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 int unknown_nmi_panic;
88 static inline void conditional_sti(struct pt_regs *regs)
90 if (regs->flags & X86_EFLAGS_IF)
94 static inline void preempt_conditional_sti(struct pt_regs *regs)
97 if (regs->flags & X86_EFLAGS_IF)
101 static inline void conditional_cli(struct pt_regs *regs)
103 if (regs->flags & X86_EFLAGS_IF)
107 static inline void preempt_conditional_cli(struct pt_regs *regs)
109 if (regs->flags & X86_EFLAGS_IF)
114 static void __kprobes
115 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
116 long error_code, siginfo_t *info)
118 struct task_struct *tsk = current;
121 if (regs->flags & X86_VM_MASK) {
123 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
124 * On nmi (interrupt 2), do_trap should not be called.
132 if (!user_mode(regs))
139 * We want error_code and trap_no set for userspace faults and
140 * kernelspace faults which result in die(), but not
141 * kernelspace faults which are fixed up. die() gives the
142 * process no chance to handle the signal and notice the
143 * kernel fault information, so that won't result in polluting
144 * the information about previously queued, but not yet
145 * delivered, faults. See also do_general_protection below.
147 tsk->thread.error_code = error_code;
148 tsk->thread.trap_no = trapnr;
151 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
152 printk_ratelimit()) {
154 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
155 tsk->comm, tsk->pid, str,
156 regs->ip, regs->sp, error_code);
157 print_vma_addr(" in ", regs->ip);
163 force_sig_info(signr, info, tsk);
165 force_sig(signr, tsk);
169 if (!fixup_exception(regs)) {
170 tsk->thread.error_code = error_code;
171 tsk->thread.trap_no = trapnr;
172 die(str, regs, error_code);
178 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
185 #define DO_ERROR(trapnr, signr, str, name) \
186 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
188 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
191 conditional_sti(regs); \
192 do_trap(trapnr, signr, str, regs, error_code, NULL); \
195 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
196 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
199 info.si_signo = signr; \
201 info.si_code = sicode; \
202 info.si_addr = (void __user *)siaddr; \
203 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
206 conditional_sti(regs); \
207 do_trap(trapnr, signr, str, regs, error_code, &info); \
210 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
211 DO_ERROR(4, SIGSEGV, "overflow", overflow)
212 DO_ERROR(5, SIGSEGV, "bounds", bounds)
213 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
214 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
215 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
216 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
218 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
220 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
223 /* Runs on IST stack */
224 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
226 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
227 12, SIGBUS) == NOTIFY_STOP)
229 preempt_conditional_sti(regs);
230 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
231 preempt_conditional_cli(regs);
234 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
236 static const char str[] = "double fault";
237 struct task_struct *tsk = current;
239 /* Return not checked because double check cannot be ignored */
240 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
242 tsk->thread.error_code = error_code;
243 tsk->thread.trap_no = 8;
246 * This is always a kernel trap and never fixable (and thus must
250 die(str, regs, error_code);
254 dotraplinkage void __kprobes
255 do_general_protection(struct pt_regs *regs, long error_code)
257 struct task_struct *tsk;
259 conditional_sti(regs);
262 if (regs->flags & X86_VM_MASK)
267 if (!user_mode(regs))
270 tsk->thread.error_code = error_code;
271 tsk->thread.trap_no = 13;
273 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
274 printk_ratelimit()) {
276 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
277 tsk->comm, task_pid_nr(tsk),
278 regs->ip, regs->sp, error_code);
279 print_vma_addr(" in ", regs->ip);
283 force_sig(SIGSEGV, tsk);
289 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
294 if (fixup_exception(regs))
297 tsk->thread.error_code = error_code;
298 tsk->thread.trap_no = 13;
299 if (notify_die(DIE_GPF, "general protection fault", regs,
300 error_code, 13, SIGSEGV) == NOTIFY_STOP)
302 die("general protection fault", regs, error_code);
305 static int __init setup_unknown_nmi_panic(char *str)
307 unknown_nmi_panic = 1;
310 __setup("unknown_nmi_panic", setup_unknown_nmi_panic);
312 static notrace __kprobes void
313 mem_parity_error(unsigned char reason, struct pt_regs *regs)
316 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
317 reason, smp_processor_id());
320 "You have some hardware problem, likely on the PCI bus.\n");
322 #if defined(CONFIG_EDAC)
323 if (edac_handler_set()) {
324 edac_atomic_assert_error();
329 if (panic_on_unrecovered_nmi)
330 panic("NMI: Not continuing");
332 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
334 /* Clear and disable the memory parity error line. */
335 reason = (reason & 0xf) | 4;
339 static notrace __kprobes void
340 io_check_error(unsigned char reason, struct pt_regs *regs)
344 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
345 show_registers(regs);
348 panic("NMI IOCK error: Not continuing");
350 /* Re-enable the IOCK line, wait for a few seconds */
351 reason = (reason & 0xf) | 8;
362 static notrace __kprobes void
363 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
365 if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
370 * Might actually be able to figure out what the guilty party
379 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
380 reason, smp_processor_id());
382 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
383 if (unknown_nmi_panic || panic_on_unrecovered_nmi)
384 panic("NMI: Not continuing");
386 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
389 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
391 unsigned char reason = 0;
394 cpu = smp_processor_id();
396 /* Only the BSP gets external NMIs from the system. */
398 reason = get_nmi_reason();
400 if (!(reason & 0xc0)) {
401 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
405 #ifdef CONFIG_X86_LOCAL_APIC
406 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT)
410 unknown_nmi_error(reason, regs);
414 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
417 /* AK: following checks seem to be broken on modern chipsets. FIXME */
419 mem_parity_error(reason, regs);
421 io_check_error(reason, regs);
424 * Reassert NMI in case it became active meanwhile
425 * as it's edge-triggered:
431 dotraplinkage notrace __kprobes void
432 do_nmi(struct pt_regs *regs, long error_code)
436 inc_irq_stat(__nmi_count);
439 default_do_nmi(regs);
449 void restart_nmi(void)
454 /* May run on IST stack. */
455 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
457 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
458 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
461 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
462 #ifdef CONFIG_KPROBES
463 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
467 if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
472 preempt_conditional_sti(regs);
473 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
474 preempt_conditional_cli(regs);
479 * Help handler running on IST stack to switch back to user stack
480 * for scheduling or signal handling. The actual stack switch is done in
483 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
485 struct pt_regs *regs = eregs;
486 /* Did already sync */
487 if (eregs == (struct pt_regs *)eregs->sp)
489 /* Exception from user space */
490 else if (user_mode(eregs))
491 regs = task_pt_regs(current);
493 * Exception from kernel and interrupts are enabled. Move to
494 * kernel process stack.
496 else if (eregs->flags & X86_EFLAGS_IF)
497 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
505 * Our handling of the processor debug registers is non-trivial.
506 * We do not clear them on entry and exit from the kernel. Therefore
507 * it is possible to get a watchpoint trap here from inside the kernel.
508 * However, the code in ./ptrace.c has ensured that the user can
509 * only set watchpoints on userspace addresses. Therefore the in-kernel
510 * watchpoint trap can only occur in code which is reading/writing
511 * from user space. Such code must not hold kernel locks (since it
512 * can equally take a page fault), therefore it is safe to call
513 * force_sig_info even though that claims and releases locks.
515 * Code in ./signal.c ensures that the debug control register
516 * is restored before we deliver any signal, and therefore that
517 * user code runs with the correct debug control register even though
520 * Being careful here means that we don't have to be as careful in a
521 * lot of more complicated places (task switching can be a bit lazy
522 * about restoring all the debug state, and ptrace doesn't have to
523 * find every occurrence of the TF bit that could be saved away even
526 * May run on IST stack.
528 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
530 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;
541 * If dr6 has no reason to give us about the origin of this trap,
542 * then it's very likely the result of an icebp/int01 trap.
543 * User wants a sigtrap for that.
545 if (!dr6 && user_mode(regs))
548 /* Catch kmemcheck conditions first of all! */
549 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
552 /* DR6 may or may not be cleared by the CPU */
556 * The processor cleared BTF, so don't mark that we need it set.
558 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
560 /* Store the virtualized DR6 value */
561 tsk->thread.debugreg6 = dr6;
563 if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
564 SIGTRAP) == NOTIFY_STOP)
567 /* It's safe to allow irq's after DR6 has been saved */
568 preempt_conditional_sti(regs);
570 if (regs->flags & X86_VM_MASK) {
571 handle_vm86_trap((struct kernel_vm86_regs *) regs,
573 preempt_conditional_cli(regs);
578 * Single-stepping through system calls: ignore any exceptions in
579 * kernel space, but re-enable TF when returning to user mode.
581 * We already checked v86 mode above, so we can check for kernel mode
582 * by just checking the CPL of CS.
584 if ((dr6 & DR_STEP) && !user_mode(regs)) {
585 tsk->thread.debugreg6 &= ~DR_STEP;
586 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
587 regs->flags &= ~X86_EFLAGS_TF;
589 si_code = get_si_code(tsk->thread.debugreg6);
590 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
591 send_sigtrap(tsk, regs, error_code, si_code);
592 preempt_conditional_cli(regs);
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(struct pt_regs *regs, int error_code, int trapnr)
604 struct task_struct *task = current;
607 char *str = (trapnr == 16) ? "fpu exception" : "simd exception";
609 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
611 conditional_sti(regs);
613 if (!user_mode_vm(regs))
615 if (!fixup_exception(regs)) {
616 task->thread.error_code = error_code;
617 task->thread.trap_no = trapnr;
618 die(str, regs, error_code);
624 * Save the info for the exception handler and clear the error.
627 task->thread.trap_no = trapnr;
628 task->thread.error_code = error_code;
629 info.si_signo = SIGFPE;
631 info.si_addr = (void __user *)regs->ip;
633 unsigned short cwd, swd;
635 * (~cwd & swd) will mask out exceptions that are not set to unmasked
636 * status. 0x3f is the exception bits in these regs, 0x200 is the
637 * C1 reg you need in case of a stack fault, 0x040 is the stack
638 * fault bit. We should only be taking one exception at a time,
639 * so if this combination doesn't produce any single exception,
640 * then we have a bad program that isn't synchronizing its FPU usage
641 * and it will suffer the consequences since we won't be able to
642 * fully reproduce the context of the exception
644 cwd = get_fpu_cwd(task);
645 swd = get_fpu_swd(task);
650 * The SIMD FPU exceptions are handled a little differently, as there
651 * is only a single status/control register. Thus, to determine which
652 * unmasked exception was caught we must mask the exception mask bits
653 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
655 unsigned short mxcsr = get_fpu_mxcsr(task);
656 err = ~(mxcsr >> 7) & mxcsr;
659 if (err & 0x001) { /* Invalid op */
661 * swd & 0x240 == 0x040: Stack Underflow
662 * swd & 0x240 == 0x240: Stack Overflow
663 * User must clear the SF bit (0x40) if set
665 info.si_code = FPE_FLTINV;
666 } else if (err & 0x004) { /* Divide by Zero */
667 info.si_code = FPE_FLTDIV;
668 } else if (err & 0x008) { /* Overflow */
669 info.si_code = FPE_FLTOVF;
670 } else if (err & 0x012) { /* Denormal, Underflow */
671 info.si_code = FPE_FLTUND;
672 } else if (err & 0x020) { /* Precision */
673 info.si_code = FPE_FLTRES;
676 * If we're using IRQ 13, or supposedly even some trap 16
677 * implementations, it's possible we get a spurious trap...
679 return; /* Spurious trap, no error */
681 force_sig_info(SIGFPE, &info, task);
684 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
690 math_error(regs, error_code, 16);
694 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
696 math_error(regs, error_code, 19);
700 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
702 conditional_sti(regs);
704 /* No need to warn about this any longer. */
705 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
709 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
713 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
718 * __math_state_restore assumes that cr0.TS is already clear and the
719 * fpu state is all ready for use. Used during context switch.
721 void __math_state_restore(void)
723 struct thread_info *thread = current_thread_info();
724 struct task_struct *tsk = thread->task;
727 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
729 if (unlikely(restore_fpu_checking(tsk))) {
731 force_sig(SIGSEGV, tsk);
735 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
740 * 'math_state_restore()' saves the current math information in the
741 * old math state array, and gets the new ones from the current task
743 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
744 * Don't touch unless you *really* know how it works.
746 * Must be called with kernel preemption disabled (in this case,
747 * local interrupts are disabled at the call-site in entry.S).
749 asmlinkage void math_state_restore(void)
751 struct thread_info *thread = current_thread_info();
752 struct task_struct *tsk = thread->task;
754 if (!tsk_used_math(tsk)) {
757 * does a slab alloc which can sleep
763 do_group_exit(SIGKILL);
769 clts(); /* Allow maths ops (or we recurse) */
771 __math_state_restore();
773 EXPORT_SYMBOL_GPL(math_state_restore);
775 dotraplinkage void __kprobes
776 do_device_not_available(struct pt_regs *regs, long error_code)
778 #ifdef CONFIG_MATH_EMULATION
779 if (read_cr0() & X86_CR0_EM) {
780 struct math_emu_info info = { };
782 conditional_sti(regs);
789 math_state_restore(); /* interrupts still off */
791 conditional_sti(regs);
796 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
801 info.si_signo = SIGILL;
803 info.si_code = ILL_BADSTK;
805 if (notify_die(DIE_TRAP, "iret exception",
806 regs, error_code, 32, SIGILL) == NOTIFY_STOP)
808 do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
812 /* Set of traps needed for early debugging. */
813 void __init early_trap_init(void)
815 set_intr_gate_ist(1, &debug, DEBUG_STACK);
816 /* int3 can be called from all */
817 set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
818 set_intr_gate(14, &page_fault);
819 load_idt(&idt_descr);
822 void __init trap_init(void)
827 void __iomem *p = early_ioremap(0x0FFFD9, 4);
829 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
834 set_intr_gate(0, ÷_error);
835 set_intr_gate_ist(2, &nmi, NMI_STACK);
836 /* int4 can be called from all */
837 set_system_intr_gate(4, &overflow);
838 set_intr_gate(5, &bounds);
839 set_intr_gate(6, &invalid_op);
840 set_intr_gate(7, &device_not_available);
842 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
844 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
846 set_intr_gate(9, &coprocessor_segment_overrun);
847 set_intr_gate(10, &invalid_TSS);
848 set_intr_gate(11, &segment_not_present);
849 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
850 set_intr_gate(13, &general_protection);
851 set_intr_gate(15, &spurious_interrupt_bug);
852 set_intr_gate(16, &coprocessor_error);
853 set_intr_gate(17, &alignment_check);
854 #ifdef CONFIG_X86_MCE
855 set_intr_gate_ist(18, &machine_check, MCE_STACK);
857 set_intr_gate(19, &simd_coprocessor_error);
859 /* Reserve all the builtin and the syscall vector: */
860 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
861 set_bit(i, used_vectors);
863 #ifdef CONFIG_IA32_EMULATION
864 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
865 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
869 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
870 set_bit(SYSCALL_VECTOR, used_vectors);
874 * Should be a barrier for any external CPU state:
878 x86_init.irqs.trap_init();