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)
111 static void __kprobes
112 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
113 long error_code, siginfo_t *info)
115 struct task_struct *tsk = current;
118 if (regs->flags & X86_VM_MASK) {
120 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
121 * On nmi (interrupt 2), do_trap should not be called.
129 if (!user_mode(regs))
136 * We want error_code and trap_no set for userspace faults and
137 * kernelspace faults which result in die(), but not
138 * kernelspace faults which are fixed up. die() gives the
139 * process no chance to handle the signal and notice the
140 * kernel fault information, so that won't result in polluting
141 * the information about previously queued, but not yet
142 * delivered, faults. See also do_general_protection below.
144 tsk->thread.error_code = error_code;
145 tsk->thread.trap_no = trapnr;
148 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
149 printk_ratelimit()) {
151 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
152 tsk->comm, tsk->pid, str,
153 regs->ip, regs->sp, error_code);
154 print_vma_addr(" in ", regs->ip);
160 force_sig_info(signr, info, tsk);
162 force_sig(signr, tsk);
166 if (!fixup_exception(regs)) {
167 tsk->thread.error_code = error_code;
168 tsk->thread.trap_no = trapnr;
169 die(str, regs, error_code);
175 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
182 #define DO_ERROR(trapnr, signr, str, name) \
183 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
185 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
188 conditional_sti(regs); \
189 do_trap(trapnr, signr, str, regs, error_code, NULL); \
192 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
193 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
196 info.si_signo = signr; \
198 info.si_code = sicode; \
199 info.si_addr = (void __user *)siaddr; \
200 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
203 conditional_sti(regs); \
204 do_trap(trapnr, signr, str, regs, error_code, &info); \
207 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
208 DO_ERROR(4, SIGSEGV, "overflow", overflow)
209 DO_ERROR(5, SIGSEGV, "bounds", bounds)
210 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
211 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
212 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
213 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
215 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
217 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
220 /* Runs on IST stack */
221 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
223 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
224 12, SIGBUS) == NOTIFY_STOP)
226 preempt_conditional_sti(regs);
227 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
228 preempt_conditional_cli(regs);
231 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
233 static const char str[] = "double fault";
234 struct task_struct *tsk = current;
236 /* Return not checked because double check cannot be ignored */
237 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
239 tsk->thread.error_code = error_code;
240 tsk->thread.trap_no = 8;
243 * This is always a kernel trap and never fixable (and thus must
247 die(str, regs, error_code);
251 dotraplinkage void __kprobes
252 do_general_protection(struct pt_regs *regs, long error_code)
254 struct task_struct *tsk;
256 conditional_sti(regs);
259 if (regs->flags & X86_VM_MASK)
264 if (!user_mode(regs))
267 tsk->thread.error_code = error_code;
268 tsk->thread.trap_no = 13;
270 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
271 printk_ratelimit()) {
273 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
274 tsk->comm, task_pid_nr(tsk),
275 regs->ip, regs->sp, error_code);
276 print_vma_addr(" in ", regs->ip);
280 force_sig(SIGSEGV, tsk);
286 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
291 if (fixup_exception(regs))
294 tsk->thread.error_code = error_code;
295 tsk->thread.trap_no = 13;
296 if (notify_die(DIE_GPF, "general protection fault", regs,
297 error_code, 13, SIGSEGV) == NOTIFY_STOP)
299 die("general protection fault", regs, error_code);
302 static notrace __kprobes void
303 mem_parity_error(unsigned char reason, struct pt_regs *regs)
306 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
307 reason, smp_processor_id());
310 "You have some hardware problem, likely on the PCI bus.\n");
312 #if defined(CONFIG_EDAC)
313 if (edac_handler_set()) {
314 edac_atomic_assert_error();
319 if (panic_on_unrecovered_nmi)
320 panic("NMI: Not continuing");
322 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
324 /* Clear and disable the memory parity error line. */
325 reason = (reason & 0xf) | 4;
329 static notrace __kprobes void
330 io_check_error(unsigned char reason, struct pt_regs *regs)
334 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
335 show_registers(regs);
338 panic("NMI IOCK error: Not continuing");
340 /* Re-enable the IOCK line, wait for a few seconds */
341 reason = (reason & 0xf) | 8;
352 static notrace __kprobes void
353 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
355 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
360 * Might actually be able to figure out what the guilty party
369 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
370 reason, smp_processor_id());
372 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
373 if (panic_on_unrecovered_nmi)
374 panic("NMI: Not continuing");
376 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
379 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
381 unsigned char reason = 0;
384 cpu = smp_processor_id();
386 /* Only the BSP gets external NMIs from the system. */
388 reason = get_nmi_reason();
390 if (!(reason & 0xc0)) {
391 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
394 #ifdef CONFIG_X86_LOCAL_APIC
396 * Ok, so this is none of the documented NMI sources,
397 * so it must be the NMI watchdog.
399 if (nmi_watchdog_tick(regs, reason))
401 if (!do_nmi_callback(regs, cpu))
402 unknown_nmi_error(reason, regs);
404 unknown_nmi_error(reason, regs);
409 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
412 /* AK: following checks seem to be broken on modern chipsets. FIXME */
414 mem_parity_error(reason, regs);
416 io_check_error(reason, regs);
419 * Reassert NMI in case it became active meanwhile
420 * as it's edge-triggered:
426 dotraplinkage notrace __kprobes void
427 do_nmi(struct pt_regs *regs, long error_code)
431 inc_irq_stat(__nmi_count);
434 default_do_nmi(regs);
445 void restart_nmi(void)
451 /* May run on IST stack. */
452 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
454 #ifdef CONFIG_KPROBES
455 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
459 if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
464 preempt_conditional_sti(regs);
465 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
466 preempt_conditional_cli(regs);
471 * Help handler running on IST stack to switch back to user stack
472 * for scheduling or signal handling. The actual stack switch is done in
475 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
477 struct pt_regs *regs = eregs;
478 /* Did already sync */
479 if (eregs == (struct pt_regs *)eregs->sp)
481 /* Exception from user space */
482 else if (user_mode(eregs))
483 regs = task_pt_regs(current);
485 * Exception from kernel and interrupts are enabled. Move to
486 * kernel process stack.
488 else if (eregs->flags & X86_EFLAGS_IF)
489 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
497 * Our handling of the processor debug registers is non-trivial.
498 * We do not clear them on entry and exit from the kernel. Therefore
499 * it is possible to get a watchpoint trap here from inside the kernel.
500 * However, the code in ./ptrace.c has ensured that the user can
501 * only set watchpoints on userspace addresses. Therefore the in-kernel
502 * watchpoint trap can only occur in code which is reading/writing
503 * from user space. Such code must not hold kernel locks (since it
504 * can equally take a page fault), therefore it is safe to call
505 * force_sig_info even though that claims and releases locks.
507 * Code in ./signal.c ensures that the debug control register
508 * is restored before we deliver any signal, and therefore that
509 * user code runs with the correct debug control register even though
512 * Being careful here means that we don't have to be as careful in a
513 * lot of more complicated places (task switching can be a bit lazy
514 * about restoring all the debug state, and ptrace doesn't have to
515 * find every occurrence of the TF bit that could be saved away even
518 * May run on IST stack.
520 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
522 struct task_struct *tsk = current;
526 get_debugreg(dr6, 6);
528 /* Filter out all the reserved bits which are preset to 1 */
529 dr6 &= ~DR6_RESERVED;
531 /* Catch kmemcheck conditions first of all! */
532 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
535 /* DR6 may or may not be cleared by the CPU */
539 * The processor cleared BTF, so don't mark that we need it set.
541 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
543 /* Store the virtualized DR6 value */
544 tsk->thread.debugreg6 = dr6;
546 if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
547 SIGTRAP) == NOTIFY_STOP)
550 /* It's safe to allow irq's after DR6 has been saved */
551 preempt_conditional_sti(regs);
553 if (regs->flags & X86_VM_MASK) {
554 handle_vm86_trap((struct kernel_vm86_regs *) regs,
560 * Single-stepping through system calls: ignore any exceptions in
561 * kernel space, but re-enable TF when returning to user mode.
563 * We already checked v86 mode above, so we can check for kernel mode
564 * by just checking the CPL of CS.
566 if ((dr6 & DR_STEP) && !user_mode(regs)) {
567 tsk->thread.debugreg6 &= ~DR_STEP;
568 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
569 regs->flags &= ~X86_EFLAGS_TF;
571 si_code = get_si_code(tsk->thread.debugreg6);
572 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS))
573 send_sigtrap(tsk, regs, error_code, si_code);
574 preempt_conditional_cli(regs);
580 * Note that we play around with the 'TS' bit in an attempt to get
581 * the correct behaviour even in the presence of the asynchronous
584 void math_error(struct pt_regs *regs, int error_code, int trapnr)
586 struct task_struct *task = current;
589 char *str = (trapnr == 16) ? "fpu exception" : "simd exception";
591 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
593 conditional_sti(regs);
595 if (!user_mode_vm(regs))
597 if (!fixup_exception(regs)) {
598 task->thread.error_code = error_code;
599 task->thread.trap_no = trapnr;
600 die(str, regs, error_code);
606 * Save the info for the exception handler and clear the error.
609 task->thread.trap_no = trapnr;
610 task->thread.error_code = error_code;
611 info.si_signo = SIGFPE;
613 info.si_addr = (void __user *)regs->ip;
615 unsigned short cwd, swd;
617 * (~cwd & swd) will mask out exceptions that are not set to unmasked
618 * status. 0x3f is the exception bits in these regs, 0x200 is the
619 * C1 reg you need in case of a stack fault, 0x040 is the stack
620 * fault bit. We should only be taking one exception at a time,
621 * so if this combination doesn't produce any single exception,
622 * then we have a bad program that isn't synchronizing its FPU usage
623 * and it will suffer the consequences since we won't be able to
624 * fully reproduce the context of the exception
626 cwd = get_fpu_cwd(task);
627 swd = get_fpu_swd(task);
632 * The SIMD FPU exceptions are handled a little differently, as there
633 * is only a single status/control register. Thus, to determine which
634 * unmasked exception was caught we must mask the exception mask bits
635 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
637 unsigned short mxcsr = get_fpu_mxcsr(task);
638 err = ~(mxcsr >> 7) & mxcsr;
641 if (err & 0x001) { /* Invalid op */
643 * swd & 0x240 == 0x040: Stack Underflow
644 * swd & 0x240 == 0x240: Stack Overflow
645 * User must clear the SF bit (0x40) if set
647 info.si_code = FPE_FLTINV;
648 } else if (err & 0x004) { /* Divide by Zero */
649 info.si_code = FPE_FLTDIV;
650 } else if (err & 0x008) { /* Overflow */
651 info.si_code = FPE_FLTOVF;
652 } else if (err & 0x012) { /* Denormal, Underflow */
653 info.si_code = FPE_FLTUND;
654 } else if (err & 0x020) { /* Precision */
655 info.si_code = FPE_FLTRES;
658 * If we're using IRQ 13, or supposedly even some trap 16
659 * implementations, it's possible we get a spurious trap...
661 return; /* Spurious trap, no error */
663 force_sig_info(SIGFPE, &info, task);
666 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
672 math_error(regs, error_code, 16);
676 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
678 math_error(regs, error_code, 19);
682 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
684 conditional_sti(regs);
686 /* No need to warn about this any longer. */
687 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
691 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
695 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
700 * __math_state_restore assumes that cr0.TS is already clear and the
701 * fpu state is all ready for use. Used during context switch.
703 void __math_state_restore(void)
705 struct thread_info *thread = current_thread_info();
706 struct task_struct *tsk = thread->task;
709 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
711 if (unlikely(restore_fpu_checking(tsk))) {
713 force_sig(SIGSEGV, tsk);
717 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
722 * 'math_state_restore()' saves the current math information in the
723 * old math state array, and gets the new ones from the current task
725 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
726 * Don't touch unless you *really* know how it works.
728 * Must be called with kernel preemption disabled (in this case,
729 * local interrupts are disabled at the call-site in entry.S).
731 asmlinkage void math_state_restore(void)
733 struct thread_info *thread = current_thread_info();
734 struct task_struct *tsk = thread->task;
736 if (!tsk_used_math(tsk)) {
739 * does a slab alloc which can sleep
745 do_group_exit(SIGKILL);
751 clts(); /* Allow maths ops (or we recurse) */
753 __math_state_restore();
755 EXPORT_SYMBOL_GPL(math_state_restore);
757 #ifndef CONFIG_MATH_EMULATION
758 void math_emulate(struct math_emu_info *info)
761 "math-emulation not enabled and no coprocessor found.\n");
762 printk(KERN_EMERG "killing %s.\n", current->comm);
763 force_sig(SIGFPE, current);
766 #endif /* CONFIG_MATH_EMULATION */
768 dotraplinkage void __kprobes
769 do_device_not_available(struct pt_regs *regs, long error_code)
772 if (read_cr0() & X86_CR0_EM) {
773 struct math_emu_info info = { };
775 conditional_sti(regs);
780 math_state_restore(); /* interrupts still off */
781 conditional_sti(regs);
784 math_state_restore();
789 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
794 info.si_signo = SIGILL;
796 info.si_code = ILL_BADSTK;
798 if (notify_die(DIE_TRAP, "iret exception",
799 regs, error_code, 32, SIGILL) == NOTIFY_STOP)
801 do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
805 void __init trap_init(void)
810 void __iomem *p = early_ioremap(0x0FFFD9, 4);
812 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
817 set_intr_gate(0, ÷_error);
818 set_intr_gate_ist(1, &debug, DEBUG_STACK);
819 set_intr_gate_ist(2, &nmi, NMI_STACK);
820 /* int3 can be called from all */
821 set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
822 /* int4 can be called from all */
823 set_system_intr_gate(4, &overflow);
824 set_intr_gate(5, &bounds);
825 set_intr_gate(6, &invalid_op);
826 set_intr_gate(7, &device_not_available);
828 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
830 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
832 set_intr_gate(9, &coprocessor_segment_overrun);
833 set_intr_gate(10, &invalid_TSS);
834 set_intr_gate(11, &segment_not_present);
835 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
836 set_intr_gate(13, &general_protection);
837 set_intr_gate(14, &page_fault);
838 set_intr_gate(15, &spurious_interrupt_bug);
839 set_intr_gate(16, &coprocessor_error);
840 set_intr_gate(17, &alignment_check);
841 #ifdef CONFIG_X86_MCE
842 set_intr_gate_ist(18, &machine_check, MCE_STACK);
844 set_intr_gate(19, &simd_coprocessor_error);
846 /* Reserve all the builtin and the syscall vector: */
847 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
848 set_bit(i, used_vectors);
850 #ifdef CONFIG_IA32_EMULATION
851 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
852 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
857 printk(KERN_INFO "Enabling fast FPU save and restore... ");
858 set_in_cr4(X86_CR4_OSFXSR);
863 "Enabling unmasked SIMD FPU exception support... ");
864 set_in_cr4(X86_CR4_OSXMMEXCPT);
868 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
869 set_bit(SYSCALL_VECTOR, used_vectors);
873 * Should be a barrier for any external CPU state:
877 x86_init.irqs.trap_init();