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.
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/context_tracking.h>
16 #include <linux/interrupt.h>
17 #include <linux/kallsyms.h>
18 #include <linux/spinlock.h>
19 #include <linux/kprobes.h>
20 #include <linux/uaccess.h>
21 #include <linux/kdebug.h>
22 #include <linux/kgdb.h>
23 #include <linux/kernel.h>
24 #include <linux/module.h>
25 #include <linux/ptrace.h>
26 #include <linux/string.h>
27 #include <linux/delay.h>
28 #include <linux/errno.h>
29 #include <linux/kexec.h>
30 #include <linux/sched.h>
31 #include <linux/timer.h>
32 #include <linux/init.h>
33 #include <linux/bug.h>
34 #include <linux/nmi.h>
36 #include <linux/smp.h>
40 #include <linux/ioport.h>
41 #include <linux/eisa.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 <linux/atomic.h>
53 #include <asm/ftrace.h>
54 #include <asm/traps.h>
57 #include <asm/fpu-internal.h>
59 #include <asm/fixmap.h>
60 #include <asm/mach_traps.h>
61 #include <asm/alternative.h>
64 #include <asm/x86_init.h>
65 #include <asm/pgalloc.h>
66 #include <asm/proto.h>
68 /* No need to be aligned, but done to keep all IDTs defined the same way. */
69 gate_desc debug_idt_table[NR_VECTORS] __page_aligned_bss;
71 #include <asm/processor-flags.h>
72 #include <asm/setup.h>
74 asmlinkage int system_call(void);
77 /* Must be page-aligned because the real IDT is used in a fixmap. */
78 gate_desc idt_table[NR_VECTORS] __page_aligned_bss;
80 DECLARE_BITMAP(used_vectors, NR_VECTORS);
81 EXPORT_SYMBOL_GPL(used_vectors);
83 static inline void conditional_sti(struct pt_regs *regs)
85 if (regs->flags & X86_EFLAGS_IF)
89 static inline void preempt_conditional_sti(struct pt_regs *regs)
92 if (regs->flags & X86_EFLAGS_IF)
96 static inline void conditional_cli(struct pt_regs *regs)
98 if (regs->flags & X86_EFLAGS_IF)
102 static inline void preempt_conditional_cli(struct pt_regs *regs)
104 if (regs->flags & X86_EFLAGS_IF)
109 static nokprobe_inline int
110 do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
111 struct pt_regs *regs, long error_code)
114 if (regs->flags & X86_VM_MASK) {
116 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
117 * On nmi (interrupt 2), do_trap should not be called.
119 if (trapnr < X86_TRAP_UD) {
120 if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
127 if (!user_mode(regs)) {
128 if (!fixup_exception(regs)) {
129 tsk->thread.error_code = error_code;
130 tsk->thread.trap_nr = trapnr;
131 die(str, regs, error_code);
140 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
141 long error_code, siginfo_t *info)
143 struct task_struct *tsk = current;
146 if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
149 * We want error_code and trap_nr set for userspace faults and
150 * kernelspace faults which result in die(), but not
151 * kernelspace faults which are fixed up. die() gives the
152 * process no chance to handle the signal and notice the
153 * kernel fault information, so that won't result in polluting
154 * the information about previously queued, but not yet
155 * delivered, faults. See also do_general_protection below.
157 tsk->thread.error_code = error_code;
158 tsk->thread.trap_nr = trapnr;
161 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
162 printk_ratelimit()) {
163 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
164 tsk->comm, tsk->pid, str,
165 regs->ip, regs->sp, error_code);
166 print_vma_addr(" in ", regs->ip);
172 force_sig_info(signr, info, tsk);
174 force_sig(signr, tsk);
176 NOKPROBE_SYMBOL(do_trap);
178 #define DO_ERROR(trapnr, signr, str, name) \
179 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
181 enum ctx_state prev_state; \
183 prev_state = exception_enter(); \
184 if (notify_die(DIE_TRAP, str, regs, error_code, \
185 trapnr, signr) == NOTIFY_STOP) { \
186 exception_exit(prev_state); \
189 conditional_sti(regs); \
190 do_trap(trapnr, signr, str, regs, error_code, NULL); \
191 exception_exit(prev_state); \
194 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
195 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
198 enum ctx_state prev_state; \
200 info.si_signo = signr; \
202 info.si_code = sicode; \
203 info.si_addr = (void __user *)siaddr; \
204 prev_state = exception_enter(); \
205 if (notify_die(DIE_TRAP, str, regs, error_code, \
206 trapnr, signr) == NOTIFY_STOP) { \
207 exception_exit(prev_state); \
210 conditional_sti(regs); \
211 do_trap(trapnr, signr, str, regs, error_code, &info); \
212 exception_exit(prev_state); \
215 DO_ERROR_INFO(X86_TRAP_DE, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip )
216 DO_ERROR (X86_TRAP_OF, SIGSEGV, "overflow", overflow )
217 DO_ERROR (X86_TRAP_BR, SIGSEGV, "bounds", bounds )
218 DO_ERROR_INFO(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip )
219 DO_ERROR (X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun )
220 DO_ERROR (X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS )
221 DO_ERROR (X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present )
223 DO_ERROR (X86_TRAP_SS, SIGBUS, "stack segment", stack_segment )
225 DO_ERROR_INFO(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0 )
228 /* Runs on IST stack */
229 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
231 enum ctx_state prev_state;
233 prev_state = exception_enter();
234 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
235 X86_TRAP_SS, SIGBUS) != NOTIFY_STOP) {
236 preempt_conditional_sti(regs);
237 do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
238 preempt_conditional_cli(regs);
240 exception_exit(prev_state);
243 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
245 static const char str[] = "double fault";
246 struct task_struct *tsk = current;
249 /* Return not checked because double check cannot be ignored */
250 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
252 tsk->thread.error_code = error_code;
253 tsk->thread.trap_nr = X86_TRAP_DF;
255 #ifdef CONFIG_DOUBLEFAULT
256 df_debug(regs, error_code);
259 * This is always a kernel trap and never fixable (and thus must
263 die(str, regs, error_code);
268 do_general_protection(struct pt_regs *regs, long error_code)
270 struct task_struct *tsk;
271 enum ctx_state prev_state;
273 prev_state = exception_enter();
274 conditional_sti(regs);
277 if (regs->flags & X86_VM_MASK) {
279 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
285 if (!user_mode(regs)) {
286 if (fixup_exception(regs))
289 tsk->thread.error_code = error_code;
290 tsk->thread.trap_nr = X86_TRAP_GP;
291 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
292 X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
293 die("general protection fault", regs, error_code);
297 tsk->thread.error_code = error_code;
298 tsk->thread.trap_nr = X86_TRAP_GP;
300 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
301 printk_ratelimit()) {
302 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
303 tsk->comm, task_pid_nr(tsk),
304 regs->ip, regs->sp, error_code);
305 print_vma_addr(" in ", regs->ip);
309 force_sig(SIGSEGV, tsk);
311 exception_exit(prev_state);
313 NOKPROBE_SYMBOL(do_general_protection);
315 /* May run on IST stack. */
316 dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code)
318 enum ctx_state prev_state;
320 #ifdef CONFIG_DYNAMIC_FTRACE
322 * ftrace must be first, everything else may cause a recursive crash.
323 * See note by declaration of modifying_ftrace_code in ftrace.c
325 if (unlikely(atomic_read(&modifying_ftrace_code)) &&
326 ftrace_int3_handler(regs))
329 if (poke_int3_handler(regs))
332 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
333 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
334 SIGTRAP) == NOTIFY_STOP)
336 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
338 #ifdef CONFIG_KPROBES
339 if (kprobe_int3_handler(regs))
342 prev_state = exception_enter();
344 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
345 SIGTRAP) == NOTIFY_STOP)
349 * Let others (NMI) know that the debug stack is in use
350 * as we may switch to the interrupt stack.
352 debug_stack_usage_inc();
353 preempt_conditional_sti(regs);
354 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
355 preempt_conditional_cli(regs);
356 debug_stack_usage_dec();
358 exception_exit(prev_state);
360 NOKPROBE_SYMBOL(do_int3);
364 * Help handler running on IST stack to switch back to user stack
365 * for scheduling or signal handling. The actual stack switch is done in
368 asmlinkage struct pt_regs *sync_regs(struct pt_regs *eregs)
370 struct pt_regs *regs = eregs;
371 /* Did already sync */
372 if (eregs == (struct pt_regs *)eregs->sp)
374 /* Exception from user space */
375 else if (user_mode(eregs))
376 regs = task_pt_regs(current);
378 * Exception from kernel and interrupts are enabled. Move to
379 * kernel process stack.
381 else if (eregs->flags & X86_EFLAGS_IF)
382 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
387 NOKPROBE_SYMBOL(sync_regs);
391 * Our handling of the processor debug registers is non-trivial.
392 * We do not clear them on entry and exit from the kernel. Therefore
393 * it is possible to get a watchpoint trap here from inside the kernel.
394 * However, the code in ./ptrace.c has ensured that the user can
395 * only set watchpoints on userspace addresses. Therefore the in-kernel
396 * watchpoint trap can only occur in code which is reading/writing
397 * from user space. Such code must not hold kernel locks (since it
398 * can equally take a page fault), therefore it is safe to call
399 * force_sig_info even though that claims and releases locks.
401 * Code in ./signal.c ensures that the debug control register
402 * is restored before we deliver any signal, and therefore that
403 * user code runs with the correct debug control register even though
406 * Being careful here means that we don't have to be as careful in a
407 * lot of more complicated places (task switching can be a bit lazy
408 * about restoring all the debug state, and ptrace doesn't have to
409 * find every occurrence of the TF bit that could be saved away even
412 * May run on IST stack.
414 dotraplinkage void do_debug(struct pt_regs *regs, long error_code)
416 struct task_struct *tsk = current;
417 enum ctx_state prev_state;
422 get_debugreg(dr6, 6);
424 /* Filter out all the reserved bits which are preset to 1 */
425 dr6 &= ~DR6_RESERVED;
428 * If dr6 has no reason to give us about the origin of this trap,
429 * then it's very likely the result of an icebp/int01 trap.
430 * User wants a sigtrap for that.
432 if (!dr6 && user_mode(regs))
435 /* Catch kmemcheck conditions first of all! */
436 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
439 /* DR6 may or may not be cleared by the CPU */
443 * The processor cleared BTF, so don't mark that we need it set.
445 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
447 /* Store the virtualized DR6 value */
448 tsk->thread.debugreg6 = dr6;
450 #ifdef CONFIG_KPROBES
451 if (kprobe_debug_handler(regs))
454 prev_state = exception_enter();
456 if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
457 SIGTRAP) == NOTIFY_STOP)
461 * Let others (NMI) know that the debug stack is in use
462 * as we may switch to the interrupt stack.
464 debug_stack_usage_inc();
466 /* It's safe to allow irq's after DR6 has been saved */
467 preempt_conditional_sti(regs);
469 if (regs->flags & X86_VM_MASK) {
470 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
472 preempt_conditional_cli(regs);
473 debug_stack_usage_dec();
478 * Single-stepping through system calls: ignore any exceptions in
479 * kernel space, but re-enable TF when returning to user mode.
481 * We already checked v86 mode above, so we can check for kernel mode
482 * by just checking the CPL of CS.
484 if ((dr6 & DR_STEP) && !user_mode(regs)) {
485 tsk->thread.debugreg6 &= ~DR_STEP;
486 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
487 regs->flags &= ~X86_EFLAGS_TF;
489 si_code = get_si_code(tsk->thread.debugreg6);
490 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
491 send_sigtrap(tsk, regs, error_code, si_code);
492 preempt_conditional_cli(regs);
493 debug_stack_usage_dec();
496 exception_exit(prev_state);
498 NOKPROBE_SYMBOL(do_debug);
501 * Note that we play around with the 'TS' bit in an attempt to get
502 * the correct behaviour even in the presence of the asynchronous
505 void math_error(struct pt_regs *regs, int error_code, int trapnr)
507 struct task_struct *task = current;
510 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
513 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
515 conditional_sti(regs);
517 if (!user_mode_vm(regs))
519 if (!fixup_exception(regs)) {
520 task->thread.error_code = error_code;
521 task->thread.trap_nr = trapnr;
522 die(str, regs, error_code);
528 * Save the info for the exception handler and clear the error.
531 task->thread.trap_nr = trapnr;
532 task->thread.error_code = error_code;
533 info.si_signo = SIGFPE;
535 info.si_addr = (void __user *)regs->ip;
536 if (trapnr == X86_TRAP_MF) {
537 unsigned short cwd, swd;
539 * (~cwd & swd) will mask out exceptions that are not set to unmasked
540 * status. 0x3f is the exception bits in these regs, 0x200 is the
541 * C1 reg you need in case of a stack fault, 0x040 is the stack
542 * fault bit. We should only be taking one exception at a time,
543 * so if this combination doesn't produce any single exception,
544 * then we have a bad program that isn't synchronizing its FPU usage
545 * and it will suffer the consequences since we won't be able to
546 * fully reproduce the context of the exception
548 cwd = get_fpu_cwd(task);
549 swd = get_fpu_swd(task);
554 * The SIMD FPU exceptions are handled a little differently, as there
555 * is only a single status/control register. Thus, to determine which
556 * unmasked exception was caught we must mask the exception mask bits
557 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
559 unsigned short mxcsr = get_fpu_mxcsr(task);
560 err = ~(mxcsr >> 7) & mxcsr;
563 if (err & 0x001) { /* Invalid op */
565 * swd & 0x240 == 0x040: Stack Underflow
566 * swd & 0x240 == 0x240: Stack Overflow
567 * User must clear the SF bit (0x40) if set
569 info.si_code = FPE_FLTINV;
570 } else if (err & 0x004) { /* Divide by Zero */
571 info.si_code = FPE_FLTDIV;
572 } else if (err & 0x008) { /* Overflow */
573 info.si_code = FPE_FLTOVF;
574 } else if (err & 0x012) { /* Denormal, Underflow */
575 info.si_code = FPE_FLTUND;
576 } else if (err & 0x020) { /* Precision */
577 info.si_code = FPE_FLTRES;
580 * If we're using IRQ 13, or supposedly even some trap
581 * X86_TRAP_MF implementations, it's possible
582 * we get a spurious trap, which is not an error.
586 force_sig_info(SIGFPE, &info, task);
589 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
591 enum ctx_state prev_state;
593 prev_state = exception_enter();
594 math_error(regs, error_code, X86_TRAP_MF);
595 exception_exit(prev_state);
599 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
601 enum ctx_state prev_state;
603 prev_state = exception_enter();
604 math_error(regs, error_code, X86_TRAP_XF);
605 exception_exit(prev_state);
609 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
611 conditional_sti(regs);
613 /* No need to warn about this any longer. */
614 pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
618 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
622 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
627 * 'math_state_restore()' saves the current math information in the
628 * old math state array, and gets the new ones from the current task
630 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
631 * Don't touch unless you *really* know how it works.
633 * Must be called with kernel preemption disabled (eg with local
634 * local interrupts as in the case of do_device_not_available).
636 void math_state_restore(void)
638 struct task_struct *tsk = current;
640 if (!tsk_used_math(tsk)) {
643 * does a slab alloc which can sleep
649 do_group_exit(SIGKILL);
655 __thread_fpu_begin(tsk);
658 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
660 if (unlikely(restore_fpu_checking(tsk))) {
662 force_sig(SIGSEGV, tsk);
666 tsk->thread.fpu_counter++;
668 EXPORT_SYMBOL_GPL(math_state_restore);
671 do_device_not_available(struct pt_regs *regs, long error_code)
673 enum ctx_state prev_state;
675 prev_state = exception_enter();
676 BUG_ON(use_eager_fpu());
678 #ifdef CONFIG_MATH_EMULATION
679 if (read_cr0() & X86_CR0_EM) {
680 struct math_emu_info info = { };
682 conditional_sti(regs);
686 exception_exit(prev_state);
690 math_state_restore(); /* interrupts still off */
692 conditional_sti(regs);
694 exception_exit(prev_state);
696 NOKPROBE_SYMBOL(do_device_not_available);
699 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
702 enum ctx_state prev_state;
704 prev_state = exception_enter();
707 info.si_signo = SIGILL;
709 info.si_code = ILL_BADSTK;
711 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
712 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
713 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
716 exception_exit(prev_state);
720 /* Set of traps needed for early debugging. */
721 void __init early_trap_init(void)
723 set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
724 /* int3 can be called from all */
725 set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
727 set_intr_gate(X86_TRAP_PF, page_fault);
729 load_idt(&idt_descr);
732 void __init early_trap_pf_init(void)
735 set_intr_gate(X86_TRAP_PF, page_fault);
739 void __init trap_init(void)
744 void __iomem *p = early_ioremap(0x0FFFD9, 4);
746 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
751 set_intr_gate(X86_TRAP_DE, divide_error);
752 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
753 /* int4 can be called from all */
754 set_system_intr_gate(X86_TRAP_OF, &overflow);
755 set_intr_gate(X86_TRAP_BR, bounds);
756 set_intr_gate(X86_TRAP_UD, invalid_op);
757 set_intr_gate(X86_TRAP_NM, device_not_available);
759 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
761 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
763 set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
764 set_intr_gate(X86_TRAP_TS, invalid_TSS);
765 set_intr_gate(X86_TRAP_NP, segment_not_present);
766 set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
767 set_intr_gate(X86_TRAP_GP, general_protection);
768 set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
769 set_intr_gate(X86_TRAP_MF, coprocessor_error);
770 set_intr_gate(X86_TRAP_AC, alignment_check);
771 #ifdef CONFIG_X86_MCE
772 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
774 set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
776 /* Reserve all the builtin and the syscall vector: */
777 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
778 set_bit(i, used_vectors);
780 #ifdef CONFIG_IA32_EMULATION
781 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
782 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
786 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
787 set_bit(SYSCALL_VECTOR, used_vectors);
791 * Set the IDT descriptor to a fixed read-only location, so that the
792 * "sidt" instruction will not leak the location of the kernel, and
793 * to defend the IDT against arbitrary memory write vulnerabilities.
794 * It will be reloaded in cpu_init() */
795 __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
796 idt_descr.address = fix_to_virt(FIX_RO_IDT);
799 * Should be a barrier for any external CPU state:
803 x86_init.irqs.trap_init();
806 memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
807 set_nmi_gate(X86_TRAP_DB, &debug);
808 set_nmi_gate(X86_TRAP_BP, &int3);