contains all sorts of tracepoints. 1509 */ 1510 dotraplinkage void notrace 1511 do_page_fault(struct pt_regs *regs, unsigned long error_code) 1512 { 1513 unsigned long address = read_cr2(); /* Get the faulting address */ 1514 enum ctx_state prev_state; 1515 1516 prev_state = exception_enter(); 1517 if (trace_pagefault_enabled()) 1518 trace_page_fault_entries(address, regs, error_code); 1519 1520 __do_page_fault(regs, error_code, address); 1521 exception_exit(prev_state); 1522 } 1523 NOKPROBE_SYMBOL(do_page_fault); 1524