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)
49 static nokprobe_inline int kprobes_fault(struct pt_regs *regs)
51 if (!kprobes_built_in())
56 * To be potentially processing a kprobe fault and to be allowed to call
57 * kprobe_running(), we have to be non-preemptible.
61 if (!kprobe_running())
63 return kprobe_fault_handler(regs, X86_TRAP_PF);
71 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
72 * Check that here and ignore it.
76 * Sometimes the CPU reports invalid exceptions on prefetch.
77 * Check that here and ignore it.
79 * Opcode checker based on code by Richard Brunner.
82 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
83 unsigned char opcode, int *prefetch)
85 unsigned char instr_hi = opcode & 0xf0;
86 unsigned char instr_lo = opcode & 0x0f;
92 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
93 * In X86_64 long mode, the CPU will signal invalid
94 * opcode if some of these prefixes are present so
95 * X86_64 will never get here anyway
97 return ((instr_lo & 7) == 0x6);
101 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
102 * Need to figure out under what instruction mode the
103 * instruction was issued. Could check the LDT for lm,
104 * but for now it's good enough to assume that long
105 * mode only uses well known segments or kernel.
107 return (!user_mode(regs) || user_64bit_mode(regs));
110 /* 0x64 thru 0x67 are valid prefixes in all modes. */
111 return (instr_lo & 0xC) == 0x4;
113 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
114 return !instr_lo || (instr_lo>>1) == 1;
116 /* Prefetch instruction is 0x0F0D or 0x0F18 */
117 if (probe_kernel_address(instr, opcode))
120 *prefetch = (instr_lo == 0xF) &&
121 (opcode == 0x0D || opcode == 0x18);
129 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
131 unsigned char *max_instr;
132 unsigned char *instr;
136 * If it was a exec (instruction fetch) fault on NX page, then
137 * do not ignore the fault:
139 if (error_code & X86_PF_INSTR)
142 instr = (void *)convert_ip_to_linear(current, regs);
143 max_instr = instr + 15;
145 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX)
148 while (instr < max_instr) {
149 unsigned char opcode;
151 if (probe_kernel_address(instr, opcode))
156 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
162 DEFINE_SPINLOCK(pgd_lock);
166 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
168 unsigned index = pgd_index(address);
175 pgd_k = init_mm.pgd + index;
177 if (!pgd_present(*pgd_k))
181 * set_pgd(pgd, *pgd_k); here would be useless on PAE
182 * and redundant with the set_pmd() on non-PAE. As would
185 p4d = p4d_offset(pgd, address);
186 p4d_k = p4d_offset(pgd_k, address);
187 if (!p4d_present(*p4d_k))
190 pud = pud_offset(p4d, address);
191 pud_k = pud_offset(p4d_k, address);
192 if (!pud_present(*pud_k))
195 pmd = pmd_offset(pud, address);
196 pmd_k = pmd_offset(pud_k, address);
197 if (!pmd_present(*pmd_k))
200 if (!pmd_present(*pmd))
201 set_pmd(pmd, *pmd_k);
203 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
208 void vmalloc_sync_all(void)
210 unsigned long address;
212 if (SHARED_KERNEL_PMD)
215 for (address = VMALLOC_START & PMD_MASK;
216 address >= TASK_SIZE_MAX && address < FIXADDR_TOP;
217 address += PMD_SIZE) {
220 spin_lock(&pgd_lock);
221 list_for_each_entry(page, &pgd_list, lru) {
222 spinlock_t *pgt_lock;
225 /* the pgt_lock only for Xen */
226 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
229 ret = vmalloc_sync_one(page_address(page), address);
230 spin_unlock(pgt_lock);
235 spin_unlock(&pgd_lock);
242 * Handle a fault on the vmalloc or module mapping area
244 static noinline int vmalloc_fault(unsigned long address)
246 unsigned long pgd_paddr;
250 /* Make sure we are in vmalloc area: */
251 if (!(address >= VMALLOC_START && address < VMALLOC_END))
255 * Synchronize this task's top level page-table
256 * with the 'reference' page table.
258 * Do _not_ use "current" here. We might be inside
259 * an interrupt in the middle of a task switch..
261 pgd_paddr = read_cr3_pa();
262 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
266 if (pmd_large(*pmd_k))
269 pte_k = pte_offset_kernel(pmd_k, address);
270 if (!pte_present(*pte_k))
275 NOKPROBE_SYMBOL(vmalloc_fault);
278 * Did it hit the DOS screen memory VA from vm86 mode?
281 check_v8086_mode(struct pt_regs *regs, unsigned long address,
282 struct task_struct *tsk)
287 if (!v8086_mode(regs) || !tsk->thread.vm86)
290 bit = (address - 0xA0000) >> PAGE_SHIFT;
292 tsk->thread.vm86->screen_bitmap |= 1 << bit;
296 static bool low_pfn(unsigned long pfn)
298 return pfn < max_low_pfn;
301 static void dump_pagetable(unsigned long address)
303 pgd_t *base = __va(read_cr3_pa());
304 pgd_t *pgd = &base[pgd_index(address)];
310 #ifdef CONFIG_X86_PAE
311 pr_info("*pdpt = %016Lx ", pgd_val(*pgd));
312 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
314 #define pr_pde pr_cont
316 #define pr_pde pr_info
318 p4d = p4d_offset(pgd, address);
319 pud = pud_offset(p4d, address);
320 pmd = pmd_offset(pud, address);
321 pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
325 * We must not directly access the pte in the highpte
326 * case if the page table is located in highmem.
327 * And let's rather not kmap-atomic the pte, just in case
328 * it's allocated already:
330 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
333 pte = pte_offset_kernel(pmd, address);
334 pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
339 #else /* CONFIG_X86_64: */
341 void vmalloc_sync_all(void)
343 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END);
349 * Handle a fault on the vmalloc area
351 static noinline int vmalloc_fault(unsigned long address)
359 /* Make sure we are in vmalloc area: */
360 if (!(address >= VMALLOC_START && address < VMALLOC_END))
364 * Copy kernel mappings over when needed. This can also
365 * happen within a race in page table update. In the later
368 pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address);
369 pgd_k = pgd_offset_k(address);
370 if (pgd_none(*pgd_k))
373 if (pgtable_l5_enabled()) {
374 if (pgd_none(*pgd)) {
375 set_pgd(pgd, *pgd_k);
376 arch_flush_lazy_mmu_mode();
378 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_k));
382 /* With 4-level paging, copying happens on the p4d level. */
383 p4d = p4d_offset(pgd, address);
384 p4d_k = p4d_offset(pgd_k, address);
385 if (p4d_none(*p4d_k))
388 if (p4d_none(*p4d) && !pgtable_l5_enabled()) {
389 set_p4d(p4d, *p4d_k);
390 arch_flush_lazy_mmu_mode();
392 BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_k));
395 BUILD_BUG_ON(CONFIG_PGTABLE_LEVELS < 4);
397 pud = pud_offset(p4d, address);
404 pmd = pmd_offset(pud, address);
411 pte = pte_offset_kernel(pmd, address);
412 if (!pte_present(*pte))
417 NOKPROBE_SYMBOL(vmalloc_fault);
419 #ifdef CONFIG_CPU_SUP_AMD
420 static const char errata93_warning[] =
422 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
423 "******* Working around it, but it may cause SEGVs or burn power.\n"
424 "******* Please consider a BIOS update.\n"
425 "******* Disabling USB legacy in the BIOS may also help.\n";
429 * No vm86 mode in 64-bit mode:
432 check_v8086_mode(struct pt_regs *regs, unsigned long address,
433 struct task_struct *tsk)
437 static int bad_address(void *p)
441 return probe_kernel_address((unsigned long *)p, dummy);
444 static void dump_pagetable(unsigned long address)
446 pgd_t *base = __va(read_cr3_pa());
447 pgd_t *pgd = base + pgd_index(address);
453 if (bad_address(pgd))
456 pr_info("PGD %lx ", pgd_val(*pgd));
458 if (!pgd_present(*pgd))
461 p4d = p4d_offset(pgd, address);
462 if (bad_address(p4d))
465 pr_cont("P4D %lx ", p4d_val(*p4d));
466 if (!p4d_present(*p4d) || p4d_large(*p4d))
469 pud = pud_offset(p4d, address);
470 if (bad_address(pud))
473 pr_cont("PUD %lx ", pud_val(*pud));
474 if (!pud_present(*pud) || pud_large(*pud))
477 pmd = pmd_offset(pud, address);
478 if (bad_address(pmd))
481 pr_cont("PMD %lx ", pmd_val(*pmd));
482 if (!pmd_present(*pmd) || pmd_large(*pmd))
485 pte = pte_offset_kernel(pmd, address);
486 if (bad_address(pte))
489 pr_cont("PTE %lx", pte_val(*pte));
497 #endif /* CONFIG_X86_64 */
500 * Workaround for K8 erratum #93 & buggy BIOS.
502 * BIOS SMM functions are required to use a specific workaround
503 * to avoid corruption of the 64bit RIP register on C stepping K8.
505 * A lot of BIOS that didn't get tested properly miss this.
507 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
508 * Try to work around it here.
510 * Note we only handle faults in kernel here.
511 * Does nothing on 32-bit.
513 static int is_errata93(struct pt_regs *regs, unsigned long address)
515 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
516 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD
517 || boot_cpu_data.x86 != 0xf)
520 if (address != regs->ip)
523 if ((address >> 32) != 0)
526 address |= 0xffffffffUL << 32;
527 if ((address >= (u64)_stext && address <= (u64)_etext) ||
528 (address >= MODULES_VADDR && address <= MODULES_END)) {
529 printk_once(errata93_warning);
538 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
539 * to illegal addresses >4GB.
541 * We catch this in the page fault handler because these addresses
542 * are not reachable. Just detect this case and return. Any code
543 * segment in LDT is compatibility mode.
545 static int is_errata100(struct pt_regs *regs, unsigned long address)
548 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
554 static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
556 #ifdef CONFIG_X86_F00F_BUG
560 * Pentium F0 0F C7 C8 bug workaround:
562 if (boot_cpu_has_bug(X86_BUG_F00F)) {
563 nr = (address - idt_descr.address) >> 3;
566 do_invalid_op(regs, 0);
574 static void show_ldttss(const struct desc_ptr *gdt, const char *name, u16 index)
576 u32 offset = (index >> 3) * sizeof(struct desc_struct);
578 struct ldttss_desc desc;
581 pr_alert("%s: NULL\n", name);
585 if (offset + sizeof(struct ldttss_desc) >= gdt->size) {
586 pr_alert("%s: 0x%hx -- out of bounds\n", name, index);
590 if (probe_kernel_read(&desc, (void *)(gdt->address + offset),
591 sizeof(struct ldttss_desc))) {
592 pr_alert("%s: 0x%hx -- GDT entry is not readable\n",
597 addr = desc.base0 | (desc.base1 << 16) | ((unsigned long)desc.base2 << 24);
599 addr |= ((u64)desc.base3 << 32);
601 pr_alert("%s: 0x%hx -- base=0x%lx limit=0x%x\n",
602 name, index, addr, (desc.limit0 | (desc.limit1 << 16)));
606 show_fault_oops(struct pt_regs *regs, unsigned long error_code, unsigned long address)
608 if (!oops_may_print())
611 if (error_code & X86_PF_INSTR) {
616 pgd = __va(read_cr3_pa());
617 pgd += pgd_index(address);
619 pte = lookup_address_in_pgd(pgd, address, &level);
621 if (pte && pte_present(*pte) && !pte_exec(*pte))
622 pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n",
623 from_kuid(&init_user_ns, current_uid()));
624 if (pte && pte_present(*pte) && pte_exec(*pte) &&
625 (pgd_flags(*pgd) & _PAGE_USER) &&
626 (__read_cr4() & X86_CR4_SMEP))
627 pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n",
628 from_kuid(&init_user_ns, current_uid()));
631 if (address < PAGE_SIZE && !user_mode(regs))
632 pr_alert("BUG: kernel NULL pointer dereference, address: %px\n",
635 pr_alert("BUG: unable to handle page fault for address: %px\n",
638 pr_alert("#PF: %s %s in %s mode\n",
639 (error_code & X86_PF_USER) ? "user" : "supervisor",
640 (error_code & X86_PF_INSTR) ? "instruction fetch" :
641 (error_code & X86_PF_WRITE) ? "write access" :
643 user_mode(regs) ? "user" : "kernel");
644 pr_alert("#PF: error_code(0x%04lx) - %s\n", error_code,
645 !(error_code & X86_PF_PROT) ? "not-present page" :
646 (error_code & X86_PF_RSVD) ? "reserved bit violation" :
647 (error_code & X86_PF_PK) ? "protection keys violation" :
648 "permissions violation");
650 if (!(error_code & X86_PF_USER) && user_mode(regs)) {
651 struct desc_ptr idt, gdt;
655 * This can happen for quite a few reasons. The more obvious
656 * ones are faults accessing the GDT, or LDT. Perhaps
657 * surprisingly, if the CPU tries to deliver a benign or
658 * contributory exception from user code and gets a page fault
659 * during delivery, the page fault can be delivered as though
660 * it originated directly from user code. This could happen
661 * due to wrong permissions on the IDT, GDT, LDT, TSS, or
662 * kernel or IST stack.
666 /* Usable even on Xen PV -- it's just slow. */
667 native_store_gdt(&gdt);
669 pr_alert("IDT: 0x%lx (limit=0x%hx) GDT: 0x%lx (limit=0x%hx)\n",
670 idt.address, idt.size, gdt.address, gdt.size);
673 show_ldttss(&gdt, "LDTR", ldtr);
676 show_ldttss(&gdt, "TR", tr);
679 dump_pagetable(address);
683 pgtable_bad(struct pt_regs *regs, unsigned long error_code,
684 unsigned long address)
686 struct task_struct *tsk;
690 flags = oops_begin();
694 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
696 dump_pagetable(address);
698 if (__die("Bad pagetable", regs, error_code))
701 oops_end(flags, regs, sig);
704 static void set_signal_archinfo(unsigned long address,
705 unsigned long error_code)
707 struct task_struct *tsk = current;
710 * To avoid leaking information about the kernel page
711 * table layout, pretend that user-mode accesses to
712 * kernel addresses are always protection faults.
714 * NB: This means that failed vsyscalls with vsyscall=none
715 * will have the PROT bit. This doesn't leak any
716 * information and does not appear to cause any problems.
718 if (address >= TASK_SIZE_MAX)
719 error_code |= X86_PF_PROT;
721 tsk->thread.trap_nr = X86_TRAP_PF;
722 tsk->thread.error_code = error_code | X86_PF_USER;
723 tsk->thread.cr2 = address;
727 no_context(struct pt_regs *regs, unsigned long error_code,
728 unsigned long address, int signal, int si_code)
730 struct task_struct *tsk = current;
734 if (user_mode(regs)) {
736 * This is an implicit supervisor-mode access from user
737 * mode. Bypass all the kernel-mode recovery code and just
743 /* Are we prepared to handle this kernel fault? */
744 if (fixup_exception(regs, X86_TRAP_PF, error_code, address)) {
746 * Any interrupt that takes a fault gets the fixup. This makes
747 * the below recursive fault logic only apply to a faults from
754 * Per the above we're !in_interrupt(), aka. task context.
756 * In this case we need to make sure we're not recursively
757 * faulting through the emulate_vsyscall() logic.
759 if (current->thread.sig_on_uaccess_err && signal) {
760 set_signal_archinfo(address, error_code);
762 /* XXX: hwpoison faults will set the wrong code. */
763 force_sig_fault(signal, si_code, (void __user *)address);
767 * Barring that, we can do the fixup and be happy.
772 #ifdef CONFIG_VMAP_STACK
774 * Stack overflow? During boot, we can fault near the initial
775 * stack in the direct map, but that's not an overflow -- check
776 * that we're in vmalloc space to avoid this.
778 if (is_vmalloc_addr((void *)address) &&
779 (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) ||
780 address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) {
781 unsigned long stack = __this_cpu_ist_top_va(DF) - sizeof(void *);
783 * We're likely to be running with very little stack space
784 * left. It's plausible that we'd hit this condition but
785 * double-fault even before we get this far, in which case
786 * we're fine: the double-fault handler will deal with it.
788 * We don't want to make it all the way into the oops code
789 * and then double-fault, though, because we're likely to
790 * break the console driver and lose most of the stack dump.
792 asm volatile ("movq %[stack], %%rsp\n\t"
793 "call handle_stack_overflow\n\t"
795 : ASM_CALL_CONSTRAINT
796 : "D" ("kernel stack overflow (page fault)"),
797 "S" (regs), "d" (address),
798 [stack] "rm" (stack));
806 * Valid to do another page fault here, because if this fault
807 * had been triggered by is_prefetch fixup_exception would have
812 * Hall of shame of CPU/BIOS bugs.
814 if (is_prefetch(regs, error_code, address))
817 if (is_errata93(regs, address))
821 * Buggy firmware could access regions which might page fault, try to
822 * recover from such faults.
824 if (IS_ENABLED(CONFIG_EFI))
825 efi_recover_from_page_fault(address);
829 * Oops. The kernel tried to access some bad page. We'll have to
830 * terminate things with extreme prejudice:
832 flags = oops_begin();
834 show_fault_oops(regs, error_code, address);
836 if (task_stack_end_corrupted(tsk))
837 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
840 if (__die("Oops", regs, error_code))
843 /* Executive summary in case the body of the oops scrolled away */
844 printk(KERN_DEFAULT "CR2: %016lx\n", address);
846 oops_end(flags, regs, sig);
850 * Print out info about fatal segfaults, if the show_unhandled_signals
854 show_signal_msg(struct pt_regs *regs, unsigned long error_code,
855 unsigned long address, struct task_struct *tsk)
857 const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG;
859 if (!unhandled_signal(tsk, SIGSEGV))
862 if (!printk_ratelimit())
865 printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx",
866 loglvl, tsk->comm, task_pid_nr(tsk), address,
867 (void *)regs->ip, (void *)regs->sp, error_code);
869 print_vma_addr(KERN_CONT " in ", regs->ip);
871 printk(KERN_CONT "\n");
873 show_opcodes(regs, loglvl);
877 * The (legacy) vsyscall page is the long page in the kernel portion
878 * of the address space that has user-accessible permissions.
880 static bool is_vsyscall_vaddr(unsigned long vaddr)
882 return unlikely((vaddr & PAGE_MASK) == VSYSCALL_ADDR);
886 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
887 unsigned long address, u32 pkey, int si_code)
889 struct task_struct *tsk = current;
891 /* User mode accesses just cause a SIGSEGV */
892 if (user_mode(regs) && (error_code & X86_PF_USER)) {
894 * It's possible to have interrupts off here:
899 * Valid to do another page fault here because this one came
902 if (is_prefetch(regs, error_code, address))
905 if (is_errata100(regs, address))
909 * To avoid leaking information about the kernel page table
910 * layout, pretend that user-mode accesses to kernel addresses
911 * are always protection faults.
913 if (address >= TASK_SIZE_MAX)
914 error_code |= X86_PF_PROT;
916 if (likely(show_unhandled_signals))
917 show_signal_msg(regs, error_code, address, tsk);
919 set_signal_archinfo(address, error_code);
921 if (si_code == SEGV_PKUERR)
922 force_sig_pkuerr((void __user *)address, pkey);
924 force_sig_fault(SIGSEGV, si_code, (void __user *)address);
929 if (is_f00f_bug(regs, address))
932 no_context(regs, error_code, address, SIGSEGV, si_code);
936 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
937 unsigned long address)
939 __bad_area_nosemaphore(regs, error_code, address, 0, SEGV_MAPERR);
943 __bad_area(struct pt_regs *regs, unsigned long error_code,
944 unsigned long address, u32 pkey, int si_code)
946 struct mm_struct *mm = current->mm;
948 * Something tried to access memory that isn't in our memory map..
949 * Fix it, but check if it's kernel or user first..
951 up_read(&mm->mmap_sem);
953 __bad_area_nosemaphore(regs, error_code, address, pkey, si_code);
957 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
959 __bad_area(regs, error_code, address, 0, SEGV_MAPERR);
962 static inline bool bad_area_access_from_pkeys(unsigned long error_code,
963 struct vm_area_struct *vma)
965 /* This code is always called on the current mm */
966 bool foreign = false;
968 if (!boot_cpu_has(X86_FEATURE_OSPKE))
970 if (error_code & X86_PF_PK)
972 /* this checks permission keys on the VMA: */
973 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
974 (error_code & X86_PF_INSTR), foreign))
980 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
981 unsigned long address, struct vm_area_struct *vma)
984 * This OSPKE check is not strictly necessary at runtime.
985 * But, doing it this way allows compiler optimizations
986 * if pkeys are compiled out.
988 if (bad_area_access_from_pkeys(error_code, vma)) {
990 * A protection key fault means that the PKRU value did not allow
991 * access to some PTE. Userspace can figure out what PKRU was
992 * from the XSAVE state. This function captures the pkey from
993 * the vma and passes it to userspace so userspace can discover
994 * which protection key was set on the PTE.
996 * If we get here, we know that the hardware signaled a X86_PF_PK
997 * fault and that there was a VMA once we got in the fault
998 * handler. It does *not* guarantee that the VMA we find here
999 * was the one that we faulted on.
1001 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4);
1002 * 2. T1 : set PKRU to deny access to pkey=4, touches page
1004 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
1005 * 5. T1 : enters fault handler, takes mmap_sem, etc...
1006 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really
1007 * faulted on a pte with its pkey=4.
1009 u32 pkey = vma_pkey(vma);
1011 __bad_area(regs, error_code, address, pkey, SEGV_PKUERR);
1013 __bad_area(regs, error_code, address, 0, SEGV_ACCERR);
1018 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
1021 /* Kernel mode? Handle exceptions or die: */
1022 if (!(error_code & X86_PF_USER)) {
1023 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
1027 /* User-space => ok to do another page fault: */
1028 if (is_prefetch(regs, error_code, address))
1031 set_signal_archinfo(address, error_code);
1033 #ifdef CONFIG_MEMORY_FAILURE
1034 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
1035 struct task_struct *tsk = current;
1039 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
1040 tsk->comm, tsk->pid, address);
1041 if (fault & VM_FAULT_HWPOISON_LARGE)
1042 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
1043 if (fault & VM_FAULT_HWPOISON)
1045 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb);
1049 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
1052 static noinline void
1053 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
1054 unsigned long address, vm_fault_t fault)
1056 if (fatal_signal_pending(current) && !(error_code & X86_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 & X86_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, fault);
1079 else if (fault & VM_FAULT_SIGSEGV)
1080 bad_area_nosemaphore(regs, error_code, address);
1086 static int spurious_kernel_fault_check(unsigned long error_code, pte_t *pte)
1088 if ((error_code & X86_PF_WRITE) && !pte_write(*pte))
1091 if ((error_code & X86_PF_INSTR) && !pte_exec(*pte))
1098 * Handle a spurious fault caused by a stale TLB entry.
1100 * This allows us to lazily refresh the TLB when increasing the
1101 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
1102 * eagerly is very expensive since that implies doing a full
1103 * cross-processor TLB flush, even if no stale TLB entries exist
1104 * on other processors.
1106 * Spurious faults may only occur if the TLB contains an entry with
1107 * fewer permission than the page table entry. Non-present (P = 0)
1108 * and reserved bit (R = 1) faults are never spurious.
1110 * There are no security implications to leaving a stale TLB when
1111 * increasing the permissions on a page.
1113 * Returns non-zero if a spurious fault was handled, zero otherwise.
1115 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
1116 * (Optional Invalidation).
1119 spurious_kernel_fault(unsigned long error_code, unsigned long address)
1129 * Only writes to RO or instruction fetches from NX may cause
1132 * These could be from user or supervisor accesses but the TLB
1133 * is only lazily flushed after a kernel mapping protection
1134 * change, so user accesses are not expected to cause spurious
1137 if (error_code != (X86_PF_WRITE | X86_PF_PROT) &&
1138 error_code != (X86_PF_INSTR | X86_PF_PROT))
1141 pgd = init_mm.pgd + pgd_index(address);
1142 if (!pgd_present(*pgd))
1145 p4d = p4d_offset(pgd, address);
1146 if (!p4d_present(*p4d))
1149 if (p4d_large(*p4d))
1150 return spurious_kernel_fault_check(error_code, (pte_t *) p4d);
1152 pud = pud_offset(p4d, address);
1153 if (!pud_present(*pud))
1156 if (pud_large(*pud))
1157 return spurious_kernel_fault_check(error_code, (pte_t *) pud);
1159 pmd = pmd_offset(pud, address);
1160 if (!pmd_present(*pmd))
1163 if (pmd_large(*pmd))
1164 return spurious_kernel_fault_check(error_code, (pte_t *) pmd);
1166 pte = pte_offset_kernel(pmd, address);
1167 if (!pte_present(*pte))
1170 ret = spurious_kernel_fault_check(error_code, pte);
1175 * Make sure we have permissions in PMD.
1176 * If not, then there's a bug in the page tables:
1178 ret = spurious_kernel_fault_check(error_code, (pte_t *) pmd);
1179 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
1183 NOKPROBE_SYMBOL(spurious_kernel_fault);
1185 int show_unhandled_signals = 1;
1188 access_error(unsigned long error_code, struct vm_area_struct *vma)
1190 /* This is only called for the current mm, so: */
1191 bool foreign = false;
1194 * Read or write was blocked by protection keys. This is
1195 * always an unconditional error and can never result in
1196 * a follow-up action to resolve the fault, like a COW.
1198 if (error_code & X86_PF_PK)
1202 * Make sure to check the VMA so that we do not perform
1203 * faults just to hit a X86_PF_PK as soon as we fill in a
1206 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
1207 (error_code & X86_PF_INSTR), foreign))
1210 if (error_code & X86_PF_WRITE) {
1211 /* write, present and write, not present: */
1212 if (unlikely(!(vma->vm_flags & VM_WRITE)))
1217 /* read, present: */
1218 if (unlikely(error_code & X86_PF_PROT))
1221 /* read, not present: */
1222 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
1228 static int fault_in_kernel_space(unsigned long address)
1231 * On 64-bit systems, the vsyscall page is at an address above
1232 * TASK_SIZE_MAX, but is not considered part of the kernel
1235 if (IS_ENABLED(CONFIG_X86_64) && is_vsyscall_vaddr(address))
1238 return address >= TASK_SIZE_MAX;
1242 * Called for all faults where 'address' is part of the kernel address
1243 * space. Might get called for faults that originate from *code* that
1244 * ran in userspace or the kernel.
1247 do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code,
1248 unsigned long address)
1251 * Protection keys exceptions only happen on user pages. We
1252 * have no user pages in the kernel portion of the address
1253 * space, so do not expect them here.
1255 WARN_ON_ONCE(hw_error_code & X86_PF_PK);
1258 * We can fault-in kernel-space virtual memory on-demand. The
1259 * 'reference' page table is init_mm.pgd.
1261 * NOTE! We MUST NOT take any locks for this case. We may
1262 * be in an interrupt or a critical region, and should
1263 * only copy the information from the master page table,
1266 * Before doing this on-demand faulting, ensure that the
1267 * fault is not any of the following:
1268 * 1. A fault on a PTE with a reserved bit set.
1269 * 2. A fault caused by a user-mode access. (Do not demand-
1270 * fault kernel memory due to user-mode accesses).
1271 * 3. A fault caused by a page-level protection violation.
1272 * (A demand fault would be on a non-present page which
1273 * would have X86_PF_PROT==0).
1275 if (!(hw_error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) {
1276 if (vmalloc_fault(address) >= 0)
1280 /* Was the fault spurious, caused by lazy TLB invalidation? */
1281 if (spurious_kernel_fault(hw_error_code, address))
1284 /* kprobes don't want to hook the spurious faults: */
1285 if (kprobes_fault(regs))
1289 * Note, despite being a "bad area", there are quite a few
1290 * acceptable reasons to get here, such as erratum fixups
1291 * and handling kernel code that can fault, like get_user().
1293 * Don't take the mm semaphore here. If we fixup a prefetch
1294 * fault we could otherwise deadlock:
1296 bad_area_nosemaphore(regs, hw_error_code, address);
1298 NOKPROBE_SYMBOL(do_kern_addr_fault);
1300 /* Handle faults in the user portion of the address space */
1302 void do_user_addr_fault(struct pt_regs *regs,
1303 unsigned long hw_error_code,
1304 unsigned long address)
1306 struct vm_area_struct *vma;
1307 struct task_struct *tsk;
1308 struct mm_struct *mm;
1309 vm_fault_t fault, major = 0;
1310 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
1315 /* kprobes don't want to hook the spurious faults: */
1316 if (unlikely(kprobes_fault(regs)))
1320 * Reserved bits are never expected to be set on
1321 * entries in the user portion of the page tables.
1323 if (unlikely(hw_error_code & X86_PF_RSVD))
1324 pgtable_bad(regs, hw_error_code, address);
1327 * If SMAP is on, check for invalid kernel (supervisor) access to user
1328 * pages in the user address space. The odd case here is WRUSS,
1329 * which, according to the preliminary documentation, does not respect
1330 * SMAP and will have the USER bit set so, in all cases, SMAP
1331 * enforcement appears to be consistent with the USER bit.
1333 if (unlikely(cpu_feature_enabled(X86_FEATURE_SMAP) &&
1334 !(hw_error_code & X86_PF_USER) &&
1335 !(regs->flags & X86_EFLAGS_AC)))
1337 bad_area_nosemaphore(regs, hw_error_code, address);
1342 * If we're in an interrupt, have no user context or are running
1343 * in a region with pagefaults disabled then we must not take the fault
1345 if (unlikely(faulthandler_disabled() || !mm)) {
1346 bad_area_nosemaphore(regs, hw_error_code, address);
1351 * It's safe to allow irq's after cr2 has been saved and the
1352 * vmalloc fault has been handled.
1354 * User-mode registers count as a user access even for any
1355 * potential system fault or CPU buglet:
1357 if (user_mode(regs)) {
1359 flags |= FAULT_FLAG_USER;
1361 if (regs->flags & X86_EFLAGS_IF)
1365 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
1367 if (hw_error_code & X86_PF_WRITE)
1368 flags |= FAULT_FLAG_WRITE;
1369 if (hw_error_code & X86_PF_INSTR)
1370 flags |= FAULT_FLAG_INSTRUCTION;
1372 #ifdef CONFIG_X86_64
1374 * Faults in the vsyscall page might need emulation. The
1375 * vsyscall page is at a high address (>PAGE_OFFSET), but is
1376 * considered to be part of the user address space.
1378 * The vsyscall page does not have a "real" VMA, so do this
1379 * emulation before we go searching for VMAs.
1381 * PKRU never rejects instruction fetches, so we don't need
1382 * to consider the PF_PK bit.
1384 if (is_vsyscall_vaddr(address)) {
1385 if (emulate_vsyscall(hw_error_code, regs, address))
1391 * Kernel-mode access to the user address space should only occur
1392 * on well-defined single instructions listed in the exception
1393 * tables. But, an erroneous kernel fault occurring outside one of
1394 * those areas which also holds mmap_sem might deadlock attempting
1395 * to validate the fault against the address space.
1397 * Only do the expensive exception table search when we might be at
1398 * risk of a deadlock. This happens if we
1399 * 1. Failed to acquire mmap_sem, and
1400 * 2. The access did not originate in userspace.
1402 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
1403 if (!user_mode(regs) && !search_exception_tables(regs->ip)) {
1405 * Fault from code in kernel from
1406 * which we do not expect faults.
1408 bad_area_nosemaphore(regs, hw_error_code, address);
1412 down_read(&mm->mmap_sem);
1415 * The above down_read_trylock() might have succeeded in
1416 * which case we'll have missed the might_sleep() from
1422 vma = find_vma(mm, address);
1423 if (unlikely(!vma)) {
1424 bad_area(regs, hw_error_code, address);
1427 if (likely(vma->vm_start <= address))
1429 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1430 bad_area(regs, hw_error_code, address);
1433 if (unlikely(expand_stack(vma, address))) {
1434 bad_area(regs, hw_error_code, address);
1439 * Ok, we have a good vm_area for this memory access, so
1440 * we can handle it..
1443 if (unlikely(access_error(hw_error_code, vma))) {
1444 bad_area_access_error(regs, hw_error_code, address, vma);
1449 * If for any reason at all we couldn't handle the fault,
1450 * make sure we exit gracefully rather than endlessly redo
1451 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1452 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
1454 * Note that handle_userfault() may also release and reacquire mmap_sem
1455 * (and not return with VM_FAULT_RETRY), when returning to userland to
1456 * repeat the page fault later with a VM_FAULT_NOPAGE retval
1457 * (potentially after handling any pending signal during the return to
1458 * userland). The return to userland is identified whenever
1459 * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags.
1461 fault = handle_mm_fault(vma, address, flags);
1462 major |= fault & VM_FAULT_MAJOR;
1465 * If we need to retry the mmap_sem has already been released,
1466 * and if there is a fatal signal pending there is no guarantee
1467 * that we made any progress. Handle this case first.
1469 if (unlikely(fault & VM_FAULT_RETRY)) {
1470 /* Retry at most once */
1471 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1472 flags &= ~FAULT_FLAG_ALLOW_RETRY;
1473 flags |= FAULT_FLAG_TRIED;
1474 if (!fatal_signal_pending(tsk))
1478 /* User mode? Just return to handle the fatal exception */
1479 if (flags & FAULT_FLAG_USER)
1482 /* Not returning to user mode? Handle exceptions or die: */
1483 no_context(regs, hw_error_code, address, SIGBUS, BUS_ADRERR);
1487 up_read(&mm->mmap_sem);
1488 if (unlikely(fault & VM_FAULT_ERROR)) {
1489 mm_fault_error(regs, hw_error_code, address, fault);
1494 * Major/minor page fault accounting. If any of the events
1495 * returned VM_FAULT_MAJOR, we account it as a major fault.
1499 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
1502 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
1505 check_v8086_mode(regs, address, tsk);
1507 NOKPROBE_SYMBOL(do_user_addr_fault);
1510 * This routine handles page faults. It determines the address,
1511 * and the problem, and then passes it off to one of the appropriate
1514 static noinline void
1515 __do_page_fault(struct pt_regs *regs, unsigned long hw_error_code,
1516 unsigned long address)
1518 prefetchw(¤t->mm->mmap_sem);
1520 if (unlikely(kmmio_fault(regs, address)))
1523 /* Was the fault on kernel-controlled part of the address space? */
1524 if (unlikely(fault_in_kernel_space(address)))
1525 do_kern_addr_fault(regs, hw_error_code, address);
1527 do_user_addr_fault(regs, hw_error_code, address);
1529 NOKPROBE_SYMBOL(__do_page_fault);
1531 static nokprobe_inline void
1532 trace_page_fault_entries(unsigned long address, struct pt_regs *regs,
1533 unsigned long error_code)
1535 if (user_mode(regs))
1536 trace_page_fault_user(address, regs, error_code);
1538 trace_page_fault_kernel(address, regs, error_code);
1542 * We must have this function blacklisted from kprobes, tagged with notrace
1543 * and call read_cr2() before calling anything else. To avoid calling any
1544 * kind of tracing machinery before we've observed the CR2 value.
1546 * exception_{enter,exit}() contains all sorts of tracepoints.
1548 dotraplinkage void notrace
1549 do_page_fault(struct pt_regs *regs, unsigned long error_code)
1551 unsigned long address = read_cr2(); /* Get the faulting address */
1552 enum ctx_state prev_state;
1554 prev_state = exception_enter();
1555 if (trace_pagefault_enabled())
1556 trace_page_fault_entries(address, regs, error_code);
1558 __do_page_fault(regs, error_code, address);
1559 exception_exit(prev_state);
1561 NOKPROBE_SYMBOL(do_page_fault);