2 * Copyright (C) 1995 Linus Torvalds
3 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs.
4 * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar
6 #include <linux/sched.h> /* test_thread_flag(), ... */
7 #include <linux/sched/task_stack.h> /* task_stack_*(), ... */
8 #include <linux/kdebug.h> /* oops_begin/end, ... */
9 #include <linux/extable.h> /* search_exception_tables */
10 #include <linux/bootmem.h> /* max_low_pfn */
11 #include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */
12 #include <linux/mmiotrace.h> /* kmmio_handler, ... */
13 #include <linux/perf_event.h> /* perf_sw_event */
14 #include <linux/hugetlb.h> /* hstate_index_to_shift */
15 #include <linux/prefetch.h> /* prefetchw */
16 #include <linux/context_tracking.h> /* exception_enter(), ... */
17 #include <linux/uaccess.h> /* faulthandler_disabled() */
19 #include <asm/cpufeature.h> /* boot_cpu_has, ... */
20 #include <asm/traps.h> /* dotraplinkage, ... */
21 #include <asm/pgalloc.h> /* pgd_*(), ... */
22 #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */
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() */
28 #define CREATE_TRACE_POINTS
29 #include <asm/trace/exceptions.h>
32 * Page fault error code bits:
34 * bit 0 == 0: no page found 1: protection fault
35 * bit 1 == 0: read access 1: write access
36 * bit 2 == 0: kernel-mode access 1: user-mode access
37 * bit 3 == 1: use of reserved bit detected
38 * bit 4 == 1: fault was an instruction fetch
39 * bit 5 == 1: protection keys block access
41 enum x86_pf_error_code {
52 * Returns 0 if mmiotrace is disabled, or if the fault is not
53 * handled by mmiotrace:
55 static nokprobe_inline int
56 kmmio_fault(struct pt_regs *regs, unsigned long addr)
58 if (unlikely(is_kmmio_active()))
59 if (kmmio_handler(regs, addr) == 1)
64 static nokprobe_inline int kprobes_fault(struct pt_regs *regs)
68 /* kprobe_running() needs smp_processor_id() */
69 if (kprobes_built_in() && !user_mode(regs)) {
71 if (kprobe_running() && kprobe_fault_handler(regs, 14))
84 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
85 * Check that here and ignore it.
89 * Sometimes the CPU reports invalid exceptions on prefetch.
90 * Check that here and ignore it.
92 * Opcode checker based on code by Richard Brunner.
95 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
96 unsigned char opcode, int *prefetch)
98 unsigned char instr_hi = opcode & 0xf0;
99 unsigned char instr_lo = opcode & 0x0f;
105 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
106 * In X86_64 long mode, the CPU will signal invalid
107 * opcode if some of these prefixes are present so
108 * X86_64 will never get here anyway
110 return ((instr_lo & 7) == 0x6);
114 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
115 * Need to figure out under what instruction mode the
116 * instruction was issued. Could check the LDT for lm,
117 * but for now it's good enough to assume that long
118 * mode only uses well known segments or kernel.
120 return (!user_mode(regs) || user_64bit_mode(regs));
123 /* 0x64 thru 0x67 are valid prefixes in all modes. */
124 return (instr_lo & 0xC) == 0x4;
126 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
127 return !instr_lo || (instr_lo>>1) == 1;
129 /* Prefetch instruction is 0x0F0D or 0x0F18 */
130 if (probe_kernel_address(instr, opcode))
133 *prefetch = (instr_lo == 0xF) &&
134 (opcode == 0x0D || opcode == 0x18);
142 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
144 unsigned char *max_instr;
145 unsigned char *instr;
149 * If it was a exec (instruction fetch) fault on NX page, then
150 * do not ignore the fault:
152 if (error_code & PF_INSTR)
155 instr = (void *)convert_ip_to_linear(current, regs);
156 max_instr = instr + 15;
158 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX)
161 while (instr < max_instr) {
162 unsigned char opcode;
164 if (probe_kernel_address(instr, opcode))
169 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
176 * A protection key fault means that the PKRU value did not allow
177 * access to some PTE. Userspace can figure out what PKRU was
178 * from the XSAVE state, and this function fills out a field in
179 * siginfo so userspace can discover which protection key was set
182 * If we get here, we know that the hardware signaled a PF_PK
183 * fault and that there was a VMA once we got in the fault
184 * handler. It does *not* guarantee that the VMA we find here
185 * was the one that we faulted on.
187 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4);
188 * 2. T1 : set PKRU to deny access to pkey=4, touches page
190 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
191 * 5. T1 : enters fault handler, takes mmap_sem, etc...
192 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really
193 * faulted on a pte with its pkey=4.
195 static void fill_sig_info_pkey(int si_code, siginfo_t *info,
196 struct vm_area_struct *vma)
198 /* This is effectively an #ifdef */
199 if (!boot_cpu_has(X86_FEATURE_OSPKE))
202 /* Fault not from Protection Keys: nothing to do */
203 if (si_code != SEGV_PKUERR)
206 * force_sig_info_fault() is called from a number of
207 * contexts, some of which have a VMA and some of which
208 * do not. The PF_PK handing happens after we have a
209 * valid VMA, so we should never reach this without a
213 WARN_ONCE(1, "PKU fault with no VMA passed in");
218 * si_pkey should be thought of as a strong hint, but not
219 * absolutely guranteed to be 100% accurate because of
220 * the race explained above.
222 info->si_pkey = vma_pkey(vma);
226 force_sig_info_fault(int si_signo, int si_code, unsigned long address,
227 struct task_struct *tsk, struct vm_area_struct *vma,
233 info.si_signo = si_signo;
235 info.si_code = si_code;
236 info.si_addr = (void __user *)address;
237 if (fault & VM_FAULT_HWPOISON_LARGE)
238 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
239 if (fault & VM_FAULT_HWPOISON)
241 info.si_addr_lsb = lsb;
243 fill_sig_info_pkey(si_code, &info, vma);
245 force_sig_info(si_signo, &info, tsk);
248 DEFINE_SPINLOCK(pgd_lock);
252 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
254 unsigned index = pgd_index(address);
261 pgd_k = init_mm.pgd + index;
263 if (!pgd_present(*pgd_k))
267 * set_pgd(pgd, *pgd_k); here would be useless on PAE
268 * and redundant with the set_pmd() on non-PAE. As would
271 p4d = p4d_offset(pgd, address);
272 p4d_k = p4d_offset(pgd_k, address);
273 if (!p4d_present(*p4d_k))
276 pud = pud_offset(p4d, address);
277 pud_k = pud_offset(p4d_k, address);
278 if (!pud_present(*pud_k))
281 pmd = pmd_offset(pud, address);
282 pmd_k = pmd_offset(pud_k, address);
283 if (!pmd_present(*pmd_k))
286 if (!pmd_present(*pmd))
287 set_pmd(pmd, *pmd_k);
289 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
294 void vmalloc_sync_all(void)
296 unsigned long address;
298 if (SHARED_KERNEL_PMD)
301 for (address = VMALLOC_START & PMD_MASK;
302 address >= TASK_SIZE_MAX && address < FIXADDR_TOP;
303 address += PMD_SIZE) {
306 spin_lock(&pgd_lock);
307 list_for_each_entry(page, &pgd_list, lru) {
308 spinlock_t *pgt_lock;
311 /* the pgt_lock only for Xen */
312 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
315 ret = vmalloc_sync_one(page_address(page), address);
316 spin_unlock(pgt_lock);
321 spin_unlock(&pgd_lock);
328 * Handle a fault on the vmalloc or module mapping area
330 static noinline int vmalloc_fault(unsigned long address)
332 unsigned long pgd_paddr;
336 /* Make sure we are in vmalloc area: */
337 if (!(address >= VMALLOC_START && address < VMALLOC_END))
340 WARN_ON_ONCE(in_nmi());
343 * Synchronize this task's top level page-table
344 * with the 'reference' page table.
346 * Do _not_ use "current" here. We might be inside
347 * an interrupt in the middle of a task switch..
349 pgd_paddr = read_cr3_pa();
350 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
354 if (pmd_huge(*pmd_k))
357 pte_k = pte_offset_kernel(pmd_k, address);
358 if (!pte_present(*pte_k))
363 NOKPROBE_SYMBOL(vmalloc_fault);
366 * Did it hit the DOS screen memory VA from vm86 mode?
369 check_v8086_mode(struct pt_regs *regs, unsigned long address,
370 struct task_struct *tsk)
375 if (!v8086_mode(regs) || !tsk->thread.vm86)
378 bit = (address - 0xA0000) >> PAGE_SHIFT;
380 tsk->thread.vm86->screen_bitmap |= 1 << bit;
384 static bool low_pfn(unsigned long pfn)
386 return pfn < max_low_pfn;
389 static void dump_pagetable(unsigned long address)
391 pgd_t *base = __va(read_cr3_pa());
392 pgd_t *pgd = &base[pgd_index(address)];
398 #ifdef CONFIG_X86_PAE
399 pr_info("*pdpt = %016Lx ", pgd_val(*pgd));
400 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
402 #define pr_pde pr_cont
404 #define pr_pde pr_info
406 p4d = p4d_offset(pgd, address);
407 pud = pud_offset(p4d, address);
408 pmd = pmd_offset(pud, address);
409 pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
413 * We must not directly access the pte in the highpte
414 * case if the page table is located in highmem.
415 * And let's rather not kmap-atomic the pte, just in case
416 * it's allocated already:
418 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
421 pte = pte_offset_kernel(pmd, address);
422 pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
427 #else /* CONFIG_X86_64: */
429 void vmalloc_sync_all(void)
431 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END);
437 * Handle a fault on the vmalloc area
439 static noinline int vmalloc_fault(unsigned long address)
441 pgd_t *pgd, *pgd_ref;
442 p4d_t *p4d, *p4d_ref;
443 pud_t *pud, *pud_ref;
444 pmd_t *pmd, *pmd_ref;
445 pte_t *pte, *pte_ref;
447 /* Make sure we are in vmalloc area: */
448 if (!(address >= VMALLOC_START && address < VMALLOC_END))
451 WARN_ON_ONCE(in_nmi());
454 * Copy kernel mappings over when needed. This can also
455 * happen within a race in page table update. In the later
458 pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address);
459 pgd_ref = pgd_offset_k(address);
460 if (pgd_none(*pgd_ref))
463 if (pgd_none(*pgd)) {
464 set_pgd(pgd, *pgd_ref);
465 arch_flush_lazy_mmu_mode();
466 } else if (CONFIG_PGTABLE_LEVELS > 4) {
468 * With folded p4d, pgd_none() is always false, so the pgd may
469 * point to an empty page table entry and pgd_page_vaddr()
470 * will return garbage.
472 * We will do the correct sanity check on the p4d level.
474 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
477 /* With 4-level paging, copying happens on the p4d level. */
478 p4d = p4d_offset(pgd, address);
479 p4d_ref = p4d_offset(pgd_ref, address);
480 if (p4d_none(*p4d_ref))
483 if (p4d_none(*p4d)) {
484 set_p4d(p4d, *p4d_ref);
485 arch_flush_lazy_mmu_mode();
487 BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_ref));
491 * Below here mismatches are bugs because these lower tables
495 pud = pud_offset(p4d, address);
496 pud_ref = pud_offset(p4d_ref, address);
497 if (pud_none(*pud_ref))
500 if (pud_none(*pud) || pud_pfn(*pud) != pud_pfn(*pud_ref))
506 pmd = pmd_offset(pud, address);
507 pmd_ref = pmd_offset(pud_ref, address);
508 if (pmd_none(*pmd_ref))
511 if (pmd_none(*pmd) || pmd_pfn(*pmd) != pmd_pfn(*pmd_ref))
517 pte_ref = pte_offset_kernel(pmd_ref, address);
518 if (!pte_present(*pte_ref))
521 pte = pte_offset_kernel(pmd, address);
524 * Don't use pte_page here, because the mappings can point
525 * outside mem_map, and the NUMA hash lookup cannot handle
528 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
533 NOKPROBE_SYMBOL(vmalloc_fault);
535 #ifdef CONFIG_CPU_SUP_AMD
536 static const char errata93_warning[] =
538 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
539 "******* Working around it, but it may cause SEGVs or burn power.\n"
540 "******* Please consider a BIOS update.\n"
541 "******* Disabling USB legacy in the BIOS may also help.\n";
545 * No vm86 mode in 64-bit mode:
548 check_v8086_mode(struct pt_regs *regs, unsigned long address,
549 struct task_struct *tsk)
553 static int bad_address(void *p)
557 return probe_kernel_address((unsigned long *)p, dummy);
560 static void dump_pagetable(unsigned long address)
562 pgd_t *base = __va(read_cr3_pa());
563 pgd_t *pgd = base + pgd_index(address);
569 if (bad_address(pgd))
572 pr_info("PGD %lx ", pgd_val(*pgd));
574 if (!pgd_present(*pgd))
577 p4d = p4d_offset(pgd, address);
578 if (bad_address(p4d))
581 pr_cont("P4D %lx ", p4d_val(*p4d));
582 if (!p4d_present(*p4d) || p4d_large(*p4d))
585 pud = pud_offset(p4d, address);
586 if (bad_address(pud))
589 pr_cont("PUD %lx ", pud_val(*pud));
590 if (!pud_present(*pud) || pud_large(*pud))
593 pmd = pmd_offset(pud, address);
594 if (bad_address(pmd))
597 pr_cont("PMD %lx ", pmd_val(*pmd));
598 if (!pmd_present(*pmd) || pmd_large(*pmd))
601 pte = pte_offset_kernel(pmd, address);
602 if (bad_address(pte))
605 pr_cont("PTE %lx", pte_val(*pte));
613 #endif /* CONFIG_X86_64 */
616 * Workaround for K8 erratum #93 & buggy BIOS.
618 * BIOS SMM functions are required to use a specific workaround
619 * to avoid corruption of the 64bit RIP register on C stepping K8.
621 * A lot of BIOS that didn't get tested properly miss this.
623 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
624 * Try to work around it here.
626 * Note we only handle faults in kernel here.
627 * Does nothing on 32-bit.
629 static int is_errata93(struct pt_regs *regs, unsigned long address)
631 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
632 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD
633 || boot_cpu_data.x86 != 0xf)
636 if (address != regs->ip)
639 if ((address >> 32) != 0)
642 address |= 0xffffffffUL << 32;
643 if ((address >= (u64)_stext && address <= (u64)_etext) ||
644 (address >= MODULES_VADDR && address <= MODULES_END)) {
645 printk_once(errata93_warning);
654 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
655 * to illegal addresses >4GB.
657 * We catch this in the page fault handler because these addresses
658 * are not reachable. Just detect this case and return. Any code
659 * segment in LDT is compatibility mode.
661 static int is_errata100(struct pt_regs *regs, unsigned long address)
664 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
670 static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
672 #ifdef CONFIG_X86_F00F_BUG
676 * Pentium F0 0F C7 C8 bug workaround:
678 if (boot_cpu_has_bug(X86_BUG_F00F)) {
679 nr = (address - idt_descr.address) >> 3;
682 do_invalid_op(regs, 0);
690 static const char nx_warning[] = KERN_CRIT
691 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
692 static const char smep_warning[] = KERN_CRIT
693 "unable to execute userspace code (SMEP?) (uid: %d)\n";
696 show_fault_oops(struct pt_regs *regs, unsigned long error_code,
697 unsigned long address)
699 if (!oops_may_print())
702 if (error_code & PF_INSTR) {
707 pgd = __va(read_cr3_pa());
708 pgd += pgd_index(address);
710 pte = lookup_address_in_pgd(pgd, address, &level);
712 if (pte && pte_present(*pte) && !pte_exec(*pte))
713 printk(nx_warning, from_kuid(&init_user_ns, current_uid()));
714 if (pte && pte_present(*pte) && pte_exec(*pte) &&
715 (pgd_flags(*pgd) & _PAGE_USER) &&
716 (__read_cr4() & X86_CR4_SMEP))
717 printk(smep_warning, from_kuid(&init_user_ns, current_uid()));
720 printk(KERN_ALERT "BUG: unable to handle kernel ");
721 if (address < PAGE_SIZE)
722 printk(KERN_CONT "NULL pointer dereference");
724 printk(KERN_CONT "paging request");
726 printk(KERN_CONT " at %p\n", (void *) address);
727 printk(KERN_ALERT "IP: %pS\n", (void *)regs->ip);
729 dump_pagetable(address);
733 pgtable_bad(struct pt_regs *regs, unsigned long error_code,
734 unsigned long address)
736 struct task_struct *tsk;
740 flags = oops_begin();
744 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
746 dump_pagetable(address);
748 tsk->thread.cr2 = address;
749 tsk->thread.trap_nr = X86_TRAP_PF;
750 tsk->thread.error_code = error_code;
752 if (__die("Bad pagetable", regs, error_code))
755 oops_end(flags, regs, sig);
759 no_context(struct pt_regs *regs, unsigned long error_code,
760 unsigned long address, int signal, int si_code)
762 struct task_struct *tsk = current;
765 /* No context means no VMA to pass down */
766 struct vm_area_struct *vma = NULL;
768 /* Are we prepared to handle this kernel fault? */
769 if (fixup_exception(regs, X86_TRAP_PF)) {
771 * Any interrupt that takes a fault gets the fixup. This makes
772 * the below recursive fault logic only apply to a faults from
779 * Per the above we're !in_interrupt(), aka. task context.
781 * In this case we need to make sure we're not recursively
782 * faulting through the emulate_vsyscall() logic.
784 if (current->thread.sig_on_uaccess_err && signal) {
785 tsk->thread.trap_nr = X86_TRAP_PF;
786 tsk->thread.error_code = error_code | PF_USER;
787 tsk->thread.cr2 = address;
789 /* XXX: hwpoison faults will set the wrong code. */
790 force_sig_info_fault(signal, si_code, address,
795 * Barring that, we can do the fixup and be happy.
800 #ifdef CONFIG_VMAP_STACK
802 * Stack overflow? During boot, we can fault near the initial
803 * stack in the direct map, but that's not an overflow -- check
804 * that we're in vmalloc space to avoid this.
806 if (is_vmalloc_addr((void *)address) &&
807 (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) ||
808 address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) {
809 unsigned long stack = this_cpu_read(orig_ist.ist[DOUBLEFAULT_STACK]) - sizeof(void *);
811 * We're likely to be running with very little stack space
812 * left. It's plausible that we'd hit this condition but
813 * double-fault even before we get this far, in which case
814 * we're fine: the double-fault handler will deal with it.
816 * We don't want to make it all the way into the oops code
817 * and then double-fault, though, because we're likely to
818 * break the console driver and lose most of the stack dump.
820 asm volatile ("movq %[stack], %%rsp\n\t"
821 "call handle_stack_overflow\n\t"
823 : ASM_CALL_CONSTRAINT
824 : "D" ("kernel stack overflow (page fault)"),
825 "S" (regs), "d" (address),
826 [stack] "rm" (stack));
834 * Valid to do another page fault here, because if this fault
835 * had been triggered by is_prefetch fixup_exception would have
840 * Hall of shame of CPU/BIOS bugs.
842 if (is_prefetch(regs, error_code, address))
845 if (is_errata93(regs, address))
849 * Oops. The kernel tried to access some bad page. We'll have to
850 * terminate things with extreme prejudice:
852 flags = oops_begin();
854 show_fault_oops(regs, error_code, address);
856 if (task_stack_end_corrupted(tsk))
857 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
859 tsk->thread.cr2 = address;
860 tsk->thread.trap_nr = X86_TRAP_PF;
861 tsk->thread.error_code = error_code;
864 if (__die("Oops", regs, error_code))
867 /* Executive summary in case the body of the oops scrolled away */
868 printk(KERN_DEFAULT "CR2: %016lx\n", address);
870 oops_end(flags, regs, sig);
874 * Print out info about fatal segfaults, if the show_unhandled_signals
878 show_signal_msg(struct pt_regs *regs, unsigned long error_code,
879 unsigned long address, struct task_struct *tsk)
881 if (!unhandled_signal(tsk, SIGSEGV))
884 if (!printk_ratelimit())
887 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
888 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
889 tsk->comm, task_pid_nr(tsk), address,
890 (void *)regs->ip, (void *)regs->sp, error_code);
892 print_vma_addr(KERN_CONT " in ", regs->ip);
894 printk(KERN_CONT "\n");
898 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
899 unsigned long address, struct vm_area_struct *vma,
902 struct task_struct *tsk = current;
904 /* User mode accesses just cause a SIGSEGV */
905 if (error_code & PF_USER) {
907 * It's possible to have interrupts off here:
912 * Valid to do another page fault here because this one came
915 if (is_prefetch(regs, error_code, address))
918 if (is_errata100(regs, address))
923 * Instruction fetch faults in the vsyscall page might need
926 if (unlikely((error_code & PF_INSTR) &&
927 ((address & ~0xfff) == VSYSCALL_ADDR))) {
928 if (emulate_vsyscall(regs, address))
934 * To avoid leaking information about the kernel page table
935 * layout, pretend that user-mode accesses to kernel addresses
936 * are always protection faults.
938 if (address >= TASK_SIZE_MAX)
939 error_code |= PF_PROT;
941 if (likely(show_unhandled_signals))
942 show_signal_msg(regs, error_code, address, tsk);
944 tsk->thread.cr2 = address;
945 tsk->thread.error_code = error_code;
946 tsk->thread.trap_nr = X86_TRAP_PF;
948 force_sig_info_fault(SIGSEGV, si_code, address, tsk, vma, 0);
953 if (is_f00f_bug(regs, address))
956 no_context(regs, error_code, address, SIGSEGV, si_code);
960 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
961 unsigned long address, struct vm_area_struct *vma)
963 __bad_area_nosemaphore(regs, error_code, address, vma, SEGV_MAPERR);
967 __bad_area(struct pt_regs *regs, unsigned long error_code,
968 unsigned long address, struct vm_area_struct *vma, int si_code)
970 struct mm_struct *mm = current->mm;
973 * Something tried to access memory that isn't in our memory map..
974 * Fix it, but check if it's kernel or user first..
976 up_read(&mm->mmap_sem);
978 __bad_area_nosemaphore(regs, error_code, address, vma, si_code);
982 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
984 __bad_area(regs, error_code, address, NULL, SEGV_MAPERR);
987 static inline bool bad_area_access_from_pkeys(unsigned long error_code,
988 struct vm_area_struct *vma)
990 /* This code is always called on the current mm */
991 bool foreign = false;
993 if (!boot_cpu_has(X86_FEATURE_OSPKE))
995 if (error_code & PF_PK)
997 /* this checks permission keys on the VMA: */
998 if (!arch_vma_access_permitted(vma, (error_code & PF_WRITE),
999 (error_code & PF_INSTR), foreign))
1004 static noinline void
1005 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
1006 unsigned long address, struct vm_area_struct *vma)
1009 * This OSPKE check is not strictly necessary at runtime.
1010 * But, doing it this way allows compiler optimizations
1011 * if pkeys are compiled out.
1013 if (bad_area_access_from_pkeys(error_code, vma))
1014 __bad_area(regs, error_code, address, vma, SEGV_PKUERR);
1016 __bad_area(regs, error_code, address, vma, SEGV_ACCERR);
1020 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
1021 struct vm_area_struct *vma, unsigned int fault)
1023 struct task_struct *tsk = current;
1024 int code = BUS_ADRERR;
1026 /* Kernel mode? Handle exceptions or die: */
1027 if (!(error_code & PF_USER)) {
1028 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
1032 /* User-space => ok to do another page fault: */
1033 if (is_prefetch(regs, error_code, address))
1036 tsk->thread.cr2 = address;
1037 tsk->thread.error_code = error_code;
1038 tsk->thread.trap_nr = X86_TRAP_PF;
1040 #ifdef CONFIG_MEMORY_FAILURE
1041 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
1043 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
1044 tsk->comm, tsk->pid, address);
1045 code = BUS_MCEERR_AR;
1048 force_sig_info_fault(SIGBUS, code, address, tsk, vma, fault);
1051 static noinline void
1052 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
1053 unsigned long address, struct vm_area_struct *vma,
1056 if (fatal_signal_pending(current) && !(error_code & PF_USER)) {
1057 no_context(regs, error_code, address, 0, 0);
1061 if (fault & VM_FAULT_OOM) {
1062 /* Kernel mode? Handle exceptions or die: */
1063 if (!(error_code & PF_USER)) {
1064 no_context(regs, error_code, address,
1065 SIGSEGV, SEGV_MAPERR);
1070 * We ran out of memory, call the OOM killer, and return the
1071 * userspace (which will retry the fault, or kill us if we got
1074 pagefault_out_of_memory();
1076 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
1077 VM_FAULT_HWPOISON_LARGE))
1078 do_sigbus(regs, error_code, address, vma, fault);
1079 else if (fault & VM_FAULT_SIGSEGV)
1080 bad_area_nosemaphore(regs, error_code, address, vma);
1086 static int spurious_fault_check(unsigned long error_code, pte_t *pte)
1088 if ((error_code & PF_WRITE) && !pte_write(*pte))
1091 if ((error_code & PF_INSTR) && !pte_exec(*pte))
1094 * Note: We do not do lazy flushing on protection key
1095 * changes, so no spurious fault will ever set PF_PK.
1097 if ((error_code & PF_PK))
1104 * Handle a spurious fault caused by a stale TLB entry.
1106 * This allows us to lazily refresh the TLB when increasing the
1107 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
1108 * eagerly is very expensive since that implies doing a full
1109 * cross-processor TLB flush, even if no stale TLB entries exist
1110 * on other processors.
1112 * Spurious faults may only occur if the TLB contains an entry with
1113 * fewer permission than the page table entry. Non-present (P = 0)
1114 * and reserved bit (R = 1) faults are never spurious.
1116 * There are no security implications to leaving a stale TLB when
1117 * increasing the permissions on a page.
1119 * Returns non-zero if a spurious fault was handled, zero otherwise.
1121 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
1122 * (Optional Invalidation).
1125 spurious_fault(unsigned long error_code, unsigned long address)
1135 * Only writes to RO or instruction fetches from NX may cause
1138 * These could be from user or supervisor accesses but the TLB
1139 * is only lazily flushed after a kernel mapping protection
1140 * change, so user accesses are not expected to cause spurious
1143 if (error_code != (PF_WRITE | PF_PROT)
1144 && error_code != (PF_INSTR | PF_PROT))
1147 pgd = init_mm.pgd + pgd_index(address);
1148 if (!pgd_present(*pgd))
1151 p4d = p4d_offset(pgd, address);
1152 if (!p4d_present(*p4d))
1155 if (p4d_large(*p4d))
1156 return spurious_fault_check(error_code, (pte_t *) p4d);
1158 pud = pud_offset(p4d, address);
1159 if (!pud_present(*pud))
1162 if (pud_large(*pud))
1163 return spurious_fault_check(error_code, (pte_t *) pud);
1165 pmd = pmd_offset(pud, address);
1166 if (!pmd_present(*pmd))
1169 if (pmd_large(*pmd))
1170 return spurious_fault_check(error_code, (pte_t *) pmd);
1172 pte = pte_offset_kernel(pmd, address);
1173 if (!pte_present(*pte))
1176 ret = spurious_fault_check(error_code, pte);
1181 * Make sure we have permissions in PMD.
1182 * If not, then there's a bug in the page tables:
1184 ret = spurious_fault_check(error_code, (pte_t *) pmd);
1185 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
1189 NOKPROBE_SYMBOL(spurious_fault);
1191 int show_unhandled_signals = 1;
1194 access_error(unsigned long error_code, struct vm_area_struct *vma)
1196 /* This is only called for the current mm, so: */
1197 bool foreign = false;
1200 * Read or write was blocked by protection keys. This is
1201 * always an unconditional error and can never result in
1202 * a follow-up action to resolve the fault, like a COW.
1204 if (error_code & PF_PK)
1208 * Make sure to check the VMA so that we do not perform
1209 * faults just to hit a PF_PK as soon as we fill in a
1212 if (!arch_vma_access_permitted(vma, (error_code & PF_WRITE),
1213 (error_code & PF_INSTR), foreign))
1216 if (error_code & PF_WRITE) {
1217 /* write, present and write, not present: */
1218 if (unlikely(!(vma->vm_flags & VM_WRITE)))
1223 /* read, present: */
1224 if (unlikely(error_code & PF_PROT))
1227 /* read, not present: */
1228 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
1234 static int fault_in_kernel_space(unsigned long address)
1236 return address >= TASK_SIZE_MAX;
1239 static inline bool smap_violation(int error_code, struct pt_regs *regs)
1241 if (!IS_ENABLED(CONFIG_X86_SMAP))
1244 if (!static_cpu_has(X86_FEATURE_SMAP))
1247 if (error_code & PF_USER)
1250 if (!user_mode(regs) && (regs->flags & X86_EFLAGS_AC))
1257 * This routine handles page faults. It determines the address,
1258 * and the problem, and then passes it off to one of the appropriate
1261 static noinline void
1262 __do_page_fault(struct pt_regs *regs, unsigned long error_code,
1263 unsigned long address)
1265 struct vm_area_struct *vma;
1266 struct task_struct *tsk;
1267 struct mm_struct *mm;
1268 int fault, major = 0;
1269 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
1275 * Detect and handle instructions that would cause a page fault for
1276 * both a tracked kernel page and a userspace page.
1278 if (kmemcheck_active(regs))
1279 kmemcheck_hide(regs);
1280 prefetchw(&mm->mmap_sem);
1282 if (unlikely(kmmio_fault(regs, address)))
1286 * We fault-in kernel-space virtual memory on-demand. The
1287 * 'reference' page table is init_mm.pgd.
1289 * NOTE! We MUST NOT take any locks for this case. We may
1290 * be in an interrupt or a critical region, and should
1291 * only copy the information from the master page table,
1294 * This verifies that the fault happens in kernel space
1295 * (error_code & 4) == 0, and that the fault was not a
1296 * protection error (error_code & 9) == 0.
1298 if (unlikely(fault_in_kernel_space(address))) {
1299 if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) {
1300 if (vmalloc_fault(address) >= 0)
1303 if (kmemcheck_fault(regs, address, error_code))
1307 /* Can handle a stale RO->RW TLB: */
1308 if (spurious_fault(error_code, address))
1311 /* kprobes don't want to hook the spurious faults: */
1312 if (kprobes_fault(regs))
1315 * Don't take the mm semaphore here. If we fixup a prefetch
1316 * fault we could otherwise deadlock:
1318 bad_area_nosemaphore(regs, error_code, address, NULL);
1323 /* kprobes don't want to hook the spurious faults: */
1324 if (unlikely(kprobes_fault(regs)))
1327 if (unlikely(error_code & PF_RSVD))
1328 pgtable_bad(regs, error_code, address);
1330 if (unlikely(smap_violation(error_code, regs))) {
1331 bad_area_nosemaphore(regs, error_code, address, NULL);
1336 * If we're in an interrupt, have no user context or are running
1337 * in a region with pagefaults disabled then we must not take the fault
1339 if (unlikely(faulthandler_disabled() || !mm)) {
1340 bad_area_nosemaphore(regs, error_code, address, NULL);
1345 * It's safe to allow irq's after cr2 has been saved and the
1346 * vmalloc fault has been handled.
1348 * User-mode registers count as a user access even for any
1349 * potential system fault or CPU buglet:
1351 if (user_mode(regs)) {
1353 error_code |= PF_USER;
1354 flags |= FAULT_FLAG_USER;
1356 if (regs->flags & X86_EFLAGS_IF)
1360 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
1362 if (error_code & PF_WRITE)
1363 flags |= FAULT_FLAG_WRITE;
1364 if (error_code & PF_INSTR)
1365 flags |= FAULT_FLAG_INSTRUCTION;
1368 * When running in the kernel we expect faults to occur only to
1369 * addresses in user space. All other faults represent errors in
1370 * the kernel and should generate an OOPS. Unfortunately, in the
1371 * case of an erroneous fault occurring in a code path which already
1372 * holds mmap_sem we will deadlock attempting to validate the fault
1373 * against the address space. Luckily the kernel only validly
1374 * references user space from well defined areas of code, which are
1375 * listed in the exceptions table.
1377 * As the vast majority of faults will be valid we will only perform
1378 * the source reference check when there is a possibility of a
1379 * deadlock. Attempt to lock the address space, if we cannot we then
1380 * validate the source. If this is invalid we can skip the address
1381 * space check, thus avoiding the deadlock:
1383 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
1384 if ((error_code & PF_USER) == 0 &&
1385 !search_exception_tables(regs->ip)) {
1386 bad_area_nosemaphore(regs, error_code, address, NULL);
1390 down_read(&mm->mmap_sem);
1393 * The above down_read_trylock() might have succeeded in
1394 * which case we'll have missed the might_sleep() from
1400 vma = find_vma(mm, address);
1401 if (unlikely(!vma)) {
1402 bad_area(regs, error_code, address);
1405 if (likely(vma->vm_start <= address))
1407 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1408 bad_area(regs, error_code, address);
1411 if (error_code & PF_USER) {
1413 * Accessing the stack below %sp is always a bug.
1414 * The large cushion allows instructions like enter
1415 * and pusha to work. ("enter $65535, $31" pushes
1416 * 32 pointers and then decrements %sp by 65535.)
1418 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) {
1419 bad_area(regs, error_code, address);
1423 if (unlikely(expand_stack(vma, address))) {
1424 bad_area(regs, error_code, address);
1429 * Ok, we have a good vm_area for this memory access, so
1430 * we can handle it..
1433 if (unlikely(access_error(error_code, vma))) {
1434 bad_area_access_error(regs, error_code, address, vma);
1439 * If for any reason at all we couldn't handle the fault,
1440 * make sure we exit gracefully rather than endlessly redo
1441 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1442 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
1444 fault = handle_mm_fault(vma, address, flags);
1445 major |= fault & VM_FAULT_MAJOR;
1448 * If we need to retry the mmap_sem has already been released,
1449 * and if there is a fatal signal pending there is no guarantee
1450 * that we made any progress. Handle this case first.
1452 if (unlikely(fault & VM_FAULT_RETRY)) {
1453 /* Retry at most once */
1454 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1455 flags &= ~FAULT_FLAG_ALLOW_RETRY;
1456 flags |= FAULT_FLAG_TRIED;
1457 if (!fatal_signal_pending(tsk))
1461 /* User mode? Just return to handle the fatal exception */
1462 if (flags & FAULT_FLAG_USER)
1465 /* Not returning to user mode? Handle exceptions or die: */
1466 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
1470 up_read(&mm->mmap_sem);
1471 if (unlikely(fault & VM_FAULT_ERROR)) {
1472 mm_fault_error(regs, error_code, address, vma, fault);
1477 * Major/minor page fault accounting. If any of the events
1478 * returned VM_FAULT_MAJOR, we account it as a major fault.
1482 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
1485 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
1488 check_v8086_mode(regs, address, tsk);
1490 NOKPROBE_SYMBOL(__do_page_fault);
1492 static nokprobe_inline void
1493 trace_page_fault_entries(unsigned long address, struct pt_regs *regs,
1494 unsigned long error_code)
1496 if (user_mode(regs))
1497 trace_page_fault_user(address, regs, error_code);
1499 trace_page_fault_kernel(address, regs, error_code);
1503 * We must have this function blacklisted from kprobes, tagged with notrace
1504 * and call read_cr2() before calling anything else. To avoid calling any
1505 * kind of tracing machinery before we've observed the CR2 value.
1507 * exception_{enter,exit}() contains all sorts of tracepoints.
1509 dotraplinkage void notrace
1510 do_page_fault(struct pt_regs *regs, unsigned long error_code)
1512 unsigned long address = read_cr2(); /* Get the faulting address */
1513 enum ctx_state prev_state;
1515 prev_state = exception_enter();
1516 if (trace_pagefault_enabled())
1517 trace_page_fault_entries(address, regs, error_code);
1519 __do_page_fault(regs, error_code, address);
1520 exception_exit(prev_state);
1522 NOKPROBE_SYMBOL(do_page_fault);