2 * Copyright (C) 1995 Linus Torvalds
3 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs.
4 * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar
6 #include <linux/magic.h> /* STACK_END_MAGIC */
7 #include <linux/sched.h> /* test_thread_flag(), ... */
8 #include <linux/kdebug.h> /* oops_begin/end, ... */
9 #include <linux/module.h> /* search_exception_table */
10 #include <linux/bootmem.h> /* max_low_pfn */
11 #include <linux/kprobes.h> /* __kprobes, ... */
12 #include <linux/mmiotrace.h> /* kmmio_handler, ... */
13 #include <linux/perf_event.h> /* perf_sw_event */
14 #include <linux/hugetlb.h> /* hstate_index_to_shift */
15 #include <linux/prefetch.h> /* prefetchw */
16 #include <linux/context_tracking.h> /* exception_enter(), ... */
18 #include <asm/traps.h> /* dotraplinkage, ... */
19 #include <asm/pgalloc.h> /* pgd_*(), ... */
20 #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */
21 #include <asm/fixmap.h> /* VSYSCALL_START */
24 * Page fault error code bits:
26 * bit 0 == 0: no page found 1: protection fault
27 * bit 1 == 0: read access 1: write access
28 * bit 2 == 0: kernel-mode access 1: user-mode access
29 * bit 3 == 1: use of reserved bit detected
30 * bit 4 == 1: fault was an instruction fetch
32 enum x86_pf_error_code {
42 * Returns 0 if mmiotrace is disabled, or if the fault is not
43 * handled by mmiotrace:
45 static inline int __kprobes
46 kmmio_fault(struct pt_regs *regs, unsigned long addr)
48 if (unlikely(is_kmmio_active()))
49 if (kmmio_handler(regs, addr) == 1)
54 static inline int __kprobes notify_page_fault(struct pt_regs *regs)
58 /* kprobe_running() needs smp_processor_id() */
59 if (kprobes_built_in() && !user_mode_vm(regs)) {
61 if (kprobe_running() && kprobe_fault_handler(regs, 14))
74 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
75 * Check that here and ignore it.
79 * Sometimes the CPU reports invalid exceptions on prefetch.
80 * Check that here and ignore it.
82 * Opcode checker based on code by Richard Brunner.
85 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
86 unsigned char opcode, int *prefetch)
88 unsigned char instr_hi = opcode & 0xf0;
89 unsigned char instr_lo = opcode & 0x0f;
95 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
96 * In X86_64 long mode, the CPU will signal invalid
97 * opcode if some of these prefixes are present so
98 * X86_64 will never get here anyway
100 return ((instr_lo & 7) == 0x6);
104 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
105 * Need to figure out under what instruction mode the
106 * instruction was issued. Could check the LDT for lm,
107 * but for now it's good enough to assume that long
108 * mode only uses well known segments or kernel.
110 return (!user_mode(regs) || user_64bit_mode(regs));
113 /* 0x64 thru 0x67 are valid prefixes in all modes. */
114 return (instr_lo & 0xC) == 0x4;
116 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
117 return !instr_lo || (instr_lo>>1) == 1;
119 /* Prefetch instruction is 0x0F0D or 0x0F18 */
120 if (probe_kernel_address(instr, opcode))
123 *prefetch = (instr_lo == 0xF) &&
124 (opcode == 0x0D || opcode == 0x18);
132 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
134 unsigned char *max_instr;
135 unsigned char *instr;
139 * If it was a exec (instruction fetch) fault on NX page, then
140 * do not ignore the fault:
142 if (error_code & PF_INSTR)
145 instr = (void *)convert_ip_to_linear(current, regs);
146 max_instr = instr + 15;
148 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
151 while (instr < max_instr) {
152 unsigned char opcode;
154 if (probe_kernel_address(instr, opcode))
159 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
166 force_sig_info_fault(int si_signo, int si_code, unsigned long address,
167 struct task_struct *tsk, int fault)
172 info.si_signo = si_signo;
174 info.si_code = si_code;
175 info.si_addr = (void __user *)address;
176 if (fault & VM_FAULT_HWPOISON_LARGE)
177 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
178 if (fault & VM_FAULT_HWPOISON)
180 info.si_addr_lsb = lsb;
182 force_sig_info(si_signo, &info, tsk);
185 DEFINE_SPINLOCK(pgd_lock);
189 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
191 unsigned index = pgd_index(address);
197 pgd_k = init_mm.pgd + index;
199 if (!pgd_present(*pgd_k))
203 * set_pgd(pgd, *pgd_k); here would be useless on PAE
204 * and redundant with the set_pmd() on non-PAE. As would
207 pud = pud_offset(pgd, address);
208 pud_k = pud_offset(pgd_k, address);
209 if (!pud_present(*pud_k))
212 pmd = pmd_offset(pud, address);
213 pmd_k = pmd_offset(pud_k, address);
214 if (!pmd_present(*pmd_k))
217 if (!pmd_present(*pmd))
218 set_pmd(pmd, *pmd_k);
220 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
225 void vmalloc_sync_all(void)
227 unsigned long address;
229 if (SHARED_KERNEL_PMD)
232 for (address = VMALLOC_START & PMD_MASK;
233 address >= TASK_SIZE && address < FIXADDR_TOP;
234 address += PMD_SIZE) {
237 spin_lock(&pgd_lock);
238 list_for_each_entry(page, &pgd_list, lru) {
239 spinlock_t *pgt_lock;
242 /* the pgt_lock only for Xen */
243 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
246 ret = vmalloc_sync_one(page_address(page), address);
247 spin_unlock(pgt_lock);
252 spin_unlock(&pgd_lock);
259 * Handle a fault on the vmalloc or module mapping area
261 static noinline __kprobes int vmalloc_fault(unsigned long address)
263 unsigned long pgd_paddr;
267 /* Make sure we are in vmalloc area: */
268 if (!(address >= VMALLOC_START && address < VMALLOC_END))
271 WARN_ON_ONCE(in_nmi());
274 * Synchronize this task's top level page-table
275 * with the 'reference' page table.
277 * Do _not_ use "current" here. We might be inside
278 * an interrupt in the middle of a task switch..
280 pgd_paddr = read_cr3();
281 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
285 pte_k = pte_offset_kernel(pmd_k, address);
286 if (!pte_present(*pte_k))
293 * Did it hit the DOS screen memory VA from vm86 mode?
296 check_v8086_mode(struct pt_regs *regs, unsigned long address,
297 struct task_struct *tsk)
301 if (!v8086_mode(regs))
304 bit = (address - 0xA0000) >> PAGE_SHIFT;
306 tsk->thread.screen_bitmap |= 1 << bit;
309 static bool low_pfn(unsigned long pfn)
311 return pfn < max_low_pfn;
314 static void dump_pagetable(unsigned long address)
316 pgd_t *base = __va(read_cr3());
317 pgd_t *pgd = &base[pgd_index(address)];
321 #ifdef CONFIG_X86_PAE
322 printk("*pdpt = %016Lx ", pgd_val(*pgd));
323 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
326 pmd = pmd_offset(pud_offset(pgd, address), address);
327 printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
330 * We must not directly access the pte in the highpte
331 * case if the page table is located in highmem.
332 * And let's rather not kmap-atomic the pte, just in case
333 * it's allocated already:
335 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
338 pte = pte_offset_kernel(pmd, address);
339 printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
344 #else /* CONFIG_X86_64: */
346 void vmalloc_sync_all(void)
348 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END);
354 * Handle a fault on the vmalloc area
356 * This assumes no large pages in there.
358 static noinline __kprobes int vmalloc_fault(unsigned long address)
360 pgd_t *pgd, *pgd_ref;
361 pud_t *pud, *pud_ref;
362 pmd_t *pmd, *pmd_ref;
363 pte_t *pte, *pte_ref;
365 /* Make sure we are in vmalloc area: */
366 if (!(address >= VMALLOC_START && address < VMALLOC_END))
369 WARN_ON_ONCE(in_nmi());
372 * Copy kernel mappings over when needed. This can also
373 * happen within a race in page table update. In the later
376 pgd = pgd_offset(current->active_mm, address);
377 pgd_ref = pgd_offset_k(address);
378 if (pgd_none(*pgd_ref))
381 if (pgd_none(*pgd)) {
382 set_pgd(pgd, *pgd_ref);
383 arch_flush_lazy_mmu_mode();
385 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
389 * Below here mismatches are bugs because these lower tables
393 pud = pud_offset(pgd, address);
394 pud_ref = pud_offset(pgd_ref, address);
395 if (pud_none(*pud_ref))
398 if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref))
401 pmd = pmd_offset(pud, address);
402 pmd_ref = pmd_offset(pud_ref, address);
403 if (pmd_none(*pmd_ref))
406 if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref))
409 pte_ref = pte_offset_kernel(pmd_ref, address);
410 if (!pte_present(*pte_ref))
413 pte = pte_offset_kernel(pmd, address);
416 * Don't use pte_page here, because the mappings can point
417 * outside mem_map, and the NUMA hash lookup cannot handle
420 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
426 #ifdef CONFIG_CPU_SUP_AMD
427 static const char errata93_warning[] =
429 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
430 "******* Working around it, but it may cause SEGVs or burn power.\n"
431 "******* Please consider a BIOS update.\n"
432 "******* Disabling USB legacy in the BIOS may also help.\n";
436 * No vm86 mode in 64-bit mode:
439 check_v8086_mode(struct pt_regs *regs, unsigned long address,
440 struct task_struct *tsk)
444 static int bad_address(void *p)
448 return probe_kernel_address((unsigned long *)p, dummy);
451 static void dump_pagetable(unsigned long address)
453 pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK);
454 pgd_t *pgd = base + pgd_index(address);
459 if (bad_address(pgd))
462 printk("PGD %lx ", pgd_val(*pgd));
464 if (!pgd_present(*pgd))
467 pud = pud_offset(pgd, address);
468 if (bad_address(pud))
471 printk("PUD %lx ", pud_val(*pud));
472 if (!pud_present(*pud) || pud_large(*pud))
475 pmd = pmd_offset(pud, address);
476 if (bad_address(pmd))
479 printk("PMD %lx ", pmd_val(*pmd));
480 if (!pmd_present(*pmd) || pmd_large(*pmd))
483 pte = pte_offset_kernel(pmd, address);
484 if (bad_address(pte))
487 printk("PTE %lx", pte_val(*pte));
495 #endif /* CONFIG_X86_64 */
498 * Workaround for K8 erratum #93 & buggy BIOS.
500 * BIOS SMM functions are required to use a specific workaround
501 * to avoid corruption of the 64bit RIP register on C stepping K8.
503 * A lot of BIOS that didn't get tested properly miss this.
505 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
506 * Try to work around it here.
508 * Note we only handle faults in kernel here.
509 * Does nothing on 32-bit.
511 static int is_errata93(struct pt_regs *regs, unsigned long address)
513 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
514 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD
515 || boot_cpu_data.x86 != 0xf)
518 if (address != regs->ip)
521 if ((address >> 32) != 0)
524 address |= 0xffffffffUL << 32;
525 if ((address >= (u64)_stext && address <= (u64)_etext) ||
526 (address >= MODULES_VADDR && address <= MODULES_END)) {
527 printk_once(errata93_warning);
536 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
537 * to illegal addresses >4GB.
539 * We catch this in the page fault handler because these addresses
540 * are not reachable. Just detect this case and return. Any code
541 * segment in LDT is compatibility mode.
543 static int is_errata100(struct pt_regs *regs, unsigned long address)
546 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
552 static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
554 #ifdef CONFIG_X86_F00F_BUG
558 * Pentium F0 0F C7 C8 bug workaround:
560 if (boot_cpu_has_bug(X86_BUG_F00F)) {
561 nr = (address - idt_descr.address) >> 3;
564 do_invalid_op(regs, 0);
572 static const char nx_warning[] = KERN_CRIT
573 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
576 show_fault_oops(struct pt_regs *regs, unsigned long error_code,
577 unsigned long address)
579 if (!oops_may_print())
582 if (error_code & PF_INSTR) {
585 pte_t *pte = lookup_address(address, &level);
587 if (pte && pte_present(*pte) && !pte_exec(*pte))
588 printk(nx_warning, from_kuid(&init_user_ns, current_uid()));
591 printk(KERN_ALERT "BUG: unable to handle kernel ");
592 if (address < PAGE_SIZE)
593 printk(KERN_CONT "NULL pointer dereference");
595 printk(KERN_CONT "paging request");
597 printk(KERN_CONT " at %p\n", (void *) address);
598 printk(KERN_ALERT "IP:");
599 printk_address(regs->ip, 1);
601 dump_pagetable(address);
605 pgtable_bad(struct pt_regs *regs, unsigned long error_code,
606 unsigned long address)
608 struct task_struct *tsk;
612 flags = oops_begin();
616 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
618 dump_pagetable(address);
620 tsk->thread.cr2 = address;
621 tsk->thread.trap_nr = X86_TRAP_PF;
622 tsk->thread.error_code = error_code;
624 if (__die("Bad pagetable", regs, error_code))
627 oops_end(flags, regs, sig);
631 no_context(struct pt_regs *regs, unsigned long error_code,
632 unsigned long address, int signal, int si_code)
634 struct task_struct *tsk = current;
635 unsigned long *stackend;
639 /* Are we prepared to handle this kernel fault? */
640 if (fixup_exception(regs)) {
641 if (current_thread_info()->sig_on_uaccess_error && signal) {
642 tsk->thread.trap_nr = X86_TRAP_PF;
643 tsk->thread.error_code = error_code | PF_USER;
644 tsk->thread.cr2 = address;
646 /* XXX: hwpoison faults will set the wrong code. */
647 force_sig_info_fault(signal, si_code, address, tsk, 0);
655 * Valid to do another page fault here, because if this fault
656 * had been triggered by is_prefetch fixup_exception would have
661 * Hall of shame of CPU/BIOS bugs.
663 if (is_prefetch(regs, error_code, address))
666 if (is_errata93(regs, address))
670 * Oops. The kernel tried to access some bad page. We'll have to
671 * terminate things with extreme prejudice:
673 flags = oops_begin();
675 show_fault_oops(regs, error_code, address);
677 stackend = end_of_stack(tsk);
678 if (tsk != &init_task && *stackend != STACK_END_MAGIC)
679 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
681 tsk->thread.cr2 = address;
682 tsk->thread.trap_nr = X86_TRAP_PF;
683 tsk->thread.error_code = error_code;
686 if (__die("Oops", regs, error_code))
689 /* Executive summary in case the body of the oops scrolled away */
690 printk(KERN_DEFAULT "CR2: %016lx\n", address);
692 oops_end(flags, regs, sig);
696 * Print out info about fatal segfaults, if the show_unhandled_signals
700 show_signal_msg(struct pt_regs *regs, unsigned long error_code,
701 unsigned long address, struct task_struct *tsk)
703 if (!unhandled_signal(tsk, SIGSEGV))
706 if (!printk_ratelimit())
709 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
710 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
711 tsk->comm, task_pid_nr(tsk), address,
712 (void *)regs->ip, (void *)regs->sp, error_code);
714 print_vma_addr(KERN_CONT " in ", regs->ip);
716 printk(KERN_CONT "\n");
720 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
721 unsigned long address, int si_code)
723 struct task_struct *tsk = current;
725 /* User mode accesses just cause a SIGSEGV */
726 if (error_code & PF_USER) {
728 * It's possible to have interrupts off here:
733 * Valid to do another page fault here because this one came
736 if (is_prefetch(regs, error_code, address))
739 if (is_errata100(regs, address))
744 * Instruction fetch faults in the vsyscall page might need
747 if (unlikely((error_code & PF_INSTR) &&
748 ((address & ~0xfff) == VSYSCALL_START))) {
749 if (emulate_vsyscall(regs, address))
753 /* Kernel addresses are always protection faults: */
754 if (address >= TASK_SIZE)
755 error_code |= PF_PROT;
757 if (likely(show_unhandled_signals))
758 show_signal_msg(regs, error_code, address, tsk);
760 tsk->thread.cr2 = address;
761 tsk->thread.error_code = error_code;
762 tsk->thread.trap_nr = X86_TRAP_PF;
764 force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0);
769 if (is_f00f_bug(regs, address))
772 no_context(regs, error_code, address, SIGSEGV, si_code);
776 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
777 unsigned long address)
779 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
783 __bad_area(struct pt_regs *regs, unsigned long error_code,
784 unsigned long address, int si_code)
786 struct mm_struct *mm = current->mm;
789 * Something tried to access memory that isn't in our memory map..
790 * Fix it, but check if it's kernel or user first..
792 up_read(&mm->mmap_sem);
794 __bad_area_nosemaphore(regs, error_code, address, si_code);
798 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
800 __bad_area(regs, error_code, address, SEGV_MAPERR);
804 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
805 unsigned long address)
807 __bad_area(regs, error_code, address, SEGV_ACCERR);
811 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
814 struct task_struct *tsk = current;
815 struct mm_struct *mm = tsk->mm;
816 int code = BUS_ADRERR;
818 up_read(&mm->mmap_sem);
820 /* Kernel mode? Handle exceptions or die: */
821 if (!(error_code & PF_USER)) {
822 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
826 /* User-space => ok to do another page fault: */
827 if (is_prefetch(regs, error_code, address))
830 tsk->thread.cr2 = address;
831 tsk->thread.error_code = error_code;
832 tsk->thread.trap_nr = X86_TRAP_PF;
834 #ifdef CONFIG_MEMORY_FAILURE
835 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
837 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
838 tsk->comm, tsk->pid, address);
839 code = BUS_MCEERR_AR;
842 force_sig_info_fault(SIGBUS, code, address, tsk, fault);
846 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
847 unsigned long address, unsigned int fault)
850 * Pagefault was interrupted by SIGKILL. We have no reason to
851 * continue pagefault.
853 if (fatal_signal_pending(current)) {
854 if (!(fault & VM_FAULT_RETRY))
855 up_read(¤t->mm->mmap_sem);
856 if (!(error_code & PF_USER))
857 no_context(regs, error_code, address, 0, 0);
860 if (!(fault & VM_FAULT_ERROR))
863 if (fault & VM_FAULT_OOM) {
864 /* Kernel mode? Handle exceptions or die: */
865 if (!(error_code & PF_USER)) {
866 up_read(¤t->mm->mmap_sem);
867 no_context(regs, error_code, address,
868 SIGSEGV, SEGV_MAPERR);
872 up_read(¤t->mm->mmap_sem);
875 * We ran out of memory, call the OOM killer, and return the
876 * userspace (which will retry the fault, or kill us if we got
879 pagefault_out_of_memory();
881 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
882 VM_FAULT_HWPOISON_LARGE))
883 do_sigbus(regs, error_code, address, fault);
890 static int spurious_fault_check(unsigned long error_code, pte_t *pte)
892 if ((error_code & PF_WRITE) && !pte_write(*pte))
895 if ((error_code & PF_INSTR) && !pte_exec(*pte))
902 * Handle a spurious fault caused by a stale TLB entry.
904 * This allows us to lazily refresh the TLB when increasing the
905 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
906 * eagerly is very expensive since that implies doing a full
907 * cross-processor TLB flush, even if no stale TLB entries exist
908 * on other processors.
910 * There are no security implications to leaving a stale TLB when
911 * increasing the permissions on a page.
913 static noinline __kprobes int
914 spurious_fault(unsigned long error_code, unsigned long address)
922 /* Reserved-bit violation or user access to kernel space? */
923 if (error_code & (PF_USER | PF_RSVD))
926 pgd = init_mm.pgd + pgd_index(address);
927 if (!pgd_present(*pgd))
930 pud = pud_offset(pgd, address);
931 if (!pud_present(*pud))
935 return spurious_fault_check(error_code, (pte_t *) pud);
937 pmd = pmd_offset(pud, address);
938 if (!pmd_present(*pmd))
942 return spurious_fault_check(error_code, (pte_t *) pmd);
944 pte = pte_offset_kernel(pmd, address);
945 if (!pte_present(*pte))
948 ret = spurious_fault_check(error_code, pte);
953 * Make sure we have permissions in PMD.
954 * If not, then there's a bug in the page tables:
956 ret = spurious_fault_check(error_code, (pte_t *) pmd);
957 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
962 int show_unhandled_signals = 1;
965 access_error(unsigned long error_code, struct vm_area_struct *vma)
967 if (error_code & PF_WRITE) {
968 /* write, present and write, not present: */
969 if (unlikely(!(vma->vm_flags & VM_WRITE)))
975 if (unlikely(error_code & PF_PROT))
978 /* read, not present: */
979 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
985 static int fault_in_kernel_space(unsigned long address)
987 return address >= TASK_SIZE_MAX;
990 static inline bool smap_violation(int error_code, struct pt_regs *regs)
992 if (error_code & PF_USER)
995 if (!user_mode_vm(regs) && (regs->flags & X86_EFLAGS_AC))
1002 * This routine handles page faults. It determines the address,
1003 * and the problem, and then passes it off to one of the appropriate
1006 static void __kprobes
1007 __do_page_fault(struct pt_regs *regs, unsigned long error_code)
1009 struct vm_area_struct *vma;
1010 struct task_struct *tsk;
1011 unsigned long address;
1012 struct mm_struct *mm;
1014 int write = error_code & PF_WRITE;
1015 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE |
1016 (write ? FAULT_FLAG_WRITE : 0);
1021 /* Get the faulting address: */
1022 address = read_cr2();
1025 * Detect and handle instructions that would cause a page fault for
1026 * both a tracked kernel page and a userspace page.
1028 if (kmemcheck_active(regs))
1029 kmemcheck_hide(regs);
1030 prefetchw(&mm->mmap_sem);
1032 if (unlikely(kmmio_fault(regs, address)))
1036 * We fault-in kernel-space virtual memory on-demand. The
1037 * 'reference' page table is init_mm.pgd.
1039 * NOTE! We MUST NOT take any locks for this case. We may
1040 * be in an interrupt or a critical region, and should
1041 * only copy the information from the master page table,
1044 * This verifies that the fault happens in kernel space
1045 * (error_code & 4) == 0, and that the fault was not a
1046 * protection error (error_code & 9) == 0.
1048 if (unlikely(fault_in_kernel_space(address))) {
1049 if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) {
1050 if (vmalloc_fault(address) >= 0)
1053 if (kmemcheck_fault(regs, address, error_code))
1057 /* Can handle a stale RO->RW TLB: */
1058 if (spurious_fault(error_code, address))
1061 /* kprobes don't want to hook the spurious faults: */
1062 if (notify_page_fault(regs))
1065 * Don't take the mm semaphore here. If we fixup a prefetch
1066 * fault we could otherwise deadlock:
1068 bad_area_nosemaphore(regs, error_code, address);
1073 /* kprobes don't want to hook the spurious faults: */
1074 if (unlikely(notify_page_fault(regs)))
1077 * It's safe to allow irq's after cr2 has been saved and the
1078 * vmalloc fault has been handled.
1080 * User-mode registers count as a user access even for any
1081 * potential system fault or CPU buglet:
1083 if (user_mode_vm(regs)) {
1085 error_code |= PF_USER;
1087 if (regs->flags & X86_EFLAGS_IF)
1091 if (unlikely(error_code & PF_RSVD))
1092 pgtable_bad(regs, error_code, address);
1094 if (static_cpu_has(X86_FEATURE_SMAP)) {
1095 if (unlikely(smap_violation(error_code, regs))) {
1096 bad_area_nosemaphore(regs, error_code, address);
1101 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
1104 * If we're in an interrupt, have no user context or are running
1105 * in an atomic region then we must not take the fault:
1107 if (unlikely(in_atomic() || !mm)) {
1108 bad_area_nosemaphore(regs, error_code, address);
1113 * When running in the kernel we expect faults to occur only to
1114 * addresses in user space. All other faults represent errors in
1115 * the kernel and should generate an OOPS. Unfortunately, in the
1116 * case of an erroneous fault occurring in a code path which already
1117 * holds mmap_sem we will deadlock attempting to validate the fault
1118 * against the address space. Luckily the kernel only validly
1119 * references user space from well defined areas of code, which are
1120 * listed in the exceptions table.
1122 * As the vast majority of faults will be valid we will only perform
1123 * the source reference check when there is a possibility of a
1124 * deadlock. Attempt to lock the address space, if we cannot we then
1125 * validate the source. If this is invalid we can skip the address
1126 * space check, thus avoiding the deadlock:
1128 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
1129 if ((error_code & PF_USER) == 0 &&
1130 !search_exception_tables(regs->ip)) {
1131 bad_area_nosemaphore(regs, error_code, address);
1135 down_read(&mm->mmap_sem);
1138 * The above down_read_trylock() might have succeeded in
1139 * which case we'll have missed the might_sleep() from
1145 vma = find_vma(mm, address);
1146 if (unlikely(!vma)) {
1147 bad_area(regs, error_code, address);
1150 if (likely(vma->vm_start <= address))
1152 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1153 bad_area(regs, error_code, address);
1156 if (error_code & PF_USER) {
1158 * Accessing the stack below %sp is always a bug.
1159 * The large cushion allows instructions like enter
1160 * and pusha to work. ("enter $65535, $31" pushes
1161 * 32 pointers and then decrements %sp by 65535.)
1163 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) {
1164 bad_area(regs, error_code, address);
1168 if (unlikely(expand_stack(vma, address))) {
1169 bad_area(regs, error_code, address);
1174 * Ok, we have a good vm_area for this memory access, so
1175 * we can handle it..
1178 if (unlikely(access_error(error_code, vma))) {
1179 bad_area_access_error(regs, error_code, address);
1184 * If for any reason at all we couldn't handle the fault,
1185 * make sure we exit gracefully rather than endlessly redo
1188 fault = handle_mm_fault(mm, vma, address, flags);
1190 if (unlikely(fault & (VM_FAULT_RETRY|VM_FAULT_ERROR))) {
1191 if (mm_fault_error(regs, error_code, address, fault))
1196 * Major/minor page fault accounting is only done on the
1197 * initial attempt. If we go through a retry, it is extremely
1198 * likely that the page will be found in page cache at that point.
1200 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1201 if (fault & VM_FAULT_MAJOR) {
1203 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
1207 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
1210 if (fault & VM_FAULT_RETRY) {
1211 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
1213 flags &= ~FAULT_FLAG_ALLOW_RETRY;
1214 flags |= FAULT_FLAG_TRIED;
1219 check_v8086_mode(regs, address, tsk);
1221 up_read(&mm->mmap_sem);
1224 dotraplinkage void __kprobes
1225 do_page_fault(struct pt_regs *regs, unsigned long error_code)
1227 enum ctx_state prev_state;
1229 prev_state = exception_enter();
1230 __do_page_fault(regs, error_code);
1231 exception_exit(prev_state);