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 */
32 #include <asm/pgtable_areas.h> /* VMALLOC_START, ... */
34 #define CREATE_TRACE_POINTS
35 #include <asm/trace/exceptions.h>
38 * Returns 0 if mmiotrace is disabled, or if the fault is not
39 * handled by mmiotrace:
41 static nokprobe_inline int
42 kmmio_fault(struct pt_regs *regs, unsigned long addr)
44 if (unlikely(is_kmmio_active()))
45 if (kmmio_handler(regs, addr) == 1)
55 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
56 * Check that here and ignore it.
60 * Sometimes the CPU reports invalid exceptions on prefetch.
61 * Check that here and ignore it.
63 * Opcode checker based on code by Richard Brunner.
66 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
67 unsigned char opcode, int *prefetch)
69 unsigned char instr_hi = opcode & 0xf0;
70 unsigned char instr_lo = opcode & 0x0f;
76 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
77 * In X86_64 long mode, the CPU will signal invalid
78 * opcode if some of these prefixes are present so
79 * X86_64 will never get here anyway
81 return ((instr_lo & 7) == 0x6);
85 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
86 * Need to figure out under what instruction mode the
87 * instruction was issued. Could check the LDT for lm,
88 * but for now it's good enough to assume that long
89 * mode only uses well known segments or kernel.
91 return (!user_mode(regs) || user_64bit_mode(regs));
94 /* 0x64 thru 0x67 are valid prefixes in all modes. */
95 return (instr_lo & 0xC) == 0x4;
97 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
98 return !instr_lo || (instr_lo>>1) == 1;
100 /* Prefetch instruction is 0x0F0D or 0x0F18 */
101 if (probe_kernel_address(instr, opcode))
104 *prefetch = (instr_lo == 0xF) &&
105 (opcode == 0x0D || opcode == 0x18);
113 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
115 unsigned char *max_instr;
116 unsigned char *instr;
120 * If it was a exec (instruction fetch) fault on NX page, then
121 * do not ignore the fault:
123 if (error_code & X86_PF_INSTR)
126 instr = (void *)convert_ip_to_linear(current, regs);
127 max_instr = instr + 15;
129 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE_MAX)
132 while (instr < max_instr) {
133 unsigned char opcode;
135 if (probe_kernel_address(instr, opcode))
140 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
146 DEFINE_SPINLOCK(pgd_lock);
150 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
152 unsigned index = pgd_index(address);
159 pgd_k = init_mm.pgd + index;
161 if (!pgd_present(*pgd_k))
165 * set_pgd(pgd, *pgd_k); here would be useless on PAE
166 * and redundant with the set_pmd() on non-PAE. As would
169 p4d = p4d_offset(pgd, address);
170 p4d_k = p4d_offset(pgd_k, address);
171 if (!p4d_present(*p4d_k))
174 pud = pud_offset(p4d, address);
175 pud_k = pud_offset(p4d_k, address);
176 if (!pud_present(*pud_k))
179 pmd = pmd_offset(pud, address);
180 pmd_k = pmd_offset(pud_k, address);
182 if (pmd_present(*pmd) != pmd_present(*pmd_k))
183 set_pmd(pmd, *pmd_k);
185 if (!pmd_present(*pmd_k))
188 BUG_ON(pmd_pfn(*pmd) != pmd_pfn(*pmd_k));
193 void arch_sync_kernel_mappings(unsigned long start, unsigned long end)
197 for (addr = start & PMD_MASK;
198 addr >= TASK_SIZE_MAX && addr < VMALLOC_END;
202 spin_lock(&pgd_lock);
203 list_for_each_entry(page, &pgd_list, lru) {
204 spinlock_t *pgt_lock;
206 /* the pgt_lock only for Xen */
207 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
210 vmalloc_sync_one(page_address(page), addr);
211 spin_unlock(pgt_lock);
213 spin_unlock(&pgd_lock);
220 * Handle a fault on the vmalloc or module mapping area
222 static noinline int vmalloc_fault(unsigned long address)
224 unsigned long pgd_paddr;
228 /* Make sure we are in vmalloc area: */
229 if (!(address >= VMALLOC_START && address < VMALLOC_END))
233 * Synchronize this task's top level page-table
234 * with the 'reference' page table.
236 * Do _not_ use "current" here. We might be inside
237 * an interrupt in the middle of a task switch..
239 pgd_paddr = read_cr3_pa();
240 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
244 if (pmd_large(*pmd_k))
247 pte_k = pte_offset_kernel(pmd_k, address);
248 if (!pte_present(*pte_k))
253 NOKPROBE_SYMBOL(vmalloc_fault);
256 * Did it hit the DOS screen memory VA from vm86 mode?
259 check_v8086_mode(struct pt_regs *regs, unsigned long address,
260 struct task_struct *tsk)
265 if (!v8086_mode(regs) || !tsk->thread.vm86)
268 bit = (address - 0xA0000) >> PAGE_SHIFT;
270 tsk->thread.vm86->screen_bitmap |= 1 << bit;
274 static bool low_pfn(unsigned long pfn)
276 return pfn < max_low_pfn;
279 static void dump_pagetable(unsigned long address)
281 pgd_t *base = __va(read_cr3_pa());
282 pgd_t *pgd = &base[pgd_index(address)];
288 #ifdef CONFIG_X86_PAE
289 pr_info("*pdpt = %016Lx ", pgd_val(*pgd));
290 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
292 #define pr_pde pr_cont
294 #define pr_pde pr_info
296 p4d = p4d_offset(pgd, address);
297 pud = pud_offset(p4d, address);
298 pmd = pmd_offset(pud, address);
299 pr_pde("*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
303 * We must not directly access the pte in the highpte
304 * case if the page table is located in highmem.
305 * And let's rather not kmap-atomic the pte, just in case
306 * it's allocated already:
308 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
311 pte = pte_offset_kernel(pmd, address);
312 pr_cont("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
317 #else /* CONFIG_X86_64: */
322 * Handle a fault on the vmalloc area
324 static noinline int vmalloc_fault(unsigned long address)
332 /* Make sure we are in vmalloc area: */
333 if (!(address >= VMALLOC_START && address < VMALLOC_END))
337 * Copy kernel mappings over when needed. This can also
338 * happen within a race in page table update. In the later
341 pgd = (pgd_t *)__va(read_cr3_pa()) + pgd_index(address);
342 pgd_k = pgd_offset_k(address);
343 if (pgd_none(*pgd_k))
346 if (pgtable_l5_enabled()) {
347 if (pgd_none(*pgd)) {
348 set_pgd(pgd, *pgd_k);
349 arch_flush_lazy_mmu_mode();
351 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_k));
355 /* With 4-level paging, copying happens on the p4d level. */
356 p4d = p4d_offset(pgd, address);
357 p4d_k = p4d_offset(pgd_k, address);
358 if (p4d_none(*p4d_k))
361 if (p4d_none(*p4d) && !pgtable_l5_enabled()) {
362 set_p4d(p4d, *p4d_k);
363 arch_flush_lazy_mmu_mode();
365 BUG_ON(p4d_pfn(*p4d) != p4d_pfn(*p4d_k));
368 BUILD_BUG_ON(CONFIG_PGTABLE_LEVELS < 4);
370 pud = pud_offset(p4d, address);
377 pmd = pmd_offset(pud, address);
384 pte = pte_offset_kernel(pmd, address);
385 if (!pte_present(*pte))
390 NOKPROBE_SYMBOL(vmalloc_fault);
392 #ifdef CONFIG_CPU_SUP_AMD
393 static const char errata93_warning[] =
395 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
396 "******* Working around it, but it may cause SEGVs or burn power.\n"
397 "******* Please consider a BIOS update.\n"
398 "******* Disabling USB legacy in the BIOS may also help.\n";
402 * No vm86 mode in 64-bit mode:
405 check_v8086_mode(struct pt_regs *regs, unsigned long address,
406 struct task_struct *tsk)
410 static int bad_address(void *p)
414 return probe_kernel_address((unsigned long *)p, dummy);
417 static void dump_pagetable(unsigned long address)
419 pgd_t *base = __va(read_cr3_pa());
420 pgd_t *pgd = base + pgd_index(address);
426 if (bad_address(pgd))
429 pr_info("PGD %lx ", pgd_val(*pgd));
431 if (!pgd_present(*pgd))
434 p4d = p4d_offset(pgd, address);
435 if (bad_address(p4d))
438 pr_cont("P4D %lx ", p4d_val(*p4d));
439 if (!p4d_present(*p4d) || p4d_large(*p4d))
442 pud = pud_offset(p4d, address);
443 if (bad_address(pud))
446 pr_cont("PUD %lx ", pud_val(*pud));
447 if (!pud_present(*pud) || pud_large(*pud))
450 pmd = pmd_offset(pud, address);
451 if (bad_address(pmd))
454 pr_cont("PMD %lx ", pmd_val(*pmd));
455 if (!pmd_present(*pmd) || pmd_large(*pmd))
458 pte = pte_offset_kernel(pmd, address);
459 if (bad_address(pte))
462 pr_cont("PTE %lx", pte_val(*pte));
470 #endif /* CONFIG_X86_64 */
473 * Workaround for K8 erratum #93 & buggy BIOS.
475 * BIOS SMM functions are required to use a specific workaround
476 * to avoid corruption of the 64bit RIP register on C stepping K8.
478 * A lot of BIOS that didn't get tested properly miss this.
480 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
481 * Try to work around it here.
483 * Note we only handle faults in kernel here.
484 * Does nothing on 32-bit.
486 static int is_errata93(struct pt_regs *regs, unsigned long address)
488 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
489 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD
490 || boot_cpu_data.x86 != 0xf)
493 if (address != regs->ip)
496 if ((address >> 32) != 0)
499 address |= 0xffffffffUL << 32;
500 if ((address >= (u64)_stext && address <= (u64)_etext) ||
501 (address >= MODULES_VADDR && address <= MODULES_END)) {
502 printk_once(errata93_warning);
511 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
512 * to illegal addresses >4GB.
514 * We catch this in the page fault handler because these addresses
515 * are not reachable. Just detect this case and return. Any code
516 * segment in LDT is compatibility mode.
518 static int is_errata100(struct pt_regs *regs, unsigned long address)
521 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
527 static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
529 #ifdef CONFIG_X86_F00F_BUG
533 * Pentium F0 0F C7 C8 bug workaround:
535 if (boot_cpu_has_bug(X86_BUG_F00F)) {
536 nr = (address - idt_descr.address) >> 3;
539 do_invalid_op(regs, 0);
547 static void show_ldttss(const struct desc_ptr *gdt, const char *name, u16 index)
549 u32 offset = (index >> 3) * sizeof(struct desc_struct);
551 struct ldttss_desc desc;
554 pr_alert("%s: NULL\n", name);
558 if (offset + sizeof(struct ldttss_desc) >= gdt->size) {
559 pr_alert("%s: 0x%hx -- out of bounds\n", name, index);
563 if (probe_kernel_read(&desc, (void *)(gdt->address + offset),
564 sizeof(struct ldttss_desc))) {
565 pr_alert("%s: 0x%hx -- GDT entry is not readable\n",
570 addr = desc.base0 | (desc.base1 << 16) | ((unsigned long)desc.base2 << 24);
572 addr |= ((u64)desc.base3 << 32);
574 pr_alert("%s: 0x%hx -- base=0x%lx limit=0x%x\n",
575 name, index, addr, (desc.limit0 | (desc.limit1 << 16)));
579 show_fault_oops(struct pt_regs *regs, unsigned long error_code, unsigned long address)
581 if (!oops_may_print())
584 if (error_code & X86_PF_INSTR) {
589 pgd = __va(read_cr3_pa());
590 pgd += pgd_index(address);
592 pte = lookup_address_in_pgd(pgd, address, &level);
594 if (pte && pte_present(*pte) && !pte_exec(*pte))
595 pr_crit("kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n",
596 from_kuid(&init_user_ns, current_uid()));
597 if (pte && pte_present(*pte) && pte_exec(*pte) &&
598 (pgd_flags(*pgd) & _PAGE_USER) &&
599 (__read_cr4() & X86_CR4_SMEP))
600 pr_crit("unable to execute userspace code (SMEP?) (uid: %d)\n",
601 from_kuid(&init_user_ns, current_uid()));
604 if (address < PAGE_SIZE && !user_mode(regs))
605 pr_alert("BUG: kernel NULL pointer dereference, address: %px\n",
608 pr_alert("BUG: unable to handle page fault for address: %px\n",
611 pr_alert("#PF: %s %s in %s mode\n",
612 (error_code & X86_PF_USER) ? "user" : "supervisor",
613 (error_code & X86_PF_INSTR) ? "instruction fetch" :
614 (error_code & X86_PF_WRITE) ? "write access" :
616 user_mode(regs) ? "user" : "kernel");
617 pr_alert("#PF: error_code(0x%04lx) - %s\n", error_code,
618 !(error_code & X86_PF_PROT) ? "not-present page" :
619 (error_code & X86_PF_RSVD) ? "reserved bit violation" :
620 (error_code & X86_PF_PK) ? "protection keys violation" :
621 "permissions violation");
623 if (!(error_code & X86_PF_USER) && user_mode(regs)) {
624 struct desc_ptr idt, gdt;
628 * This can happen for quite a few reasons. The more obvious
629 * ones are faults accessing the GDT, or LDT. Perhaps
630 * surprisingly, if the CPU tries to deliver a benign or
631 * contributory exception from user code and gets a page fault
632 * during delivery, the page fault can be delivered as though
633 * it originated directly from user code. This could happen
634 * due to wrong permissions on the IDT, GDT, LDT, TSS, or
635 * kernel or IST stack.
639 /* Usable even on Xen PV -- it's just slow. */
640 native_store_gdt(&gdt);
642 pr_alert("IDT: 0x%lx (limit=0x%hx) GDT: 0x%lx (limit=0x%hx)\n",
643 idt.address, idt.size, gdt.address, gdt.size);
646 show_ldttss(&gdt, "LDTR", ldtr);
649 show_ldttss(&gdt, "TR", tr);
652 dump_pagetable(address);
656 pgtable_bad(struct pt_regs *regs, unsigned long error_code,
657 unsigned long address)
659 struct task_struct *tsk;
663 flags = oops_begin();
667 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
669 dump_pagetable(address);
671 if (__die("Bad pagetable", regs, error_code))
674 oops_end(flags, regs, sig);
677 static void set_signal_archinfo(unsigned long address,
678 unsigned long error_code)
680 struct task_struct *tsk = current;
683 * To avoid leaking information about the kernel page
684 * table layout, pretend that user-mode accesses to
685 * kernel addresses are always protection faults.
687 * NB: This means that failed vsyscalls with vsyscall=none
688 * will have the PROT bit. This doesn't leak any
689 * information and does not appear to cause any problems.
691 if (address >= TASK_SIZE_MAX)
692 error_code |= X86_PF_PROT;
694 tsk->thread.trap_nr = X86_TRAP_PF;
695 tsk->thread.error_code = error_code | X86_PF_USER;
696 tsk->thread.cr2 = address;
700 no_context(struct pt_regs *regs, unsigned long error_code,
701 unsigned long address, int signal, int si_code)
703 struct task_struct *tsk = current;
707 if (user_mode(regs)) {
709 * This is an implicit supervisor-mode access from user
710 * mode. Bypass all the kernel-mode recovery code and just
716 /* Are we prepared to handle this kernel fault? */
717 if (fixup_exception(regs, X86_TRAP_PF, error_code, address)) {
719 * Any interrupt that takes a fault gets the fixup. This makes
720 * the below recursive fault logic only apply to a faults from
727 * Per the above we're !in_interrupt(), aka. task context.
729 * In this case we need to make sure we're not recursively
730 * faulting through the emulate_vsyscall() logic.
732 if (current->thread.sig_on_uaccess_err && signal) {
733 set_signal_archinfo(address, error_code);
735 /* XXX: hwpoison faults will set the wrong code. */
736 force_sig_fault(signal, si_code, (void __user *)address);
740 * Barring that, we can do the fixup and be happy.
745 #ifdef CONFIG_VMAP_STACK
747 * Stack overflow? During boot, we can fault near the initial
748 * stack in the direct map, but that's not an overflow -- check
749 * that we're in vmalloc space to avoid this.
751 if (is_vmalloc_addr((void *)address) &&
752 (((unsigned long)tsk->stack - 1 - address < PAGE_SIZE) ||
753 address - ((unsigned long)tsk->stack + THREAD_SIZE) < PAGE_SIZE)) {
754 unsigned long stack = __this_cpu_ist_top_va(DF) - sizeof(void *);
756 * We're likely to be running with very little stack space
757 * left. It's plausible that we'd hit this condition but
758 * double-fault even before we get this far, in which case
759 * we're fine: the double-fault handler will deal with it.
761 * We don't want to make it all the way into the oops code
762 * and then double-fault, though, because we're likely to
763 * break the console driver and lose most of the stack dump.
765 asm volatile ("movq %[stack], %%rsp\n\t"
766 "call handle_stack_overflow\n\t"
768 : ASM_CALL_CONSTRAINT
769 : "D" ("kernel stack overflow (page fault)"),
770 "S" (regs), "d" (address),
771 [stack] "rm" (stack));
779 * Valid to do another page fault here, because if this fault
780 * had been triggered by is_prefetch fixup_exception would have
785 * Hall of shame of CPU/BIOS bugs.
787 if (is_prefetch(regs, error_code, address))
790 if (is_errata93(regs, address))
794 * Buggy firmware could access regions which might page fault, try to
795 * recover from such faults.
797 if (IS_ENABLED(CONFIG_EFI))
798 efi_recover_from_page_fault(address);
802 * Oops. The kernel tried to access some bad page. We'll have to
803 * terminate things with extreme prejudice:
805 flags = oops_begin();
807 show_fault_oops(regs, error_code, address);
809 if (task_stack_end_corrupted(tsk))
810 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
813 if (__die("Oops", regs, error_code))
816 /* Executive summary in case the body of the oops scrolled away */
817 printk(KERN_DEFAULT "CR2: %016lx\n", address);
819 oops_end(flags, regs, sig);
823 * Print out info about fatal segfaults, if the show_unhandled_signals
827 show_signal_msg(struct pt_regs *regs, unsigned long error_code,
828 unsigned long address, struct task_struct *tsk)
830 const char *loglvl = task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG;
832 if (!unhandled_signal(tsk, SIGSEGV))
835 if (!printk_ratelimit())
838 printk("%s%s[%d]: segfault at %lx ip %px sp %px error %lx",
839 loglvl, tsk->comm, task_pid_nr(tsk), address,
840 (void *)regs->ip, (void *)regs->sp, error_code);
842 print_vma_addr(KERN_CONT " in ", regs->ip);
844 printk(KERN_CONT "\n");
846 show_opcodes(regs, loglvl);
850 * The (legacy) vsyscall page is the long page in the kernel portion
851 * of the address space that has user-accessible permissions.
853 static bool is_vsyscall_vaddr(unsigned long vaddr)
855 return unlikely((vaddr & PAGE_MASK) == VSYSCALL_ADDR);
859 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
860 unsigned long address, u32 pkey, int si_code)
862 struct task_struct *tsk = current;
864 /* User mode accesses just cause a SIGSEGV */
865 if (user_mode(regs) && (error_code & X86_PF_USER)) {
867 * It's possible to have interrupts off here:
872 * Valid to do another page fault here because this one came
875 if (is_prefetch(regs, error_code, address))
878 if (is_errata100(regs, address))
882 * To avoid leaking information about the kernel page table
883 * layout, pretend that user-mode accesses to kernel addresses
884 * are always protection faults.
886 if (address >= TASK_SIZE_MAX)
887 error_code |= X86_PF_PROT;
889 if (likely(show_unhandled_signals))
890 show_signal_msg(regs, error_code, address, tsk);
892 set_signal_archinfo(address, error_code);
894 if (si_code == SEGV_PKUERR)
895 force_sig_pkuerr((void __user *)address, pkey);
897 force_sig_fault(SIGSEGV, si_code, (void __user *)address);
902 if (is_f00f_bug(regs, address))
905 no_context(regs, error_code, address, SIGSEGV, si_code);
909 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
910 unsigned long address)
912 __bad_area_nosemaphore(regs, error_code, address, 0, SEGV_MAPERR);
916 __bad_area(struct pt_regs *regs, unsigned long error_code,
917 unsigned long address, u32 pkey, int si_code)
919 struct mm_struct *mm = current->mm;
921 * Something tried to access memory that isn't in our memory map..
922 * Fix it, but check if it's kernel or user first..
924 up_read(&mm->mmap_sem);
926 __bad_area_nosemaphore(regs, error_code, address, pkey, si_code);
930 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
932 __bad_area(regs, error_code, address, 0, SEGV_MAPERR);
935 static inline bool bad_area_access_from_pkeys(unsigned long error_code,
936 struct vm_area_struct *vma)
938 /* This code is always called on the current mm */
939 bool foreign = false;
941 if (!boot_cpu_has(X86_FEATURE_OSPKE))
943 if (error_code & X86_PF_PK)
945 /* this checks permission keys on the VMA: */
946 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
947 (error_code & X86_PF_INSTR), foreign))
953 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
954 unsigned long address, struct vm_area_struct *vma)
957 * This OSPKE check is not strictly necessary at runtime.
958 * But, doing it this way allows compiler optimizations
959 * if pkeys are compiled out.
961 if (bad_area_access_from_pkeys(error_code, vma)) {
963 * A protection key fault means that the PKRU value did not allow
964 * access to some PTE. Userspace can figure out what PKRU was
965 * from the XSAVE state. This function captures the pkey from
966 * the vma and passes it to userspace so userspace can discover
967 * which protection key was set on the PTE.
969 * If we get here, we know that the hardware signaled a X86_PF_PK
970 * fault and that there was a VMA once we got in the fault
971 * handler. It does *not* guarantee that the VMA we find here
972 * was the one that we faulted on.
974 * 1. T1 : mprotect_key(foo, PAGE_SIZE, pkey=4);
975 * 2. T1 : set PKRU to deny access to pkey=4, touches page
977 * 4. T2: mprotect_key(foo, PAGE_SIZE, pkey=5);
978 * 5. T1 : enters fault handler, takes mmap_sem, etc...
979 * 6. T1 : reaches here, sees vma_pkey(vma)=5, when we really
980 * faulted on a pte with its pkey=4.
982 u32 pkey = vma_pkey(vma);
984 __bad_area(regs, error_code, address, pkey, SEGV_PKUERR);
986 __bad_area(regs, error_code, address, 0, SEGV_ACCERR);
991 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
994 /* Kernel mode? Handle exceptions or die: */
995 if (!(error_code & X86_PF_USER)) {
996 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
1000 /* User-space => ok to do another page fault: */
1001 if (is_prefetch(regs, error_code, address))
1004 set_signal_archinfo(address, error_code);
1006 #ifdef CONFIG_MEMORY_FAILURE
1007 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
1008 struct task_struct *tsk = current;
1012 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
1013 tsk->comm, tsk->pid, address);
1014 if (fault & VM_FAULT_HWPOISON_LARGE)
1015 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
1016 if (fault & VM_FAULT_HWPOISON)
1018 force_sig_mceerr(BUS_MCEERR_AR, (void __user *)address, lsb);
1022 force_sig_fault(SIGBUS, BUS_ADRERR, (void __user *)address);
1025 static noinline void
1026 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
1027 unsigned long address, vm_fault_t fault)
1029 if (fatal_signal_pending(current) && !(error_code & X86_PF_USER)) {
1030 no_context(regs, error_code, address, 0, 0);
1034 if (fault & VM_FAULT_OOM) {
1035 /* Kernel mode? Handle exceptions or die: */
1036 if (!(error_code & X86_PF_USER)) {
1037 no_context(regs, error_code, address,
1038 SIGSEGV, SEGV_MAPERR);
1043 * We ran out of memory, call the OOM killer, and return the
1044 * userspace (which will retry the fault, or kill us if we got
1047 pagefault_out_of_memory();
1049 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
1050 VM_FAULT_HWPOISON_LARGE))
1051 do_sigbus(regs, error_code, address, fault);
1052 else if (fault & VM_FAULT_SIGSEGV)
1053 bad_area_nosemaphore(regs, error_code, address);
1059 static int spurious_kernel_fault_check(unsigned long error_code, pte_t *pte)
1061 if ((error_code & X86_PF_WRITE) && !pte_write(*pte))
1064 if ((error_code & X86_PF_INSTR) && !pte_exec(*pte))
1071 * Handle a spurious fault caused by a stale TLB entry.
1073 * This allows us to lazily refresh the TLB when increasing the
1074 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
1075 * eagerly is very expensive since that implies doing a full
1076 * cross-processor TLB flush, even if no stale TLB entries exist
1077 * on other processors.
1079 * Spurious faults may only occur if the TLB contains an entry with
1080 * fewer permission than the page table entry. Non-present (P = 0)
1081 * and reserved bit (R = 1) faults are never spurious.
1083 * There are no security implications to leaving a stale TLB when
1084 * increasing the permissions on a page.
1086 * Returns non-zero if a spurious fault was handled, zero otherwise.
1088 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
1089 * (Optional Invalidation).
1092 spurious_kernel_fault(unsigned long error_code, unsigned long address)
1102 * Only writes to RO or instruction fetches from NX may cause
1105 * These could be from user or supervisor accesses but the TLB
1106 * is only lazily flushed after a kernel mapping protection
1107 * change, so user accesses are not expected to cause spurious
1110 if (error_code != (X86_PF_WRITE | X86_PF_PROT) &&
1111 error_code != (X86_PF_INSTR | X86_PF_PROT))
1114 pgd = init_mm.pgd + pgd_index(address);
1115 if (!pgd_present(*pgd))
1118 p4d = p4d_offset(pgd, address);
1119 if (!p4d_present(*p4d))
1122 if (p4d_large(*p4d))
1123 return spurious_kernel_fault_check(error_code, (pte_t *) p4d);
1125 pud = pud_offset(p4d, address);
1126 if (!pud_present(*pud))
1129 if (pud_large(*pud))
1130 return spurious_kernel_fault_check(error_code, (pte_t *) pud);
1132 pmd = pmd_offset(pud, address);
1133 if (!pmd_present(*pmd))
1136 if (pmd_large(*pmd))
1137 return spurious_kernel_fault_check(error_code, (pte_t *) pmd);
1139 pte = pte_offset_kernel(pmd, address);
1140 if (!pte_present(*pte))
1143 ret = spurious_kernel_fault_check(error_code, pte);
1148 * Make sure we have permissions in PMD.
1149 * If not, then there's a bug in the page tables:
1151 ret = spurious_kernel_fault_check(error_code, (pte_t *) pmd);
1152 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
1156 NOKPROBE_SYMBOL(spurious_kernel_fault);
1158 int show_unhandled_signals = 1;
1161 access_error(unsigned long error_code, struct vm_area_struct *vma)
1163 /* This is only called for the current mm, so: */
1164 bool foreign = false;
1167 * Read or write was blocked by protection keys. This is
1168 * always an unconditional error and can never result in
1169 * a follow-up action to resolve the fault, like a COW.
1171 if (error_code & X86_PF_PK)
1175 * Make sure to check the VMA so that we do not perform
1176 * faults just to hit a X86_PF_PK as soon as we fill in a
1179 if (!arch_vma_access_permitted(vma, (error_code & X86_PF_WRITE),
1180 (error_code & X86_PF_INSTR), foreign))
1183 if (error_code & X86_PF_WRITE) {
1184 /* write, present and write, not present: */
1185 if (unlikely(!(vma->vm_flags & VM_WRITE)))
1190 /* read, present: */
1191 if (unlikely(error_code & X86_PF_PROT))
1194 /* read, not present: */
1195 if (unlikely(!vma_is_accessible(vma)))
1201 static int fault_in_kernel_space(unsigned long address)
1204 * On 64-bit systems, the vsyscall page is at an address above
1205 * TASK_SIZE_MAX, but is not considered part of the kernel
1208 if (IS_ENABLED(CONFIG_X86_64) && is_vsyscall_vaddr(address))
1211 return address >= TASK_SIZE_MAX;
1215 * Called for all faults where 'address' is part of the kernel address
1216 * space. Might get called for faults that originate from *code* that
1217 * ran in userspace or the kernel.
1220 do_kern_addr_fault(struct pt_regs *regs, unsigned long hw_error_code,
1221 unsigned long address)
1224 * Protection keys exceptions only happen on user pages. We
1225 * have no user pages in the kernel portion of the address
1226 * space, so do not expect them here.
1228 WARN_ON_ONCE(hw_error_code & X86_PF_PK);
1231 * We can fault-in kernel-space virtual memory on-demand. The
1232 * 'reference' page table is init_mm.pgd.
1234 * NOTE! We MUST NOT take any locks for this case. We may
1235 * be in an interrupt or a critical region, and should
1236 * only copy the information from the master page table,
1239 * Before doing this on-demand faulting, ensure that the
1240 * fault is not any of the following:
1241 * 1. A fault on a PTE with a reserved bit set.
1242 * 2. A fault caused by a user-mode access. (Do not demand-
1243 * fault kernel memory due to user-mode accesses).
1244 * 3. A fault caused by a page-level protection violation.
1245 * (A demand fault would be on a non-present page which
1246 * would have X86_PF_PROT==0).
1248 if (!(hw_error_code & (X86_PF_RSVD | X86_PF_USER | X86_PF_PROT))) {
1249 if (vmalloc_fault(address) >= 0)
1253 /* Was the fault spurious, caused by lazy TLB invalidation? */
1254 if (spurious_kernel_fault(hw_error_code, address))
1257 /* kprobes don't want to hook the spurious faults: */
1258 if (kprobe_page_fault(regs, X86_TRAP_PF))
1262 * Note, despite being a "bad area", there are quite a few
1263 * acceptable reasons to get here, such as erratum fixups
1264 * and handling kernel code that can fault, like get_user().
1266 * Don't take the mm semaphore here. If we fixup a prefetch
1267 * fault we could otherwise deadlock:
1269 bad_area_nosemaphore(regs, hw_error_code, address);
1271 NOKPROBE_SYMBOL(do_kern_addr_fault);
1273 /* Handle faults in the user portion of the address space */
1275 void do_user_addr_fault(struct pt_regs *regs,
1276 unsigned long hw_error_code,
1277 unsigned long address)
1279 struct vm_area_struct *vma;
1280 struct task_struct *tsk;
1281 struct mm_struct *mm;
1282 vm_fault_t fault, major = 0;
1283 unsigned int flags = FAULT_FLAG_DEFAULT;
1288 /* kprobes don't want to hook the spurious faults: */
1289 if (unlikely(kprobe_page_fault(regs, X86_TRAP_PF)))
1293 * Reserved bits are never expected to be set on
1294 * entries in the user portion of the page tables.
1296 if (unlikely(hw_error_code & X86_PF_RSVD))
1297 pgtable_bad(regs, hw_error_code, address);
1300 * If SMAP is on, check for invalid kernel (supervisor) access to user
1301 * pages in the user address space. The odd case here is WRUSS,
1302 * which, according to the preliminary documentation, does not respect
1303 * SMAP and will have the USER bit set so, in all cases, SMAP
1304 * enforcement appears to be consistent with the USER bit.
1306 if (unlikely(cpu_feature_enabled(X86_FEATURE_SMAP) &&
1307 !(hw_error_code & X86_PF_USER) &&
1308 !(regs->flags & X86_EFLAGS_AC)))
1310 bad_area_nosemaphore(regs, hw_error_code, address);
1315 * If we're in an interrupt, have no user context or are running
1316 * in a region with pagefaults disabled then we must not take the fault
1318 if (unlikely(faulthandler_disabled() || !mm)) {
1319 bad_area_nosemaphore(regs, hw_error_code, address);
1324 * It's safe to allow irq's after cr2 has been saved and the
1325 * vmalloc fault has been handled.
1327 * User-mode registers count as a user access even for any
1328 * potential system fault or CPU buglet:
1330 if (user_mode(regs)) {
1332 flags |= FAULT_FLAG_USER;
1334 if (regs->flags & X86_EFLAGS_IF)
1338 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
1340 if (hw_error_code & X86_PF_WRITE)
1341 flags |= FAULT_FLAG_WRITE;
1342 if (hw_error_code & X86_PF_INSTR)
1343 flags |= FAULT_FLAG_INSTRUCTION;
1345 #ifdef CONFIG_X86_64
1347 * Faults in the vsyscall page might need emulation. The
1348 * vsyscall page is at a high address (>PAGE_OFFSET), but is
1349 * considered to be part of the user address space.
1351 * The vsyscall page does not have a "real" VMA, so do this
1352 * emulation before we go searching for VMAs.
1354 * PKRU never rejects instruction fetches, so we don't need
1355 * to consider the PF_PK bit.
1357 if (is_vsyscall_vaddr(address)) {
1358 if (emulate_vsyscall(hw_error_code, regs, address))
1364 * Kernel-mode access to the user address space should only occur
1365 * on well-defined single instructions listed in the exception
1366 * tables. But, an erroneous kernel fault occurring outside one of
1367 * those areas which also holds mmap_sem might deadlock attempting
1368 * to validate the fault against the address space.
1370 * Only do the expensive exception table search when we might be at
1371 * risk of a deadlock. This happens if we
1372 * 1. Failed to acquire mmap_sem, and
1373 * 2. The access did not originate in userspace.
1375 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
1376 if (!user_mode(regs) && !search_exception_tables(regs->ip)) {
1378 * Fault from code in kernel from
1379 * which we do not expect faults.
1381 bad_area_nosemaphore(regs, hw_error_code, address);
1385 down_read(&mm->mmap_sem);
1388 * The above down_read_trylock() might have succeeded in
1389 * which case we'll have missed the might_sleep() from
1395 vma = find_vma(mm, address);
1396 if (unlikely(!vma)) {
1397 bad_area(regs, hw_error_code, address);
1400 if (likely(vma->vm_start <= address))
1402 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1403 bad_area(regs, hw_error_code, address);
1406 if (unlikely(expand_stack(vma, address))) {
1407 bad_area(regs, hw_error_code, address);
1412 * Ok, we have a good vm_area for this memory access, so
1413 * we can handle it..
1416 if (unlikely(access_error(hw_error_code, vma))) {
1417 bad_area_access_error(regs, hw_error_code, address, vma);
1422 * If for any reason at all we couldn't handle the fault,
1423 * make sure we exit gracefully rather than endlessly redo
1424 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1425 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
1427 * Note that handle_userfault() may also release and reacquire mmap_sem
1428 * (and not return with VM_FAULT_RETRY), when returning to userland to
1429 * repeat the page fault later with a VM_FAULT_NOPAGE retval
1430 * (potentially after handling any pending signal during the return to
1431 * userland). The return to userland is identified whenever
1432 * FAULT_FLAG_USER|FAULT_FLAG_KILLABLE are both set in flags.
1434 fault = handle_mm_fault(vma, address, flags);
1435 major |= fault & VM_FAULT_MAJOR;
1437 /* Quick path to respond to signals */
1438 if (fault_signal_pending(fault, regs)) {
1439 if (!user_mode(regs))
1440 no_context(regs, hw_error_code, address, SIGBUS,
1446 * If we need to retry the mmap_sem has already been released,
1447 * and if there is a fatal signal pending there is no guarantee
1448 * that we made any progress. Handle this case first.
1450 if (unlikely((fault & VM_FAULT_RETRY) &&
1451 (flags & FAULT_FLAG_ALLOW_RETRY))) {
1452 flags |= FAULT_FLAG_TRIED;
1456 up_read(&mm->mmap_sem);
1457 if (unlikely(fault & VM_FAULT_ERROR)) {
1458 mm_fault_error(regs, hw_error_code, address, fault);
1463 * Major/minor page fault accounting. If any of the events
1464 * returned VM_FAULT_MAJOR, we account it as a major fault.
1468 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, regs, address);
1471 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, regs, address);
1474 check_v8086_mode(regs, address, tsk);
1476 NOKPROBE_SYMBOL(do_user_addr_fault);
1478 static __always_inline void
1479 trace_page_fault_entries(struct pt_regs *regs, unsigned long error_code,
1480 unsigned long address)
1482 if (!trace_pagefault_enabled())
1485 if (user_mode(regs))
1486 trace_page_fault_user(address, regs, error_code);
1488 trace_page_fault_kernel(address, regs, error_code);
1492 do_page_fault(struct pt_regs *regs, unsigned long hw_error_code,
1493 unsigned long address)
1495 prefetchw(¤t->mm->mmap_sem);
1496 trace_page_fault_entries(regs, hw_error_code, address);
1498 if (unlikely(kmmio_fault(regs, address)))
1501 /* Was the fault on kernel-controlled part of the address space? */
1502 if (unlikely(fault_in_kernel_space(address)))
1503 do_kern_addr_fault(regs, hw_error_code, address);
1505 do_user_addr_fault(regs, hw_error_code, address);
1507 NOKPROBE_SYMBOL(do_page_fault);