1 // SPDX-License-Identifier: GPL-2.0
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
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>
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() */
29 #define CREATE_TRACE_POINTS
30 #include <asm/trace/exceptions.h>
33 * Returns 0 if mmiotrace is disabled, or if the fault is not
34 * handled by mmiotrace:
36 static nokprobe_inline int
37 kmmio_fault(struct pt_regs *regs, unsigned long addr)
39 if (unlikely(is_kmmio_active()))
40 if (kmmio_handler(regs, addr) == 1)
45 static nokprobe_inline int kprobes_fault(struct pt_regs *regs)
49 /* kprobe_running() needs smp_processor_id() */
50 if (kprobes_built_in() && !user_mode(regs)) {
52 if (kprobe_running() && kprobe_fault_handler(regs, 14))
65 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
66 * Check that here and ignore it.
70 * Sometimes the CPU reports invalid exceptions on prefetch.
71 * Check that here and ignore it.
73 * Opcode checker based on code by Richard Brunner.
76 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
77 unsigned char opcode, int *prefetch)
79 unsigned char instr_hi = opcode & 0xf0;
80 unsigned char instr_lo = opcode & 0x0f;
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
91 return ((instr_lo & 7) == 0x6);
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.
101 return (!user_mode(regs) || user_64bit_mode(regs));
104 /* 0x64 thru 0x67 are valid prefixes in all modes. */
105 return (instr_lo & 0xC) == 0x4;
107 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
108 return !instr_lo || (instr_lo>>1) == 1;
110 /* Prefetch instruction is 0x0F0D or 0x0F18 */
111 if (probe_kernel_address(instr, opcode))
114 *prefetch = (instr_lo == 0xF) &&
115 (opcode == 0x0D || opcode == 0x18);
123 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
125 unsigned char *max_instr;
126 unsigned char *instr;
130 * If it was a exec (instruction fetch) fault on NX page, then
131 * do not ignore the fault:
133 if (error_code & X86_PF_INSTR)
136 instr = (void *)convert_ip_to_linear(current, regs);
137 max_instr = instr + 15;
139 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX)
142 while (instr < max_instr) {
143 unsigned char opcode;
145 if (probe_kernel_address(instr, opcode))
150 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
156 DEFINE_SPINLOCK(pgd_lock);
160 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
162 unsigned index = pgd_index(address);
169 pgd_k = init_mm.pgd + index;
171 if (!pgd_present(*pgd_k))
175 * set_pgd(pgd, *pgd_k); here would be useless on PAE
176 * and redundant with the set_pmd() on non-PAE. As would
179 p4d = p4d_offset(pgd, address);
180 p4d_k = p4d_offset(pgd_k, address);
181 if (!p4d_present(*p4d_k))
184 pud = pud_offset(p4d, address);
185 pud_k = pud_offset(p4d_k, address);
186 if (!pud_present(*pud_k))
189 pmd = pmd_offset(pud, address);
190 pmd_k = pmd_offset(pud_k, address);
191 if (!pmd_present(*pmd_k))
194 if (!pmd_present(*pmd))
195 set_pmd(pmd, *pmd_k);
197 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
202 void vmalloc_sync_all(void)
204 unsigned long address;
206 if (SHARED_KERNEL_PMD)
209 for (address = VMALLOC_START & PMD_MASK;
210 address >= TASK_SIZE_MAX && address < FIXADDR_TOP;
211 address += PMD_SIZE) {
214 spin_lock(&pgd_lock);
215 list_for_each_entry(page, &pgd_list, lru) {
216 spinlock_t *pgt_lock;
219 /* the pgt_lock only for Xen */
220 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
223 ret = vmalloc_sync_one(page_address(page), address);
224 spin_unlock(pgt_lock);
229 spin_unlock(&pgd_lock);
236 * Handle a fault on the vmalloc or module mapping area
238 static noinline int vmalloc_fault(unsigned long address)
240 unsigned long pgd_paddr;
244 /* Make sure we are in vmalloc area: */
245 if (!(address >= VMALLOC_START && address < VMALLOC_END))
249 * Synchronize this task's top level page-table
250 * with the 'reference' page table.
252 * Do _not_ use "current" here. We might be inside
253 * an interrupt in the middle of a task switch..
255 pgd_paddr = read_cr3_pa();
256 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
260 if (pmd_large(*pmd_k))
263 pte_k = pte_offset_kernel(pmd_k, address);
264 if (!pte_present(*pte_k))
269 NOKPROBE_SYMBOL(vmalloc_fault);
272 * Did it hit the DOS screen memory VA from vm86 mode?
275 check_v8086_mode(struct pt_regs *regs, unsigned long address,
276 struct task_struct *tsk)
281 if (!v8086_mode(regs) || !tsk->thread.vm86)
284 bit = (address - 0xA0000) >> PAGE_SHIFT;
286 tsk->thread.vm86->screen_bitmap |= 1 << bit;
290 static bool low_pfn(unsigned long pfn)
292 return pfn < max_low_pfn;
295 static void dump_pagetable(unsigned long address)
297 pgd_t *base = __va(read_cr3_pa());
298 pgd_t *pgd = &base[pgd_index(address)];
304 #ifdef CONFIG_X86_PAE
305 pr_info("*pdpt = %016Lx ", pgd_val(*pgd));
306 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
308 #define pr_pde pr_cont
310 #define pr_pde pr_info
312 p4d = p4d_offset(pgd, address);
313 pud = pud_offset(p4d, address);
314 pmd = pmd_offset(pud, address);
315 pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
319 * We must not directly access the pte in the highpte
320 * case if the page table is located in highmem.
321 * And let's rather not kmap-atomic the pte, just in case
322 * it's allocated already:
324 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
327 pte = pte_offset_kernel(pmd, address);
328 pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
333 #else /* CONFIG_X86_64: */
335 void vmalloc_sync_all(void)
337 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END);
343 * Handle a fault on the vmalloc area
345 static noinline int vmalloc_fault(unsigned long address)
353 /* Make sure we are in vmalloc area: */
354 if (!(address >= VMALLOC_START && address < VMALLOC_END))
357 WARN_ON_ONCE(in_nmi());
360 * Copy kernel mappings over when needed. This can also
361 * happen within a race in page table update. In the later
364 pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address);
365 pgd_k = pgd_offset_k(address);
366 if (pgd_none(*pgd_k))
369 if (pgtable_l5_enabled()) {
370 if (pgd_none(*pgd)) {
371 set_pgd(pgd, *pgd_k);
372 arch_flush_lazy_mmu_mode();
374 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_k));
378 /* With 4-level paging, copying happens on the p4d level. */
379 p4d = p4d_offset(pgd, address);
380 p4d_k = p4d_offset(pgd_k, address);
381 if (p4d_none(*p4d_k))
384 if (p4d_none(*p4d) && !pgtable_l5_enabled()) {
385 set_p4d(p4d, *p4d_k);
386 arch_flush_lazy_mmu_mode();
388 BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_k));
391 BUILD_BUG_ON(CONFIG_PGTABLE_LEVELS < 4);
393 pud = pud_offset(p4d, address);
400 pmd = pmd_offset(pud, address);
407 pte = pte_offset_kernel(pmd, address);
408 if (!pte_present(*pte))
413 NOKPROBE_SYMBOL(vmalloc_fault);
415 #ifdef CONFIG_CPU_SUP_AMD
416 static const char errata93_warning[] =
418 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
419 "******* Working around it, but it may cause SEGVs or burn power.\n"
420 "******* Please consider a BIOS update.\n"
421 "******* Disabling USB legacy in the BIOS may also help.\n";
425 * No vm86 mode in 64-bit mode:
428 check_v8086_mode(struct pt_regs *regs, unsigned long address,
429 struct task_struct *tsk)
433 static int bad_address(void *p)
437 return probe_kernel_address((unsigned long *)p, dummy);
440 static void dump_pagetable(unsigned long address)
442 pgd_t *base = __va(read_cr3_pa());
443 pgd_t *pgd = base + pgd_index(address);
449 if (bad_address(pgd))
452 pr_info("PGD %lx ", pgd_val(*pgd));
454 if (!pgd_present(*pgd))
457 p4d = p4d_offset(pgd, address);
458 if (bad_address(p4d))
461 pr_cont("P4D %lx ", p4d_val(*p4d));
462 if (!p4d_present(*p4d) || p4d_large(*p4d))
465 pud = pud_offset(p4d, address);
466 if (bad_address(pud))
469 pr_cont("PUD %lx ", pud_val(*pud));
470 if (!pud_present(*pud) || pud_large(*pud))
473 pmd = pmd_offset(pud, address);
474 if (bad_address(pmd))
477 pr_cont("PMD %lx ", pmd_val(*pmd));
478 if (!pmd_present(*pmd) || pmd_large(*pmd))
481 pte = pte_offset_kernel(pmd, address);
482 if (bad_address(pte))
485 pr_cont("PTE %lx", pte_val(*pte));
493 #endif /* CONFIG_X86_64 */
496 * Workaround for K8 erratum #93 & buggy BIOS.
498 * BIOS SMM functions are required to use a specific workaround
499 * to avoid corruption of the 64bit RIP register on C stepping K8.
501 * A lot of BIOS that didn't get tested properly miss this.
503 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
504 * Try to work around it here.
506 * Note we only handle faults in kernel here.
507 * Does nothing on 32-bit.
509 static int is_errata93(struct pt_regs *regs, unsigned long address)
511 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
512 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD
513 || boot_cpu_data.x86 != 0xf)
516 if (address != regs->ip)
519 if ((address >> 32) != 0)
522 address |= 0xffffffffUL << 32;
523 if ((address >= (u64)_stext && address <= (u64)_etext) ||
524 (address >= MODULES_VADDR && address <= MODULES_END)) {
525 printk_once(errata93_warning);
534 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
535 * to illegal addresses >4GB.
537 * We catch this in the page fault handler because these addresses
538 * are not reachable. Just detect this case and return. Any code
539 * segment in LDT is compatibility mode.
541 static int is_errata100(struct pt_regs *regs, unsigned long address)
544 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
550 static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
552 #ifdef CONFIG_X86_F00F_BUG
556 * Pentium F0 0F C7 C8 bug workaround:
558 if (boot_cpu_has_bug(X86_BUG_F00F)) {
559 nr = (address - idt_descr.address) >> 3;
562 do_invalid_op(regs, 0);
571 show_fault_oops(struct pt_regs *regs, unsigned long error_code,
572 unsigned long address)
574 if (!oops_may_print())
577 if (error_code & X86_PF_INSTR) {
582 pgd = __va(read_cr3_pa());
583 pgd += pgd_index(address);
585 pte = lookup_address_in_pgd(pgd, address, &level);
587 if (pte && pte_present(*pte) && !pte_exec(*pte))
588 pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n",
589 from_kuid(&init_user_ns, current_uid()));
590 if (pte && pte_present(*pte) && pte_exec(*pte) &&
591 (pgd_flags(*pgd) & _PAGE_USER) &&
592 (__read_cr4() & X86_CR4_SMEP))
593 pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n",
594 from_kuid(&init_user_ns, current_uid()));
597 pr_alert("BUG: unable to handle kernel %s at %px\n",
598 address < PAGE_SIZE ? "NULL pointer dereference" : "paging request",
601 dump_pagetable(address);
605 pgtable_bad(struct pt_regs *regs, unsigned long error_code,
606 unsigned long address)
608 struct task_struct *tsk;
612 flags = oops_begin();
616 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
618 dump_pagetable(address);
620 tsk->thread.cr2 = address;
621 tsk->thread.trap_nr = X86_TRAP_PF;
622 tsk->thread.error_code = error_code;
624 if (__die("Bad pagetable", regs, error_code))
627 oops_end(flags, regs, sig);
631 no_context(struct pt_regs *regs, unsigned long error_code,
632 unsigned long address, int signal, int si_code)
634 struct task_struct *tsk = current;
638 /* Are we prepared to handle this kernel fault? */
639 if (fixup_exception(regs, X86_TRAP_PF)) {
641 * Any interrupt that takes a fault gets the fixup. This makes
642 * the below recursive fault logic only apply to a faults from
649 * Per the above we're !in_interrupt(), aka. task context.
651 * In this case we need to make sure we're not recursively
652 * faulting through the emulate_vsyscall() logic.
654 if (current->thread.sig_on_uaccess_err && signal) {
655 tsk->thread.trap_nr = X86_TRAP_PF;
656 tsk->thread.error_code = error_code | X86_PF_USER;
657 tsk->thread.cr2 = address;
659 /* XXX: hwpoison faults will set the wrong code. */
660 force_sig_fault(signal, si_code, (void __user *)address,
665 * Barring that, we can do the fixup and be happy.
670 #ifdef CONFIG_VMAP_STACK
672 * Stack overflow? During boot, we can fault near the initial
673 * stack in the direct map, but that's not an overflow -- check
674 * that we're in vmalloc space to avoid this.
676 if (is_vmalloc_addr((void *)address) &&
677 (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) ||
678 address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) {
679 unsigned long stack = this_cpu_read(orig_ist.ist[DOUBLEFAULT_STACK]) - sizeof(void *);
681 * We're likely to be running with very little stack space
682 * left. It's plausible that we'd hit this condition but
683 * double-fault even before we get this far, in which case
684 * we're fine: the double-fault handler will deal with it.
686 * We don't want to make it all the way into the oops code
687 * and then double-fault, though, because we're likely to
688 * break the console driver and lose most of the stack dump.
690 asm volatile ("movq %[stack], %%rsp\n\t"
691 "call handle_stack_overflow\n\t"
693 : ASM_CALL_CONSTRAINT
694 : "D" ("kernel stack overflow (page fault)"),
695 "S" (regs), "d" (address),
696 [stack] "rm" (stack));
704 * Valid to do another page fault here, because if this fault
705 * had been triggered by is_prefetch fixup_exception would have
710 * Hall of shame of CPU/BIOS bugs.
712 if (is_prefetch(regs, error_code, address))
715 if (is_errata93(regs, address))
719 * Oops. The kernel tried to access some bad page. We'll have to
720 * terminate things with extreme prejudice:
722 flags = oops_begin();
724 show_fault_oops(regs, error_code, address);
726 if (task_stack_end_corrupted(tsk))
727 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
729 tsk->thread.cr2 = address;
730 tsk->thread.trap_nr = X86_TRAP_PF;
731 tsk->thread.error_code = error_code;
734 if (__die("Oops", regs, error_code))
737 /* Executive summary in case the body of the oops scrolled away */
738 printk(KERN_DEFAULT "CR2: %016lx\n", address);
740 oops_end(flags, regs, sig);
744 * Print out info about fatal segfaults, if the show_unhandled_signals
748 show_signal_msg(struct pt_regs *regs, unsigned long error_code,
749 unsigned long address, struct task_struct *tsk)
751 const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG;
753 if (!unhandled_signal(tsk, SIGSEGV))
756 if (!printk_ratelimit())
759 printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx",
760 loglvl, tsk->comm, task_pid_nr(tsk), address,
761 (void *)regs->ip, (void *)regs->sp, error_code);
763 print_vma_addr(KERN_CONT " in ", regs->ip);
765 printk(KERN_CONT "\n");
767 show_opcodes((u8 *)regs->ip, loglvl);
771 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
772 unsigned long address, u32 pkey, int si_code)
774 struct task_struct *tsk = current;
776 /* User mode accesses just cause a SIGSEGV */
777 if (error_code & X86_PF_USER) {
779 * It's possible to have interrupts off here:
784 * Valid to do another page fault here because this one came
787 if (is_prefetch(regs, error_code, address))
790 if (is_errata100(regs, address))
795 * Instruction fetch faults in the vsyscall page might need
798 if (unlikely((error_code & X86_PF_INSTR) &&
799 ((address & ~0xfff) == VSYSCALL_ADDR))) {
800 if (emulate_vsyscall(regs, address))
806 * To avoid leaking information about the kernel page table
807 * layout, pretend that user-mode accesses to kernel addresses
808 * are always protection faults.
810 if (address >= TASK_SIZE_MAX)
811 error_code |= X86_PF_PROT;
813 if (likely(show_unhandled_signals))
814 show_signal_msg(regs, error_code, address, tsk);
816 tsk->thread.cr2 = address;
817 tsk->thread.error_code = error_code;
818 tsk->thread.trap_nr = X86_TRAP_PF;
820 if (si_code == SEGV_PKUERR)
821 force_sig_pkuerr((void __user *)address, pkey);
823 force_sig_fault(SIGSEGV, si_code, (void __user *)address, tsk);
828 if (is_f00f_bug(regs, address))
831 no_context(regs, error_code, address, SIGSEGV, si_code);
835 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
836 unsigned long address)
838 __bad_area_nosemaphore(regs, error_code, address, 0, SEGV_MAPERR);
842 __bad_area(struct pt_regs *regs, unsigned long error_code,
843 unsigned long address, u32 pkey, int si_code)
845 struct mm_struct *mm = current->mm;
847 * Something tried to access memory that isn't in our memory map..
848 * Fix it, but check if it's kernel or user first..
850 up_read(&mm->mmap_sem);
852 __bad_area_nosemaphore(regs, error_code, address, pkey, si_code);
856 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
858 __bad_area(regs, error_code, address, 0, SEGV_MAPERR);
861 static inline bool bad_area_access_from_pkeys(unsigned long error_code,
862 struct vm_area_struct *vma)
864 /* This code is always called on the current mm */
865 bool foreign = false;
867 if (!boot_cpu_has(X86_FEATURE_OSPKE))
869 if (error_code & X86_PF_PK)
871 /* this checks permission keys on the VMA: */
872 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
873 (error_code & X86_PF_INSTR), foreign))
879 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
880 unsigned long address, struct vm_area_struct *vma)
883 * This OSPKE check is not strictly necessary at runtime.
884 * But, doing it this way allows compiler optimizations
885 * if pkeys are compiled out.
887 if (bad_area_access_from_pkeys(error_code, vma)) {
889 * A protection key fault means that the PKRU value did not allow
890 * access to some PTE. Userspace can figure out what PKRU was
891 * from the XSAVE state. This function captures the pkey from
892 * the vma and passes it to userspace so userspace can discover
893 * which protection key was set on the PTE.
895 * If we get here, we know that the hardware signaled a X86_PF_PK
896 * fault and that there was a VMA once we got in the fault
897 * handler. It does *not* guarantee that the VMA we find here
898 * was the one that we faulted on.
900 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4);
901 * 2. T1 : set PKRU to deny access to pkey=4, touches page
903 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
904 * 5. T1 : enters fault handler, takes mmap_sem, etc...
905 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really
906 * faulted on a pte with its pkey=4.
908 u32 pkey = vma_pkey(vma);
910 __bad_area(regs, error_code, address, pkey, SEGV_PKUERR);
912 __bad_area(regs, error_code, address, 0, SEGV_ACCERR);
917 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
920 struct task_struct *tsk = current;
922 /* Kernel mode? Handle exceptions or die: */
923 if (!(error_code & X86_PF_USER)) {
924 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
928 /* User-space => ok to do another page fault: */
929 if (is_prefetch(regs, error_code, address))
932 tsk->thread.cr2 = address;
933 tsk->thread.error_code = error_code;
934 tsk->thread.trap_nr = X86_TRAP_PF;
936 #ifdef CONFIG_MEMORY_FAILURE
937 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
941 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
942 tsk->comm, tsk->pid, address);
943 if (fault & VM_FAULT_HWPOISON_LARGE)
944 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
945 if (fault & VM_FAULT_HWPOISON)
947 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb, tsk);
951 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address, tsk);
955 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
956 unsigned long address, vm_fault_t fault)
958 if (fatal_signal_pending(current) && !(error_code & X86_PF_USER)) {
959 no_context(regs, error_code, address, 0, 0);
963 if (fault & VM_FAULT_OOM) {
964 /* Kernel mode? Handle exceptions or die: */
965 if (!(error_code & X86_PF_USER)) {
966 no_context(regs, error_code, address,
967 SIGSEGV, SEGV_MAPERR);
972 * We ran out of memory, call the OOM killer, and return the
973 * userspace (which will retry the fault, or kill us if we got
976 pagefault_out_of_memory();
978 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
979 VM_FAULT_HWPOISON_LARGE))
980 do_sigbus(regs, error_code, address, fault);
981 else if (fault & VM_FAULT_SIGSEGV)
982 bad_area_nosemaphore(regs, error_code, address);
988 static int spurious_fault_check(unsigned long error_code, pte_t *pte)
990 if ((error_code & X86_PF_WRITE) && !pte_write(*pte))
993 if ((error_code & X86_PF_INSTR) && !pte_exec(*pte))
996 * Note: We do not do lazy flushing on protection key
997 * changes, so no spurious fault will ever set X86_PF_PK.
999 if ((error_code & X86_PF_PK))
1006 * Handle a spurious fault caused by a stale TLB entry.
1008 * This allows us to lazily refresh the TLB when increasing the
1009 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
1010 * eagerly is very expensive since that implies doing a full
1011 * cross-processor TLB flush, even if no stale TLB entries exist
1012 * on other processors.
1014 * Spurious faults may only occur if the TLB contains an entry with
1015 * fewer permission than the page table entry. Non-present (P = 0)
1016 * and reserved bit (R = 1) faults are never spurious.
1018 * There are no security implications to leaving a stale TLB when
1019 * increasing the permissions on a page.
1021 * Returns non-zero if a spurious fault was handled, zero otherwise.
1023 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
1024 * (Optional Invalidation).
1027 spurious_fault(unsigned long error_code, unsigned long address)
1037 * Only writes to RO or instruction fetches from NX may cause
1040 * These could be from user or supervisor accesses but the TLB
1041 * is only lazily flushed after a kernel mapping protection
1042 * change, so user accesses are not expected to cause spurious
1045 if (error_code != (X86_PF_WRITE | X86_PF_PROT) &&
1046 error_code != (X86_PF_INSTR | X86_PF_PROT))
1049 pgd = init_mm.pgd + pgd_index(address);
1050 if (!pgd_present(*pgd))
1053 p4d = p4d_offset(pgd, address);
1054 if (!p4d_present(*p4d))
1057 if (p4d_large(*p4d))
1058 return spurious_fault_check(error_code, (pte_t *) p4d);
1060 pud = pud_offset(p4d, address);
1061 if (!pud_present(*pud))
1064 if (pud_large(*pud))
1065 return spurious_fault_check(error_code, (pte_t *) pud);
1067 pmd = pmd_offset(pud, address);
1068 if (!pmd_present(*pmd))
1071 if (pmd_large(*pmd))
1072 return spurious_fault_check(error_code, (pte_t *) pmd);
1074 pte = pte_offset_kernel(pmd, address);
1075 if (!pte_present(*pte))
1078 ret = spurious_fault_check(error_code, pte);
1083 * Make sure we have permissions in PMD.
1084 * If not, then there's a bug in the page tables:
1086 ret = spurious_fault_check(error_code, (pte_t *) pmd);
1087 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
1091 NOKPROBE_SYMBOL(spurious_fault);
1093 int show_unhandled_signals = 1;
1096 access_error(unsigned long error_code, struct vm_area_struct *vma)
1098 /* This is only called for the current mm, so: */
1099 bool foreign = false;
1102 * Read or write was blocked by protection keys. This is
1103 * always an unconditional error and can never result in
1104 * a follow-up action to resolve the fault, like a COW.
1106 if (error_code & X86_PF_PK)
1110 * Make sure to check the VMA so that we do not perform
1111 * faults just to hit a X86_PF_PK as soon as we fill in a
1114 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
1115 (error_code & X86_PF_INSTR), foreign))
1118 if (error_code & X86_PF_WRITE) {
1119 /* write, present and write, not present: */
1120 if (unlikely(!(vma->vm_flags & VM_WRITE)))
1125 /* read, present: */
1126 if (unlikely(error_code & X86_PF_PROT))
1129 /* read, not present: */
1130 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
1136 static int fault_in_kernel_space(unsigned long address)
1138 return address >= TASK_SIZE_MAX;
1141 static inline bool smap_violation(int error_code, struct pt_regs *regs)
1143 if (!IS_ENABLED(CONFIG_X86_SMAP))
1146 if (!static_cpu_has(X86_FEATURE_SMAP))
1149 if (error_code & X86_PF_USER)
1152 if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC))
1159 * This routine handles page faults. It determines the address,
1160 * and the problem, and then passes it off to one of the appropriate
1163 static noinline void
1164 __do_page_fault(struct pt_regs *regs, unsigned long error_code,
1165 unsigned long address)
1167 struct vm_area_struct *vma;
1168 struct task_struct *tsk;
1169 struct mm_struct *mm;
1170 vm_fault_t fault, major = 0;
1171 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
1176 prefetchw(&mm->mmap_sem);
1178 if (unlikely(kmmio_fault(regs, address)))
1182 * We fault-in kernel-space virtual memory on-demand. The
1183 * 'reference' page table is init_mm.pgd.
1185 * NOTE! We MUST NOT take any locks for this case. We may
1186 * be in an interrupt or a critical region, and should
1187 * only copy the information from the master page table,
1190 * This verifies that the fault happens in kernel space
1191 * (error_code & 4) == 0, and that the fault was not a
1192 * protection error (error_code & 9) == 0.
1194 if (unlikely(fault_in_kernel_space(address))) {
1195 if (!(error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) {
1196 if (vmalloc_fault(address) >= 0)
1200 /* Can handle a stale RO->RW TLB: */
1201 if (spurious_fault(error_code, address))
1204 /* kprobes don't want to hook the spurious faults: */
1205 if (kprobes_fault(regs))
1208 * Don't take the mm semaphore here. If we fixup a prefetch
1209 * fault we could otherwise deadlock:
1211 bad_area_nosemaphore(regs, error_code, address);
1216 /* kprobes don't want to hook the spurious faults: */
1217 if (unlikely(kprobes_fault(regs)))
1220 if (unlikely(error_code & X86_PF_RSVD))
1221 pgtable_bad(regs, error_code, address);
1223 if (unlikely(smap_violation(error_code, regs))) {
1224 bad_area_nosemaphore(regs, error_code, address);
1229 * If we're in an interrupt, have no user context or are running
1230 * in a region with pagefaults disabled then we must not take the fault
1232 if (unlikely(faulthandler_disabled() || !mm)) {
1233 bad_area_nosemaphore(regs, error_code, address);
1238 * It's safe to allow irq's after cr2 has been saved and the
1239 * vmalloc fault has been handled.
1241 * User-mode registers count as a user access even for any
1242 * potential system fault or CPU buglet:
1244 if (user_mode(regs)) {
1246 error_code |= X86_PF_USER;
1247 flags |= FAULT_FLAG_USER;
1249 if (regs->flags & X86_EFLAGS_IF)
1253 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
1255 if (error_code & X86_PF_WRITE)
1256 flags |= FAULT_FLAG_WRITE;
1257 if (error_code & X86_PF_INSTR)
1258 flags |= FAULT_FLAG_INSTRUCTION;
1261 * When running in the kernel we expect faults to occur only to
1262 * addresses in user space. All other faults represent errors in
1263 * the kernel and should generate an OOPS. Unfortunately, in the
1264 * case of an erroneous fault occurring in a code path which already
1265 * holds mmap_sem we will deadlock attempting to validate the fault
1266 * against the address space. Luckily the kernel only validly
1267 * references user space from well defined areas of code, which are
1268 * listed in the exceptions table.
1270 * As the vast majority of faults will be valid we will only perform
1271 * the source reference check when there is a possibility of a
1272 * deadlock. Attempt to lock the address space, if we cannot we then
1273 * validate the source. If this is invalid we can skip the address
1274 * space check, thus avoiding the deadlock:
1276 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
1277 if (!(error_code & X86_PF_USER) &&
1278 !search_exception_tables(regs->ip)) {
1279 bad_area_nosemaphore(regs, error_code, address);
1283 down_read(&mm->mmap_sem);
1286 * The above down_read_trylock() might have succeeded in
1287 * which case we'll have missed the might_sleep() from
1293 vma = find_vma(mm, address);
1294 if (unlikely(!vma)) {
1295 bad_area(regs, error_code, address);
1298 if (likely(vma->vm_start <= address))
1300 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1301 bad_area(regs, error_code, address);
1304 if (error_code & X86_PF_USER) {
1306 * Accessing the stack below %sp is always a bug.
1307 * The large cushion allows instructions like enter
1308 * and pusha to work. ("enter $65535, $31" pushes
1309 * 32 pointers and then decrements %sp by 65535.)
1311 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) {
1312 bad_area(regs, error_code, address);
1316 if (unlikely(expand_stack(vma, address))) {
1317 bad_area(regs, error_code, address);
1322 * Ok, we have a good vm_area for this memory access, so
1323 * we can handle it..
1326 if (unlikely(access_error(error_code, vma))) {
1327 bad_area_access_error(regs, error_code, address, vma);
1332 * If for any reason at all we couldn't handle the fault,
1333 * make sure we exit gracefully rather than endlessly redo
1334 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1335 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
1337 * Note that handle_userfault() may also release and reacquire mmap_sem
1338 * (and not return with VM_FAULT_RETRY), when returning to userland to
1339 * repeat the page fault later with a VM_FAULT_NOPAGE retval
1340 * (potentially after handling any pending signal during the return to
1341 * userland). The return to userland is identified whenever
1342 * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags.
1344 fault = handle_mm_fault(vma, address, flags);
1345 major |= fault & VM_FAULT_MAJOR;
1348 * If we need to retry the mmap_sem has already been released,
1349 * and if there is a fatal signal pending there is no guarantee
1350 * that we made any progress. Handle this case first.
1352 if (unlikely(fault & VM_FAULT_RETRY)) {
1353 /* Retry at most once */
1354 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1355 flags &= ~FAULT_FLAG_ALLOW_RETRY;
1356 flags |= FAULT_FLAG_TRIED;
1357 if (!fatal_signal_pending(tsk))
1361 /* User mode? Just return to handle the fatal exception */
1362 if (flags & FAULT_FLAG_USER)
1365 /* Not returning to user mode? Handle exceptions or die: */
1366 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
1370 up_read(&mm->mmap_sem);
1371 if (unlikely(fault & VM_FAULT_ERROR)) {
1372 mm_fault_error(regs, error_code, address, fault);
1377 * Major/minor page fault accounting. If any of the events
1378 * returned VM_FAULT_MAJOR, we account it as a major fault.
1382 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
1385 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
1388 check_v8086_mode(regs, address, tsk);
1390 NOKPROBE_SYMBOL(__do_page_fault);
1392 static nokprobe_inline void
1393 trace_page_fault_entries(unsigned long address, struct pt_regs *regs,
1394 unsigned long error_code)
1396 if (user_mode(regs))
1397 trace_page_fault_user(address, regs, error_code);
1399 trace_page_fault_kernel(address, regs, error_code);
1403 * We must have this function blacklisted from kprobes, tagged with notrace
1404 * and call read_cr2() before calling anything else. To avoid calling any
1405 * kind of tracing machinery before we've observed the CR2 value.
1407 * exception_{enter,exit}() contains all sorts of tracepoints.
1409 dotraplinkage void notrace
1410 do_page_fault(struct pt_regs *regs, unsigned long error_code)
1412 unsigned long address = read_cr2(); /* Get the faulting address */
1413 enum ctx_state prev_state;
1415 prev_state = exception_enter();
1416 if (trace_pagefault_enabled())
1417 trace_page_fault_entries(address, regs, error_code);
1419 __do_page_fault(regs, error_code, address);
1420 exception_exit(prev_state);
1422 NOKPROBE_SYMBOL(do_page_fault);