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/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>
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 */
33 #define CREATE_TRACE_POINTS
34 #include <asm/trace/exceptions.h>
37 * Returns 0 if mmiotrace is disabled, or if the fault is not
38 * handled by mmiotrace:
40 static nokprobe_inline int
41 kmmio_fault(struct pt_regs *regs, unsigned long addr)
43 if (unlikely(is_kmmio_active()))
44 if (kmmio_handler(regs, addr) == 1)
54 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
55 * Check that here and ignore it.
59 * Sometimes the CPU reports invalid exceptions on prefetch.
60 * Check that here and ignore it.
62 * Opcode checker based on code by Richard Brunner.
65 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
66 unsigned char opcode, int *prefetch)
68 unsigned char instr_hi = opcode & 0xf0;
69 unsigned char instr_lo = opcode & 0x0f;
75 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
76 * In X86_64 long mode, the CPU will signal invalid
77 * opcode if some of these prefixes are present so
78 * X86_64 will never get here anyway
80 return ((instr_lo & 7) == 0x6);
84 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
85 * Need to figure out under what instruction mode the
86 * instruction was issued. Could check the LDT for lm,
87 * but for now it's good enough to assume that long
88 * mode only uses well known segments or kernel.
90 return (!user_mode(regs) || user_64bit_mode(regs));
93 /* 0x64 thru 0x67 are valid prefixes in all modes. */
94 return (instr_lo & 0xC) == 0x4;
96 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
97 return !instr_lo || (instr_lo>>1) == 1;
99 /* Prefetch instruction is 0x0F0D or 0x0F18 */
100 if (probe_kernel_address(instr, opcode))
103 *prefetch = (instr_lo == 0xF) &&
104 (opcode == 0x0D || opcode == 0x18);
112 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
114 unsigned char *max_instr;
115 unsigned char *instr;
119 * If it was a exec (instruction fetch) fault on NX page, then
120 * do not ignore the fault:
122 if (error_code & X86_PF_INSTR)
125 instr = (void *)convert_ip_to_linear(current, regs);
126 max_instr = instr + 15;
128 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX)
131 while (instr < max_instr) {
132 unsigned char opcode;
134 if (probe_kernel_address(instr, opcode))
139 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
145 DEFINE_SPINLOCK(pgd_lock);
149 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
151 unsigned index = pgd_index(address);
158 pgd_k = init_mm.pgd + index;
160 if (!pgd_present(*pgd_k))
164 * set_pgd(pgd, *pgd_k); here would be useless on PAE
165 * and redundant with the set_pmd() on non-PAE. As would
168 p4d = p4d_offset(pgd, address);
169 p4d_k = p4d_offset(pgd_k, address);
170 if (!p4d_present(*p4d_k))
173 pud = pud_offset(p4d, address);
174 pud_k = pud_offset(p4d_k, address);
175 if (!pud_present(*pud_k))
178 pmd = pmd_offset(pud, address);
179 pmd_k = pmd_offset(pud_k, address);
180 if (!pmd_present(*pmd_k))
183 if (!pmd_present(*pmd))
184 set_pmd(pmd, *pmd_k);
186 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
191 void vmalloc_sync_all(void)
193 unsigned long address;
195 if (SHARED_KERNEL_PMD)
198 for (address = VMALLOC_START & PMD_MASK;
199 address >= TASK_SIZE_MAX && address < FIXADDR_TOP;
200 address += PMD_SIZE) {
203 spin_lock(&pgd_lock);
204 list_for_each_entry(page, &pgd_list, lru) {
205 spinlock_t *pgt_lock;
208 /* the pgt_lock only for Xen */
209 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
212 ret = vmalloc_sync_one(page_address(page), address);
213 spin_unlock(pgt_lock);
218 spin_unlock(&pgd_lock);
225 * Handle a fault on the vmalloc or module mapping area
227 static noinline int vmalloc_fault(unsigned long address)
229 unsigned long pgd_paddr;
233 /* Make sure we are in vmalloc area: */
234 if (!(address >= VMALLOC_START && address < VMALLOC_END))
238 * Synchronize this task's top level page-table
239 * with the 'reference' page table.
241 * Do _not_ use "current" here. We might be inside
242 * an interrupt in the middle of a task switch..
244 pgd_paddr = read_cr3_pa();
245 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
249 if (pmd_large(*pmd_k))
252 pte_k = pte_offset_kernel(pmd_k, address);
253 if (!pte_present(*pte_k))
258 NOKPROBE_SYMBOL(vmalloc_fault);
261 * Did it hit the DOS screen memory VA from vm86 mode?
264 check_v8086_mode(struct pt_regs *regs, unsigned long address,
265 struct task_struct *tsk)
270 if (!v8086_mode(regs) || !tsk->thread.vm86)
273 bit = (address - 0xA0000) >> PAGE_SHIFT;
275 tsk->thread.vm86->screen_bitmap |= 1 << bit;
279 static bool low_pfn(unsigned long pfn)
281 return pfn < max_low_pfn;
284 static void dump_pagetable(unsigned long address)
286 pgd_t *base = __va(read_cr3_pa());
287 pgd_t *pgd = &base[pgd_index(address)];
293 #ifdef CONFIG_X86_PAE
294 pr_info("*pdpt = %016Lx ", pgd_val(*pgd));
295 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
297 #define pr_pde pr_cont
299 #define pr_pde pr_info
301 p4d = p4d_offset(pgd, address);
302 pud = pud_offset(p4d, address);
303 pmd = pmd_offset(pud, address);
304 pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
308 * We must not directly access the pte in the highpte
309 * case if the page table is located in highmem.
310 * And let's rather not kmap-atomic the pte, just in case
311 * it's allocated already:
313 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
316 pte = pte_offset_kernel(pmd, address);
317 pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
322 #else /* CONFIG_X86_64: */
324 void vmalloc_sync_all(void)
326 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END);
332 * Handle a fault on the vmalloc area
334 static noinline int vmalloc_fault(unsigned long address)
342 /* Make sure we are in vmalloc area: */
343 if (!(address >= VMALLOC_START && address < VMALLOC_END))
347 * Copy kernel mappings over when needed. This can also
348 * happen within a race in page table update. In the later
351 pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address);
352 pgd_k = pgd_offset_k(address);
353 if (pgd_none(*pgd_k))
356 if (pgtable_l5_enabled()) {
357 if (pgd_none(*pgd)) {
358 set_pgd(pgd, *pgd_k);
359 arch_flush_lazy_mmu_mode();
361 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_k));
365 /* With 4-level paging, copying happens on the p4d level. */
366 p4d = p4d_offset(pgd, address);
367 p4d_k = p4d_offset(pgd_k, address);
368 if (p4d_none(*p4d_k))
371 if (p4d_none(*p4d) && !pgtable_l5_enabled()) {
372 set_p4d(p4d, *p4d_k);
373 arch_flush_lazy_mmu_mode();
375 BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_k));
378 BUILD_BUG_ON(CONFIG_PGTABLE_LEVELS < 4);
380 pud = pud_offset(p4d, address);
387 pmd = pmd_offset(pud, address);
394 pte = pte_offset_kernel(pmd, address);
395 if (!pte_present(*pte))
400 NOKPROBE_SYMBOL(vmalloc_fault);
402 #ifdef CONFIG_CPU_SUP_AMD
403 static const char errata93_warning[] =
405 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
406 "******* Working around it, but it may cause SEGVs or burn power.\n"
407 "******* Please consider a BIOS update.\n"
408 "******* Disabling USB legacy in the BIOS may also help.\n";
412 * No vm86 mode in 64-bit mode:
415 check_v8086_mode(struct pt_regs *regs, unsigned long address,
416 struct task_struct *tsk)
420 static int bad_address(void *p)
424 return probe_kernel_address((unsigned long *)p, dummy);
427 static void dump_pagetable(unsigned long address)
429 pgd_t *base = __va(read_cr3_pa());
430 pgd_t *pgd = base + pgd_index(address);
436 if (bad_address(pgd))
439 pr_info("PGD %lx ", pgd_val(*pgd));
441 if (!pgd_present(*pgd))
444 p4d = p4d_offset(pgd, address);
445 if (bad_address(p4d))
448 pr_cont("P4D %lx ", p4d_val(*p4d));
449 if (!p4d_present(*p4d) || p4d_large(*p4d))
452 pud = pud_offset(p4d, address);
453 if (bad_address(pud))
456 pr_cont("PUD %lx ", pud_val(*pud));
457 if (!pud_present(*pud) || pud_large(*pud))
460 pmd = pmd_offset(pud, address);
461 if (bad_address(pmd))
464 pr_cont("PMD %lx ", pmd_val(*pmd));
465 if (!pmd_present(*pmd) || pmd_large(*pmd))
468 pte = pte_offset_kernel(pmd, address);
469 if (bad_address(pte))
472 pr_cont("PTE %lx", pte_val(*pte));
480 #endif /* CONFIG_X86_64 */
483 * Workaround for K8 erratum #93 & buggy BIOS.
485 * BIOS SMM functions are required to use a specific workaround
486 * to avoid corruption of the 64bit RIP register on C stepping K8.
488 * A lot of BIOS that didn't get tested properly miss this.
490 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
491 * Try to work around it here.
493 * Note we only handle faults in kernel here.
494 * Does nothing on 32-bit.
496 static int is_errata93(struct pt_regs *regs, unsigned long address)
498 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
499 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD
500 || boot_cpu_data.x86 != 0xf)
503 if (address != regs->ip)
506 if ((address >> 32) != 0)
509 address |= 0xffffffffUL << 32;
510 if ((address >= (u64)_stext && address <= (u64)_etext) ||
511 (address >= MODULES_VADDR && address <= MODULES_END)) {
512 printk_once(errata93_warning);
521 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
522 * to illegal addresses >4GB.
524 * We catch this in the page fault handler because these addresses
525 * are not reachable. Just detect this case and return. Any code
526 * segment in LDT is compatibility mode.
528 static int is_errata100(struct pt_regs *regs, unsigned long address)
531 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
537 static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
539 #ifdef CONFIG_X86_F00F_BUG
543 * Pentium F0 0F C7 C8 bug workaround:
545 if (boot_cpu_has_bug(X86_BUG_F00F)) {
546 nr = (address - idt_descr.address) >> 3;
549 do_invalid_op(regs, 0);
557 static void show_ldttss(const struct desc_ptr *gdt, const char *name, u16 index)
559 u32 offset = (index >> 3) * sizeof(struct desc_struct);
561 struct ldttss_desc desc;
564 pr_alert("%s: NULL\n", name);
568 if (offset + sizeof(struct ldttss_desc) >= gdt->size) {
569 pr_alert("%s: 0x%hx -- out of bounds\n", name, index);
573 if (probe_kernel_read(&desc, (void *)(gdt->address + offset),
574 sizeof(struct ldttss_desc))) {
575 pr_alert("%s: 0x%hx -- GDT entry is not readable\n",
580 addr = desc.base0 | (desc.base1 << 16) | ((unsigned long)desc.base2 << 24);
582 addr |= ((u64)desc.base3 << 32);
584 pr_alert("%s: 0x%hx -- base=0x%lx limit=0x%x\n",
585 name, index, addr, (desc.limit0 | (desc.limit1 << 16)));
589 show_fault_oops(struct pt_regs *regs, unsigned long error_code, unsigned long address)
591 if (!oops_may_print())
594 if (error_code & X86_PF_INSTR) {
599 pgd = __va(read_cr3_pa());
600 pgd += pgd_index(address);
602 pte = lookup_address_in_pgd(pgd, address, &level);
604 if (pte && pte_present(*pte) && !pte_exec(*pte))
605 pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n",
606 from_kuid(&init_user_ns, current_uid()));
607 if (pte && pte_present(*pte) && pte_exec(*pte) &&
608 (pgd_flags(*pgd) & _PAGE_USER) &&
609 (__read_cr4() & X86_CR4_SMEP))
610 pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n",
611 from_kuid(&init_user_ns, current_uid()));
614 if (address < PAGE_SIZE && !user_mode(regs))
615 pr_alert("BUG: kernel NULL pointer dereference, address: %px\n",
618 pr_alert("BUG: unable to handle page fault for address: %px\n",
621 pr_alert("#PF: %s %s in %s mode\n",
622 (error_code & X86_PF_USER) ? "user" : "supervisor",
623 (error_code & X86_PF_INSTR) ? "instruction fetch" :
624 (error_code & X86_PF_WRITE) ? "write access" :
626 user_mode(regs) ? "user" : "kernel");
627 pr_alert("#PF: error_code(0x%04lx) - %s\n", error_code,
628 !(error_code & X86_PF_PROT) ? "not-present page" :
629 (error_code & X86_PF_RSVD) ? "reserved bit violation" :
630 (error_code & X86_PF_PK) ? "protection keys violation" :
631 "permissions violation");
633 if (!(error_code & X86_PF_USER) && user_mode(regs)) {
634 struct desc_ptr idt, gdt;
638 * This can happen for quite a few reasons. The more obvious
639 * ones are faults accessing the GDT, or LDT. Perhaps
640 * surprisingly, if the CPU tries to deliver a benign or
641 * contributory exception from user code and gets a page fault
642 * during delivery, the page fault can be delivered as though
643 * it originated directly from user code. This could happen
644 * due to wrong permissions on the IDT, GDT, LDT, TSS, or
645 * kernel or IST stack.
649 /* Usable even on Xen PV -- it's just slow. */
650 native_store_gdt(&gdt);
652 pr_alert("IDT: 0x%lx (limit=0x%hx) GDT: 0x%lx (limit=0x%hx)\n",
653 idt.address, idt.size, gdt.address, gdt.size);
656 show_ldttss(&gdt, "LDTR", ldtr);
659 show_ldttss(&gdt, "TR", tr);
662 dump_pagetable(address);
666 pgtable_bad(struct pt_regs *regs, unsigned long error_code,
667 unsigned long address)
669 struct task_struct *tsk;
673 flags = oops_begin();
677 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
679 dump_pagetable(address);
681 if (__die("Bad pagetable", regs, error_code))
684 oops_end(flags, regs, sig);
687 static void set_signal_archinfo(unsigned long address,
688 unsigned long error_code)
690 struct task_struct *tsk = current;
693 * To avoid leaking information about the kernel page
694 * table layout, pretend that user-mode accesses to
695 * kernel addresses are always protection faults.
697 * NB: This means that failed vsyscalls with vsyscall=none
698 * will have the PROT bit. This doesn't leak any
699 * information and does not appear to cause any problems.
701 if (address >= TASK_SIZE_MAX)
702 error_code |= X86_PF_PROT;
704 tsk->thread.trap_nr = X86_TRAP_PF;
705 tsk->thread.error_code = error_code | X86_PF_USER;
706 tsk->thread.cr2 = address;
710 no_context(struct pt_regs *regs, unsigned long error_code,
711 unsigned long address, int signal, int si_code)
713 struct task_struct *tsk = current;
717 if (user_mode(regs)) {
719 * This is an implicit supervisor-mode access from user
720 * mode. Bypass all the kernel-mode recovery code and just
726 /* Are we prepared to handle this kernel fault? */
727 if (fixup_exception(regs, X86_TRAP_PF, error_code, address)) {
729 * Any interrupt that takes a fault gets the fixup. This makes
730 * the below recursive fault logic only apply to a faults from
737 * Per the above we're !in_interrupt(), aka. task context.
739 * In this case we need to make sure we're not recursively
740 * faulting through the emulate_vsyscall() logic.
742 if (current->thread.sig_on_uaccess_err && signal) {
743 set_signal_archinfo(address, error_code);
745 /* XXX: hwpoison faults will set the wrong code. */
746 force_sig_fault(signal, si_code, (void __user *)address);
750 * Barring that, we can do the fixup and be happy.
755 #ifdef CONFIG_VMAP_STACK
757 * Stack overflow? During boot, we can fault near the initial
758 * stack in the direct map, but that's not an overflow -- check
759 * that we're in vmalloc space to avoid this.
761 if (is_vmalloc_addr((void *)address) &&
762 (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) ||
763 address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) {
764 unsigned long stack = __this_cpu_ist_top_va(DF) - sizeof(void *);
766 * We're likely to be running with very little stack space
767 * left. It's plausible that we'd hit this condition but
768 * double-fault even before we get this far, in which case
769 * we're fine: the double-fault handler will deal with it.
771 * We don't want to make it all the way into the oops code
772 * and then double-fault, though, because we're likely to
773 * break the console driver and lose most of the stack dump.
775 asm volatile ("movq %[stack], %%rsp\n\t"
776 "call handle_stack_overflow\n\t"
778 : ASM_CALL_CONSTRAINT
779 : "D" ("kernel stack overflow (page fault)"),
780 "S" (regs), "d" (address),
781 [stack] "rm" (stack));
789 * Valid to do another page fault here, because if this fault
790 * had been triggered by is_prefetch fixup_exception would have
795 * Hall of shame of CPU/BIOS bugs.
797 if (is_prefetch(regs, error_code, address))
800 if (is_errata93(regs, address))
804 * Buggy firmware could access regions which might page fault, try to
805 * recover from such faults.
807 if (IS_ENABLED(CONFIG_EFI))
808 efi_recover_from_page_fault(address);
812 * Oops. The kernel tried to access some bad page. We'll have to
813 * terminate things with extreme prejudice:
815 flags = oops_begin();
817 show_fault_oops(regs, error_code, address);
819 if (task_stack_end_corrupted(tsk))
820 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
823 if (__die("Oops", regs, error_code))
826 /* Executive summary in case the body of the oops scrolled away */
827 printk(KERN_DEFAULT "CR2: %016lx\n", address);
829 oops_end(flags, regs, sig);
833 * Print out info about fatal segfaults, if the show_unhandled_signals
837 show_signal_msg(struct pt_regs *regs, unsigned long error_code,
838 unsigned long address, struct task_struct *tsk)
840 const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG;
842 if (!unhandled_signal(tsk, SIGSEGV))
845 if (!printk_ratelimit())
848 printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx",
849 loglvl, tsk->comm, task_pid_nr(tsk), address,
850 (void *)regs->ip, (void *)regs->sp, error_code);
852 print_vma_addr(KERN_CONT " in ", regs->ip);
854 printk(KERN_CONT "\n");
856 show_opcodes(regs, loglvl);
860 * The (legacy) vsyscall page is the long page in the kernel portion
861 * of the address space that has user-accessible permissions.
863 static bool is_vsyscall_vaddr(unsigned long vaddr)
865 return unlikely((vaddr & PAGE_MASK) == VSYSCALL_ADDR);
869 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
870 unsigned long address, u32 pkey, int si_code)
872 struct task_struct *tsk = current;
874 /* User mode accesses just cause a SIGSEGV */
875 if (user_mode(regs) && (error_code & X86_PF_USER)) {
877 * It's possible to have interrupts off here:
882 * Valid to do another page fault here because this one came
885 if (is_prefetch(regs, error_code, address))
888 if (is_errata100(regs, address))
892 * To avoid leaking information about the kernel page table
893 * layout, pretend that user-mode accesses to kernel addresses
894 * are always protection faults.
896 if (address >= TASK_SIZE_MAX)
897 error_code |= X86_PF_PROT;
899 if (likely(show_unhandled_signals))
900 show_signal_msg(regs, error_code, address, tsk);
902 set_signal_archinfo(address, error_code);
904 if (si_code == SEGV_PKUERR)
905 force_sig_pkuerr((void __user *)address, pkey);
907 force_sig_fault(SIGSEGV, si_code, (void __user *)address);
912 if (is_f00f_bug(regs, address))
915 no_context(regs, error_code, address, SIGSEGV, si_code);
919 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
920 unsigned long address)
922 __bad_area_nosemaphore(regs, error_code, address, 0, SEGV_MAPERR);
926 __bad_area(struct pt_regs *regs, unsigned long error_code,
927 unsigned long address, u32 pkey, int si_code)
929 struct mm_struct *mm = current->mm;
931 * Something tried to access memory that isn't in our memory map..
932 * Fix it, but check if it's kernel or user first..
934 up_read(&mm->mmap_sem);
936 __bad_area_nosemaphore(regs, error_code, address, pkey, si_code);
940 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
942 __bad_area(regs, error_code, address, 0, SEGV_MAPERR);
945 static inline bool bad_area_access_from_pkeys(unsigned long error_code,
946 struct vm_area_struct *vma)
948 /* This code is always called on the current mm */
949 bool foreign = false;
951 if (!boot_cpu_has(X86_FEATURE_OSPKE))
953 if (error_code & X86_PF_PK)
955 /* this checks permission keys on the VMA: */
956 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
957 (error_code & X86_PF_INSTR), foreign))
963 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
964 unsigned long address, struct vm_area_struct *vma)
967 * This OSPKE check is not strictly necessary at runtime.
968 * But, doing it this way allows compiler optimizations
969 * if pkeys are compiled out.
971 if (bad_area_access_from_pkeys(error_code, vma)) {
973 * A protection key fault means that the PKRU value did not allow
974 * access to some PTE. Userspace can figure out what PKRU was
975 * from the XSAVE state. This function captures the pkey from
976 * the vma and passes it to userspace so userspace can discover
977 * which protection key was set on the PTE.
979 * If we get here, we know that the hardware signaled a X86_PF_PK
980 * fault and that there was a VMA once we got in the fault
981 * handler. It does *not* guarantee that the VMA we find here
982 * was the one that we faulted on.
984 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4);
985 * 2. T1 : set PKRU to deny access to pkey=4, touches page
987 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
988 * 5. T1 : enters fault handler, takes mmap_sem, etc...
989 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really
990 * faulted on a pte with its pkey=4.
992 u32 pkey = vma_pkey(vma);
994 __bad_area(regs, error_code, address, pkey, SEGV_PKUERR);
996 __bad_area(regs, error_code, address, 0, SEGV_ACCERR);
1001 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
1004 /* Kernel mode? Handle exceptions or die: */
1005 if (!(error_code & X86_PF_USER)) {
1006 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
1010 /* User-space => ok to do another page fault: */
1011 if (is_prefetch(regs, error_code, address))
1014 set_signal_archinfo(address, error_code);
1016 #ifdef CONFIG_MEMORY_FAILURE
1017 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
1018 struct task_struct *tsk = current;
1022 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
1023 tsk->comm, tsk->pid, address);
1024 if (fault & VM_FAULT_HWPOISON_LARGE)
1025 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
1026 if (fault & VM_FAULT_HWPOISON)
1028 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb);
1032 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
1035 static noinline void
1036 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
1037 unsigned long address, vm_fault_t fault)
1039 if (fatal_signal_pending(current) && !(error_code & X86_PF_USER)) {
1040 no_context(regs, error_code, address, 0, 0);
1044 if (fault & VM_FAULT_OOM) {
1045 /* Kernel mode? Handle exceptions or die: */
1046 if (!(error_code & X86_PF_USER)) {
1047 no_context(regs, error_code, address,
1048 SIGSEGV, SEGV_MAPERR);
1053 * We ran out of memory, call the OOM killer, and return the
1054 * userspace (which will retry the fault, or kill us if we got
1057 pagefault_out_of_memory();
1059 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
1060 VM_FAULT_HWPOISON_LARGE))
1061 do_sigbus(regs, error_code, address, fault);
1062 else if (fault & VM_FAULT_SIGSEGV)
1063 bad_area_nosemaphore(regs, error_code, address);
1069 static int spurious_kernel_fault_check(unsigned long error_code, pte_t *pte)
1071 if ((error_code & X86_PF_WRITE) && !pte_write(*pte))
1074 if ((error_code & X86_PF_INSTR) && !pte_exec(*pte))
1081 * Handle a spurious fault caused by a stale TLB entry.
1083 * This allows us to lazily refresh the TLB when increasing the
1084 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
1085 * eagerly is very expensive since that implies doing a full
1086 * cross-processor TLB flush, even if no stale TLB entries exist
1087 * on other processors.
1089 * Spurious faults may only occur if the TLB contains an entry with
1090 * fewer permission than the page table entry. Non-present (P = 0)
1091 * and reserved bit (R = 1) faults are never spurious.
1093 * There are no security implications to leaving a stale TLB when
1094 * increasing the permissions on a page.
1096 * Returns non-zero if a spurious fault was handled, zero otherwise.
1098 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
1099 * (Optional Invalidation).
1102 spurious_kernel_fault(unsigned long error_code, unsigned long address)
1112 * Only writes to RO or instruction fetches from NX may cause
1115 * These could be from user or supervisor accesses but the TLB
1116 * is only lazily flushed after a kernel mapping protection
1117 * change, so user accesses are not expected to cause spurious
1120 if (error_code != (X86_PF_WRITE | X86_PF_PROT) &&
1121 error_code != (X86_PF_INSTR | X86_PF_PROT))
1124 pgd = init_mm.pgd + pgd_index(address);
1125 if (!pgd_present(*pgd))
1128 p4d = p4d_offset(pgd, address);
1129 if (!p4d_present(*p4d))
1132 if (p4d_large(*p4d))
1133 return spurious_kernel_fault_check(error_code, (pte_t *) p4d);
1135 pud = pud_offset(p4d, address);
1136 if (!pud_present(*pud))
1139 if (pud_large(*pud))
1140 return spurious_kernel_fault_check(error_code, (pte_t *) pud);
1142 pmd = pmd_offset(pud, address);
1143 if (!pmd_present(*pmd))
1146 if (pmd_large(*pmd))
1147 return spurious_kernel_fault_check(error_code, (pte_t *) pmd);
1149 pte = pte_offset_kernel(pmd, address);
1150 if (!pte_present(*pte))
1153 ret = spurious_kernel_fault_check(error_code, pte);
1158 * Make sure we have permissions in PMD.
1159 * If not, then there's a bug in the page tables:
1161 ret = spurious_kernel_fault_check(error_code, (pte_t *) pmd);
1162 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
1166 NOKPROBE_SYMBOL(spurious_kernel_fault);
1168 int show_unhandled_signals = 1;
1171 access_error(unsigned long error_code, struct vm_area_struct *vma)
1173 /* This is only called for the current mm, so: */
1174 bool foreign = false;
1177 * Read or write was blocked by protection keys. This is
1178 * always an unconditional error and can never result in
1179 * a follow-up action to resolve the fault, like a COW.
1181 if (error_code & X86_PF_PK)
1185 * Make sure to check the VMA so that we do not perform
1186 * faults just to hit a X86_PF_PK as soon as we fill in a
1189 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
1190 (error_code & X86_PF_INSTR), foreign))
1193 if (error_code & X86_PF_WRITE) {
1194 /* write, present and write, not present: */
1195 if (unlikely(!(vma->vm_flags & VM_WRITE)))
1200 /* read, present: */
1201 if (unlikely(error_code & X86_PF_PROT))
1204 /* read, not present: */
1205 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
1211 static int fault_in_kernel_space(unsigned long address)
1214 * On 64-bit systems, the vsyscall page is at an address above
1215 * TASK_SIZE_MAX, but is not considered part of the kernel
1218 if (IS_ENABLED(CONFIG_X86_64) && is_vsyscall_vaddr(address))
1221 return address >= TASK_SIZE_MAX;
1225 * Called for all faults where 'address' is part of the kernel address
1226 * space. Might get called for faults that originate from *code* that
1227 * ran in userspace or the kernel.
1230 do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code,
1231 unsigned long address)
1234 * Protection keys exceptions only happen on user pages. We
1235 * have no user pages in the kernel portion of the address
1236 * space, so do not expect them here.
1238 WARN_ON_ONCE(hw_error_code & X86_PF_PK);
1241 * We can fault-in kernel-space virtual memory on-demand. The
1242 * 'reference' page table is init_mm.pgd.
1244 * NOTE! We MUST NOT take any locks for this case. We may
1245 * be in an interrupt or a critical region, and should
1246 * only copy the information from the master page table,
1249 * Before doing this on-demand faulting, ensure that the
1250 * fault is not any of the following:
1251 * 1. A fault on a PTE with a reserved bit set.
1252 * 2. A fault caused by a user-mode access. (Do not demand-
1253 * fault kernel memory due to user-mode accesses).
1254 * 3. A fault caused by a page-level protection violation.
1255 * (A demand fault would be on a non-present page which
1256 * would have X86_PF_PROT==0).
1258 if (!(hw_error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) {
1259 if (vmalloc_fault(address) >= 0)
1263 /* Was the fault spurious, caused by lazy TLB invalidation? */
1264 if (spurious_kernel_fault(hw_error_code, address))
1267 /* kprobes don't want to hook the spurious faults: */
1268 if (kprobe_page_fault(regs, X86_TRAP_PF))
1272 * Note, despite being a "bad area", there are quite a few
1273 * acceptable reasons to get here, such as erratum fixups
1274 * and handling kernel code that can fault, like get_user().
1276 * Don't take the mm semaphore here. If we fixup a prefetch
1277 * fault we could otherwise deadlock:
1279 bad_area_nosemaphore(regs, hw_error_code, address);
1281 NOKPROBE_SYMBOL(do_kern_addr_fault);
1283 /* Handle faults in the user portion of the address space */
1285 void do_user_addr_fault(struct pt_regs *regs,
1286 unsigned long hw_error_code,
1287 unsigned long address)
1289 struct vm_area_struct *vma;
1290 struct task_struct *tsk;
1291 struct mm_struct *mm;
1292 vm_fault_t fault, major = 0;
1293 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
1298 /* kprobes don't want to hook the spurious faults: */
1299 if (unlikely(kprobe_page_fault(regs, X86_TRAP_PF)))
1303 * Reserved bits are never expected to be set on
1304 * entries in the user portion of the page tables.
1306 if (unlikely(hw_error_code & X86_PF_RSVD))
1307 pgtable_bad(regs, hw_error_code, address);
1310 * If SMAP is on, check for invalid kernel (supervisor) access to user
1311 * pages in the user address space. The odd case here is WRUSS,
1312 * which, according to the preliminary documentation, does not respect
1313 * SMAP and will have the USER bit set so, in all cases, SMAP
1314 * enforcement appears to be consistent with the USER bit.
1316 if (unlikely(cpu_feature_enabled(X86_FEATURE_SMAP) &&
1317 !(hw_error_code & X86_PF_USER) &&
1318 !(regs->flags & X86_EFLAGS_AC)))
1320 bad_area_nosemaphore(regs, hw_error_code, address);
1325 * If we're in an interrupt, have no user context or are running
1326 * in a region with pagefaults disabled then we must not take the fault
1328 if (unlikely(faulthandler_disabled() || !mm)) {
1329 bad_area_nosemaphore(regs, hw_error_code, address);
1334 * It's safe to allow irq's after cr2 has been saved and the
1335 * vmalloc fault has been handled.
1337 * User-mode registers count as a user access even for any
1338 * potential system fault or CPU buglet:
1340 if (user_mode(regs)) {
1342 flags |= FAULT_FLAG_USER;
1344 if (regs->flags & X86_EFLAGS_IF)
1348 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
1350 if (hw_error_code & X86_PF_WRITE)
1351 flags |= FAULT_FLAG_WRITE;
1352 if (hw_error_code & X86_PF_INSTR)
1353 flags |= FAULT_FLAG_INSTRUCTION;
1355 #ifdef CONFIG_X86_64
1357 * Faults in the vsyscall page might need emulation. The
1358 * vsyscall page is at a high address (>PAGE_OFFSET), but is
1359 * considered to be part of the user address space.
1361 * The vsyscall page does not have a "real" VMA, so do this
1362 * emulation before we go searching for VMAs.
1364 * PKRU never rejects instruction fetches, so we don't need
1365 * to consider the PF_PK bit.
1367 if (is_vsyscall_vaddr(address)) {
1368 if (emulate_vsyscall(hw_error_code, regs, address))
1374 * Kernel-mode access to the user address space should only occur
1375 * on well-defined single instructions listed in the exception
1376 * tables. But, an erroneous kernel fault occurring outside one of
1377 * those areas which also holds mmap_sem might deadlock attempting
1378 * to validate the fault against the address space.
1380 * Only do the expensive exception table search when we might be at
1381 * risk of a deadlock. This happens if we
1382 * 1. Failed to acquire mmap_sem, and
1383 * 2. The access did not originate in userspace.
1385 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
1386 if (!user_mode(regs) && !search_exception_tables(regs->ip)) {
1388 * Fault from code in kernel from
1389 * which we do not expect faults.
1391 bad_area_nosemaphore(regs, hw_error_code, address);
1395 down_read(&mm->mmap_sem);
1398 * The above down_read_trylock() might have succeeded in
1399 * which case we'll have missed the might_sleep() from
1405 vma = find_vma(mm, address);
1406 if (unlikely(!vma)) {
1407 bad_area(regs, hw_error_code, address);
1410 if (likely(vma->vm_start <= address))
1412 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1413 bad_area(regs, hw_error_code, address);
1416 if (unlikely(expand_stack(vma, address))) {
1417 bad_area(regs, hw_error_code, address);
1422 * Ok, we have a good vm_area for this memory access, so
1423 * we can handle it..
1426 if (unlikely(access_error(hw_error_code, vma))) {
1427 bad_area_access_error(regs, hw_error_code, address, vma);
1432 * If for any reason at all we couldn't handle the fault,
1433 * make sure we exit gracefully rather than endlessly redo
1434 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1435 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
1437 * Note that handle_userfault() may also release and reacquire mmap_sem
1438 * (and not return with VM_FAULT_RETRY), when returning to userland to
1439 * repeat the page fault later with a VM_FAULT_NOPAGE retval
1440 * (potentially after handling any pending signal during the return to
1441 * userland). The return to userland is identified whenever
1442 * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags.
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, hw_error_code, address, SIGBUS, BUS_ADRERR);
1470 up_read(&mm->mmap_sem);
1471 if (unlikely(fault & VM_FAULT_ERROR)) {
1472 mm_fault_error(regs, hw_error_code, address, 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_user_addr_fault);
1493 * This routine handles page faults. It determines the address,
1494 * and the problem, and then passes it off to one of the appropriate
1497 static noinline void
1498 __do_page_fault(struct pt_regs *regs, unsigned long hw_error_code,
1499 unsigned long address)
1501 prefetchw(¤t->mm->mmap_sem);
1503 if (unlikely(kmmio_fault(regs, address)))
1506 /* Was the fault on kernel-controlled part of the address space? */
1507 if (unlikely(fault_in_kernel_space(address)))
1508 do_kern_addr_fault(regs, hw_error_code, address);
1510 do_user_addr_fault(regs, hw_error_code, address);
1512 NOKPROBE_SYMBOL(__do_page_fault);
1514 static nokprobe_inline void
1515 trace_page_fault_entries(unsigned long address, struct pt_regs *regs,
1516 unsigned long error_code)
1518 if (user_mode(regs))
1519 trace_page_fault_user(address, regs, error_code);
1521 trace_page_fault_kernel(address, regs, error_code);
1525 * We must have this function blacklisted from kprobes, tagged with notrace
1526 * and call read_cr2() before calling anything else. To avoid calling any
1527 * kind of tracing machinery before we've observed the CR2 value.
1529 * exception_{enter,exit}() contains all sorts of tracepoints.
1531 dotraplinkage void notrace
1532 do_page_fault(struct pt_regs *regs, unsigned long error_code)
1534 unsigned long address = read_cr2(); /* Get the faulting address */
1535 enum ctx_state prev_state;
1537 prev_state = exception_enter();
1538 if (trace_pagefault_enabled())
1539 trace_page_fault_entries(address, regs, error_code);
1541 __do_page_fault(regs, error_code, address);
1542 exception_exit(prev_state);
1544 NOKPROBE_SYMBOL(do_page_fault);