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 20 #include <asm/cpufeature.h> /* boot_cpu_has, ... */ 21 #include <asm/traps.h> /* dotraplinkage, ... */ 22 #include <asm/pgalloc.h> /* pgd_*(), ... */ 23 #include <asm/fixmap.h> /* VSYSCALL_ADDR */ 24 #include <asm/vsyscall.h> /* emulate_vsyscall */ 25 #include <asm/vm86.h> /* struct vm86 */ 26 #include <asm/mmu_context.h> /* vma_pkey() */ 27 28 #define CREATE_TRACE_POINTS 29 #include <asm/trace/exceptions.h> 30 31 /* 32 * Returns 0 if mmiotrace is disabled, or if the fault is not 33 * handled by mmiotrace: 34 */ 35 static nokprobe_inline int 36 kmmio_fault(struct pt_regs *regs, unsigned long addr) 37 { 38 if (unlikely(is_kmmio_active())) 39 if (kmmio_handler(regs, addr) == 1) 40 return -1; 41 return 0; 42 } 43 44 static nokprobe_inline int kprobes_fault(struct pt_regs *regs) 45 { 46 int ret = 0; 47 48 /* kprobe_running() needs smp_processor_id() */ 49 if (kprobes_built_in() && !user_mode(regs)) { 50 preempt_disable(); 51 if (kprobe_running() && kprobe_fault_handler(regs, 14)) 52 ret = 1; 53 preempt_enable(); 54 } 55 56 return ret; 57 } 58 59 /* 60 * Prefetch quirks: 61 * 62 * 32-bit mode: 63 * 64 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. 65 * Check that here and ignore it. 66 * 67 * 64-bit mode: 68 * 69 * Sometimes the CPU reports invalid exceptions on prefetch. 70 * Check that here and ignore it. 71 * 72 * Opcode checker based on code by Richard Brunner. 73 */ 74 static inline int 75 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, 76 unsigned char opcode, int *prefetch) 77 { 78 unsigned char instr_hi = opcode & 0xf0; 79 unsigned char instr_lo = opcode & 0x0f; 80 81 switch (instr_hi) { 82 case 0x20: 83 case 0x30: 84 /* 85 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. 86 * In X86_64 long mode, the CPU will signal invalid 87 * opcode if some of these prefixes are present so 88 * X86_64 will never get here anyway 89 */ 90 return ((instr_lo & 7) == 0x6); 91 #ifdef CONFIG_X86_64 92 case 0x40: 93 /* 94 * In AMD64 long mode 0x40..0x4F are valid REX prefixes 95 * Need to figure out under what instruction mode the 96 * instruction was issued. Could check the LDT for lm, 97 * but for now it's good enough to assume that long 98 * mode only uses well known segments or kernel. 99 */ 100 return (!user_mode(regs) || user_64bit_mode(regs)); 101 #endif 102 case 0x60: 103 /* 0x64 thru 0x67 are valid prefixes in all modes. */ 104 return (instr_lo & 0xC) == 0x4; 105 case 0xF0: 106 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ 107 return !instr_lo || (instr_lo>>1) == 1; 108 case 0x00: 109 /* Prefetch instruction is 0x0F0D or 0x0F18 */ 110 if (probe_kernel_address(instr, opcode)) 111 return 0; 112 113 *prefetch = (instr_lo == 0xF) && 114 (opcode == 0x0D || opcode == 0x18); 115 return 0; 116 default: 117 return 0; 118 } 119 } 120 121 static int 122 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) 123 { 124 unsigned char *max_instr; 125 unsigned char *instr; 126 int prefetch = 0; 127 128 /* 129 * If it was a exec (instruction fetch) fault on NX page, then 130 * do not ignore the fault: 131 */ 132 if (error_code & X86_PF_INSTR) 133 return 0; 134 135 instr = (void *)convert_ip_to_linear(current, regs); 136 max_instr = instr + 15; 137 138 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX) 139 return 0; 140 141 while (instr < max_instr) { 142 unsigned char opcode; 143 144 if (probe_kernel_address(instr, opcode)) 145 break; 146 147 instr++; 148 149 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) 150 break; 151 } 152 return prefetch; 153 } 154 155 /* 156 * A protection key fault means that the PKRU value did not allow 157 * access to some PTE. Userspace can figure out what PKRU was 158 * from the XSAVE state, and this function fills out a field in 159 * siginfo so userspace can discover which protection key was set 160 * on the PTE. 161 * 162 * If we get here, we know that the hardware signaled a X86_PF_PK 163 * fault and that there was a VMA once we got in the fault 164 * handler. It does *not* guarantee that the VMA we find here 165 * was the one that we faulted on. 166 * 167 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4); 168 * 2. T1 : set PKRU to deny access to pkey=4, touches page 169 * 3. T1 : faults... 170 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5); 171 * 5. T1 : enters fault handler, takes mmap_sem, etc... 172 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really 173 * faulted on a pte with its pkey=4. 174 */ 175 static void fill_sig_info_pkey(int si_signo, int si_code, siginfo_t *info, 176 u32 *pkey) 177 { 178 /* This is effectively an #ifdef */ 179 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 180 return; 181 182 /* Fault not from Protection Keys: nothing to do */ 183 if ((si_code != SEGV_PKUERR) || (si_signo != SIGSEGV)) 184 return; 185 /* 186 * force_sig_info_fault() is called from a number of 187 * contexts, some of which have a VMA and some of which 188 * do not. The X86_PF_PK handing happens after we have a 189 * valid VMA, so we should never reach this without a 190 * valid VMA. 191 */ 192 if (!pkey) { 193 WARN_ONCE(1, "PKU fault with no VMA passed in"); 194 info->si_pkey = 0; 195 return; 196 } 197 /* 198 * si_pkey should be thought of as a strong hint, but not 199 * absolutely guranteed to be 100% accurate because of 200 * the race explained above. 201 */ 202 info->si_pkey = *pkey; 203 } 204 205 static void 206 force_sig_info_fault(int si_signo, int si_code, unsigned long address, 207 struct task_struct *tsk, u32 *pkey, int fault) 208 { 209 unsigned lsb = 0; 210 siginfo_t info; 211 212 clear_siginfo(&info); 213 info.si_signo = si_signo; 214 info.si_errno = 0; 215 info.si_code = si_code; 216 info.si_addr = (void __user *)address; 217 if (fault & VM_FAULT_HWPOISON_LARGE) 218 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); 219 if (fault & VM_FAULT_HWPOISON) 220 lsb = PAGE_SHIFT; 221 info.si_addr_lsb = lsb; 222 223 fill_sig_info_pkey(si_signo, si_code, &info, pkey); 224 225 force_sig_info(si_signo, &info, tsk); 226 } 227 228 DEFINE_SPINLOCK(pgd_lock); 229 LIST_HEAD(pgd_list); 230 231 #ifdef CONFIG_X86_32 232 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) 233 { 234 unsigned index = pgd_index(address); 235 pgd_t *pgd_k; 236 p4d_t *p4d, *p4d_k; 237 pud_t *pud, *pud_k; 238 pmd_t *pmd, *pmd_k; 239 240 pgd += index; 241 pgd_k = init_mm.pgd + index; 242 243 if (!pgd_present(*pgd_k)) 244 return NULL; 245 246 /* 247 * set_pgd(pgd, *pgd_k); here would be useless on PAE 248 * and redundant with the set_pmd() on non-PAE. As would 249 * set_p4d/set_pud. 250 */ 251 p4d = p4d_offset(pgd, address); 252 p4d_k = p4d_offset(pgd_k, address); 253 if (!p4d_present(*p4d_k)) 254 return NULL; 255 256 pud = pud_offset(p4d, address); 257 pud_k = pud_offset(p4d_k, address); 258 if (!pud_present(*pud_k)) 259 return NULL; 260 261 pmd = pmd_offset(pud, address); 262 pmd_k = pmd_offset(pud_k, address); 263 if (!pmd_present(*pmd_k)) 264 return NULL; 265 266 if (!pmd_present(*pmd)) 267 set_pmd(pmd, *pmd_k); 268 else 269 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); 270 271 return pmd_k; 272 } 273 274 void vmalloc_sync_all(void) 275 { 276 unsigned long address; 277 278 if (SHARED_KERNEL_PMD) 279 return; 280 281 for (address = VMALLOC_START & PMD_MASK; 282 address >= TASK_SIZE_MAX && address < FIXADDR_TOP; 283 address += PMD_SIZE) { 284 struct page *page; 285 286 spin_lock(&pgd_lock); 287 list_for_each_entry(page, &pgd_list, lru) { 288 spinlock_t *pgt_lock; 289 pmd_t *ret; 290 291 /* the pgt_lock only for Xen */ 292 pgt_lock = &pgd_page_get_mm(page)->page_table_lock; 293 294 spin_lock(pgt_lock); 295 ret = vmalloc_sync_one(page_address(page), address); 296 spin_unlock(pgt_lock); 297 298 if (!ret) 299 break; 300 } 301 spin_unlock(&pgd_lock); 302 } 303 } 304 305 /* 306 * 32-bit: 307 * 308 * Handle a fault on the vmalloc or module mapping area 309 */ 310 static noinline int vmalloc_fault(unsigned long address) 311 { 312 unsigned long pgd_paddr; 313 pmd_t *pmd_k; 314 pte_t *pte_k; 315 316 /* Make sure we are in vmalloc area: */ 317 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 318 return -1; 319 320 WARN_ON_ONCE(in_nmi()); 321 322 /* 323 * Synchronize this task's top level page-table 324 * with the 'reference' page table. 325 * 326 * Do _not_ use "current" here. We might be inside 327 * an interrupt in the middle of a task switch.. 328 */ 329 pgd_paddr = read_cr3_pa(); 330 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); 331 if (!pmd_k) 332 return -1; 333 334 if (pmd_large(*pmd_k)) 335 return 0; 336 337 pte_k = pte_offset_kernel(pmd_k, address); 338 if (!pte_present(*pte_k)) 339 return -1; 340 341 return 0; 342 } 343 NOKPROBE_SYMBOL(vmalloc_fault); 344 345 /* 346 * Did it hit the DOS screen memory VA from vm86 mode? 347 */ 348 static inline void 349 check_v8086_mode(struct pt_regs *regs, unsigned long address, 350 struct task_struct *tsk) 351 { 352 #ifdef CONFIG_VM86 353 unsigned long bit; 354 355 if (!v8086_mode(regs) || !tsk->thread.vm86) 356 return; 357 358 bit = (address - 0xA0000) >> PAGE_SHIFT; 359 if (bit < 32) 360 tsk->thread.vm86->screen_bitmap |= 1 << bit; 361 #endif 362 } 363 364 static bool low_pfn(unsigned long pfn) 365 { 366 return pfn < max_low_pfn; 367 } 368 369 static void dump_pagetable(unsigned long address) 370 { 371 pgd_t *base = __va(read_cr3_pa()); 372 pgd_t *pgd = &base[pgd_index(address)]; 373 p4d_t *p4d; 374 pud_t *pud; 375 pmd_t *pmd; 376 pte_t *pte; 377 378 #ifdef CONFIG_X86_PAE 379 pr_info("*pdpt = %016Lx ", pgd_val(*pgd)); 380 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) 381 goto out; 382 #define pr_pde pr_cont 383 #else 384 #define pr_pde pr_info 385 #endif 386 p4d = p4d_offset(pgd, address); 387 pud = pud_offset(p4d, address); 388 pmd = pmd_offset(pud, address); 389 pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); 390 #undef pr_pde 391 392 /* 393 * We must not directly access the pte in the highpte 394 * case if the page table is located in highmem. 395 * And let's rather not kmap-atomic the pte, just in case 396 * it's allocated already: 397 */ 398 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) 399 goto out; 400 401 pte = pte_offset_kernel(pmd, address); 402 pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); 403 out: 404 pr_cont("\n"); 405 } 406 407 #else /* CONFIG_X86_64: */ 408 409 void vmalloc_sync_all(void) 410 { 411 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END); 412 } 413 414 /* 415 * 64-bit: 416 * 417 * Handle a fault on the vmalloc area 418 */ 419 static noinline int vmalloc_fault(unsigned long address) 420 { 421 pgd_t *pgd, *pgd_k; 422 p4d_t *p4d, *p4d_k; 423 pud_t *pud; 424 pmd_t *pmd; 425 pte_t *pte; 426 427 /* Make sure we are in vmalloc area: */ 428 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 429 return -1; 430 431 WARN_ON_ONCE(in_nmi()); 432 433 /* 434 * Copy kernel mappings over when needed. This can also 435 * happen within a race in page table update. In the later 436 * case just flush: 437 */ 438 pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address); 439 pgd_k = pgd_offset_k(address); 440 if (pgd_none(*pgd_k)) 441 return -1; 442 443 if (pgtable_l5_enabled()) { 444 if (pgd_none(*pgd)) { 445 set_pgd(pgd, *pgd_k); 446 arch_flush_lazy_mmu_mode(); 447 } else { 448 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_k)); 449 } 450 } 451 452 /* With 4-level paging, copying happens on the p4d level. */ 453 p4d = p4d_offset(pgd, address); 454 p4d_k = p4d_offset(pgd_k, address); 455 if (p4d_none(*p4d_k)) 456 return -1; 457 458 if (p4d_none(*p4d) && !pgtable_l5_enabled()) { 459 set_p4d(p4d, *p4d_k); 460 arch_flush_lazy_mmu_mode(); 461 } else { 462 BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_k)); 463 } 464 465 BUILD_BUG_ON(CONFIG_PGTABLE_LEVELS < 4); 466 467 pud = pud_offset(p4d, address); 468 if (pud_none(*pud)) 469 return -1; 470 471 if (pud_large(*pud)) 472 return 0; 473 474 pmd = pmd_offset(pud, address); 475 if (pmd_none(*pmd)) 476 return -1; 477 478 if (pmd_large(*pmd)) 479 return 0; 480 481 pte = pte_offset_kernel(pmd, address); 482 if (!pte_present(*pte)) 483 return -1; 484 485 return 0; 486 } 487 NOKPROBE_SYMBOL(vmalloc_fault); 488 489 #ifdef CONFIG_CPU_SUP_AMD 490 static const char errata93_warning[] = 491 KERN_ERR 492 "******* Your BIOS seems to not contain a fix for K8 errata #93\n" 493 "******* Working around it, but it may cause SEGVs or burn power.\n" 494 "******* Please consider a BIOS update.\n" 495 "******* Disabling USB legacy in the BIOS may also help.\n"; 496 #endif 497 498 /* 499 * No vm86 mode in 64-bit mode: 500 */ 501 static inline void 502 check_v8086_mode(struct pt_regs *regs, unsigned long address, 503 struct task_struct *tsk) 504 { 505 } 506 507 static int bad_address(void *p) 508 { 509 unsigned long dummy; 510 511 return probe_kernel_address((unsigned long *)p, dummy); 512 } 513 514 static void dump_pagetable(unsigned long address) 515 { 516 pgd_t *base = __va(read_cr3_pa()); 517 pgd_t *pgd = base + pgd_index(address); 518 p4d_t *p4d; 519 pud_t *pud; 520 pmd_t *pmd; 521 pte_t *pte; 522 523 if (bad_address(pgd)) 524 goto bad; 525 526 pr_info("PGD %lx ", pgd_val(*pgd)); 527 528 if (!pgd_present(*pgd)) 529 goto out; 530 531 p4d = p4d_offset(pgd, address); 532 if (bad_address(p4d)) 533 goto bad; 534 535 pr_cont("P4D %lx ", p4d_val(*p4d)); 536 if (!p4d_present(*p4d) || p4d_large(*p4d)) 537 goto out; 538 539 pud = pud_offset(p4d, address); 540 if (bad_address(pud)) 541 goto bad; 542 543 pr_cont("PUD %lx ", pud_val(*pud)); 544 if (!pud_present(*pud) || pud_large(*pud)) 545 goto out; 546 547 pmd = pmd_offset(pud, address); 548 if (bad_address(pmd)) 549 goto bad; 550 551 pr_cont("PMD %lx ", pmd_val(*pmd)); 552 if (!pmd_present(*pmd) || pmd_large(*pmd)) 553 goto out; 554 555 pte = pte_offset_kernel(pmd, address); 556 if (bad_address(pte)) 557 goto bad; 558 559 pr_cont("PTE %lx", pte_val(*pte)); 560 out: 561 pr_cont("\n"); 562 return; 563 bad: 564 pr_info("BAD\n"); 565 } 566 567 #endif /* CONFIG_X86_64 */ 568 569 /* 570 * Workaround for K8 erratum #93 & buggy BIOS. 571 * 572 * BIOS SMM functions are required to use a specific workaround 573 * to avoid corruption of the 64bit RIP register on C stepping K8. 574 * 575 * A lot of BIOS that didn't get tested properly miss this. 576 * 577 * The OS sees this as a page fault with the upper 32bits of RIP cleared. 578 * Try to work around it here. 579 * 580 * Note we only handle faults in kernel here. 581 * Does nothing on 32-bit. 582 */ 583 static int is_errata93(struct pt_regs *regs, unsigned long address) 584 { 585 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) 586 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD 587 || boot_cpu_data.x86 != 0xf) 588 return 0; 589 590 if (address != regs->ip) 591 return 0; 592 593 if ((address >> 32) != 0) 594 return 0; 595 596 address |= 0xffffffffUL << 32; 597 if ((address >= (u64)_stext && address <= (u64)_etext) || 598 (address >= MODULES_VADDR && address <= MODULES_END)) { 599 printk_once(errata93_warning); 600 regs->ip = address; 601 return 1; 602 } 603 #endif 604 return 0; 605 } 606 607 /* 608 * Work around K8 erratum #100 K8 in compat mode occasionally jumps 609 * to illegal addresses >4GB. 610 * 611 * We catch this in the page fault handler because these addresses 612 * are not reachable. Just detect this case and return. Any code 613 * segment in LDT is compatibility mode. 614 */ 615 static int is_errata100(struct pt_regs *regs, unsigned long address) 616 { 617 #ifdef CONFIG_X86_64 618 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) 619 return 1; 620 #endif 621 return 0; 622 } 623 624 static int is_f00f_bug(struct pt_regs *regs, unsigned long address) 625 { 626 #ifdef CONFIG_X86_F00F_BUG 627 unsigned long nr; 628 629 /* 630 * Pentium F0 0F C7 C8 bug workaround: 631 */ 632 if (boot_cpu_has_bug(X86_BUG_F00F)) { 633 nr = (address - idt_descr.address) >> 3; 634 635 if (nr == 6) { 636 do_invalid_op(regs, 0); 637 return 1; 638 } 639 } 640 #endif 641 return 0; 642 } 643 644 static void 645 show_fault_oops(struct pt_regs *regs, unsigned long error_code, 646 unsigned long address) 647 { 648 if (!oops_may_print()) 649 return; 650 651 if (error_code & X86_PF_INSTR) { 652 unsigned int level; 653 pgd_t *pgd; 654 pte_t *pte; 655 656 pgd = __va(read_cr3_pa()); 657 pgd += pgd_index(address); 658 659 pte = lookup_address_in_pgd(pgd, address, &level); 660 661 if (pte && pte_present(*pte) && !pte_exec(*pte)) 662 pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n", 663 from_kuid(&init_user_ns, current_uid())); 664 if (pte && pte_present(*pte) && pte_exec(*pte) && 665 (pgd_flags(*pgd) & _PAGE_USER) && 666 (__read_cr4() & X86_CR4_SMEP)) 667 pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n", 668 from_kuid(&init_user_ns, current_uid())); 669 } 670 671 pr_alert("BUG: unable to handle kernel %s at %px\n", 672 address < PAGE_SIZE ? "NULL pointer dereference" : "paging request", 673 (void *)address); 674 675 dump_pagetable(address); 676 } 677 678 static noinline void 679 pgtable_bad(struct pt_regs *regs, unsigned long error_code, 680 unsigned long address) 681 { 682 struct task_struct *tsk; 683 unsigned long flags; 684 int sig; 685 686 flags = oops_begin(); 687 tsk = current; 688 sig = SIGKILL; 689 690 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", 691 tsk->comm, address); 692 dump_pagetable(address); 693 694 tsk->thread.cr2 = address; 695 tsk->thread.trap_nr = X86_TRAP_PF; 696 tsk->thread.error_code = error_code; 697 698 if (__die("Bad pagetable", regs, error_code)) 699 sig = 0; 700 701 oops_end(flags, regs, sig); 702 } 703 704 static noinline void 705 no_context(struct pt_regs *regs, unsigned long error_code, 706 unsigned long address, int signal, int si_code) 707 { 708 struct task_struct *tsk = current; 709 unsigned long flags; 710 int sig; 711 712 /* Are we prepared to handle this kernel fault? */ 713 if (fixup_exception(regs, X86_TRAP_PF)) { 714 /* 715 * Any interrupt that takes a fault gets the fixup. This makes 716 * the below recursive fault logic only apply to a faults from 717 * task context. 718 */ 719 if (in_interrupt()) 720 return; 721 722 /* 723 * Per the above we're !in_interrupt(), aka. task context. 724 * 725 * In this case we need to make sure we're not recursively 726 * faulting through the emulate_vsyscall() logic. 727 */ 728 if (current->thread.sig_on_uaccess_err && signal) { 729 tsk->thread.trap_nr = X86_TRAP_PF; 730 tsk->thread.error_code = error_code | X86_PF_USER; 731 tsk->thread.cr2 = address; 732 733 /* XXX: hwpoison faults will set the wrong code. */ 734 force_sig_info_fault(signal, si_code, address, 735 tsk, NULL, 0); 736 } 737 738 /* 739 * Barring that, we can do the fixup and be happy. 740 */ 741 return; 742 } 743 744 #ifdef CONFIG_VMAP_STACK 745 /* 746 * Stack overflow? During boot, we can fault near the initial 747 * stack in the direct map, but that's not an overflow -- check 748 * that we're in vmalloc space to avoid this. 749 */ 750 if (is_vmalloc_addr((void *)address) && 751 (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) || 752 address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) { 753 unsigned long stack = this_cpu_read(orig_ist.ist[DOUBLEFAULT_STACK]) - sizeof(void *); 754 /* 755 * We're likely to be running with very little stack space 756 * left. It's plausible that we'd hit this condition but 757 * double-fault even before we get this far, in which case 758 * we're fine: the double-fault handler will deal with it. 759 * 760 * We don't want to make it all the way into the oops code 761 * and then double-fault, though, because we're likely to 762 * break the console driver and lose most of the stack dump. 763 */ 764 asm volatile ("movq %[stack], %%rsp\n\t" 765 "call handle_stack_overflow\n\t" 766 "1: jmp 1b" 767 : ASM_CALL_CONSTRAINT 768 : "D" ("kernel stack overflow (page fault)"), 769 "S" (regs), "d" (address), 770 [stack] "rm" (stack)); 771 unreachable(); 772 } 773 #endif 774 775 /* 776 * 32-bit: 777 * 778 * Valid to do another page fault here, because if this fault 779 * had been triggered by is_prefetch fixup_exception would have 780 * handled it. 781 * 782 * 64-bit: 783 * 784 * Hall of shame of CPU/BIOS bugs. 785 */ 786 if (is_prefetch(regs, error_code, address)) 787 return; 788 789 if (is_errata93(regs, address)) 790 return; 791 792 /* 793 * Oops. The kernel tried to access some bad page. We'll have to 794 * terminate things with extreme prejudice: 795 */ 796 flags = oops_begin(); 797 798 show_fault_oops(regs, error_code, address); 799 800 if (task_stack_end_corrupted(tsk)) 801 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); 802 803 tsk->thread.cr2 = address; 804 tsk->thread.trap_nr = X86_TRAP_PF; 805 tsk->thread.error_code = error_code; 806 807 sig = SIGKILL; 808 if (__die("Oops", regs, error_code)) 809 sig = 0; 810 811 /* Executive summary in case the body of the oops scrolled away */ 812 printk(KERN_DEFAULT "CR2: %016lx\n", address); 813 814 oops_end(flags, regs, sig); 815 } 816 817 /* 818 * Print out info about fatal segfaults, if the show_unhandled_signals 819 * sysctl is set: 820 */ 821 static inline void 822 show_signal_msg(struct pt_regs *regs, unsigned long error_code, 823 unsigned long address, struct task_struct *tsk) 824 { 825 const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG; 826 827 if (!unhandled_signal(tsk, SIGSEGV)) 828 return; 829 830 if (!printk_ratelimit()) 831 return; 832 833 printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx", 834 loglvl, tsk->comm, task_pid_nr(tsk), address, 835 (void *)regs->ip, (void *)regs->sp, error_code); 836 837 print_vma_addr(KERN_CONT " in ", regs->ip); 838 839 printk(KERN_CONT "\n"); 840 841 show_opcodes((u8 *)regs->ip, loglvl); 842 } 843 844 static void 845 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 846 unsigned long address, u32 *pkey, int si_code) 847 { 848 struct task_struct *tsk = current; 849 850 /* User mode accesses just cause a SIGSEGV */ 851 if (error_code & X86_PF_USER) { 852 /* 853 * It's possible to have interrupts off here: 854 */ 855 local_irq_enable(); 856 857 /* 858 * Valid to do another page fault here because this one came 859 * from user space: 860 */ 861 if (is_prefetch(regs, error_code, address)) 862 return; 863 864 if (is_errata100(regs, address)) 865 return; 866 867 #ifdef CONFIG_X86_64 868 /* 869 * Instruction fetch faults in the vsyscall page might need 870 * emulation. 871 */ 872 if (unlikely((error_code & X86_PF_INSTR) && 873 ((address & ~0xfff) == VSYSCALL_ADDR))) { 874 if (emulate_vsyscall(regs, address)) 875 return; 876 } 877 #endif 878 879 /* 880 * To avoid leaking information about the kernel page table 881 * layout, pretend that user-mode accesses to kernel addresses 882 * are always protection faults. 883 */ 884 if (address >= TASK_SIZE_MAX) 885 error_code |= X86_PF_PROT; 886 887 if (likely(show_unhandled_signals)) 888 show_signal_msg(regs, error_code, address, tsk); 889 890 tsk->thread.cr2 = address; 891 tsk->thread.error_code = error_code; 892 tsk->thread.trap_nr = X86_TRAP_PF; 893 894 force_sig_info_fault(SIGSEGV, si_code, address, tsk, pkey, 0); 895 896 return; 897 } 898 899 if (is_f00f_bug(regs, address)) 900 return; 901 902 no_context(regs, error_code, address, SIGSEGV, si_code); 903 } 904 905 static noinline void 906 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 907 unsigned long address, u32 *pkey) 908 { 909 __bad_area_nosemaphore(regs, error_code, address, pkey, SEGV_MAPERR); 910 } 911 912 static void 913 __bad_area(struct pt_regs *regs, unsigned long error_code, 914 unsigned long address, struct vm_area_struct *vma, int si_code) 915 { 916 struct mm_struct *mm = current->mm; 917 u32 pkey; 918 919 if (vma) 920 pkey = vma_pkey(vma); 921 922 /* 923 * Something tried to access memory that isn't in our memory map.. 924 * Fix it, but check if it's kernel or user first.. 925 */ 926 up_read(&mm->mmap_sem); 927 928 __bad_area_nosemaphore(regs, error_code, address, 929 (vma) ? &pkey : NULL, si_code); 930 } 931 932 static noinline void 933 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) 934 { 935 __bad_area(regs, error_code, address, NULL, SEGV_MAPERR); 936 } 937 938 static inline bool bad_area_access_from_pkeys(unsigned long error_code, 939 struct vm_area_struct *vma) 940 { 941 /* This code is always called on the current mm */ 942 bool foreign = false; 943 944 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 945 return false; 946 if (error_code & X86_PF_PK) 947 return true; 948 /* this checks permission keys on the VMA: */ 949 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), 950 (error_code & X86_PF_INSTR), foreign)) 951 return true; 952 return false; 953 } 954 955 static noinline void 956 bad_area_access_error(struct pt_regs *regs, unsigned long error_code, 957 unsigned long address, struct vm_area_struct *vma) 958 { 959 /* 960 * This OSPKE check is not strictly necessary at runtime. 961 * But, doing it this way allows compiler optimizations 962 * if pkeys are compiled out. 963 */ 964 if (bad_area_access_from_pkeys(error_code, vma)) 965 __bad_area(regs, error_code, address, vma, SEGV_PKUERR); 966 else 967 __bad_area(regs, error_code, address, vma, SEGV_ACCERR); 968 } 969 970 static void 971 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, 972 u32 *pkey, unsigned int fault) 973 { 974 struct task_struct *tsk = current; 975 int code = BUS_ADRERR; 976 977 /* Kernel mode? Handle exceptions or die: */ 978 if (!(error_code & X86_PF_USER)) { 979 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); 980 return; 981 } 982 983 /* User-space => ok to do another page fault: */ 984 if (is_prefetch(regs, error_code, address)) 985 return; 986 987 tsk->thread.cr2 = address; 988 tsk->thread.error_code = error_code; 989 tsk->thread.trap_nr = X86_TRAP_PF; 990 991 #ifdef CONFIG_MEMORY_FAILURE 992 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { 993 printk(KERN_ERR 994 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", 995 tsk->comm, tsk->pid, address); 996 code = BUS_MCEERR_AR; 997 } 998 #endif 999 force_sig_info_fault(SIGBUS, code, address, tsk, pkey, fault); 1000 } 1001 1002 static noinline void 1003 mm_fault_error(struct pt_regs *regs, unsigned long error_code, 1004 unsigned long address, u32 *pkey, unsigned int fault) 1005 { 1006 if (fatal_signal_pending(current) && !(error_code & X86_PF_USER)) { 1007 no_context(regs, error_code, address, 0, 0); 1008 return; 1009 } 1010 1011 if (fault & VM_FAULT_OOM) { 1012 /* Kernel mode? Handle exceptions or die: */ 1013 if (!(error_code & X86_PF_USER)) { 1014 no_context(regs, error_code, address, 1015 SIGSEGV, SEGV_MAPERR); 1016 return; 1017 } 1018 1019 /* 1020 * We ran out of memory, call the OOM killer, and return the 1021 * userspace (which will retry the fault, or kill us if we got 1022 * oom-killed): 1023 */ 1024 pagefault_out_of_memory(); 1025 } else { 1026 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| 1027 VM_FAULT_HWPOISON_LARGE)) 1028 do_sigbus(regs, error_code, address, pkey, fault); 1029 else if (fault & VM_FAULT_SIGSEGV) 1030 bad_area_nosemaphore(regs, error_code, address, pkey); 1031 else 1032 BUG(); 1033 } 1034 } 1035 1036 static int spurious_fault_check(unsigned long error_code, pte_t *pte) 1037 { 1038 if ((error_code & X86_PF_WRITE) && !pte_write(*pte)) 1039 return 0; 1040 1041 if ((error_code & X86_PF_INSTR) && !pte_exec(*pte)) 1042 return 0; 1043 /* 1044 * Note: We do not do lazy flushing on protection key 1045 * changes, so no spurious fault will ever set X86_PF_PK. 1046 */ 1047 if ((error_code & X86_PF_PK)) 1048 return 1; 1049 1050 return 1; 1051 } 1052 1053 /* 1054 * Handle a spurious fault caused by a stale TLB entry. 1055 * 1056 * This allows us to lazily refresh the TLB when increasing the 1057 * permissions of a kernel page (RO -> RW or NX -> X). Doing it 1058 * eagerly is very expensive since that implies doing a full 1059 * cross-processor TLB flush, even if no stale TLB entries exist 1060 * on other processors. 1061 * 1062 * Spurious faults may only occur if the TLB contains an entry with 1063 * fewer permission than the page table entry. Non-present (P = 0) 1064 * and reserved bit (R = 1) faults are never spurious. 1065 * 1066 * There are no security implications to leaving a stale TLB when 1067 * increasing the permissions on a page. 1068 * 1069 * Returns non-zero if a spurious fault was handled, zero otherwise. 1070 * 1071 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 1072 * (Optional Invalidation). 1073 */ 1074 static noinline int 1075 spurious_fault(unsigned long error_code, unsigned long address) 1076 { 1077 pgd_t *pgd; 1078 p4d_t *p4d; 1079 pud_t *pud; 1080 pmd_t *pmd; 1081 pte_t *pte; 1082 int ret; 1083 1084 /* 1085 * Only writes to RO or instruction fetches from NX may cause 1086 * spurious faults. 1087 * 1088 * These could be from user or supervisor accesses but the TLB 1089 * is only lazily flushed after a kernel mapping protection 1090 * change, so user accesses are not expected to cause spurious 1091 * faults. 1092 */ 1093 if (error_code != (X86_PF_WRITE | X86_PF_PROT) && 1094 error_code != (X86_PF_INSTR | X86_PF_PROT)) 1095 return 0; 1096 1097 pgd = init_mm.pgd + pgd_index(address); 1098 if (!pgd_present(*pgd)) 1099 return 0; 1100 1101 p4d = p4d_offset(pgd, address); 1102 if (!p4d_present(*p4d)) 1103 return 0; 1104 1105 if (p4d_large(*p4d)) 1106 return spurious_fault_check(error_code, (pte_t *) p4d); 1107 1108 pud = pud_offset(p4d, address); 1109 if (!pud_present(*pud)) 1110 return 0; 1111 1112 if (pud_large(*pud)) 1113 return spurious_fault_check(error_code, (pte_t *) pud); 1114 1115 pmd = pmd_offset(pud, address); 1116 if (!pmd_present(*pmd)) 1117 return 0; 1118 1119 if (pmd_large(*pmd)) 1120 return spurious_fault_check(error_code, (pte_t *) pmd); 1121 1122 pte = pte_offset_kernel(pmd, address); 1123 if (!pte_present(*pte)) 1124 return 0; 1125 1126 ret = spurious_fault_check(error_code, pte); 1127 if (!ret) 1128 return 0; 1129 1130 /* 1131 * Make sure we have permissions in PMD. 1132 * If not, then there's a bug in the page tables: 1133 */ 1134 ret = spurious_fault_check(error_code, (pte_t *) pmd); 1135 WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); 1136 1137 return ret; 1138 } 1139 NOKPROBE_SYMBOL(spurious_fault); 1140 1141 int show_unhandled_signals = 1; 1142 1143 static inline int 1144 access_error(unsigned long error_code, struct vm_area_struct *vma) 1145 { 1146 /* This is only called for the current mm, so: */ 1147 bool foreign = false; 1148 1149 /* 1150 * Read or write was blocked by protection keys. This is 1151 * always an unconditional error and can never result in 1152 * a follow-up action to resolve the fault, like a COW. 1153 */ 1154 if (error_code & X86_PF_PK) 1155 return 1; 1156 1157 /* 1158 * Make sure to check the VMA so that we do not perform 1159 * faults just to hit a X86_PF_PK as soon as we fill in a 1160 * page. 1161 */ 1162 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), 1163 (error_code & X86_PF_INSTR), foreign)) 1164 return 1; 1165 1166 if (error_code & X86_PF_WRITE) { 1167 /* write, present and write, not present: */ 1168 if (unlikely(!(vma->vm_flags & VM_WRITE))) 1169 return 1; 1170 return 0; 1171 } 1172 1173 /* read, present: */ 1174 if (unlikely(error_code & X86_PF_PROT)) 1175 return 1; 1176 1177 /* read, not present: */ 1178 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) 1179 return 1; 1180 1181 return 0; 1182 } 1183 1184 static int fault_in_kernel_space(unsigned long address) 1185 { 1186 return address >= TASK_SIZE_MAX; 1187 } 1188 1189 static inline bool smap_violation(int error_code, struct pt_regs *regs) 1190 { 1191 if (!IS_ENABLED(CONFIG_X86_SMAP)) 1192 return false; 1193 1194 if (!static_cpu_has(X86_FEATURE_SMAP)) 1195 return false; 1196 1197 if (error_code & X86_PF_USER) 1198 return false; 1199 1200 if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC)) 1201 return false; 1202 1203 return true; 1204 } 1205 1206 /* 1207 * This routine handles page faults. It determines the address, 1208 * and the problem, and then passes it off to one of the appropriate 1209 * routines. 1210 */ 1211 static noinline void 1212 __do_page_fault(struct pt_regs *regs, unsigned long error_code, 1213 unsigned long address) 1214 { 1215 struct vm_area_struct *vma; 1216 struct task_struct *tsk; 1217 struct mm_struct *mm; 1218 int fault, major = 0; 1219 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; 1220 u32 pkey; 1221 1222 tsk = current; 1223 mm = tsk->mm; 1224 1225 prefetchw(&mm->mmap_sem); 1226 1227 if (unlikely(kmmio_fault(regs, address))) 1228 return; 1229 1230 /* 1231 * We fault-in kernel-space virtual memory on-demand. The 1232 * 'reference' page table is init_mm.pgd. 1233 * 1234 * NOTE! We MUST NOT take any locks for this case. We may 1235 * be in an interrupt or a critical region, and should 1236 * only copy the information from the master page table, 1237 * nothing more. 1238 * 1239 * This verifies that the fault happens in kernel space 1240 * (error_code & 4) == 0, and that the fault was not a 1241 * protection error (error_code & 9) == 0. 1242 */ 1243 if (unlikely(fault_in_kernel_space(address))) { 1244 if (!(error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) { 1245 if (vmalloc_fault(address) >= 0) 1246 return; 1247 } 1248 1249 /* Can handle a stale RO->RW TLB: */ 1250 if (spurious_fault(error_code, address)) 1251 return; 1252 1253 /* kprobes don't want to hook the spurious faults: */ 1254 if (kprobes_fault(regs)) 1255 return; 1256 /* 1257 * Don't take the mm semaphore here. If we fixup a prefetch 1258 * fault we could otherwise deadlock: 1259 */ 1260 bad_area_nosemaphore(regs, error_code, address, NULL); 1261 1262 return; 1263 } 1264 1265 /* kprobes don't want to hook the spurious faults: */ 1266 if (unlikely(kprobes_fault(regs))) 1267 return; 1268 1269 if (unlikely(error_code & X86_PF_RSVD)) 1270 pgtable_bad(regs, error_code, address); 1271 1272 if (unlikely(smap_violation(error_code, regs))) { 1273 bad_area_nosemaphore(regs, error_code, address, NULL); 1274 return; 1275 } 1276 1277 /* 1278 * If we're in an interrupt, have no user context or are running 1279 * in a region with pagefaults disabled then we must not take the fault 1280 */ 1281 if (unlikely(faulthandler_disabled() || !mm)) { 1282 bad_area_nosemaphore(regs, error_code, address, NULL); 1283 return; 1284 } 1285 1286 /* 1287 * It's safe to allow irq's after cr2 has been saved and the 1288 * vmalloc fault has been handled. 1289 * 1290 * User-mode registers count as a user access even for any 1291 * potential system fault or CPU buglet: 1292 */ 1293 if (user_mode(regs)) { 1294 local_irq_enable(); 1295 error_code |= X86_PF_USER; 1296 flags |= FAULT_FLAG_USER; 1297 } else { 1298 if (regs->flags & X86_EFLAGS_IF) 1299 local_irq_enable(); 1300 } 1301 1302 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 1303 1304 if (error_code & X86_PF_WRITE) 1305 flags |= FAULT_FLAG_WRITE; 1306 if (error_code & X86_PF_INSTR) 1307 flags |= FAULT_FLAG_INSTRUCTION; 1308 1309 /* 1310 * When running in the kernel we expect faults to occur only to 1311 * addresses in user space. All other faults represent errors in 1312 * the kernel and should generate an OOPS. Unfortunately, in the 1313 * case of an erroneous fault occurring in a code path which already 1314 * holds mmap_sem we will deadlock attempting to validate the fault 1315 * against the address space. Luckily the kernel only validly 1316 * references user space from well defined areas of code, which are 1317 * listed in the exceptions table. 1318 * 1319 * As the vast majority of faults will be valid we will only perform 1320 * the source reference check when there is a possibility of a 1321 * deadlock. Attempt to lock the address space, if we cannot we then 1322 * validate the source. If this is invalid we can skip the address 1323 * space check, thus avoiding the deadlock: 1324 */ 1325 if (unlikely(!down_read_trylock(&mm->mmap_sem))) { 1326 if (!(error_code & X86_PF_USER) && 1327 !search_exception_tables(regs->ip)) { 1328 bad_area_nosemaphore(regs, error_code, address, NULL); 1329 return; 1330 } 1331 retry: 1332 down_read(&mm->mmap_sem); 1333 } else { 1334 /* 1335 * The above down_read_trylock() might have succeeded in 1336 * which case we'll have missed the might_sleep() from 1337 * down_read(): 1338 */ 1339 might_sleep(); 1340 } 1341 1342 vma = find_vma(mm, address); 1343 if (unlikely(!vma)) { 1344 bad_area(regs, error_code, address); 1345 return; 1346 } 1347 if (likely(vma->vm_start <= address)) 1348 goto good_area; 1349 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { 1350 bad_area(regs, error_code, address); 1351 return; 1352 } 1353 if (error_code & X86_PF_USER) { 1354 /* 1355 * Accessing the stack below %sp is always a bug. 1356 * The large cushion allows instructions like enter 1357 * and pusha to work. ("enter $65535, $31" pushes 1358 * 32 pointers and then decrements %sp by 65535.) 1359 */ 1360 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { 1361 bad_area(regs, error_code, address); 1362 return; 1363 } 1364 } 1365 if (unlikely(expand_stack(vma, address))) { 1366 bad_area(regs, error_code, address); 1367 return; 1368 } 1369 1370 /* 1371 * Ok, we have a good vm_area for this memory access, so 1372 * we can handle it.. 1373 */ 1374 good_area: 1375 if (unlikely(access_error(error_code, vma))) { 1376 bad_area_access_error(regs, error_code, address, vma); 1377 return; 1378 } 1379 1380 /* 1381 * If for any reason at all we couldn't handle the fault, 1382 * make sure we exit gracefully rather than endlessly redo 1383 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if 1384 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked. 1385 * 1386 * Note that handle_userfault() may also release and reacquire mmap_sem 1387 * (and not return with VM_FAULT_RETRY), when returning to userland to 1388 * repeat the page fault later with a VM_FAULT_NOPAGE retval 1389 * (potentially after handling any pending signal during the return to 1390 * userland). The return to userland is identified whenever 1391 * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags. 1392 * Thus we have to be careful about not touching vma after handling the 1393 * fault, so we read the pkey beforehand. 1394 */ 1395 pkey = vma_pkey(vma); 1396 fault = handle_mm_fault(vma, address, flags); 1397 major |= fault & VM_FAULT_MAJOR; 1398 1399 /* 1400 * If we need to retry the mmap_sem has already been released, 1401 * and if there is a fatal signal pending there is no guarantee 1402 * that we made any progress. Handle this case first. 1403 */ 1404 if (unlikely(fault & VM_FAULT_RETRY)) { 1405 /* Retry at most once */ 1406 if (flags & FAULT_FLAG_ALLOW_RETRY) { 1407 flags &= ~FAULT_FLAG_ALLOW_RETRY; 1408 flags |= FAULT_FLAG_TRIED; 1409 if (!fatal_signal_pending(tsk)) 1410 goto retry; 1411 } 1412 1413 /* User mode? Just return to handle the fatal exception */ 1414 if (flags & FAULT_FLAG_USER) 1415 return; 1416 1417 /* Not returning to user mode? Handle exceptions or die: */ 1418 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); 1419 return; 1420 } 1421 1422 up_read(&mm->mmap_sem); 1423 if (unlikely(fault & VM_FAULT_ERROR)) { 1424 mm_fault_error(regs, error_code, address, &pkey, fault); 1425 return; 1426 } 1427 1428 /* 1429 * Major/minor page fault accounting. If any of the events 1430 * returned VM_FAULT_MAJOR, we account it as a major fault. 1431 */ 1432 if (major) { 1433 tsk->maj_flt++; 1434 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); 1435 } else { 1436 tsk->min_flt++; 1437 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); 1438 } 1439 1440 check_v8086_mode(regs, address, tsk); 1441 } 1442 NOKPROBE_SYMBOL(__do_page_fault); 1443 1444 static nokprobe_inline void 1445 trace_page_fault_entries(unsigned long address, struct pt_regs *regs, 1446 unsigned long error_code) 1447 { 1448 if (user_mode(regs)) 1449 trace_page_fault_user(address, regs, error_code); 1450 else 1451 trace_page_fault_kernel(address, regs, error_code); 1452 } 1453 1454 /* 1455 * We must have this function blacklisted from kprobes, tagged with notrace 1456 * and call read_cr2() before calling anything else. To avoid calling any 1457 * kind of tracing machinery before we've observed the CR2 value. 1458 * 1459 * exception_{enter,exit}() contains all sorts of tracepoints. 1460 */ 1461 dotraplinkage void notrace 1462 do_page_fault(struct pt_regs *regs, unsigned long error_code) 1463 { 1464 unsigned long address = read_cr2(); /* Get the faulting address */ 1465 enum ctx_state prev_state; 1466 1467 prev_state = exception_enter(); 1468 if (trace_pagefault_enabled()) 1469 trace_page_fault_entries(address, regs, error_code); 1470 1471 __do_page_fault(regs, error_code, address); 1472 exception_exit(prev_state); 1473 } 1474 NOKPROBE_SYMBOL(do_page_fault); 1475