2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
10 * Handle hardware traps and faults.
12 #include <linux/interrupt.h>
13 #include <linux/kallsyms.h>
14 #include <linux/spinlock.h>
15 #include <linux/kprobes.h>
16 #include <linux/uaccess.h>
17 #include <linux/kdebug.h>
18 #include <linux/kernel.h>
19 #include <linux/module.h>
20 #include <linux/ptrace.h>
21 #include <linux/string.h>
22 #include <linux/delay.h>
23 #include <linux/errno.h>
24 #include <linux/kexec.h>
25 #include <linux/sched.h>
26 #include <linux/timer.h>
27 #include <linux/init.h>
28 #include <linux/bug.h>
29 #include <linux/nmi.h>
31 #include <linux/smp.h>
35 #include <linux/ioport.h>
36 #include <linux/eisa.h>
40 #include <linux/mca.h>
43 #if defined(CONFIG_EDAC)
44 #include <linux/edac.h>
47 #include <asm/kmemcheck.h>
48 #include <asm/stacktrace.h>
49 #include <asm/processor.h>
50 #include <asm/debugreg.h>
51 #include <asm/atomic.h>
52 #include <asm/system.h>
53 #include <asm/traps.h>
58 #include <asm/mach_traps.h>
61 #include <asm/x86_init.h>
62 #include <asm/pgalloc.h>
63 #include <asm/proto.h>
65 #include <asm/processor-flags.h>
66 #include <asm/setup.h>
68 asmlinkage int system_call(void);
70 /* Do we ignore FPU interrupts ? */
74 * The IDT has to be page-aligned to simplify the Pentium
75 * F0 0F bug workaround.
77 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
80 DECLARE_BITMAP(used_vectors, NR_VECTORS);
81 EXPORT_SYMBOL_GPL(used_vectors);
83 static int ignore_nmis;
85 static inline void conditional_sti(struct pt_regs *regs)
87 if (regs->flags & X86_EFLAGS_IF)
91 static inline void preempt_conditional_sti(struct pt_regs *regs)
94 if (regs->flags & X86_EFLAGS_IF)
98 static inline void conditional_cli(struct pt_regs *regs)
100 if (regs->flags & X86_EFLAGS_IF)
104 static inline void preempt_conditional_cli(struct pt_regs *regs)
106 if (regs->flags & X86_EFLAGS_IF)
113 die_if_kernel(const char *str, struct pt_regs *regs, long err)
115 if (!user_mode_vm(regs))
120 static void __kprobes
121 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
122 long error_code, siginfo_t *info)
124 struct task_struct *tsk = current;
127 if (regs->flags & X86_VM_MASK) {
129 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
130 * On nmi (interrupt 2), do_trap should not be called.
138 if (!user_mode(regs))
145 * We want error_code and trap_no set for userspace faults and
146 * kernelspace faults which result in die(), but not
147 * kernelspace faults which are fixed up. die() gives the
148 * process no chance to handle the signal and notice the
149 * kernel fault information, so that won't result in polluting
150 * the information about previously queued, but not yet
151 * delivered, faults. See also do_general_protection below.
153 tsk->thread.error_code = error_code;
154 tsk->thread.trap_no = trapnr;
157 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
158 printk_ratelimit()) {
160 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
161 tsk->comm, tsk->pid, str,
162 regs->ip, regs->sp, error_code);
163 print_vma_addr(" in ", regs->ip);
169 force_sig_info(signr, info, tsk);
171 force_sig(signr, tsk);
175 if (!fixup_exception(regs)) {
176 tsk->thread.error_code = error_code;
177 tsk->thread.trap_no = trapnr;
178 die(str, regs, error_code);
184 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
191 #define DO_ERROR(trapnr, signr, str, name) \
192 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
194 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
197 conditional_sti(regs); \
198 do_trap(trapnr, signr, str, regs, error_code, NULL); \
201 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
202 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
205 info.si_signo = signr; \
207 info.si_code = sicode; \
208 info.si_addr = (void __user *)siaddr; \
209 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
212 conditional_sti(regs); \
213 do_trap(trapnr, signr, str, regs, error_code, &info); \
216 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
217 DO_ERROR(4, SIGSEGV, "overflow", overflow)
218 DO_ERROR(5, SIGSEGV, "bounds", bounds)
219 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
220 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
221 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
222 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
224 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
226 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
229 /* Runs on IST stack */
230 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
232 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
233 12, SIGBUS) == NOTIFY_STOP)
235 preempt_conditional_sti(regs);
236 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
237 preempt_conditional_cli(regs);
240 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
242 static const char str[] = "double fault";
243 struct task_struct *tsk = current;
245 /* Return not checked because double check cannot be ignored */
246 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
248 tsk->thread.error_code = error_code;
249 tsk->thread.trap_no = 8;
252 * This is always a kernel trap and never fixable (and thus must
256 die(str, regs, error_code);
260 dotraplinkage void __kprobes
261 do_general_protection(struct pt_regs *regs, long error_code)
263 struct task_struct *tsk;
265 conditional_sti(regs);
268 if (regs->flags & X86_VM_MASK)
273 if (!user_mode(regs))
276 tsk->thread.error_code = error_code;
277 tsk->thread.trap_no = 13;
279 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
280 printk_ratelimit()) {
282 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
283 tsk->comm, task_pid_nr(tsk),
284 regs->ip, regs->sp, error_code);
285 print_vma_addr(" in ", regs->ip);
289 force_sig(SIGSEGV, tsk);
295 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
300 if (fixup_exception(regs))
303 tsk->thread.error_code = error_code;
304 tsk->thread.trap_no = 13;
305 if (notify_die(DIE_GPF, "general protection fault", regs,
306 error_code, 13, SIGSEGV) == NOTIFY_STOP)
308 die("general protection fault", regs, error_code);
311 static notrace __kprobes void
312 mem_parity_error(unsigned char reason, struct pt_regs *regs)
315 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
316 reason, smp_processor_id());
319 "You have some hardware problem, likely on the PCI bus.\n");
321 #if defined(CONFIG_EDAC)
322 if (edac_handler_set()) {
323 edac_atomic_assert_error();
328 if (panic_on_unrecovered_nmi)
329 panic("NMI: Not continuing");
331 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
333 /* Clear and disable the memory parity error line. */
334 reason = (reason & 0xf) | 4;
338 static notrace __kprobes void
339 io_check_error(unsigned char reason, struct pt_regs *regs)
343 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
344 show_registers(regs);
347 panic("NMI IOCK error: Not continuing");
349 /* Re-enable the IOCK line, wait for a few seconds */
350 reason = (reason & 0xf) | 8;
361 static notrace __kprobes void
362 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
364 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
369 * Might actually be able to figure out what the guilty party
378 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
379 reason, smp_processor_id());
381 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
382 if (panic_on_unrecovered_nmi)
383 panic("NMI: Not continuing");
385 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
388 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
390 unsigned char reason = 0;
393 cpu = smp_processor_id();
395 /* Only the BSP gets external NMIs from the system. */
397 reason = get_nmi_reason();
399 if (!(reason & 0xc0)) {
400 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
403 #ifdef CONFIG_X86_LOCAL_APIC
405 * Ok, so this is none of the documented NMI sources,
406 * so it must be the NMI watchdog.
408 if (nmi_watchdog_tick(regs, reason))
410 if (!do_nmi_callback(regs, cpu))
411 unknown_nmi_error(reason, regs);
413 unknown_nmi_error(reason, regs);
418 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
421 /* AK: following checks seem to be broken on modern chipsets. FIXME */
423 mem_parity_error(reason, regs);
425 io_check_error(reason, regs);
428 * Reassert NMI in case it became active meanwhile
429 * as it's edge-triggered:
435 dotraplinkage notrace __kprobes void
436 do_nmi(struct pt_regs *regs, long error_code)
440 inc_irq_stat(__nmi_count);
443 default_do_nmi(regs);
454 void restart_nmi(void)
460 /* May run on IST stack. */
461 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
463 #ifdef CONFIG_KPROBES
464 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
468 if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
473 preempt_conditional_sti(regs);
474 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
475 preempt_conditional_cli(regs);
480 * Help handler running on IST stack to switch back to user stack
481 * for scheduling or signal handling. The actual stack switch is done in
484 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
486 struct pt_regs *regs = eregs;
487 /* Did already sync */
488 if (eregs == (struct pt_regs *)eregs->sp)
490 /* Exception from user space */
491 else if (user_mode(eregs))
492 regs = task_pt_regs(current);
494 * Exception from kernel and interrupts are enabled. Move to
495 * kernel process stack.
497 else if (eregs->flags & X86_EFLAGS_IF)
498 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
506 * Our handling of the processor debug registers is non-trivial.
507 * We do not clear them on entry and exit from the kernel. Therefore
508 * it is possible to get a watchpoint trap here from inside the kernel.
509 * However, the code in ./ptrace.c has ensured that the user can
510 * only set watchpoints on userspace addresses. Therefore the in-kernel
511 * watchpoint trap can only occur in code which is reading/writing
512 * from user space. Such code must not hold kernel locks (since it
513 * can equally take a page fault), therefore it is safe to call
514 * force_sig_info even though that claims and releases locks.
516 * Code in ./signal.c ensures that the debug control register
517 * is restored before we deliver any signal, and therefore that
518 * user code runs with the correct debug control register even though
521 * Being careful here means that we don't have to be as careful in a
522 * lot of more complicated places (task switching can be a bit lazy
523 * about restoring all the debug state, and ptrace doesn't have to
524 * find every occurrence of the TF bit that could be saved away even
527 * May run on IST stack.
529 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
531 struct task_struct *tsk = current;
535 get_debugreg(dr6, 6);
537 /* Filter out all the reserved bits which are preset to 1 */
538 dr6 &= ~DR6_RESERVED;
540 /* Catch kmemcheck conditions first of all! */
541 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
544 /* DR6 may or may not be cleared by the CPU */
547 /* Store the virtualized DR6 value */
548 tsk->thread.debugreg6 = dr6;
550 if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
551 SIGTRAP) == NOTIFY_STOP)
554 /* It's safe to allow irq's after DR6 has been saved */
555 preempt_conditional_sti(regs);
557 if (regs->flags & X86_VM_MASK) {
558 handle_vm86_trap((struct kernel_vm86_regs *) regs,
564 * Single-stepping through system calls: ignore any exceptions in
565 * kernel space, but re-enable TF when returning to user mode.
567 * We already checked v86 mode above, so we can check for kernel mode
568 * by just checking the CPL of CS.
570 if ((dr6 & DR_STEP) && !user_mode(regs)) {
571 tsk->thread.debugreg6 &= ~DR_STEP;
572 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
573 regs->flags &= ~X86_EFLAGS_TF;
575 si_code = get_si_code(tsk->thread.debugreg6);
576 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS))
577 send_sigtrap(tsk, regs, error_code, si_code);
578 preempt_conditional_cli(regs);
584 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
586 if (fixup_exception(regs))
589 notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
590 /* Illegal floating point operation in the kernel */
591 current->thread.trap_no = trapnr;
598 * Note that we play around with the 'TS' bit in an attempt to get
599 * the correct behaviour even in the presence of the asynchronous
602 void math_error(void __user *ip)
604 struct task_struct *task;
606 unsigned short cwd, swd, err;
609 * Save the info for the exception handler and clear the error.
613 task->thread.trap_no = 16;
614 task->thread.error_code = 0;
615 info.si_signo = SIGFPE;
619 * (~cwd & swd) will mask out exceptions that are not set to unmasked
620 * status. 0x3f is the exception bits in these regs, 0x200 is the
621 * C1 reg you need in case of a stack fault, 0x040 is the stack
622 * fault bit. We should only be taking one exception at a time,
623 * so if this combination doesn't produce any single exception,
624 * then we have a bad program that isn't synchronizing its FPU usage
625 * and it will suffer the consequences since we won't be able to
626 * fully reproduce the context of the exception
628 cwd = get_fpu_cwd(task);
629 swd = get_fpu_swd(task);
633 if (err & 0x001) { /* Invalid op */
635 * swd & 0x240 == 0x040: Stack Underflow
636 * swd & 0x240 == 0x240: Stack Overflow
637 * User must clear the SF bit (0x40) if set
639 info.si_code = FPE_FLTINV;
640 } else if (err & 0x004) { /* Divide by Zero */
641 info.si_code = FPE_FLTDIV;
642 } else if (err & 0x008) { /* Overflow */
643 info.si_code = FPE_FLTOVF;
644 } else if (err & 0x012) { /* Denormal, Underflow */
645 info.si_code = FPE_FLTUND;
646 } else if (err & 0x020) { /* Precision */
647 info.si_code = FPE_FLTRES;
650 * If we're using IRQ 13, or supposedly even some trap 16
651 * implementations, it's possible we get a spurious trap...
653 return; /* Spurious trap, no error */
655 force_sig_info(SIGFPE, &info, task);
658 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
660 conditional_sti(regs);
665 if (!user_mode(regs) &&
666 kernel_math_error(regs, "kernel x87 math error", 16))
670 math_error((void __user *)regs->ip);
673 static void simd_math_error(void __user *ip)
675 struct task_struct *task;
677 unsigned short mxcsr;
680 * Save the info for the exception handler and clear the error.
684 task->thread.trap_no = 19;
685 task->thread.error_code = 0;
686 info.si_signo = SIGFPE;
688 info.si_code = __SI_FAULT;
691 * The SIMD FPU exceptions are handled a little differently, as there
692 * is only a single status/control register. Thus, to determine which
693 * unmasked exception was caught we must mask the exception mask bits
694 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
696 mxcsr = get_fpu_mxcsr(task);
697 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
701 case 0x001: /* Invalid Op */
702 info.si_code = FPE_FLTINV;
704 case 0x002: /* Denormalize */
705 case 0x010: /* Underflow */
706 info.si_code = FPE_FLTUND;
708 case 0x004: /* Zero Divide */
709 info.si_code = FPE_FLTDIV;
711 case 0x008: /* Overflow */
712 info.si_code = FPE_FLTOVF;
714 case 0x020: /* Precision */
715 info.si_code = FPE_FLTRES;
718 force_sig_info(SIGFPE, &info, task);
722 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
724 conditional_sti(regs);
728 /* Handle SIMD FPU exceptions on PIII+ processors. */
730 simd_math_error((void __user *)regs->ip);
734 * Handle strange cache flush from user space exception
735 * in all other cases. This is undocumented behaviour.
737 if (regs->flags & X86_VM_MASK) {
738 handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
741 current->thread.trap_no = 19;
742 current->thread.error_code = error_code;
743 die_if_kernel("cache flush denied", regs, error_code);
744 force_sig(SIGSEGV, current);
746 if (!user_mode(regs) &&
747 kernel_math_error(regs, "kernel simd math error", 19))
749 simd_math_error((void __user *)regs->ip);
754 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
756 conditional_sti(regs);
758 /* No need to warn about this any longer. */
759 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
763 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
767 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
772 * __math_state_restore assumes that cr0.TS is already clear and the
773 * fpu state is all ready for use. Used during context switch.
775 void __math_state_restore(void)
777 struct thread_info *thread = current_thread_info();
778 struct task_struct *tsk = thread->task;
781 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
783 if (unlikely(restore_fpu_checking(tsk))) {
785 force_sig(SIGSEGV, tsk);
789 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
794 * 'math_state_restore()' saves the current math information in the
795 * old math state array, and gets the new ones from the current task
797 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
798 * Don't touch unless you *really* know how it works.
800 * Must be called with kernel preemption disabled (in this case,
801 * local interrupts are disabled at the call-site in entry.S).
803 asmlinkage void math_state_restore(void)
805 struct thread_info *thread = current_thread_info();
806 struct task_struct *tsk = thread->task;
808 if (!tsk_used_math(tsk)) {
811 * does a slab alloc which can sleep
817 do_group_exit(SIGKILL);
823 clts(); /* Allow maths ops (or we recurse) */
825 __math_state_restore();
827 EXPORT_SYMBOL_GPL(math_state_restore);
829 #ifndef CONFIG_MATH_EMULATION
830 void math_emulate(struct math_emu_info *info)
833 "math-emulation not enabled and no coprocessor found.\n");
834 printk(KERN_EMERG "killing %s.\n", current->comm);
835 force_sig(SIGFPE, current);
838 #endif /* CONFIG_MATH_EMULATION */
840 dotraplinkage void __kprobes
841 do_device_not_available(struct pt_regs *regs, long error_code)
844 if (read_cr0() & X86_CR0_EM) {
845 struct math_emu_info info = { };
847 conditional_sti(regs);
852 math_state_restore(); /* interrupts still off */
853 conditional_sti(regs);
856 math_state_restore();
861 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
866 info.si_signo = SIGILL;
868 info.si_code = ILL_BADSTK;
870 if (notify_die(DIE_TRAP, "iret exception",
871 regs, error_code, 32, SIGILL) == NOTIFY_STOP)
873 do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
877 void __init trap_init(void)
882 void __iomem *p = early_ioremap(0x0FFFD9, 4);
884 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
889 set_intr_gate(0, ÷_error);
890 set_intr_gate_ist(1, &debug, DEBUG_STACK);
891 set_intr_gate_ist(2, &nmi, NMI_STACK);
892 /* int3 can be called from all */
893 set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
894 /* int4 can be called from all */
895 set_system_intr_gate(4, &overflow);
896 set_intr_gate(5, &bounds);
897 set_intr_gate(6, &invalid_op);
898 set_intr_gate(7, &device_not_available);
900 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
902 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
904 set_intr_gate(9, &coprocessor_segment_overrun);
905 set_intr_gate(10, &invalid_TSS);
906 set_intr_gate(11, &segment_not_present);
907 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
908 set_intr_gate(13, &general_protection);
909 set_intr_gate(14, &page_fault);
910 set_intr_gate(15, &spurious_interrupt_bug);
911 set_intr_gate(16, &coprocessor_error);
912 set_intr_gate(17, &alignment_check);
913 #ifdef CONFIG_X86_MCE
914 set_intr_gate_ist(18, &machine_check, MCE_STACK);
916 set_intr_gate(19, &simd_coprocessor_error);
918 /* Reserve all the builtin and the syscall vector: */
919 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
920 set_bit(i, used_vectors);
922 #ifdef CONFIG_IA32_EMULATION
923 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
924 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
929 printk(KERN_INFO "Enabling fast FPU save and restore... ");
930 set_in_cr4(X86_CR4_OSFXSR);
935 "Enabling unmasked SIMD FPU exception support... ");
936 set_in_cr4(X86_CR4_OSXMMEXCPT);
940 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
941 set_bit(SYSCALL_VECTOR, used_vectors);
945 * Should be a barrier for any external CPU state:
949 x86_init.irqs.trap_init();