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/kgdb.h>
19 #include <linux/kernel.h>
20 #include <linux/module.h>
21 #include <linux/ptrace.h>
22 #include <linux/string.h>
23 #include <linux/delay.h>
24 #include <linux/errno.h>
25 #include <linux/kexec.h>
26 #include <linux/sched.h>
27 #include <linux/timer.h>
28 #include <linux/init.h>
29 #include <linux/bug.h>
30 #include <linux/nmi.h>
32 #include <linux/smp.h>
36 #include <linux/ioport.h>
37 #include <linux/eisa.h>
41 #include <linux/mca.h>
44 #if defined(CONFIG_EDAC)
45 #include <linux/edac.h>
48 #include <asm/kmemcheck.h>
49 #include <asm/stacktrace.h>
50 #include <asm/processor.h>
51 #include <asm/debugreg.h>
52 #include <asm/atomic.h>
53 #include <asm/system.h>
54 #include <asm/traps.h>
59 #include <asm/mach_traps.h>
62 #include <asm/x86_init.h>
63 #include <asm/pgalloc.h>
64 #include <asm/proto.h>
66 #include <asm/processor-flags.h>
67 #include <asm/setup.h>
69 asmlinkage int system_call(void);
71 /* Do we ignore FPU interrupts ? */
75 * The IDT has to be page-aligned to simplify the Pentium
76 * F0 0F bug workaround.
78 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
81 DECLARE_BITMAP(used_vectors, NR_VECTORS);
82 EXPORT_SYMBOL_GPL(used_vectors);
84 static int ignore_nmis;
86 static inline void conditional_sti(struct pt_regs *regs)
88 if (regs->flags & X86_EFLAGS_IF)
92 static inline void preempt_conditional_sti(struct pt_regs *regs)
95 if (regs->flags & X86_EFLAGS_IF)
99 static inline void conditional_cli(struct pt_regs *regs)
101 if (regs->flags & X86_EFLAGS_IF)
105 static inline void preempt_conditional_cli(struct pt_regs *regs)
107 if (regs->flags & X86_EFLAGS_IF)
112 static void __kprobes
113 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
114 long error_code, siginfo_t *info)
116 struct task_struct *tsk = current;
119 if (regs->flags & X86_VM_MASK) {
121 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
122 * On nmi (interrupt 2), do_trap should not be called.
130 if (!user_mode(regs))
137 * We want error_code and trap_no set for userspace faults and
138 * kernelspace faults which result in die(), but not
139 * kernelspace faults which are fixed up. die() gives the
140 * process no chance to handle the signal and notice the
141 * kernel fault information, so that won't result in polluting
142 * the information about previously queued, but not yet
143 * delivered, faults. See also do_general_protection below.
145 tsk->thread.error_code = error_code;
146 tsk->thread.trap_no = trapnr;
149 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
150 printk_ratelimit()) {
152 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
153 tsk->comm, tsk->pid, str,
154 regs->ip, regs->sp, error_code);
155 print_vma_addr(" in ", regs->ip);
161 force_sig_info(signr, info, tsk);
163 force_sig(signr, tsk);
167 if (!fixup_exception(regs)) {
168 tsk->thread.error_code = error_code;
169 tsk->thread.trap_no = trapnr;
170 die(str, regs, error_code);
176 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
183 #define DO_ERROR(trapnr, signr, str, name) \
184 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
186 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
189 conditional_sti(regs); \
190 do_trap(trapnr, signr, str, regs, error_code, NULL); \
193 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
194 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
197 info.si_signo = signr; \
199 info.si_code = sicode; \
200 info.si_addr = (void __user *)siaddr; \
201 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
204 conditional_sti(regs); \
205 do_trap(trapnr, signr, str, regs, error_code, &info); \
208 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
209 DO_ERROR(4, SIGSEGV, "overflow", overflow)
210 DO_ERROR(5, SIGSEGV, "bounds", bounds)
211 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
212 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
213 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
214 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
216 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
218 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
221 /* Runs on IST stack */
222 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
224 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
225 12, SIGBUS) == NOTIFY_STOP)
227 preempt_conditional_sti(regs);
228 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
229 preempt_conditional_cli(regs);
232 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
234 static const char str[] = "double fault";
235 struct task_struct *tsk = current;
237 /* Return not checked because double check cannot be ignored */
238 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
240 tsk->thread.error_code = error_code;
241 tsk->thread.trap_no = 8;
244 * This is always a kernel trap and never fixable (and thus must
248 die(str, regs, error_code);
252 dotraplinkage void __kprobes
253 do_general_protection(struct pt_regs *regs, long error_code)
255 struct task_struct *tsk;
257 conditional_sti(regs);
260 if (regs->flags & X86_VM_MASK)
265 if (!user_mode(regs))
268 tsk->thread.error_code = error_code;
269 tsk->thread.trap_no = 13;
271 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
272 printk_ratelimit()) {
274 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
275 tsk->comm, task_pid_nr(tsk),
276 regs->ip, regs->sp, error_code);
277 print_vma_addr(" in ", regs->ip);
281 force_sig(SIGSEGV, tsk);
287 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
292 if (fixup_exception(regs))
295 tsk->thread.error_code = error_code;
296 tsk->thread.trap_no = 13;
297 if (notify_die(DIE_GPF, "general protection fault", regs,
298 error_code, 13, SIGSEGV) == NOTIFY_STOP)
300 die("general protection fault", regs, error_code);
303 static notrace __kprobes void
304 mem_parity_error(unsigned char reason, struct pt_regs *regs)
307 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
308 reason, smp_processor_id());
311 "You have some hardware problem, likely on the PCI bus.\n");
313 #if defined(CONFIG_EDAC)
314 if (edac_handler_set()) {
315 edac_atomic_assert_error();
320 if (panic_on_unrecovered_nmi)
321 panic("NMI: Not continuing");
323 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
325 /* Clear and disable the memory parity error line. */
326 reason = (reason & 0xf) | 4;
330 static notrace __kprobes void
331 io_check_error(unsigned char reason, struct pt_regs *regs)
335 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
336 show_registers(regs);
339 panic("NMI IOCK error: Not continuing");
341 /* Re-enable the IOCK line, wait for a few seconds */
342 reason = (reason & 0xf) | 8;
353 static notrace __kprobes void
354 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
356 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
361 * Might actually be able to figure out what the guilty party
370 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
371 reason, smp_processor_id());
373 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
374 if (panic_on_unrecovered_nmi)
375 panic("NMI: Not continuing");
377 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
380 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
382 unsigned char reason = 0;
385 cpu = smp_processor_id();
387 /* Only the BSP gets external NMIs from the system. */
389 reason = get_nmi_reason();
391 if (!(reason & 0xc0)) {
392 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
395 #ifdef CONFIG_X86_LOCAL_APIC
397 * Ok, so this is none of the documented NMI sources,
398 * so it must be the NMI watchdog.
400 if (nmi_watchdog_tick(regs, reason))
402 if (!do_nmi_callback(regs, cpu))
403 unknown_nmi_error(reason, regs);
405 unknown_nmi_error(reason, regs);
410 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
413 /* AK: following checks seem to be broken on modern chipsets. FIXME */
415 mem_parity_error(reason, regs);
417 io_check_error(reason, regs);
420 * Reassert NMI in case it became active meanwhile
421 * as it's edge-triggered:
427 dotraplinkage notrace __kprobes void
428 do_nmi(struct pt_regs *regs, long error_code)
432 inc_irq_stat(__nmi_count);
435 default_do_nmi(regs);
446 void restart_nmi(void)
452 /* May run on IST stack. */
453 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
455 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
456 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
459 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
460 #ifdef CONFIG_KPROBES
461 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
465 if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
470 preempt_conditional_sti(regs);
471 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
472 preempt_conditional_cli(regs);
477 * Help handler running on IST stack to switch back to user stack
478 * for scheduling or signal handling. The actual stack switch is done in
481 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
483 struct pt_regs *regs = eregs;
484 /* Did already sync */
485 if (eregs == (struct pt_regs *)eregs->sp)
487 /* Exception from user space */
488 else if (user_mode(eregs))
489 regs = task_pt_regs(current);
491 * Exception from kernel and interrupts are enabled. Move to
492 * kernel process stack.
494 else if (eregs->flags & X86_EFLAGS_IF)
495 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
503 * Our handling of the processor debug registers is non-trivial.
504 * We do not clear them on entry and exit from the kernel. Therefore
505 * it is possible to get a watchpoint trap here from inside the kernel.
506 * However, the code in ./ptrace.c has ensured that the user can
507 * only set watchpoints on userspace addresses. Therefore the in-kernel
508 * watchpoint trap can only occur in code which is reading/writing
509 * from user space. Such code must not hold kernel locks (since it
510 * can equally take a page fault), therefore it is safe to call
511 * force_sig_info even though that claims and releases locks.
513 * Code in ./signal.c ensures that the debug control register
514 * is restored before we deliver any signal, and therefore that
515 * user code runs with the correct debug control register even though
518 * Being careful here means that we don't have to be as careful in a
519 * lot of more complicated places (task switching can be a bit lazy
520 * about restoring all the debug state, and ptrace doesn't have to
521 * find every occurrence of the TF bit that could be saved away even
524 * May run on IST stack.
526 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
528 struct task_struct *tsk = current;
532 get_debugreg(dr6, 6);
534 /* Filter out all the reserved bits which are preset to 1 */
535 dr6 &= ~DR6_RESERVED;
537 /* Catch kmemcheck conditions first of all! */
538 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
541 /* DR6 may or may not be cleared by the CPU */
545 * The processor cleared BTF, so don't mark that we need it set.
547 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
549 /* Store the virtualized DR6 value */
550 tsk->thread.debugreg6 = dr6;
552 if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
553 SIGTRAP) == NOTIFY_STOP)
556 /* It's safe to allow irq's after DR6 has been saved */
557 preempt_conditional_sti(regs);
559 if (regs->flags & X86_VM_MASK) {
560 handle_vm86_trap((struct kernel_vm86_regs *) regs,
566 * Single-stepping through system calls: ignore any exceptions in
567 * kernel space, but re-enable TF when returning to user mode.
569 * We already checked v86 mode above, so we can check for kernel mode
570 * by just checking the CPL of CS.
572 if ((dr6 & DR_STEP) && !user_mode(regs)) {
573 tsk->thread.debugreg6 &= ~DR_STEP;
574 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
575 regs->flags &= ~X86_EFLAGS_TF;
577 si_code = get_si_code(tsk->thread.debugreg6);
578 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS))
579 send_sigtrap(tsk, regs, error_code, si_code);
580 preempt_conditional_cli(regs);
586 * Note that we play around with the 'TS' bit in an attempt to get
587 * the correct behaviour even in the presence of the asynchronous
590 void math_error(struct pt_regs *regs, int error_code, int trapnr)
592 struct task_struct *task = current;
595 char *str = (trapnr == 16) ? "fpu exception" : "simd exception";
597 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
599 conditional_sti(regs);
601 if (!user_mode_vm(regs))
603 if (!fixup_exception(regs)) {
604 task->thread.error_code = error_code;
605 task->thread.trap_no = trapnr;
606 die(str, regs, error_code);
612 * Save the info for the exception handler and clear the error.
615 task->thread.trap_no = trapnr;
616 task->thread.error_code = error_code;
617 info.si_signo = SIGFPE;
619 info.si_addr = (void __user *)regs->ip;
621 unsigned short cwd, swd;
623 * (~cwd & swd) will mask out exceptions that are not set to unmasked
624 * status. 0x3f is the exception bits in these regs, 0x200 is the
625 * C1 reg you need in case of a stack fault, 0x040 is the stack
626 * fault bit. We should only be taking one exception at a time,
627 * so if this combination doesn't produce any single exception,
628 * then we have a bad program that isn't synchronizing its FPU usage
629 * and it will suffer the consequences since we won't be able to
630 * fully reproduce the context of the exception
632 cwd = get_fpu_cwd(task);
633 swd = get_fpu_swd(task);
638 * The SIMD FPU exceptions are handled a little differently, as there
639 * is only a single status/control register. Thus, to determine which
640 * unmasked exception was caught we must mask the exception mask bits
641 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
643 unsigned short mxcsr = get_fpu_mxcsr(task);
644 err = ~(mxcsr >> 7) & mxcsr;
647 if (err & 0x001) { /* Invalid op */
649 * swd & 0x240 == 0x040: Stack Underflow
650 * swd & 0x240 == 0x240: Stack Overflow
651 * User must clear the SF bit (0x40) if set
653 info.si_code = FPE_FLTINV;
654 } else if (err & 0x004) { /* Divide by Zero */
655 info.si_code = FPE_FLTDIV;
656 } else if (err & 0x008) { /* Overflow */
657 info.si_code = FPE_FLTOVF;
658 } else if (err & 0x012) { /* Denormal, Underflow */
659 info.si_code = FPE_FLTUND;
660 } else if (err & 0x020) { /* Precision */
661 info.si_code = FPE_FLTRES;
664 * If we're using IRQ 13, or supposedly even some trap 16
665 * implementations, it's possible we get a spurious trap...
667 return; /* Spurious trap, no error */
669 force_sig_info(SIGFPE, &info, task);
672 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
678 math_error(regs, error_code, 16);
682 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
684 math_error(regs, error_code, 19);
688 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
690 conditional_sti(regs);
692 /* No need to warn about this any longer. */
693 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
697 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
701 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
706 * __math_state_restore assumes that cr0.TS is already clear and the
707 * fpu state is all ready for use. Used during context switch.
709 void __math_state_restore(void)
711 struct thread_info *thread = current_thread_info();
712 struct task_struct *tsk = thread->task;
715 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
717 if (unlikely(restore_fpu_checking(tsk))) {
719 force_sig(SIGSEGV, tsk);
723 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
728 * 'math_state_restore()' saves the current math information in the
729 * old math state array, and gets the new ones from the current task
731 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
732 * Don't touch unless you *really* know how it works.
734 * Must be called with kernel preemption disabled (in this case,
735 * local interrupts are disabled at the call-site in entry.S).
737 asmlinkage void math_state_restore(void)
739 struct thread_info *thread = current_thread_info();
740 struct task_struct *tsk = thread->task;
742 if (!tsk_used_math(tsk)) {
745 * does a slab alloc which can sleep
751 do_group_exit(SIGKILL);
757 clts(); /* Allow maths ops (or we recurse) */
759 __math_state_restore();
761 EXPORT_SYMBOL_GPL(math_state_restore);
763 #ifndef CONFIG_MATH_EMULATION
764 void math_emulate(struct math_emu_info *info)
767 "math-emulation not enabled and no coprocessor found.\n");
768 printk(KERN_EMERG "killing %s.\n", current->comm);
769 force_sig(SIGFPE, current);
772 #endif /* CONFIG_MATH_EMULATION */
774 dotraplinkage void __kprobes
775 do_device_not_available(struct pt_regs *regs, long error_code)
778 if (read_cr0() & X86_CR0_EM) {
779 struct math_emu_info info = { };
781 conditional_sti(regs);
786 math_state_restore(); /* interrupts still off */
787 conditional_sti(regs);
790 math_state_restore();
795 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
800 info.si_signo = SIGILL;
802 info.si_code = ILL_BADSTK;
804 if (notify_die(DIE_TRAP, "iret exception",
805 regs, error_code, 32, SIGILL) == NOTIFY_STOP)
807 do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
811 void __init trap_init(void)
816 void __iomem *p = early_ioremap(0x0FFFD9, 4);
818 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
823 set_intr_gate(0, ÷_error);
824 set_intr_gate_ist(1, &debug, DEBUG_STACK);
825 set_intr_gate_ist(2, &nmi, NMI_STACK);
826 /* int3 can be called from all */
827 set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
828 /* int4 can be called from all */
829 set_system_intr_gate(4, &overflow);
830 set_intr_gate(5, &bounds);
831 set_intr_gate(6, &invalid_op);
832 set_intr_gate(7, &device_not_available);
834 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
836 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
838 set_intr_gate(9, &coprocessor_segment_overrun);
839 set_intr_gate(10, &invalid_TSS);
840 set_intr_gate(11, &segment_not_present);
841 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
842 set_intr_gate(13, &general_protection);
843 set_intr_gate(14, &page_fault);
844 set_intr_gate(15, &spurious_interrupt_bug);
845 set_intr_gate(16, &coprocessor_error);
846 set_intr_gate(17, &alignment_check);
847 #ifdef CONFIG_X86_MCE
848 set_intr_gate_ist(18, &machine_check, MCE_STACK);
850 set_intr_gate(19, &simd_coprocessor_error);
852 /* Reserve all the builtin and the syscall vector: */
853 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
854 set_bit(i, used_vectors);
856 #ifdef CONFIG_IA32_EMULATION
857 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
858 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
863 printk(KERN_INFO "Enabling fast FPU save and restore... ");
864 set_in_cr4(X86_CR4_OSFXSR);
869 "Enabling unmasked SIMD FPU exception support... ");
870 set_in_cr4(X86_CR4_OSXMMEXCPT);
874 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
875 set_bit(SYSCALL_VECTOR, used_vectors);
879 * Should be a barrier for any external CPU state:
883 x86_init.irqs.trap_init();