1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 1995 Linus Torvalds 4 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs. 5 * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar 6 */ 7 #include <linux/sched.h> /* test_thread_flag(), ... */ 8 #include <linux/sched/task_stack.h> /* task_stack_*(), ... */ 9 #include <linux/kdebug.h> /* oops_begin/end, ... */ 10 #include <linux/extable.h> /* search_exception_tables */ 11 #include <linux/memblock.h> /* max_low_pfn */ 12 #include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */ 13 #include <linux/mmiotrace.h> /* kmmio_handler, ... */ 14 #include <linux/perf_event.h> /* perf_sw_event */ 15 #include <linux/hugetlb.h> /* hstate_index_to_shift */ 16 #include <linux/prefetch.h> /* prefetchw */ 17 #include <linux/context_tracking.h> /* exception_enter(), ... */ 18 #include <linux/uaccess.h> /* faulthandler_disabled() */ 19 #include <linux/efi.h> /* efi_recover_from_page_fault()*/ 20 #include <linux/mm_types.h> 21 22 #include <asm/cpufeature.h> /* boot_cpu_has, ... */ 23 #include <asm/traps.h> /* dotraplinkage, ... */ 24 #include <asm/pgalloc.h> /* pgd_*(), ... */ 25 #include <asm/fixmap.h> /* VSYSCALL_ADDR */ 26 #include <asm/vsyscall.h> /* emulate_vsyscall */ 27 #include <asm/vm86.h> /* struct vm86 */ 28 #include <asm/mmu_context.h> /* vma_pkey() */ 29 #include <asm/efi.h> /* efi_recover_from_page_fault()*/ 30 #include <asm/desc.h> /* store_idt(), ... */ 31 #include <asm/cpu_entry_area.h> /* exception stack */ 32 #include <asm/pgtable_areas.h> /* VMALLOC_START, ... */ 33 34 #define CREATE_TRACE_POINTS 35 #include <asm/trace/exceptions.h> 36 37 /* 38 * Returns 0 if mmiotrace is disabled, or if the fault is not 39 * handled by mmiotrace: 40 */ 41 static nokprobe_inline int 42 kmmio_fault(struct pt_regs *regs, unsigned long addr) 43 { 44 if (unlikely(is_kmmio_active())) 45 if (kmmio_handler(regs, addr) == 1) 46 return -1; 47 return 0; 48 } 49 50 /* 51 * Prefetch quirks: 52 * 53 * 32-bit mode: 54 * 55 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. 56 * Check that here and ignore it. 57 * 58 * 64-bit mode: 59 * 60 * Sometimes the CPU reports invalid exceptions on prefetch. 61 * Check that here and ignore it. 62 * 63 * Opcode checker based on code by Richard Brunner. 64 */ 65 static inline int 66 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, 67 unsigned char opcode, int *prefetch) 68 { 69 unsigned char instr_hi = opcode & 0xf0; 70 unsigned char instr_lo = opcode & 0x0f; 71 72 switch (instr_hi) { 73 case 0x20: 74 case 0x30: 75 /* 76 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. 77 * In X86_64 long mode, the CPU will signal invalid 78 * opcode if some of these prefixes are present so 79 * X86_64 will never get here anyway 80 */ 81 return ((instr_lo & 7) == 0x6); 82 #ifdef CONFIG_X86_64 83 case 0x40: 84 /* 85 * In AMD64 long mode 0x40..0x4F are valid REX prefixes 86 * Need to figure out under what instruction mode the 87 * instruction was issued. Could check the LDT for lm, 88 * but for now it's good enough to assume that long 89 * mode only uses well known segments or kernel. 90 */ 91 return (!user_mode(regs) || user_64bit_mode(regs)); 92 #endif 93 case 0x60: 94 /* 0x64 thru 0x67 are valid prefixes in all modes. */ 95 return (instr_lo & 0xC) == 0x4; 96 case 0xF0: 97 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ 98 return !instr_lo || (instr_lo>>1) == 1; 99 case 0x00: 100 /* Prefetch instruction is 0x0F0D or 0x0F18 */ 101 if (probe_kernel_address(instr, opcode)) 102 return 0; 103 104 *prefetch = (instr_lo == 0xF) && 105 (opcode == 0x0D || opcode == 0x18); 106 return 0; 107 default: 108 return 0; 109 } 110 } 111 112 static int 113 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) 114 { 115 unsigned char *max_instr; 116 unsigned char *instr; 117 int prefetch = 0; 118 119 /* 120 * If it was a exec (instruction fetch) fault on NX page, then 121 * do not ignore the fault: 122 */ 123 if (error_code & X86_PF_INSTR) 124 return 0; 125 126 instr = (void *)convert_ip_to_linear(current, regs); 127 max_instr = instr + 15; 128 129 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX) 130 return 0; 131 132 while (instr < max_instr) { 133 unsigned char opcode; 134 135 if (probe_kernel_address(instr, opcode)) 136 break; 137 138 instr++; 139 140 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) 141 break; 142 } 143 return prefetch; 144 } 145 146 DEFINE_SPINLOCK(pgd_lock); 147 LIST_HEAD(pgd_list); 148 149 #ifdef CONFIG_X86_32 150 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) 151 { 152 unsigned index = pgd_index(address); 153 pgd_t *pgd_k; 154 p4d_t *p4d, *p4d_k; 155 pud_t *pud, *pud_k; 156 pmd_t *pmd, *pmd_k; 157 158 pgd += index; 159 pgd_k = init_mm.pgd + index; 160 161 if (!pgd_present(*pgd_k)) 162 return NULL; 163 164 /* 165 * set_pgd(pgd, *pgd_k); here would be useless on PAE 166 * and redundant with the set_pmd() on non-PAE. As would 167 * set_p4d/set_pud. 168 */ 169 p4d = p4d_offset(pgd, address); 170 p4d_k = p4d_offset(pgd_k, address); 171 if (!p4d_present(*p4d_k)) 172 return NULL; 173 174 pud = pud_offset(p4d, address); 175 pud_k = pud_offset(p4d_k, address); 176 if (!pud_present(*pud_k)) 177 return NULL; 178 179 pmd = pmd_offset(pud, address); 180 pmd_k = pmd_offset(pud_k, address); 181 182 if (pmd_present(*pmd) != pmd_present(*pmd_k)) 183 set_pmd(pmd, *pmd_k); 184 185 if (!pmd_present(*pmd_k)) 186 return NULL; 187 else 188 BUG_ON(pmd_pfn(*pmd) != pmd_pfn(*pmd_k)); 189 190 return pmd_k; 191 } 192 193 void vmalloc_sync_all(void) 194 { 195 unsigned long address; 196 197 if (SHARED_KERNEL_PMD) 198 return; 199 200 for (address = VMALLOC_START & PMD_MASK; 201 address >= TASK_SIZE_MAX && address < VMALLOC_END; 202 address += PMD_SIZE) { 203 struct page *page; 204 205 spin_lock(&pgd_lock); 206 list_for_each_entry(page, &pgd_list, lru) { 207 spinlock_t *pgt_lock; 208 209 /* the pgt_lock only for Xen */ 210 pgt_lock = &pgd_page_get_mm(page)->page_table_lock; 211 212 spin_lock(pgt_lock); 213 vmalloc_sync_one(page_address(page), address); 214 spin_unlock(pgt_lock); 215 } 216 spin_unlock(&pgd_lock); 217 } 218 } 219 220 /* 221 * 32-bit: 222 * 223 * Handle a fault on the vmalloc or module mapping area 224 */ 225 static noinline int vmalloc_fault(unsigned long address) 226 { 227 unsigned long pgd_paddr; 228 pmd_t *pmd_k; 229 pte_t *pte_k; 230 231 /* Make sure we are in vmalloc area: */ 232 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 233 return -1; 234 235 /* 236 * Synchronize this task's top level page-table 237 * with the 'reference' page table. 238 * 239 * Do _not_ use "current" here. We might be inside 240 * an interrupt in the middle of a task switch.. 241 */ 242 pgd_paddr = read_cr3_pa(); 243 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); 244 if (!pmd_k) 245 return -1; 246 247 if (pmd_large(*pmd_k)) 248 return 0; 249 250 pte_k = pte_offset_kernel(pmd_k, address); 251 if (!pte_present(*pte_k)) 252 return -1; 253 254 return 0; 255 } 256 NOKPROBE_SYMBOL(vmalloc_fault); 257 258 /* 259 * Did it hit the DOS screen memory VA from vm86 mode? 260 */ 261 static inline void 262 check_v8086_mode(struct pt_regs *regs, unsigned long address, 263 struct task_struct *tsk) 264 { 265 #ifdef CONFIG_VM86 266 unsigned long bit; 267 268 if (!v8086_mode(regs) || !tsk->thread.vm86) 269 return; 270 271 bit = (address - 0xA0000) >> PAGE_SHIFT; 272 if (bit < 32) 273 tsk->thread.vm86->screen_bitmap |= 1 << bit; 274 #endif 275 } 276 277 static bool low_pfn(unsigned long pfn) 278 { 279 return pfn < max_low_pfn; 280 } 281 282 static void dump_pagetable(unsigned long address) 283 { 284 pgd_t *base = __va(read_cr3_pa()); 285 pgd_t *pgd = &base[pgd_index(address)]; 286 p4d_t *p4d; 287 pud_t *pud; 288 pmd_t *pmd; 289 pte_t *pte; 290 291 #ifdef CONFIG_X86_PAE 292 pr_info("*pdpt = %016Lx ", pgd_val(*pgd)); 293 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) 294 goto out; 295 #define pr_pde pr_cont 296 #else 297 #define pr_pde pr_info 298 #endif 299 p4d = p4d_offset(pgd, address); 300 pud = pud_offset(p4d, address); 301 pmd = pmd_offset(pud, address); 302 pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); 303 #undef pr_pde 304 305 /* 306 * We must not directly access the pte in the highpte 307 * case if the page table is located in highmem. 308 * And let's rather not kmap-atomic the pte, just in case 309 * it's allocated already: 310 */ 311 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) 312 goto out; 313 314 pte = pte_offset_kernel(pmd, address); 315 pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); 316 out: 317 pr_cont("\n"); 318 } 319 320 #else /* CONFIG_X86_64: */ 321 322 void vmalloc_sync_all(void) 323 { 324 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END); 325 } 326 327 /* 328 * 64-bit: 329 * 330 * Handle a fault on the vmalloc area 331 */ 332 static noinline int vmalloc_fault(unsigned long address) 333 { 334 pgd_t *pgd, *pgd_k; 335 p4d_t *p4d, *p4d_k; 336 pud_t *pud; 337 pmd_t *pmd; 338 pte_t *pte; 339 340 /* Make sure we are in vmalloc area: */ 341 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 342 return -1; 343 344 /* 345 * Copy kernel mappings over when needed. This can also 346 * happen within a race in page table update. In the later 347 * case just flush: 348 */ 349 pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address); 350 pgd_k = pgd_offset_k(address); 351 if (pgd_none(*pgd_k)) 352 return -1; 353 354 if (pgtable_l5_enabled()) { 355 if (pgd_none(*pgd)) { 356 set_pgd(pgd, *pgd_k); 357 arch_flush_lazy_mmu_mode(); 358 } else { 359 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_k)); 360 } 361 } 362 363 /* With 4-level paging, copying happens on the p4d level. */ 364 p4d = p4d_offset(pgd, address); 365 p4d_k = p4d_offset(pgd_k, address); 366 if (p4d_none(*p4d_k)) 367 return -1; 368 369 if (p4d_none(*p4d) && !pgtable_l5_enabled()) { 370 set_p4d(p4d, *p4d_k); 371 arch_flush_lazy_mmu_mode(); 372 } else { 373 BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_k)); 374 } 375 376 BUILD_BUG_ON(CONFIG_PGTABLE_LEVELS < 4); 377 378 pud = pud_offset(p4d, address); 379 if (pud_none(*pud)) 380 return -1; 381 382 if (pud_large(*pud)) 383 return 0; 384 385 pmd = pmd_offset(pud, address); 386 if (pmd_none(*pmd)) 387 return -1; 388 389 if (pmd_large(*pmd)) 390 return 0; 391 392 pte = pte_offset_kernel(pmd, address); 393 if (!pte_present(*pte)) 394 return -1; 395 396 return 0; 397 } 398 NOKPROBE_SYMBOL(vmalloc_fault); 399 400 #ifdef CONFIG_CPU_SUP_AMD 401 static const char errata93_warning[] = 402 KERN_ERR 403 "******* Your BIOS seems to not contain a fix for K8 errata #93\n" 404 "******* Working around it, but it may cause SEGVs or burn power.\n" 405 "******* Please consider a BIOS update.\n" 406 "******* Disabling USB legacy in the BIOS may also help.\n"; 407 #endif 408 409 /* 410 * No vm86 mode in 64-bit mode: 411 */ 412 static inline void 413 check_v8086_mode(struct pt_regs *regs, unsigned long address, 414 struct task_struct *tsk) 415 { 416 } 417 418 static int bad_address(void *p) 419 { 420 unsigned long dummy; 421 422 return probe_kernel_address((unsigned long *)p, dummy); 423 } 424 425 static void dump_pagetable(unsigned long address) 426 { 427 pgd_t *base = __va(read_cr3_pa()); 428 pgd_t *pgd = base + pgd_index(address); 429 p4d_t *p4d; 430 pud_t *pud; 431 pmd_t *pmd; 432 pte_t *pte; 433 434 if (bad_address(pgd)) 435 goto bad; 436 437 pr_info("PGD %lx ", pgd_val(*pgd)); 438 439 if (!pgd_present(*pgd)) 440 goto out; 441 442 p4d = p4d_offset(pgd, address); 443 if (bad_address(p4d)) 444 goto bad; 445 446 pr_cont("P4D %lx ", p4d_val(*p4d)); 447 if (!p4d_present(*p4d) || p4d_large(*p4d)) 448 goto out; 449 450 pud = pud_offset(p4d, address); 451 if (bad_address(pud)) 452 goto bad; 453 454 pr_cont("PUD %lx ", pud_val(*pud)); 455 if (!pud_present(*pud) || pud_large(*pud)) 456 goto out; 457 458 pmd = pmd_offset(pud, address); 459 if (bad_address(pmd)) 460 goto bad; 461 462 pr_cont("PMD %lx ", pmd_val(*pmd)); 463 if (!pmd_present(*pmd) || pmd_large(*pmd)) 464 goto out; 465 466 pte = pte_offset_kernel(pmd, address); 467 if (bad_address(pte)) 468 goto bad; 469 470 pr_cont("PTE %lx", pte_val(*pte)); 471 out: 472 pr_cont("\n"); 473 return; 474 bad: 475 pr_info("BAD\n"); 476 } 477 478 #endif /* CONFIG_X86_64 */ 479 480 /* 481 * Workaround for K8 erratum #93 & buggy BIOS. 482 * 483 * BIOS SMM functions are required to use a specific workaround 484 * to avoid corruption of the 64bit RIP register on C stepping K8. 485 * 486 * A lot of BIOS that didn't get tested properly miss this. 487 * 488 * The OS sees this as a page fault with the upper 32bits of RIP cleared. 489 * Try to work around it here. 490 * 491 * Note we only handle faults in kernel here. 492 * Does nothing on 32-bit. 493 */ 494 static int is_errata93(struct pt_regs *regs, unsigned long address) 495 { 496 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) 497 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD 498 || boot_cpu_data.x86 != 0xf) 499 return 0; 500 501 if (address != regs->ip) 502 return 0; 503 504 if ((address >> 32) != 0) 505 return 0; 506 507 address |= 0xffffffffUL << 32; 508 if ((address >= (u64)_stext && address <= (u64)_etext) || 509 (address >= MODULES_VADDR && address <= MODULES_END)) { 510 printk_once(errata93_warning); 511 regs->ip = address; 512 return 1; 513 } 514 #endif 515 return 0; 516 } 517 518 /* 519 * Work around K8 erratum #100 K8 in compat mode occasionally jumps 520 * to illegal addresses >4GB. 521 * 522 * We catch this in the page fault handler because these addresses 523 * are not reachable. Just detect this case and return. Any code 524 * segment in LDT is compatibility mode. 525 */ 526 static int is_errata100(struct pt_regs *regs, unsigned long address) 527 { 528 #ifdef CONFIG_X86_64 529 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) 530 return 1; 531 #endif 532 return 0; 533 } 534 535 static int is_f00f_bug(struct pt_regs *regs, unsigned long address) 536 { 537 #ifdef CONFIG_X86_F00F_BUG 538 unsigned long nr; 539 540 /* 541 * Pentium F0 0F C7 C8 bug workaround: 542 */ 543 if (boot_cpu_has_bug(X86_BUG_F00F)) { 544 nr = (address - idt_descr.address) >> 3; 545 546 if (nr == 6) { 547 do_invalid_op(regs, 0); 548 return 1; 549 } 550 } 551 #endif 552 return 0; 553 } 554 555 static void show_ldttss(const struct desc_ptr *gdt, const char *name, u16 index) 556 { 557 u32 offset = (index >> 3) * sizeof(struct desc_struct); 558 unsigned long addr; 559 struct ldttss_desc desc; 560 561 if (index == 0) { 562 pr_alert("%s: NULL\n", name); 563 return; 564 } 565 566 if (offset + sizeof(struct ldttss_desc) >= gdt->size) { 567 pr_alert("%s: 0x%hx -- out of bounds\n", name, index); 568 return; 569 } 570 571 if (probe_kernel_read(&desc, (void *)(gdt->address + offset), 572 sizeof(struct ldttss_desc))) { 573 pr_alert("%s: 0x%hx -- GDT entry is not readable\n", 574 name, index); 575 return; 576 } 577 578 addr = desc.base0 | (desc.base1 << 16) | ((unsigned long)desc.base2 << 24); 579 #ifdef CONFIG_X86_64 580 addr |= ((u64)desc.base3 << 32); 581 #endif 582 pr_alert("%s: 0x%hx -- base=0x%lx limit=0x%x\n", 583 name, index, addr, (desc.limit0 | (desc.limit1 << 16))); 584 } 585 586 static void 587 show_fault_oops(struct pt_regs *regs, unsigned long error_code, unsigned long address) 588 { 589 if (!oops_may_print()) 590 return; 591 592 if (error_code & X86_PF_INSTR) { 593 unsigned int level; 594 pgd_t *pgd; 595 pte_t *pte; 596 597 pgd = __va(read_cr3_pa()); 598 pgd += pgd_index(address); 599 600 pte = lookup_address_in_pgd(pgd, address, &level); 601 602 if (pte && pte_present(*pte) && !pte_exec(*pte)) 603 pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n", 604 from_kuid(&init_user_ns, current_uid())); 605 if (pte && pte_present(*pte) && pte_exec(*pte) && 606 (pgd_flags(*pgd) & _PAGE_USER) && 607 (__read_cr4() & X86_CR4_SMEP)) 608 pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n", 609 from_kuid(&init_user_ns, current_uid())); 610 } 611 612 if (address < PAGE_SIZE && !user_mode(regs)) 613 pr_alert("BUG: kernel NULL pointer dereference, address: %px\n", 614 (void *)address); 615 else 616 pr_alert("BUG: unable to handle page fault for address: %px\n", 617 (void *)address); 618 619 pr_alert("#PF: %s %s in %s mode\n", 620 (error_code & X86_PF_USER) ? "user" : "supervisor", 621 (error_code & X86_PF_INSTR) ? "instruction fetch" : 622 (error_code & X86_PF_WRITE) ? "write access" : 623 "read access", 624 user_mode(regs) ? "user" : "kernel"); 625 pr_alert("#PF: error_code(0x%04lx) - %s\n", error_code, 626 !(error_code & X86_PF_PROT) ? "not-present page" : 627 (error_code & X86_PF_RSVD) ? "reserved bit violation" : 628 (error_code & X86_PF_PK) ? "protection keys violation" : 629 "permissions violation"); 630 631 if (!(error_code & X86_PF_USER) && user_mode(regs)) { 632 struct desc_ptr idt, gdt; 633 u16 ldtr, tr; 634 635 /* 636 * This can happen for quite a few reasons. The more obvious 637 * ones are faults accessing the GDT, or LDT. Perhaps 638 * surprisingly, if the CPU tries to deliver a benign or 639 * contributory exception from user code and gets a page fault 640 * during delivery, the page fault can be delivered as though 641 * it originated directly from user code. This could happen 642 * due to wrong permissions on the IDT, GDT, LDT, TSS, or 643 * kernel or IST stack. 644 */ 645 store_idt(&idt); 646 647 /* Usable even on Xen PV -- it's just slow. */ 648 native_store_gdt(&gdt); 649 650 pr_alert("IDT: 0x%lx (limit=0x%hx) GDT: 0x%lx (limit=0x%hx)\n", 651 idt.address, idt.size, gdt.address, gdt.size); 652 653 store_ldt(ldtr); 654 show_ldttss(&gdt, "LDTR", ldtr); 655 656 store_tr(tr); 657 show_ldttss(&gdt, "TR", tr); 658 } 659 660 dump_pagetable(address); 661 } 662 663 static noinline void 664 pgtable_bad(struct pt_regs *regs, unsigned long error_code, 665 unsigned long address) 666 { 667 struct task_struct *tsk; 668 unsigned long flags; 669 int sig; 670 671 flags = oops_begin(); 672 tsk = current; 673 sig = SIGKILL; 674 675 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", 676 tsk->comm, address); 677 dump_pagetable(address); 678 679 if (__die("Bad pagetable", regs, error_code)) 680 sig = 0; 681 682 oops_end(flags, regs, sig); 683 } 684 685 static void set_signal_archinfo(unsigned long address, 686 unsigned long error_code) 687 { 688 struct task_struct *tsk = current; 689 690 /* 691 * To avoid leaking information about the kernel page 692 * table layout, pretend that user-mode accesses to 693 * kernel addresses are always protection faults. 694 * 695 * NB: This means that failed vsyscalls with vsyscall=none 696 * will have the PROT bit. This doesn't leak any 697 * information and does not appear to cause any problems. 698 */ 699 if (address >= TASK_SIZE_MAX) 700 error_code |= X86_PF_PROT; 701 702 tsk->thread.trap_nr = X86_TRAP_PF; 703 tsk->thread.error_code = error_code | X86_PF_USER; 704 tsk->thread.cr2 = address; 705 } 706 707 static noinline void 708 no_context(struct pt_regs *regs, unsigned long error_code, 709 unsigned long address, int signal, int si_code) 710 { 711 struct task_struct *tsk = current; 712 unsigned long flags; 713 int sig; 714 715 if (user_mode(regs)) { 716 /* 717 * This is an implicit supervisor-mode access from user 718 * mode. Bypass all the kernel-mode recovery code and just 719 * OOPS. 720 */ 721 goto oops; 722 } 723 724 /* Are we prepared to handle this kernel fault? */ 725 if (fixup_exception(regs, X86_TRAP_PF, error_code, address)) { 726 /* 727 * Any interrupt that takes a fault gets the fixup. This makes 728 * the below recursive fault logic only apply to a faults from 729 * task context. 730 */ 731 if (in_interrupt()) 732 return; 733 734 /* 735 * Per the above we're !in_interrupt(), aka. task context. 736 * 737 * In this case we need to make sure we're not recursively 738 * faulting through the emulate_vsyscall() logic. 739 */ 740 if (current->thread.sig_on_uaccess_err && signal) { 741 set_signal_archinfo(address, error_code); 742 743 /* XXX: hwpoison faults will set the wrong code. */ 744 force_sig_fault(signal, si_code, (void __user *)address); 745 } 746 747 /* 748 * Barring that, we can do the fixup and be happy. 749 */ 750 return; 751 } 752 753 #ifdef CONFIG_VMAP_STACK 754 /* 755 * Stack overflow? During boot, we can fault near the initial 756 * stack in the direct map, but that's not an overflow -- check 757 * that we're in vmalloc space to avoid this. 758 */ 759 if (is_vmalloc_addr((void *)address) && 760 (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) || 761 address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) { 762 unsigned long stack = __this_cpu_ist_top_va(DF) - sizeof(void *); 763 /* 764 * We're likely to be running with very little stack space 765 * left. It's plausible that we'd hit this condition but 766 * double-fault even before we get this far, in which case 767 * we're fine: the double-fault handler will deal with it. 768 * 769 * We don't want to make it all the way into the oops code 770 * and then double-fault, though, because we're likely to 771 * break the console driver and lose most of the stack dump. 772 */ 773 asm volatile ("movq %[stack], %%rsp\n\t" 774 "call handle_stack_overflow\n\t" 775 "1: jmp 1b" 776 : ASM_CALL_CONSTRAINT 777 : "D" ("kernel stack overflow (page fault)"), 778 "S" (regs), "d" (address), 779 [stack] "rm" (stack)); 780 unreachable(); 781 } 782 #endif 783 784 /* 785 * 32-bit: 786 * 787 * Valid to do another page fault here, because if this fault 788 * had been triggered by is_prefetch fixup_exception would have 789 * handled it. 790 * 791 * 64-bit: 792 * 793 * Hall of shame of CPU/BIOS bugs. 794 */ 795 if (is_prefetch(regs, error_code, address)) 796 return; 797 798 if (is_errata93(regs, address)) 799 return; 800 801 /* 802 * Buggy firmware could access regions which might page fault, try to 803 * recover from such faults. 804 */ 805 if (IS_ENABLED(CONFIG_EFI)) 806 efi_recover_from_page_fault(address); 807 808 oops: 809 /* 810 * Oops. The kernel tried to access some bad page. We'll have to 811 * terminate things with extreme prejudice: 812 */ 813 flags = oops_begin(); 814 815 show_fault_oops(regs, error_code, address); 816 817 if (task_stack_end_corrupted(tsk)) 818 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); 819 820 sig = SIGKILL; 821 if (__die("Oops", regs, error_code)) 822 sig = 0; 823 824 /* Executive summary in case the body of the oops scrolled away */ 825 printk(KERN_DEFAULT "CR2: %016lx\n", address); 826 827 oops_end(flags, regs, sig); 828 } 829 830 /* 831 * Print out info about fatal segfaults, if the show_unhandled_signals 832 * sysctl is set: 833 */ 834 static inline void 835 show_signal_msg(struct pt_regs *regs, unsigned long error_code, 836 unsigned long address, struct task_struct *tsk) 837 { 838 const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG; 839 840 if (!unhandled_signal(tsk, SIGSEGV)) 841 return; 842 843 if (!printk_ratelimit()) 844 return; 845 846 printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx", 847 loglvl, tsk->comm, task_pid_nr(tsk), address, 848 (void *)regs->ip, (void *)regs->sp, error_code); 849 850 print_vma_addr(KERN_CONT " in ", regs->ip); 851 852 printk(KERN_CONT "\n"); 853 854 show_opcodes(regs, loglvl); 855 } 856 857 /* 858 * The (legacy) vsyscall page is the long page in the kernel portion 859 * of the address space that has user-accessible permissions. 860 */ 861 static bool is_vsyscall_vaddr(unsigned long vaddr) 862 { 863 return unlikely((vaddr & PAGE_MASK) == VSYSCALL_ADDR); 864 } 865 866 static void 867 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 868 unsigned long address, u32 pkey, int si_code) 869 { 870 struct task_struct *tsk = current; 871 872 /* User mode accesses just cause a SIGSEGV */ 873 if (user_mode(regs) && (error_code & X86_PF_USER)) { 874 /* 875 * It's possible to have interrupts off here: 876 */ 877 local_irq_enable(); 878 879 /* 880 * Valid to do another page fault here because this one came 881 * from user space: 882 */ 883 if (is_prefetch(regs, error_code, address)) 884 return; 885 886 if (is_errata100(regs, address)) 887 return; 888 889 /* 890 * To avoid leaking information about the kernel page table 891 * layout, pretend that user-mode accesses to kernel addresses 892 * are always protection faults. 893 */ 894 if (address >= TASK_SIZE_MAX) 895 error_code |= X86_PF_PROT; 896 897 if (likely(show_unhandled_signals)) 898 show_signal_msg(regs, error_code, address, tsk); 899 900 set_signal_archinfo(address, error_code); 901 902 if (si_code == SEGV_PKUERR) 903 force_sig_pkuerr((void __user *)address, pkey); 904 905 force_sig_fault(SIGSEGV, si_code, (void __user *)address); 906 907 return; 908 } 909 910 if (is_f00f_bug(regs, address)) 911 return; 912 913 no_context(regs, error_code, address, SIGSEGV, si_code); 914 } 915 916 static noinline void 917 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 918 unsigned long address) 919 { 920 __bad_area_nosemaphore(regs, error_code, address, 0, SEGV_MAPERR); 921 } 922 923 static void 924 __bad_area(struct pt_regs *regs, unsigned long error_code, 925 unsigned long address, u32 pkey, int si_code) 926 { 927 struct mm_struct *mm = current->mm; 928 /* 929 * Something tried to access memory that isn't in our memory map.. 930 * Fix it, but check if it's kernel or user first.. 931 */ 932 up_read(&mm->mmap_sem); 933 934 __bad_area_nosemaphore(regs, error_code, address, pkey, si_code); 935 } 936 937 static noinline void 938 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) 939 { 940 __bad_area(regs, error_code, address, 0, SEGV_MAPERR); 941 } 942 943 static inline bool bad_area_access_from_pkeys(unsigned long error_code, 944 struct vm_area_struct *vma) 945 { 946 /* This code is always called on the current mm */ 947 bool foreign = false; 948 949 if (!boot_cpu_has(X86_FEATURE_OSPKE)) 950 return false; 951 if (error_code & X86_PF_PK) 952 return true; 953 /* this checks permission keys on the VMA: */ 954 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), 955 (error_code & X86_PF_INSTR), foreign)) 956 return true; 957 return false; 958 } 959 960 static noinline void 961 bad_area_access_error(struct pt_regs *regs, unsigned long error_code, 962 unsigned long address, struct vm_area_struct *vma) 963 { 964 /* 965 * This OSPKE check is not strictly necessary at runtime. 966 * But, doing it this way allows compiler optimizations 967 * if pkeys are compiled out. 968 */ 969 if (bad_area_access_from_pkeys(error_code, vma)) { 970 /* 971 * A protection key fault means that the PKRU value did not allow 972 * access to some PTE. Userspace can figure out what PKRU was 973 * from the XSAVE state. This function captures the pkey from 974 * the vma and passes it to userspace so userspace can discover 975 * which protection key was set on the PTE. 976 * 977 * If we get here, we know that the hardware signaled a X86_PF_PK 978 * fault and that there was a VMA once we got in the fault 979 * handler. It does *not* guarantee that the VMA we find here 980 * was the one that we faulted on. 981 * 982 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4); 983 * 2. T1 : set PKRU to deny access to pkey=4, touches page 984 * 3. T1 : faults... 985 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5); 986 * 5. T1 : enters fault handler, takes mmap_sem, etc... 987 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really 988 * faulted on a pte with its pkey=4. 989 */ 990 u32 pkey = vma_pkey(vma); 991 992 __bad_area(regs, error_code, address, pkey, SEGV_PKUERR); 993 } else { 994 __bad_area(regs, error_code, address, 0, SEGV_ACCERR); 995 } 996 } 997 998 static void 999 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, 1000 vm_fault_t fault) 1001 { 1002 /* Kernel mode? Handle exceptions or die: */ 1003 if (!(error_code & X86_PF_USER)) { 1004 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); 1005 return; 1006 } 1007 1008 /* User-space => ok to do another page fault: */ 1009 if (is_prefetch(regs, error_code, address)) 1010 return; 1011 1012 set_signal_archinfo(address, error_code); 1013 1014 #ifdef CONFIG_MEMORY_FAILURE 1015 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { 1016 struct task_struct *tsk = current; 1017 unsigned lsb = 0; 1018 1019 pr_err( 1020 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", 1021 tsk->comm, tsk->pid, address); 1022 if (fault & VM_FAULT_HWPOISON_LARGE) 1023 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); 1024 if (fault & VM_FAULT_HWPOISON) 1025 lsb = PAGE_SHIFT; 1026 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb); 1027 return; 1028 } 1029 #endif 1030 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address); 1031 } 1032 1033 static noinline void 1034 mm_fault_error(struct pt_regs *regs, unsigned long error_code, 1035 unsigned long address, vm_fault_t fault) 1036 { 1037 if (fatal_signal_pending(current) && !(error_code & X86_PF_USER)) { 1038 no_context(regs, error_code, address, 0, 0); 1039 return; 1040 } 1041 1042 if (fault & VM_FAULT_OOM) { 1043 /* Kernel mode? Handle exceptions or die: */ 1044 if (!(error_code & X86_PF_USER)) { 1045 no_context(regs, error_code, address, 1046 SIGSEGV, SEGV_MAPERR); 1047 return; 1048 } 1049 1050 /* 1051 * We ran out of memory, call the OOM killer, and return the 1052 * userspace (which will retry the fault, or kill us if we got 1053 * oom-killed): 1054 */ 1055 pagefault_out_of_memory(); 1056 } else { 1057 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| 1058 VM_FAULT_HWPOISON_LARGE)) 1059 do_sigbus(regs, error_code, address, fault); 1060 else if (fault & VM_FAULT_SIGSEGV) 1061 bad_area_nosemaphore(regs, error_code, address); 1062 else 1063 BUG(); 1064 } 1065 } 1066 1067 static int spurious_kernel_fault_check(unsigned long error_code, pte_t *pte) 1068 { 1069 if ((error_code & X86_PF_WRITE) && !pte_write(*pte)) 1070 return 0; 1071 1072 if ((error_code & X86_PF_INSTR) && !pte_exec(*pte)) 1073 return 0; 1074 1075 return 1; 1076 } 1077 1078 /* 1079 * Handle a spurious fault caused by a stale TLB entry. 1080 * 1081 * This allows us to lazily refresh the TLB when increasing the 1082 * permissions of a kernel page (RO -> RW or NX -> X). Doing it 1083 * eagerly is very expensive since that implies doing a full 1084 * cross-processor TLB flush, even if no stale TLB entries exist 1085 * on other processors. 1086 * 1087 * Spurious faults may only occur if the TLB contains an entry with 1088 * fewer permission than the page table entry. Non-present (P = 0) 1089 * and reserved bit (R = 1) faults are never spurious. 1090 * 1091 * There are no security implications to leaving a stale TLB when 1092 * increasing the permissions on a page. 1093 * 1094 * Returns non-zero if a spurious fault was handled, zero otherwise. 1095 * 1096 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 1097 * (Optional Invalidation). 1098 */ 1099 static noinline int 1100 spurious_kernel_fault(unsigned long error_code, unsigned long address) 1101 { 1102 pgd_t *pgd; 1103 p4d_t *p4d; 1104 pud_t *pud; 1105 pmd_t *pmd; 1106 pte_t *pte; 1107 int ret; 1108 1109 /* 1110 * Only writes to RO or instruction fetches from NX may cause 1111 * spurious faults. 1112 * 1113 * These could be from user or supervisor accesses but the TLB 1114 * is only lazily flushed after a kernel mapping protection 1115 * change, so user accesses are not expected to cause spurious 1116 * faults. 1117 */ 1118 if (error_code != (X86_PF_WRITE | X86_PF_PROT) && 1119 error_code != (X86_PF_INSTR | X86_PF_PROT)) 1120 return 0; 1121 1122 pgd = init_mm.pgd + pgd_index(address); 1123 if (!pgd_present(*pgd)) 1124 return 0; 1125 1126 p4d = p4d_offset(pgd, address); 1127 if (!p4d_present(*p4d)) 1128 return 0; 1129 1130 if (p4d_large(*p4d)) 1131 return spurious_kernel_fault_check(error_code, (pte_t *) p4d); 1132 1133 pud = pud_offset(p4d, address); 1134 if (!pud_present(*pud)) 1135 return 0; 1136 1137 if (pud_large(*pud)) 1138 return spurious_kernel_fault_check(error_code, (pte_t *) pud); 1139 1140 pmd = pmd_offset(pud, address); 1141 if (!pmd_present(*pmd)) 1142 return 0; 1143 1144 if (pmd_large(*pmd)) 1145 return spurious_kernel_fault_check(error_code, (pte_t *) pmd); 1146 1147 pte = pte_offset_kernel(pmd, address); 1148 if (!pte_present(*pte)) 1149 return 0; 1150 1151 ret = spurious_kernel_fault_check(error_code, pte); 1152 if (!ret) 1153 return 0; 1154 1155 /* 1156 * Make sure we have permissions in PMD. 1157 * If not, then there's a bug in the page tables: 1158 */ 1159 ret = spurious_kernel_fault_check(error_code, (pte_t *) pmd); 1160 WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); 1161 1162 return ret; 1163 } 1164 NOKPROBE_SYMBOL(spurious_kernel_fault); 1165 1166 int show_unhandled_signals = 1; 1167 1168 static inline int 1169 access_error(unsigned long error_code, struct vm_area_struct *vma) 1170 { 1171 /* This is only called for the current mm, so: */ 1172 bool foreign = false; 1173 1174 /* 1175 * Read or write was blocked by protection keys. This is 1176 * always an unconditional error and can never result in 1177 * a follow-up action to resolve the fault, like a COW. 1178 */ 1179 if (error_code & X86_PF_PK) 1180 return 1; 1181 1182 /* 1183 * Make sure to check the VMA so that we do not perform 1184 * faults just to hit a X86_PF_PK as soon as we fill in a 1185 * page. 1186 */ 1187 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE), 1188 (error_code & X86_PF_INSTR), foreign)) 1189 return 1; 1190 1191 if (error_code & X86_PF_WRITE) { 1192 /* write, present and write, not present: */ 1193 if (unlikely(!(vma->vm_flags & VM_WRITE))) 1194 return 1; 1195 return 0; 1196 } 1197 1198 /* read, present: */ 1199 if (unlikely(error_code & X86_PF_PROT)) 1200 return 1; 1201 1202 /* read, not present: */ 1203 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) 1204 return 1; 1205 1206 return 0; 1207 } 1208 1209 static int fault_in_kernel_space(unsigned long address) 1210 { 1211 /* 1212 * On 64-bit systems, the vsyscall page is at an address above 1213 * TASK_SIZE_MAX, but is not considered part of the kernel 1214 * address space. 1215 */ 1216 if (IS_ENABLED(CONFIG_X86_64) && is_vsyscall_vaddr(address)) 1217 return false; 1218 1219 return address >= TASK_SIZE_MAX; 1220 } 1221 1222 /* 1223 * Called for all faults where 'address' is part of the kernel address 1224 * space. Might get called for faults that originate from *code* that 1225 * ran in userspace or the kernel. 1226 */ 1227 static void 1228 do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code, 1229 unsigned long address) 1230 { 1231 /* 1232 * Protection keys exceptions only happen on user pages. We 1233 * have no user pages in the kernel portion of the address 1234 * space, so do not expect them here. 1235 */ 1236 WARN_ON_ONCE(hw_error_code & X86_PF_PK); 1237 1238 /* 1239 * We can fault-in kernel-space virtual memory on-demand. The 1240 * 'reference' page table is init_mm.pgd. 1241 * 1242 * NOTE! We MUST NOT take any locks for this case. We may 1243 * be in an interrupt or a critical region, and should 1244 * only copy the information from the master page table, 1245 * nothing more. 1246 * 1247 * Before doing this on-demand faulting, ensure that the 1248 * fault is not any of the following: 1249 * 1. A fault on a PTE with a reserved bit set. 1250 * 2. A fault caused by a user-mode access. (Do not demand- 1251 * fault kernel memory due to user-mode accesses). 1252 * 3. A fault caused by a page-level protection violation. 1253 * (A demand fault would be on a non-present page which 1254 * would have X86_PF_PROT==0). 1255 */ 1256 if (!(hw_error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) { 1257 if (vmalloc_fault(address) >= 0) 1258 return; 1259 } 1260 1261 /* Was the fault spurious, caused by lazy TLB invalidation? */ 1262 if (spurious_kernel_fault(hw_error_code, address)) 1263 return; 1264 1265 /* kprobes don't want to hook the spurious faults: */ 1266 if (kprobe_page_fault(regs, X86_TRAP_PF)) 1267 return; 1268 1269 /* 1270 * Note, despite being a "bad area", there are quite a few 1271 * acceptable reasons to get here, such as erratum fixups 1272 * and handling kernel code that can fault, like get_user(). 1273 * 1274 * Don't take the mm semaphore here. If we fixup a prefetch 1275 * fault we could otherwise deadlock: 1276 */ 1277 bad_area_nosemaphore(regs, hw_error_code, address); 1278 } 1279 NOKPROBE_SYMBOL(do_kern_addr_fault); 1280 1281 /* Handle faults in the user portion of the address space */ 1282 static inline 1283 void do_user_addr_fault(struct pt_regs *regs, 1284 unsigned long hw_error_code, 1285 unsigned long address) 1286 { 1287 struct vm_area_struct *vma; 1288 struct task_struct *tsk; 1289 struct mm_struct *mm; 1290 vm_fault_t fault, major = 0; 1291 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; 1292 1293 tsk = current; 1294 mm = tsk->mm; 1295 1296 /* kprobes don't want to hook the spurious faults: */ 1297 if (unlikely(kprobe_page_fault(regs, X86_TRAP_PF))) 1298 return; 1299 1300 /* 1301 * Reserved bits are never expected to be set on 1302 * entries in the user portion of the page tables. 1303 */ 1304 if (unlikely(hw_error_code & X86_PF_RSVD)) 1305 pgtable_bad(regs, hw_error_code, address); 1306 1307 /* 1308 * If SMAP is on, check for invalid kernel (supervisor) access to user 1309 * pages in the user address space. The odd case here is WRUSS, 1310 * which, according to the preliminary documentation, does not respect 1311 * SMAP and will have the USER bit set so, in all cases, SMAP 1312 * enforcement appears to be consistent with the USER bit. 1313 */ 1314 if (unlikely(cpu_feature_enabled(X86_FEATURE_SMAP) && 1315 !(hw_error_code & X86_PF_USER) && 1316 !(regs->flags & X86_EFLAGS_AC))) 1317 { 1318 bad_area_nosemaphore(regs, hw_error_code, address); 1319 return; 1320 } 1321 1322 /* 1323 * If we're in an interrupt, have no user context or are running 1324 * in a region with pagefaults disabled then we must not take the fault 1325 */ 1326 if (unlikely(faulthandler_disabled() || !mm)) { 1327 bad_area_nosemaphore(regs, hw_error_code, address); 1328 return; 1329 } 1330 1331 /* 1332 * It's safe to allow irq's after cr2 has been saved and the 1333 * vmalloc fault has been handled. 1334 * 1335 * User-mode registers count as a user access even for any 1336 * potential system fault or CPU buglet: 1337 */ 1338 if (user_mode(regs)) { 1339 local_irq_enable(); 1340 flags |= FAULT_FLAG_USER; 1341 } else { 1342 if (regs->flags & X86_EFLAGS_IF) 1343 local_irq_enable(); 1344 } 1345 1346 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 1347 1348 if (hw_error_code & X86_PF_WRITE) 1349 flags |= FAULT_FLAG_WRITE; 1350 if (hw_error_code & X86_PF_INSTR) 1351 flags |= FAULT_FLAG_INSTRUCTION; 1352 1353 #ifdef CONFIG_X86_64 1354 /* 1355 * Faults in the vsyscall page might need emulation. The 1356 * vsyscall page is at a high address (>PAGE_OFFSET), but is 1357 * considered to be part of the user address space. 1358 * 1359 * The vsyscall page does not have a "real" VMA, so do this 1360 * emulation before we go searching for VMAs. 1361 * 1362 * PKRU never rejects instruction fetches, so we don't need 1363 * to consider the PF_PK bit. 1364 */ 1365 if (is_vsyscall_vaddr(address)) { 1366 if (emulate_vsyscall(hw_error_code, regs, address)) 1367 return; 1368 } 1369 #endif 1370 1371 /* 1372 * Kernel-mode access to the user address space should only occur 1373 * on well-defined single instructions listed in the exception 1374 * tables. But, an erroneous kernel fault occurring outside one of 1375 * those areas which also holds mmap_sem might deadlock attempting 1376 * to validate the fault against the address space. 1377 * 1378 * Only do the expensive exception table search when we might be at 1379 * risk of a deadlock. This happens if we 1380 * 1. Failed to acquire mmap_sem, and 1381 * 2. The access did not originate in userspace. 1382 */ 1383 if (unlikely(!down_read_trylock(&mm->mmap_sem))) { 1384 if (!user_mode(regs) && !search_exception_tables(regs->ip)) { 1385 /* 1386 * Fault from code in kernel from 1387 * which we do not expect faults. 1388 */ 1389 bad_area_nosemaphore(regs, hw_error_code, address); 1390 return; 1391 } 1392 retry: 1393 down_read(&mm->mmap_sem); 1394 } else { 1395 /* 1396 * The above down_read_trylock() might have succeeded in 1397 * which case we'll have missed the might_sleep() from 1398 * down_read(): 1399 */ 1400 might_sleep(); 1401 } 1402 1403 vma = find_vma(mm, address); 1404 if (unlikely(!vma)) { 1405 bad_area(regs, hw_error_code, address); 1406 return; 1407 } 1408 if (likely(vma->vm_start <= address)) 1409 goto good_area; 1410 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { 1411 bad_area(regs, hw_error_code, address); 1412 return; 1413 } 1414 if (unlikely(expand_stack(vma, address))) { 1415 bad_area(regs, hw_error_code, address); 1416 return; 1417 } 1418 1419 /* 1420 * Ok, we have a good vm_area for this memory access, so 1421 * we can handle it.. 1422 */ 1423 good_area: 1424 if (unlikely(access_error(hw_error_code, vma))) { 1425 bad_area_access_error(regs, hw_error_code, address, vma); 1426 return; 1427 } 1428 1429 /* 1430 * If for any reason at all we couldn't handle the fault, 1431 * make sure we exit gracefully rather than endlessly redo 1432 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if 1433 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked. 1434 * 1435 * Note that handle_userfault() may also release and reacquire mmap_sem 1436 * (and not return with VM_FAULT_RETRY), when returning to userland to 1437 * repeat the page fault later with a VM_FAULT_NOPAGE retval 1438 * (potentially after handling any pending signal during the return to 1439 * userland). The return to userland is identified whenever 1440 * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags. 1441 */ 1442 fault = handle_mm_fault(vma, address, flags); 1443 major |= fault & VM_FAULT_MAJOR; 1444 1445 /* 1446 * If we need to retry the mmap_sem has already been released, 1447 * and if there is a fatal signal pending there is no guarantee 1448 * that we made any progress. Handle this case first. 1449 */ 1450 if (unlikely(fault & VM_FAULT_RETRY)) { 1451 /* Retry at most once */ 1452 if (flags & FAULT_FLAG_ALLOW_RETRY) { 1453 flags &= ~FAULT_FLAG_ALLOW_RETRY; 1454 flags |= FAULT_FLAG_TRIED; 1455 if (!fatal_signal_pending(tsk)) 1456 goto retry; 1457 } 1458 1459 /* User mode? Just return to handle the fatal exception */ 1460 if (flags & FAULT_FLAG_USER) 1461 return; 1462 1463 /* Not returning to user mode? Handle exceptions or die: */ 1464 no_context(regs, hw_error_code, address, SIGBUS, BUS_ADRERR); 1465 return; 1466 } 1467 1468 up_read(&mm->mmap_sem); 1469 if (unlikely(fault & VM_FAULT_ERROR)) { 1470 mm_fault_error(regs, hw_error_code, address, fault); 1471 return; 1472 } 1473 1474 /* 1475 * Major/minor page fault accounting. If any of the events 1476 * returned VM_FAULT_MAJOR, we account it as a major fault. 1477 */ 1478 if (major) { 1479 tsk->maj_flt++; 1480 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address); 1481 } else { 1482 tsk->min_flt++; 1483 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address); 1484 } 1485 1486 check_v8086_mode(regs, address, tsk); 1487 } 1488 NOKPROBE_SYMBOL(do_user_addr_fault); 1489 1490 static __always_inline void 1491 trace_page_fault_entries(struct pt_regs *regs, unsigned long error_code, 1492 unsigned long address) 1493 { 1494 if (!trace_pagefault_enabled()) 1495 return; 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 dotraplinkage void 1504 do_page_fault(struct pt_regs *regs, unsigned long hw_error_code, 1505 unsigned long address) 1506 { 1507 prefetchw(¤t->mm->mmap_sem); 1508 trace_page_fault_entries(regs, hw_error_code, address); 1509 1510 if (unlikely(kmmio_fault(regs, address))) 1511 return; 1512 1513 /* Was the fault on kernel-controlled part of the address space? */ 1514 if (unlikely(fault_in_kernel_space(address))) 1515 do_kern_addr_fault(regs, hw_error_code, address); 1516 else 1517 do_user_addr_fault(regs, hw_error_code, address); 1518 } 1519 NOKPROBE_SYMBOL(do_page_fault); 1520