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/bootmem.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/mm_types.h> 20 21 #include <asm/cpufeature.h> /* boot_cpu_has, ... */ 22 #include <asm/traps.h> /* dotraplinkage, ... */ 23 #include <asm/pgalloc.h> /* pgd_*(), ... */ 24 #include <asm/fixmap.h> /* VSYSCALL_ADDR */ 25 #include <asm/vsyscall.h> /* emulate_vsyscall */ 26 #include <asm/vm86.h> /* struct vm86 */ 27 #include <asm/mmu_context.h> /* vma_pkey() */ 28 29 #define CREATE_TRACE_POINTS 30 #include <asm/trace/exceptions.h> 31 32 /* 33 * Returns 0 if mmiotrace is disabled, or if the fault is not 34 * handled by mmiotrace: 35 */ 36 static nokprobe_inline int 37 kmmio_fault(struct pt_regs *regs, unsigned long addr) 38 { 39 if (unlikely(is_kmmio_active())) 40 if (kmmio_handler(regs, addr) == 1) 41 return -1; 42 return 0; 43 } 44 45 static nokprobe_inline int kprobes_fault(struct pt_regs *regs) 46 { 47 int ret = 0; 48 49 /* kprobe_running() needs smp_processor_id() */ 50 if (kprobes_built_in() && !user_mode(regs)) { 51 preempt_disable(); 52 if (kprobe_running() && kprobe_fault_handler(regs, 14)) 53 ret = 1; 54 preempt_enable(); 55 } 56 57 return ret; 58 } 59 60 /* 61 * Prefetch quirks: 62 * 63 * 32-bit mode: 64 * 65 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. 66 * Check that here and ignore it. 67 * 68 * 64-bit mode: 69 * 70 * Sometimes the CPU reports invalid exceptions on prefetch. 71 * Check that here and ignore it. 72 * 73 * Opcode checker based on code by Richard Brunner. 74 */ 75 static inline int 76 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, 77 unsigned char opcode, int *prefetch) 78 { 79 unsigned char instr_hi = opcode & 0xf0; 80 unsigned char instr_lo = opcode & 0x0f; 81 82 switch (instr_hi) { 83 case 0x20: 84 case 0x30: 85 /* 86 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. 87 * In X86_64 long mode, the CPU will signal invalid 88 * opcode if some of these prefixes are present so 89 * X86_64 will never get here anyway 90 */ 91 return ((instr_lo & 7) == 0x6); 92 #ifdef CONFIG_X86_64 93 case 0x40: 94 /* 95 * In AMD64 long mode 0x40..0x4F are valid REX prefixes 96 * Need to figure out under what instruction mode the 97 * instruction was issued. Could check the LDT for lm, 98 * but for now it's good enough to assume that long 99 * mode only uses well known segments or kernel. 100 */ 101 return (!user_mode(regs) || user_64bit_mode(regs)); 102 #endif 103 case 0x60: 104 /* 0x64 thru 0x67 are valid prefixes in all modes. */ 105 return (instr_lo & 0xC) == 0x4; 106 case 0xF0: 107 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ 108 return !instr_lo || (instr_lo>>1) == 1; 109 case 0x00: 110 /* Prefetch instruction is 0x0F0D or 0x0F18 */ 111 if (probe_kernel_address(instr, opcode)) 112 return 0; 113 114 *prefetch = (instr_lo == 0xF) && 115 (opcode == 0x0D || opcode == 0x18); 116 return 0; 117 default: 118 return 0; 119 } 120 } 121 122 static int 123 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) 124 { 125 unsigned char *max_instr; 126 unsigned char *instr; 127 int prefetch = 0; 128 129 /* 130 * If it was a exec (instruction fetch) fault on NX page, then 131 * do not ignore the fault: 132 */ 133 if (error_code & X86_PF_INSTR) 134 return 0; 135 136 instr = (void *)convert_ip_to_linear(current, regs); 137 max_instr = instr + 15; 138 139 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX) 140 return 0; 141 142 while (instr < max_instr) { 143 unsigned char opcode; 144 145 if (probe_kernel_address(instr, opcode)) 146 break; 147 148 instr++; 149 150 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) 151 break; 152 } 153 return prefetch; 154 } 155 156 /* 157 * A protection key fault means that the PKRU value did not allow 158 * access to some PTE. Userspace can figure out what PKRU was 159 * from the XSAVE state, and this function fills out a field in 160 * siginfo so userspace can discover which protection key was set 161 * on the PTE. 162 * 163 * If we get here, we know that the hardware signaled a X86_PF_PK 164 * fault and that there was a VMA once we got in the fault 165 * handler. It does *not* guarantee that the VMA we find here 166 * was the one that we faulted on. 167 * 168 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4); 169 * 2. T1 : set PKRU to deny access to pkey=4, touches page 170 * 3. T1 : faults... 171 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5); 172 * 5. T1 : enters fault handler, takes mmap_sem, etc... 173 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really 174 * faulted on a pte with its pkey=4. 175 */ 176 static void fill_sig_info_pkey(int si_signo, int si_code, siginfo_t *info, 177 u32 *pkey) 178 { 179 /* This is effectively an #ifdef */ 180 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 181 return; 182 183 /* Fault not from Protection Keys: nothing to do */ 184 if ((si_code != SEGV_PKUERR) || (si_signo != SIGSEGV)) 185 return; 186 /* 187 * force_sig_info_fault() is called from a number of 188 * contexts, some of which have a VMA and some of which 189 * do not. The X86_PF_PK handing happens after we have a 190 * valid VMA, so we should never reach this without a 191 * valid VMA. 192 */ 193 if (!pkey) { 194 WARN_ONCE(1, "PKU fault with no VMA passed in"); 195 info->si_pkey = 0; 196 return; 197 } 198 /* 199 * si_pkey should be thought of as a strong hint, but not 200 * absolutely guranteed to be 100% accurate because of 201 * the race explained above. 202 */ 203 info->si_pkey = *pkey; 204 } 205 206 static void 207 force_sig_info_fault(int si_signo, int si_code, unsigned long address, 208 struct task_struct *tsk, u32 *pkey, int fault) 209 { 210 unsigned lsb = 0; 211 siginfo_t info; 212 213 clear_siginfo(&info); 214 info.si_signo = si_signo; 215 info.si_errno = 0; 216 info.si_code = si_code; 217 info.si_addr = (void __user *)address; 218 if (fault & VM_FAULT_HWPOISON_LARGE) 219 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); 220 if (fault & VM_FAULT_HWPOISON) 221 lsb = PAGE_SHIFT; 222 info.si_addr_lsb = lsb; 223 224 fill_sig_info_pkey(si_signo, si_code, &info, pkey); 225 226 force_sig_info(si_signo, &info, tsk); 227 } 228 229 DEFINE_SPINLOCK(pgd_lock); 230 LIST_HEAD(pgd_list); 231 232 #ifdef CONFIG_X86_32 233 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) 234 { 235 unsigned index = pgd_index(address); 236 pgd_t *pgd_k; 237 p4d_t *p4d, *p4d_k; 238 pud_t *pud, *pud_k; 239 pmd_t *pmd, *pmd_k; 240 241 pgd += index; 242 pgd_k = init_mm.pgd + index; 243 244 if (!pgd_present(*pgd_k)) 245 return NULL; 246 247 /* 248 * set_pgd(pgd, *pgd_k); here would be useless on PAE 249 * and redundant with the set_pmd() on non-PAE. As would 250 * set_p4d/set_pud. 251 */ 252 p4d = p4d_offset(pgd, address); 253 p4d_k = p4d_offset(pgd_k, address); 254 if (!p4d_present(*p4d_k)) 255 return NULL; 256 257 pud = pud_offset(p4d, address); 258 pud_k = pud_offset(p4d_k, address); 259 if (!pud_present(*pud_k)) 260 return NULL; 261 262 pmd = pmd_offset(pud, address); 263 pmd_k = pmd_offset(pud_k, address); 264 if (!pmd_present(*pmd_k)) 265 return NULL; 266 267 if (!pmd_present(*pmd)) 268 set_pmd(pmd, *pmd_k); 269 else 270 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); 271 272 return pmd_k; 273 } 274 275 void vmalloc_sync_all(void) 276 { 277 unsigned long address; 278 279 if (SHARED_KERNEL_PMD) 280 return; 281 282 for (address = VMALLOC_START & PMD_MASK; 283 address >= TASK_SIZE_MAX && address < FIXADDR_TOP; 284 address += PMD_SIZE) { 285 struct page *page; 286 287 spin_lock(&pgd_lock); 288 list_for_each_entry(page, &pgd_list, lru) { 289 spinlock_t *pgt_lock; 290 pmd_t *ret; 291 292 /* the pgt_lock only for Xen */ 293 pgt_lock = &pgd_page_get_mm(page)->page_table_lock; 294 295 spin_lock(pgt_lock); 296 ret = vmalloc_sync_one(page_address(page), address); 297 spin_unlock(pgt_lock); 298 299 if (!ret) 300 break; 301 } 302 spin_unlock(&pgd_lock); 303 } 304 } 305 306 /* 307 * 32-bit: 308 * 309 * Handle a fault on the vmalloc or module mapping area 310 */ 311 static noinline int vmalloc_fault(unsigned long address) 312 { 313 unsigned long pgd_paddr; 314 pmd_t *pmd_k; 315 pte_t *pte_k; 316 317 /* Make sure we are in vmalloc area: */ 318 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 319 return -1; 320 321 /* 322 * Synchronize this task's top level page-table 323 * with the 'reference' page table. 324 * 325 * Do _not_ use "current" here. We might be inside 326 * an interrupt in the middle of a task switch.. 327 */ 328 pgd_paddr = read_cr3_pa(); 329 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); 330 if (!pmd_k) 331 return -1; 332 333 if (pmd_large(*pmd_k)) 334 return 0; 335 336 pte_k = pte_offset_kernel(pmd_k, address); 337 if (!pte_present(*pte_k)) 338 return -1; 339 340 return 0; 341 } 342 NOKPROBE_SYMBOL(vmalloc_fault); 343 344 /* 345 * Did it hit the DOS screen memory VA from vm86 mode? 346 */ 347 static inline void 348 check_v8086_mode(struct pt_regs *regs, unsigned long address, 349 struct task_struct *tsk) 350 { 351 #ifdef CONFIG_VM86 352 unsigned long bit; 353 354 if (!v8086_mode(regs) || !tsk->thread.vm86) 355 return; 356 357 bit = (address - 0xA0000) >> PAGE_SHIFT; 358 if (bit < 32) 359 tsk->thread.vm86->screen_bitmap |= 1 << bit; 360 #endif 361 } 362 363 static bool low_pfn(unsigned long pfn) 364 { 365 return pfn < max_low_pfn; 366 } 367 368 static void dump_pagetable(unsigned long address) 369 { 370 pgd_t *base = __va(read_cr3_pa()); 371 pgd_t *pgd = &base[pgd_index(address)]; 372 p4d_t *p4d; 373 pud_t *pud; 374 pmd_t *pmd; 375 pte_t *pte; 376 377 #ifdef CONFIG_X86_PAE 378 pr_info("*pdpt = %016Lx ", pgd_val(*pgd)); 379 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) 380 goto out; 381 #define pr_pde pr_cont 382 #else 383 #define pr_pde pr_info 384 #endif 385 p4d = p4d_offset(pgd, address); 386 pud = pud_offset(p4d, address); 387 pmd = pmd_offset(pud, address); 388 pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); 389 #undef pr_pde 390 391 /* 392 * We must not directly access the pte in the highpte 393 * case if the page table is located in highmem. 394 * And let's rather not kmap-atomic the pte, just in case 395 * it's allocated already: 396 */ 397 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) 398 goto out; 399 400 pte = pte_offset_kernel(pmd, address); 401 pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); 402 out: 403 pr_cont("\n"); 404 } 405 406 #else /* CONFIG_X86_64: */ 407 408 void vmalloc_sync_all(void) 409 { 410 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END); 411 } 412 413 /* 414 * 64-bit: 415 * 416 * Handle a fault on the vmalloc area 417 */ 418 static noinline int vmalloc_fault(unsigned long address) 419 { 420 pgd_t *pgd, *pgd_k; 421 p4d_t *p4d, *p4d_k; 422 pud_t *pud; 423 pmd_t *pmd; 424 pte_t *pte; 425 426 /* Make sure we are in vmalloc area: */ 427 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 428 return -1; 429 430 WARN_ON_ONCE(in_nmi()); 431 432 /* 433 * Copy kernel mappings over when needed. This can also 434 * happen within a race in page table update. In the later 435 * case just flush: 436 */ 437 pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address); 438 pgd_k = pgd_offset_k(address); 439 if (pgd_none(*pgd_k)) 440 return -1; 441 442 if (pgtable_l5_enabled()) { 443 if (pgd_none(*pgd)) { 444 set_pgd(pgd, *pgd_k); 445 arch_flush_lazy_mmu_mode(); 446 } else { 447 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_k)); 448 } 449 } 450 451 /* With 4-level paging, copying happens on the p4d level. */ 452 p4d = p4d_offset(pgd, address); 453 p4d_k = p4d_offset(pgd_k, address); 454 if (p4d_none(*p4d_k)) 455 return -1; 456 457 if (p4d_none(*p4d) && !pgtable_l5_enabled()) { 458 set_p4d(p4d, *p4d_k); 459 arch_flush_lazy_mmu_mode(); 460 } else { 461 BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_k)); 462 } 463 464 BUILD_BUG_ON(CONFIG_PGTABLE_LEVELS < 4); 465 466 pud = pud_offset(p4d, address); 467 if (pud_none(*pud)) 468 return -1; 469 470 if (pud_large(*pud)) 471 return 0; 472 473 pmd = pmd_offset(pud, address); 474 if (pmd_none(*pmd)) 475 return -1; 476 477 if (pmd_large(*pmd)) 478 return 0; 479 480 pte = pte_offset_kernel(pmd, address); 481 if (!pte_present(*pte)) 482 return -1; 483 484 return 0; 485 } 486 NOKPROBE_SYMBOL(vmalloc_fault); 487 488 #ifdef CONFIG_CPU_SUP_AMD 489 static const char errata93_warning[] = 490 KERN_ERR 491 "******* Your BIOS seems to not contain a fix for K8 errata #93\n" 492 "******* Working around it, but it may cause SEGVs or burn power.\n" 493 "******* Please consider a BIOS update.\n" 494 "******* Disabling USB legacy in the BIOS may also help.\n"; 495 #endif 496 497 /* 498 * No vm86 mode in 64-bit mode: 499 */ 500 static inline void 501 check_v8086_mode(struct pt_regs *regs, unsigned long address, 502 struct task_struct *tsk) 503 { 504 } 505 506 static int bad_address(void *p) 507 { 508 unsigned long dummy; 509 510 return probe_kernel_address((unsigned long *)p, dummy); 511 } 512 513 static void dump_pagetable(unsigned long address) 514 { 515 pgd_t *base = __va(read_cr3_pa()); 516 pgd_t *pgd = base + pgd_index(address); 517 p4d_t *p4d; 518 pud_t *pud; 519 pmd_t *pmd; 520 pte_t *pte; 521 522 if (bad_address(pgd)) 523 goto bad; 524 525 pr_info("PGD %lx ", pgd_val(*pgd)); 526 527 if (!pgd_present(*pgd)) 528 goto out; 529 530 p4d = p4d_offset(pgd, address); 531 if (bad_address(p4d)) 532 goto bad; 533 534 pr_cont("P4D %lx ", p4d_val(*p4d)); 535 if (!p4d_present(*p4d) || p4d_large(*p4d)) 536 goto out; 537 538 pud = pud_offset(p4d, address); 539 if (bad_address(pud)) 540 goto bad; 541 542 pr_cont("PUD %lx ", pud_val(*pud)); 543 if (!pud_present(*pud) || pud_large(*pud)) 544 goto out; 545 546 pmd = pmd_offset(pud, address); 547 if (bad_address(pmd)) 548 goto bad; 549 550 pr_cont("PMD %lx ", pmd_val(*pmd)); 551 if (!pmd_present(*pmd) || pmd_large(*pmd)) 552 goto out; 553 554 pte = pte_offset_kernel(pmd, address); 555 if (bad_address(pte)) 556 goto bad; 557 558 pr_cont("PTE %lx", pte_val(*pte)); 559 out: 560 pr_cont("\n"); 561 return; 562 bad: 563 pr_info("BAD\n"); 564 } 565 566 #endif /* CONFIG_X86_64 */ 567 568 /* 569 * Workaround for K8 erratum #93 & buggy BIOS. 570 * 571 * BIOS SMM functions are required to use a specific workaround 572 * to avoid corruption of the 64bit RIP register on C stepping K8. 573 * 574 * A lot of BIOS that didn't get tested properly miss this. 575 * 576 * The OS sees this as a page fault with the upper 32bits of RIP cleared. 577 * Try to work around it here. 578 * 579 * Note we only handle faults in kernel here. 580 * Does nothing on 32-bit. 581 */ 582 static int is_errata93(struct pt_regs *regs, unsigned long address) 583 { 584 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) 585 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD 586 || boot_cpu_data.x86 != 0xf) 587 return 0; 588 589 if (address != regs->ip) 590 return 0; 591 592 if ((address >> 32) != 0) 593 return 0; 594 595 address |= 0xffffffffUL << 32; 596 if ((address >= (u64)_stext && address <= (u64)_etext) || 597 (address >= MODULES_VADDR && address <= MODULES_END)) { 598 printk_once(errata93_warning); 599 regs->ip = address; 600 return 1; 601 } 602 #endif 603 return 0; 604 } 605 606 /* 607 * Work around K8 erratum #100 K8 in compat mode occasionally jumps 608 * to illegal addresses >4GB. 609 * 610 * We catch this in the page fault handler because these addresses 611 * are not reachable. Just detect this case and return. Any code 612 * segment in LDT is compatibility mode. 613 */ 614 static int is_errata100(struct pt_regs *regs, unsigned long address) 615 { 616 #ifdef CONFIG_X86_64 617 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) 618 return 1; 619 #endif 620 return 0; 621 } 622 623 static int is_f00f_bug(struct pt_regs *regs, unsigned long address) 624 { 625 #ifdef CONFIG_X86_F00F_BUG 626 unsigned long nr; 627 628 /* 629 * Pentium F0 0F C7 C8 bug workaround: 630 */ 631 if (boot_cpu_has_bug(X86_BUG_F00F)) { 632 nr = (address - idt_descr.address) >> 3; 633 634 if (nr == 6) { 635 do_invalid_op(regs, 0); 636 return 1; 637 } 638 } 639 #endif 640 return 0; 641 } 642 643 static void 644 show_fault_oops(struct pt_regs *regs, unsigned long error_code, 645 unsigned long address) 646 { 647 if (!oops_may_print()) 648 return; 649 650 if (error_code & X86_PF_INSTR) { 651 unsigned int level; 652 pgd_t *pgd; 653 pte_t *pte; 654 655 pgd = __va(read_cr3_pa()); 656 pgd += pgd_index(address); 657 658 pte = lookup_address_in_pgd(pgd, address, &level); 659 660 if (pte && pte_present(*pte) && !pte_exec(*pte)) 661 pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n", 662 from_kuid(&init_user_ns, current_uid())); 663 if (pte && pte_present(*pte) && pte_exec(*pte) && 664 (pgd_flags(*pgd) & _PAGE_USER) && 665 (__read_cr4() & X86_CR4_SMEP)) 666 pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n", 667 from_kuid(&init_user_ns, current_uid())); 668 } 669 670 pr_alert("BUG: unable to handle kernel %s at %px\n", 671 address < PAGE_SIZE ? "NULL pointer dereference" : "paging request", 672 (void *)address); 673 674 dump_pagetable(address); 675 } 676 677 static noinline void 678 pgtable_bad(struct pt_regs *regs, unsigned long error_code, 679 unsigned long address) 680 { 681 struct task_struct *tsk; 682 unsigned long flags; 683 int sig; 684 685 flags = oops_begin(); 686 tsk = current; 687 sig = SIGKILL; 688 689 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", 690 tsk->comm, address); 691 dump_pagetable(address); 692 693 tsk->thread.cr2 = address; 694 tsk->thread.trap_nr = X86_TRAP_PF; 695 tsk->thread.error_code = error_code; 696 697 if (__die("Bad pagetable", regs, error_code)) 698 sig = 0; 699 700 oops_end(flags, regs, sig); 701 } 702 703 static noinline void 704 no_context(struct pt_regs *regs, unsigned long error_code, 705 unsigned long address, int signal, int si_code) 706 { 707 struct task_struct *tsk = current; 708 unsigned long flags; 709 int sig; 710 711 /* Are we prepared to handle this kernel fault? */ 712 if (fixup_exception(regs, X86_TRAP_PF)) { 713 /* 714 * Any interrupt that takes a fault gets the fixup. This makes 715 * the below recursive fault logic only apply to a faults from 716 * task context. 717 */ 718 if (in_interrupt()) 719 return; 720 721 /* 722 * Per the above we're !in_interrupt(), aka. task context. 723 * 724 * In this case we need to make sure we're not recursively 725 * faulting through the emulate_vsyscall() logic. 726 */ 727 if (current->thread.sig_on_uaccess_err && signal) { 728 tsk->thread.trap_nr = X86_TRAP_PF; 729 tsk->thread.error_code = error_code | X86_PF_USER; 730 tsk->thread.cr2 = address; 731 732 /* XXX: hwpoison faults will set the wrong code. */ 733 force_sig_info_fault(signal, si_code, address, 734 tsk, NULL, 0); 735 } 736 737 /* 738 * Barring that, we can do the fixup and be happy. 739 */ 740 return; 741 } 742 743 #ifdef CONFIG_VMAP_STACK 744 /* 745 * Stack overflow? During boot, we can fault near the initial 746 * stack in the direct map, but that's not an overflow -- check 747 * that we're in vmalloc space to avoid this. 748 */ 749 if (is_vmalloc_addr((void *)address) && 750 (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) || 751 address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) { 752 unsigned long stack = this_cpu_read(orig_ist.ist[DOUBLEFAULT_STACK]) - sizeof(void *); 753 /* 754 * We're likely to be running with very little stack space 755 * left. It's plausible that we'd hit this condition but 756 * double-fault even before we get this far, in which case 757 * we're fine: the double-fault handler will deal with it. 758 * 759 * We don't want to make it all the way into the oops code 760 * and then double-fault, though, because we're likely to 761 * break the console driver and lose most of the stack dump. 762 */ 763 asm volatile ("movq %[stack], %%rsp\n\t" 764 "call handle_stack_overflow\n\t" 765 "1: jmp 1b" 766 : ASM_CALL_CONSTRAINT 767 : "D" ("kernel stack overflow (page fault)"), 768 "S" (regs), "d" (address), 769 [stack] "rm" (stack)); 770 unreachable(); 771 } 772 #endif 773 774 /* 775 * 32-bit: 776 * 777 * Valid to do another page fault here, because if this fault 778 * had been triggered by is_prefetch fixup_exception would have 779 * handled it. 780 * 781 * 64-bit: 782 * 783 * Hall of shame of CPU/BIOS bugs. 784 */ 785 if (is_prefetch(regs, error_code, address)) 786 return; 787 788 if (is_errata93(regs, address)) 789 return; 790 791 /* 792 * Oops. The kernel tried to access some bad page. We'll have to 793 * terminate things with extreme prejudice: 794 */ 795 flags = oops_begin(); 796 797 show_fault_oops(regs, error_code, address); 798 799 if (task_stack_end_corrupted(tsk)) 800 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); 801 802 tsk->thread.cr2 = address; 803 tsk->thread.trap_nr = X86_TRAP_PF; 804 tsk->thread.error_code = error_code; 805 806 sig = SIGKILL; 807 if (__die("Oops", regs, error_code)) 808 sig = 0; 809 810 /* Executive summary in case the body of the oops scrolled away */ 811 printk(KERN_DEFAULT "CR2: %016lx\n", address); 812 813 oops_end(flags, regs, sig); 814 } 815 816 /* 817 * Print out info about fatal segfaults, if the show_unhandled_signals 818 * sysctl is set: 819 */ 820 static inline void 821 show_signal_msg(struct pt_regs *regs, unsigned long error_code, 822 unsigned long address, struct task_struct *tsk) 823 { 824 const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG; 825 826 if (!unhandled_signal(tsk, SIGSEGV)) 827 return; 828 829 if (!printk_ratelimit()) 830 return; 831 832 printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx", 833 loglvl, tsk->comm, task_pid_nr(tsk), address, 834 (void *)regs->ip, (void *)regs->sp, error_code); 835 836 print_vma_addr(KERN_CONT " in ", regs->ip); 837 838 printk(KERN_CONT "\n"); 839 840 show_opcodes(regs, loglvl); 841 } 842 843 static void 844 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 845 unsigned long address, u32 *pkey, int si_code) 846 { 847 struct task_struct *tsk = current; 848 849 /* User mode accesses just cause a SIGSEGV */ 850 if (error_code & X86_PF_USER) { 851 /* 852 * It's possible to have interrupts off here: 853 */ 854 local_irq_enable(); 855 856 /* 857 * Valid to do another page fault here because this one came 858 * from user space: 859 */ 860 if (is_prefetch(regs, error_code, address)) 861 return; 862 863 if (is_errata100(regs, address)) 864 return; 865 866 #ifdef CONFIG_X86_64 867 /* 868 * Instruction fetch faults in the vsyscall page might need 869 * emulation. 870 */ 871 if (unlikely((error_code & X86_PF_INSTR) && 872 ((address & ~0xfff) == VSYSCALL_ADDR))) { 873 if (emulate_vsyscall(regs, address)) 874 return; 875 } 876 #endif 877 878 /* 879 * To avoid leaking information about the kernel page table 880 * layout, pretend that user-mode accesses to kernel addresses 881 * are always protection faults. 882 */ 883 if (address >= TASK_SIZE_MAX) 884 error_code |= X86_PF_PROT; 885 886 if (likely(show_unhandled_signals)) 887 show_signal_msg(regs, error_code, address, tsk); 888 889 tsk->thread.cr2 = address; 890 tsk->thread.error_code = error_code; 891 tsk->thread.trap_nr = X86_TRAP_PF; 892 893 force_sig_info_fault(SIGSEGV, si_code, address, tsk, pkey, 0); 894 895 return; 896 } 897 898 if (is_f00f_bug(regs, address)) 899 return; 900 901 no_context(regs, error_code, address, SIGSEGV, si_code); 902 } 903 904 static noinline void 905 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 906 unsigned long address, u32 *pkey) 907 { 908 __bad_area_nosemaphore(regs, error_code, address, pkey, SEGV_MAPERR); 909 } 910 911 static void 912 __bad_area(struct pt_regs *regs, unsigned long error_code, 913 unsigned long address, struct vm_area_struct *vma, int si_code) 914 { 915 struct mm_struct *mm = current->mm; 916 u32 pkey; 917 918 if (vma) 919 pkey = vma_pkey(vma); 920 921 /* 922 * Something tried to access memory that isn't in our memory map.. 923 * Fix it, but check if it's kernel or user first.. 924 */ 925 up_read(&mm->mmap_sem); 926 927 __bad_area_nosemaphore(regs, error_code, address, 928 (vma) ? &pkey : NULL, si_code); 929 } 930 931 static noinline void 932 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) 933 { 934 __bad_area(regs, error_code, address, NULL, SEGV_MAPERR); 935 } 936 937 static inline bool bad_area_access_from_pkeys(unsigned long error_code, 938 struct vm_area_struct *vma) 939 { 940 /* This code is always called on the current mm */ 941 bool foreign = false; 942 943 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 944 return false; 945 if (error_code & X86_PF_PK) 946 return true; 947 /* this checks permission keys on the VMA: */ 948 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), 949 (error_code & X86_PF_INSTR), foreign)) 950 return true; 951 return false; 952 } 953 954 static noinline void 955 bad_area_access_error(struct pt_regs *regs, unsigned long error_code, 956 unsigned long address, struct vm_area_struct *vma) 957 { 958 /* 959 * This OSPKE check is not strictly necessary at runtime. 960 * But, doing it this way allows compiler optimizations 961 * if pkeys are compiled out. 962 */ 963 if (bad_area_access_from_pkeys(error_code, vma)) 964 __bad_area(regs, error_code, address, vma, SEGV_PKUERR); 965 else 966 __bad_area(regs, error_code, address, vma, SEGV_ACCERR); 967 } 968 969 static void 970 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, 971 u32 *pkey, unsigned int fault) 972 { 973 struct task_struct *tsk = current; 974 int code = BUS_ADRERR; 975 976 /* Kernel mode? Handle exceptions or die: */ 977 if (!(error_code & X86_PF_USER)) { 978 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); 979 return; 980 } 981 982 /* User-space => ok to do another page fault: */ 983 if (is_prefetch(regs, error_code, address)) 984 return; 985 986 tsk->thread.cr2 = address; 987 tsk->thread.error_code = error_code; 988 tsk->thread.trap_nr = X86_TRAP_PF; 989 990 #ifdef CONFIG_MEMORY_FAILURE 991 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { 992 printk(KERN_ERR 993 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", 994 tsk->comm, tsk->pid, address); 995 code = BUS_MCEERR_AR; 996 } 997 #endif 998 force_sig_info_fault(SIGBUS, code, address, tsk, pkey, fault); 999 } 1000 1001 static noinline void 1002 mm_fault_error(struct pt_regs *regs, unsigned long error_code, 1003 unsigned long address, u32 *pkey, vm_fault_t fault) 1004 { 1005 if (fatal_signal_pending(current) && !(error_code & X86_PF_USER)) { 1006 no_context(regs, error_code, address, 0, 0); 1007 return; 1008 } 1009 1010 if (fault & VM_FAULT_OOM) { 1011 /* Kernel mode? Handle exceptions or die: */ 1012 if (!(error_code & X86_PF_USER)) { 1013 no_context(regs, error_code, address, 1014 SIGSEGV, SEGV_MAPERR); 1015 return; 1016 } 1017 1018 /* 1019 * We ran out of memory, call the OOM killer, and return the 1020 * userspace (which will retry the fault, or kill us if we got 1021 * oom-killed): 1022 */ 1023 pagefault_out_of_memory(); 1024 } else { 1025 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| 1026 VM_FAULT_HWPOISON_LARGE)) 1027 do_sigbus(regs, error_code, address, pkey, fault); 1028 else if (fault & VM_FAULT_SIGSEGV) 1029 bad_area_nosemaphore(regs, error_code, address, pkey); 1030 else 1031 BUG(); 1032 } 1033 } 1034 1035 static int spurious_fault_check(unsigned long error_code, pte_t *pte) 1036 { 1037 if ((error_code & X86_PF_WRITE) && !pte_write(*pte)) 1038 return 0; 1039 1040 if ((error_code & X86_PF_INSTR) && !pte_exec(*pte)) 1041 return 0; 1042 /* 1043 * Note: We do not do lazy flushing on protection key 1044 * changes, so no spurious fault will ever set X86_PF_PK. 1045 */ 1046 if ((error_code & X86_PF_PK)) 1047 return 1; 1048 1049 return 1; 1050 } 1051 1052 /* 1053 * Handle a spurious fault caused by a stale TLB entry. 1054 * 1055 * This allows us to lazily refresh the TLB when increasing the 1056 * permissions of a kernel page (RO -> RW or NX -> X). Doing it 1057 * eagerly is very expensive since that implies doing a full 1058 * cross-processor TLB flush, even if no stale TLB entries exist 1059 * on other processors. 1060 * 1061 * Spurious faults may only occur if the TLB contains an entry with 1062 * fewer permission than the page table entry. Non-present (P = 0) 1063 * and reserved bit (R = 1) faults are never spurious. 1064 * 1065 * There are no security implications to leaving a stale TLB when 1066 * increasing the permissions on a page. 1067 * 1068 * Returns non-zero if a spurious fault was handled, zero otherwise. 1069 * 1070 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 1071 * (Optional Invalidation). 1072 */ 1073 static noinline int 1074 spurious_fault(unsigned long error_code, unsigned long address) 1075 { 1076 pgd_t *pgd; 1077 p4d_t *p4d; 1078 pud_t *pud; 1079 pmd_t *pmd; 1080 pte_t *pte; 1081 int ret; 1082 1083 /* 1084 * Only writes to RO or instruction fetches from NX may cause 1085 * spurious faults. 1086 * 1087 * These could be from user or supervisor accesses but the TLB 1088 * is only lazily flushed after a kernel mapping protection 1089 * change, so user accesses are not expected to cause spurious 1090 * faults. 1091 */ 1092 if (error_code != (X86_PF_WRITE | X86_PF_PROT) && 1093 error_code != (X86_PF_INSTR | X86_PF_PROT)) 1094 return 0; 1095 1096 pgd = init_mm.pgd + pgd_index(address); 1097 if (!pgd_present(*pgd)) 1098 return 0; 1099 1100 p4d = p4d_offset(pgd, address); 1101 if (!p4d_present(*p4d)) 1102 return 0; 1103 1104 if (p4d_large(*p4d)) 1105 return spurious_fault_check(error_code, (pte_t *) p4d); 1106 1107 pud = pud_offset(p4d, address); 1108 if (!pud_present(*pud)) 1109 return 0; 1110 1111 if (pud_large(*pud)) 1112 return spurious_fault_check(error_code, (pte_t *) pud); 1113 1114 pmd = pmd_offset(pud, address); 1115 if (!pmd_present(*pmd)) 1116 return 0; 1117 1118 if (pmd_large(*pmd)) 1119 return spurious_fault_check(error_code, (pte_t *) pmd); 1120 1121 pte = pte_offset_kernel(pmd, address); 1122 if (!pte_present(*pte)) 1123 return 0; 1124 1125 ret = spurious_fault_check(error_code, pte); 1126 if (!ret) 1127 return 0; 1128 1129 /* 1130 * Make sure we have permissions in PMD. 1131 * If not, then there's a bug in the page tables: 1132 */ 1133 ret = spurious_fault_check(error_code, (pte_t *) pmd); 1134 WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); 1135 1136 return ret; 1137 } 1138 NOKPROBE_SYMBOL(spurious_fault); 1139 1140 int show_unhandled_signals = 1; 1141 1142 static inline int 1143 access_error(unsigned long error_code, struct vm_area_struct *vma) 1144 { 1145 /* This is only called for the current mm, so: */ 1146 bool foreign = false; 1147 1148 /* 1149 * Read or write was blocked by protection keys. This is 1150 * always an unconditional error and can never result in 1151 * a follow-up action to resolve the fault, like a COW. 1152 */ 1153 if (error_code & X86_PF_PK) 1154 return 1; 1155 1156 /* 1157 * Make sure to check the VMA so that we do not perform 1158 * faults just to hit a X86_PF_PK as soon as we fill in a 1159 * page. 1160 */ 1161 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), 1162 (error_code & X86_PF_INSTR), foreign)) 1163 return 1; 1164 1165 if (error_code & X86_PF_WRITE) { 1166 /* write, present and write, not present: */ 1167 if (unlikely(!(vma->vm_flags & VM_WRITE))) 1168 return 1; 1169 return 0; 1170 } 1171 1172 /* read, present: */ 1173 if (unlikely(error_code & X86_PF_PROT)) 1174 return 1; 1175 1176 /* read, not present: */ 1177 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) 1178 return 1; 1179 1180 return 0; 1181 } 1182 1183 static int fault_in_kernel_space(unsigned long address) 1184 { 1185 return address >= TASK_SIZE_MAX; 1186 } 1187 1188 static inline bool smap_violation(int error_code, struct pt_regs *regs) 1189 { 1190 if (!IS_ENABLED(CONFIG_X86_SMAP)) 1191 return false; 1192 1193 if (!static_cpu_has(X86_FEATURE_SMAP)) 1194 return false; 1195 1196 if (error_code & X86_PF_USER) 1197 return false; 1198 1199 if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC)) 1200 return false; 1201 1202 return true; 1203 } 1204 1205 /* 1206 * This routine handles page faults. It determines the address, 1207 * and the problem, and then passes it off to one of the appropriate 1208 * routines. 1209 */ 1210 static noinline void 1211 __do_page_fault(struct pt_regs *regs, unsigned long error_code, 1212 unsigned long address) 1213 { 1214 struct vm_area_struct *vma; 1215 struct task_struct *tsk; 1216 struct mm_struct *mm; 1217 vm_fault_t fault, major = 0; 1218 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; 1219 u32 pkey; 1220 1221 tsk = current; 1222 mm = tsk->mm; 1223 1224 prefetchw(&mm->mmap_sem); 1225 1226 if (unlikely(kmmio_fault(regs, address))) 1227 return; 1228 1229 /* 1230 * We fault-in kernel-space virtual memory on-demand. The 1231 * 'reference' page table is init_mm.pgd. 1232 * 1233 * NOTE! We MUST NOT take any locks for this case. We may 1234 * be in an interrupt or a critical region, and should 1235 * only copy the information from the master page table, 1236 * nothing more. 1237 * 1238 * This verifies that the fault happens in kernel space 1239 * (error_code & 4) == 0, and that the fault was not a 1240 * protection error (error_code & 9) == 0. 1241 */ 1242 if (unlikely(fault_in_kernel_space(address))) { 1243 if (!(error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) { 1244 if (vmalloc_fault(address) >= 0) 1245 return; 1246 } 1247 1248 /* Can handle a stale RO->RW TLB: */ 1249 if (spurious_fault(error_code, address)) 1250 return; 1251 1252 /* kprobes don't want to hook the spurious faults: */ 1253 if (kprobes_fault(regs)) 1254 return; 1255 /* 1256 * Don't take the mm semaphore here. If we fixup a prefetch 1257 * fault we could otherwise deadlock: 1258 */ 1259 bad_area_nosemaphore(regs, error_code, address, NULL); 1260 1261 return; 1262 } 1263 1264 /* kprobes don't want to hook the spurious faults: */ 1265 if (unlikely(kprobes_fault(regs))) 1266 return; 1267 1268 if (unlikely(error_code & X86_PF_RSVD)) 1269 pgtable_bad(regs, error_code, address); 1270 1271 if (unlikely(smap_violation(error_code, regs))) { 1272 bad_area_nosemaphore(regs, error_code, address, NULL); 1273 return; 1274 } 1275 1276 /* 1277 * If we're in an interrupt, have no user context or are running 1278 * in a region with pagefaults disabled then we must not take the fault 1279 */ 1280 if (unlikely(faulthandler_disabled() || !mm)) { 1281 bad_area_nosemaphore(regs, error_code, address, NULL); 1282 return; 1283 } 1284 1285 /* 1286 * It's safe to allow irq's after cr2 has been saved and the 1287 * vmalloc fault has been handled. 1288 * 1289 * User-mode registers count as a user access even for any 1290 * potential system fault or CPU buglet: 1291 */ 1292 if (user_mode(regs)) { 1293 local_irq_enable(); 1294 error_code |= X86_PF_USER; 1295 flags |= FAULT_FLAG_USER; 1296 } else { 1297 if (regs->flags & X86_EFLAGS_IF) 1298 local_irq_enable(); 1299 } 1300 1301 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 1302 1303 if (error_code & X86_PF_WRITE) 1304 flags |= FAULT_FLAG_WRITE; 1305 if (error_code & X86_PF_INSTR) 1306 flags |= FAULT_FLAG_INSTRUCTION; 1307 1308 /* 1309 * When running in the kernel we expect faults to occur only to 1310 * addresses in user space. All other faults represent errors in 1311 * the kernel and should generate an OOPS. Unfortunately, in the 1312 * case of an erroneous fault occurring in a code path which already 1313 * holds mmap_sem we will deadlock attempting to validate the fault 1314 * against the address space. Luckily the kernel only validly 1315 * references user space from well defined areas of code, which are 1316 * listed in the exceptions table. 1317 * 1318 * As the vast majority of faults will be valid we will only perform 1319 * the source reference check when there is a possibility of a 1320 * deadlock. Attempt to lock the address space, if we cannot we then 1321 * validate the source. If this is invalid we can skip the address 1322 * space check, thus avoiding the deadlock: 1323 */ 1324 if (unlikely(!down_read_trylock(&mm->mmap_sem))) { 1325 if (!(error_code & X86_PF_USER) && 1326 !search_exception_tables(regs->ip)) { 1327 bad_area_nosemaphore(regs, error_code, address, NULL); 1328 return; 1329 } 1330 retry: 1331 down_read(&mm->mmap_sem); 1332 } else { 1333 /* 1334 * The above down_read_trylock() might have succeeded in 1335 * which case we'll have missed the might_sleep() from 1336 * down_read(): 1337 */ 1338 might_sleep(); 1339 } 1340 1341 vma = find_vma(mm, address); 1342 if (unlikely(!vma)) { 1343 bad_area(regs, error_code, address); 1344 return; 1345 } 1346 if (likely(vma->vm_start <= address)) 1347 goto good_area; 1348 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { 1349 bad_area(regs, error_code, address); 1350 return; 1351 } 1352 if (error_code & X86_PF_USER) { 1353 /* 1354 * Accessing the stack below %sp is always a bug. 1355 * The large cushion allows instructions like enter 1356 * and pusha to work. ("enter $65535, $31" pushes 1357 * 32 pointers and then decrements %sp by 65535.) 1358 */ 1359 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { 1360 bad_area(regs, error_code, address); 1361 return; 1362 } 1363 } 1364 if (unlikely(expand_stack(vma, address))) { 1365 bad_area(regs, error_code, address); 1366 return; 1367 } 1368 1369 /* 1370 * Ok, we have a good vm_area for this memory access, so 1371 * we can handle it.. 1372 */ 1373 good_area: 1374 if (unlikely(access_error(error_code, vma))) { 1375 bad_area_access_error(regs, error_code, address, vma); 1376 return; 1377 } 1378 1379 /* 1380 * If for any reason at all we couldn't handle the fault, 1381 * make sure we exit gracefully rather than endlessly redo 1382 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if 1383 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked. 1384 * 1385 * Note that handle_userfault() may also release and reacquire mmap_sem 1386 * (and not return with VM_FAULT_RETRY), when returning to userland to 1387 * repeat the page fault later with a VM_FAULT_NOPAGE retval 1388 * (potentially after handling any pending signal during the return to 1389 * userland). The return to userland is identified whenever 1390 * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags. 1391 * Thus we have to be careful about not touching vma after handling the 1392 * fault, so we read the pkey beforehand. 1393 */ 1394 pkey = vma_pkey(vma); 1395 fault = handle_mm_fault(vma, address, flags); 1396 major |= fault & VM_FAULT_MAJOR; 1397 1398 /* 1399 * If we need to retry the mmap_sem has already been released, 1400 * and if there is a fatal signal pending there is no guarantee 1401 * that we made any progress. Handle this case first. 1402 */ 1403 if (unlikely(fault & VM_FAULT_RETRY)) { 1404 /* Retry at most once */ 1405 if (flags & FAULT_FLAG_ALLOW_RETRY) { 1406 flags &= ~FAULT_FLAG_ALLOW_RETRY; 1407 flags |= FAULT_FLAG_TRIED; 1408 if (!fatal_signal_pending(tsk)) 1409 goto retry; 1410 } 1411 1412 /* User mode? Just return to handle the fatal exception */ 1413 if (flags & FAULT_FLAG_USER) 1414 return; 1415 1416 /* Not returning to user mode? Handle exceptions or die: */ 1417 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); 1418 return; 1419 } 1420 1421 up_read(&mm->mmap_sem); 1422 if (unlikely(fault & VM_FAULT_ERROR)) { 1423 mm_fault_error(regs, error_code, address, &pkey, fault); 1424 return; 1425 } 1426 1427 /* 1428 * Major/minor page fault accounting. If any of the events 1429 * returned VM_FAULT_MAJOR, we account it as a major fault. 1430 */ 1431 if (major) { 1432 tsk->maj_flt++; 1433 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); 1434 } else { 1435 tsk->min_flt++; 1436 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); 1437 } 1438 1439 check_v8086_mode(regs, address, tsk); 1440 } 1441 NOKPROBE_SYMBOL(__do_page_fault); 1442 1443 static nokprobe_inline void 1444 trace_page_fault_entries(unsigned long address, struct pt_regs *regs, 1445 unsigned long error_code) 1446 { 1447 if (user_mode(regs)) 1448 trace_page_fault_user(address, regs, error_code); 1449 else 1450 trace_page_fault_kernel(address, regs, error_code); 1451 } 1452 1453 /* 1454 * We must have this function blacklisted from kprobes, tagged with notrace 1455 * and call read_cr2() before calling anything else. To avoid calling any 1456 * kind of tracing machinery before we've observed the CR2 value. 1457 * 1458 * exception_{enter,exit}() contains all sorts of tracepoints. 1459 */ 1460 dotraplinkage void notrace 1461 do_page_fault(struct pt_regs *regs, unsigned long error_code) 1462 { 1463 unsigned long address = read_cr2(); /* Get the faulting address */ 1464 enum ctx_state prev_state; 1465 1466 prev_state = exception_enter(); 1467 if (trace_pagefault_enabled()) 1468 trace_page_fault_entries(address, regs, error_code); 1469 1470 __do_page_fault(regs, error_code, address); 1471 exception_exit(prev_state); 1472 } 1473 NOKPROBE_SYMBOL(do_page_fault); 1474