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/uprobes.h>
27 #include <linux/string.h>
28 #include <linux/delay.h>
29 #include <linux/errno.h>
30 #include <linux/kexec.h>
31 #include <linux/sched.h>
32 #include <linux/timer.h>
33 #include <linux/init.h>
34 #include <linux/bug.h>
35 #include <linux/nmi.h>
37 #include <linux/smp.h>
41 #include <linux/ioport.h>
42 #include <linux/eisa.h>
45 #if defined(CONFIG_EDAC)
46 #include <linux/edac.h>
49 #include <asm/kmemcheck.h>
50 #include <asm/stacktrace.h>
51 #include <asm/processor.h>
52 #include <asm/debugreg.h>
53 #include <linux/atomic.h>
54 #include <asm/ftrace.h>
55 #include <asm/traps.h>
58 #include <asm/fpu-internal.h>
60 #include <asm/fixmap.h>
61 #include <asm/mach_traps.h>
62 #include <asm/alternative.h>
65 #include <asm/x86_init.h>
66 #include <asm/pgalloc.h>
67 #include <asm/proto.h>
69 /* No need to be aligned, but done to keep all IDTs defined the same way. */
70 gate_desc debug_idt_table[NR_VECTORS] __page_aligned_bss;
72 #include <asm/processor-flags.h>
73 #include <asm/setup.h>
75 asmlinkage int system_call(void);
78 /* Must be page-aligned because the real IDT is used in a fixmap. */
79 gate_desc idt_table[NR_VECTORS] __page_aligned_bss;
81 DECLARE_BITMAP(used_vectors, NR_VECTORS);
82 EXPORT_SYMBOL_GPL(used_vectors);
84 static inline void conditional_sti(struct pt_regs *regs)
86 if (regs->flags & X86_EFLAGS_IF)
90 static inline void preempt_conditional_sti(struct pt_regs *regs)
93 if (regs->flags & X86_EFLAGS_IF)
97 static inline void conditional_cli(struct pt_regs *regs)
99 if (regs->flags & X86_EFLAGS_IF)
103 static inline void preempt_conditional_cli(struct pt_regs *regs)
105 if (regs->flags & X86_EFLAGS_IF)
110 static nokprobe_inline int
111 do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
112 struct pt_regs *regs, long error_code)
115 if (regs->flags & X86_VM_MASK) {
117 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
118 * On nmi (interrupt 2), do_trap should not be called.
120 if (trapnr < X86_TRAP_UD) {
121 if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
128 if (!user_mode(regs)) {
129 if (!fixup_exception(regs)) {
130 tsk->thread.error_code = error_code;
131 tsk->thread.trap_nr = trapnr;
132 die(str, regs, error_code);
140 static siginfo_t *fill_trap_info(struct pt_regs *regs, int signr, int trapnr,
143 unsigned long siaddr;
148 return SEND_SIG_PRIV;
152 siaddr = uprobe_get_trap_addr(regs);
156 siaddr = uprobe_get_trap_addr(regs);
164 info->si_signo = signr;
166 info->si_code = sicode;
167 info->si_addr = (void __user *)siaddr;
172 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
173 long error_code, siginfo_t *info)
175 struct task_struct *tsk = current;
178 if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
181 * We want error_code and trap_nr set for userspace faults and
182 * kernelspace faults which result in die(), but not
183 * kernelspace faults which are fixed up. die() gives the
184 * process no chance to handle the signal and notice the
185 * kernel fault information, so that won't result in polluting
186 * the information about previously queued, but not yet
187 * delivered, faults. See also do_general_protection below.
189 tsk->thread.error_code = error_code;
190 tsk->thread.trap_nr = trapnr;
193 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
194 printk_ratelimit()) {
195 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
196 tsk->comm, tsk->pid, str,
197 regs->ip, regs->sp, error_code);
198 print_vma_addr(" in ", regs->ip);
203 force_sig_info(signr, info ?: SEND_SIG_PRIV, tsk);
205 NOKPROBE_SYMBOL(do_trap);
207 static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
208 unsigned long trapnr, int signr)
210 enum ctx_state prev_state = exception_enter();
213 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
215 conditional_sti(regs);
216 do_trap(trapnr, signr, str, regs, error_code,
217 fill_trap_info(regs, signr, trapnr, &info));
220 exception_exit(prev_state);
223 #define DO_ERROR(trapnr, signr, str, name) \
224 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
226 do_error_trap(regs, error_code, str, trapnr, signr); \
229 DO_ERROR(X86_TRAP_DE, SIGFPE, "divide error", divide_error)
230 DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
231 DO_ERROR(X86_TRAP_BR, SIGSEGV, "bounds", bounds)
232 DO_ERROR(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op)
233 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",coprocessor_segment_overrun)
234 DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
235 DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
237 DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
239 DO_ERROR(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check)
242 /* Runs on IST stack */
243 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
245 enum ctx_state prev_state;
247 prev_state = exception_enter();
248 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
249 X86_TRAP_SS, SIGBUS) != NOTIFY_STOP) {
250 preempt_conditional_sti(regs);
251 do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
252 preempt_conditional_cli(regs);
254 exception_exit(prev_state);
257 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
259 static const char str[] = "double fault";
260 struct task_struct *tsk = current;
263 /* Return not checked because double check cannot be ignored */
264 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
266 tsk->thread.error_code = error_code;
267 tsk->thread.trap_nr = X86_TRAP_DF;
269 #ifdef CONFIG_DOUBLEFAULT
270 df_debug(regs, error_code);
273 * This is always a kernel trap and never fixable (and thus must
277 die(str, regs, error_code);
282 do_general_protection(struct pt_regs *regs, long error_code)
284 struct task_struct *tsk;
285 enum ctx_state prev_state;
287 prev_state = exception_enter();
288 conditional_sti(regs);
291 if (regs->flags & X86_VM_MASK) {
293 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
299 if (!user_mode(regs)) {
300 if (fixup_exception(regs))
303 tsk->thread.error_code = error_code;
304 tsk->thread.trap_nr = X86_TRAP_GP;
305 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
306 X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
307 die("general protection fault", regs, error_code);
311 tsk->thread.error_code = error_code;
312 tsk->thread.trap_nr = X86_TRAP_GP;
314 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
315 printk_ratelimit()) {
316 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
317 tsk->comm, task_pid_nr(tsk),
318 regs->ip, regs->sp, error_code);
319 print_vma_addr(" in ", regs->ip);
323 force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
325 exception_exit(prev_state);
327 NOKPROBE_SYMBOL(do_general_protection);
329 /* May run on IST stack. */
330 dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code)
332 enum ctx_state prev_state;
334 #ifdef CONFIG_DYNAMIC_FTRACE
336 * ftrace must be first, everything else may cause a recursive crash.
337 * See note by declaration of modifying_ftrace_code in ftrace.c
339 if (unlikely(atomic_read(&modifying_ftrace_code)) &&
340 ftrace_int3_handler(regs))
343 if (poke_int3_handler(regs))
346 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
347 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
348 SIGTRAP) == NOTIFY_STOP)
350 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
352 #ifdef CONFIG_KPROBES
353 if (kprobe_int3_handler(regs))
356 prev_state = exception_enter();
358 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
359 SIGTRAP) == NOTIFY_STOP)
363 * Let others (NMI) know that the debug stack is in use
364 * as we may switch to the interrupt stack.
366 debug_stack_usage_inc();
367 preempt_conditional_sti(regs);
368 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
369 preempt_conditional_cli(regs);
370 debug_stack_usage_dec();
372 exception_exit(prev_state);
374 NOKPROBE_SYMBOL(do_int3);
378 * Help handler running on IST stack to switch back to user stack
379 * for scheduling or signal handling. The actual stack switch is done in
382 asmlinkage __visible struct pt_regs *sync_regs(struct pt_regs *eregs)
384 struct pt_regs *regs = eregs;
385 /* Did already sync */
386 if (eregs == (struct pt_regs *)eregs->sp)
388 /* Exception from user space */
389 else if (user_mode(eregs))
390 regs = task_pt_regs(current);
392 * Exception from kernel and interrupts are enabled. Move to
393 * kernel process stack.
395 else if (eregs->flags & X86_EFLAGS_IF)
396 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
401 NOKPROBE_SYMBOL(sync_regs);
405 * Our handling of the processor debug registers is non-trivial.
406 * We do not clear them on entry and exit from the kernel. Therefore
407 * it is possible to get a watchpoint trap here from inside the kernel.
408 * However, the code in ./ptrace.c has ensured that the user can
409 * only set watchpoints on userspace addresses. Therefore the in-kernel
410 * watchpoint trap can only occur in code which is reading/writing
411 * from user space. Such code must not hold kernel locks (since it
412 * can equally take a page fault), therefore it is safe to call
413 * force_sig_info even though that claims and releases locks.
415 * Code in ./signal.c ensures that the debug control register
416 * is restored before we deliver any signal, and therefore that
417 * user code runs with the correct debug control register even though
420 * Being careful here means that we don't have to be as careful in a
421 * lot of more complicated places (task switching can be a bit lazy
422 * about restoring all the debug state, and ptrace doesn't have to
423 * find every occurrence of the TF bit that could be saved away even
426 * May run on IST stack.
428 dotraplinkage void do_debug(struct pt_regs *regs, long error_code)
430 struct task_struct *tsk = current;
431 enum ctx_state prev_state;
436 get_debugreg(dr6, 6);
438 /* Filter out all the reserved bits which are preset to 1 */
439 dr6 &= ~DR6_RESERVED;
442 * If dr6 has no reason to give us about the origin of this trap,
443 * then it's very likely the result of an icebp/int01 trap.
444 * User wants a sigtrap for that.
446 if (!dr6 && user_mode(regs))
449 /* Catch kmemcheck conditions first of all! */
450 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
453 /* DR6 may or may not be cleared by the CPU */
457 * The processor cleared BTF, so don't mark that we need it set.
459 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
461 /* Store the virtualized DR6 value */
462 tsk->thread.debugreg6 = dr6;
464 #ifdef CONFIG_KPROBES
465 if (kprobe_debug_handler(regs))
468 prev_state = exception_enter();
470 if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
471 SIGTRAP) == NOTIFY_STOP)
475 * Let others (NMI) know that the debug stack is in use
476 * as we may switch to the interrupt stack.
478 debug_stack_usage_inc();
480 /* It's safe to allow irq's after DR6 has been saved */
481 preempt_conditional_sti(regs);
483 if (regs->flags & X86_VM_MASK) {
484 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
486 preempt_conditional_cli(regs);
487 debug_stack_usage_dec();
492 * Single-stepping through system calls: ignore any exceptions in
493 * kernel space, but re-enable TF when returning to user mode.
495 * We already checked v86 mode above, so we can check for kernel mode
496 * by just checking the CPL of CS.
498 if ((dr6 & DR_STEP) && !user_mode(regs)) {
499 tsk->thread.debugreg6 &= ~DR_STEP;
500 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
501 regs->flags &= ~X86_EFLAGS_TF;
503 si_code = get_si_code(tsk->thread.debugreg6);
504 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
505 send_sigtrap(tsk, regs, error_code, si_code);
506 preempt_conditional_cli(regs);
507 debug_stack_usage_dec();
510 exception_exit(prev_state);
512 NOKPROBE_SYMBOL(do_debug);
515 * Note that we play around with the 'TS' bit in an attempt to get
516 * the correct behaviour even in the presence of the asynchronous
519 static void math_error(struct pt_regs *regs, int error_code, int trapnr)
521 struct task_struct *task = current;
524 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
527 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
529 conditional_sti(regs);
531 if (!user_mode_vm(regs))
533 if (!fixup_exception(regs)) {
534 task->thread.error_code = error_code;
535 task->thread.trap_nr = trapnr;
536 die(str, regs, error_code);
542 * Save the info for the exception handler and clear the error.
545 task->thread.trap_nr = trapnr;
546 task->thread.error_code = error_code;
547 info.si_signo = SIGFPE;
549 info.si_addr = (void __user *)uprobe_get_trap_addr(regs);
550 if (trapnr == X86_TRAP_MF) {
551 unsigned short cwd, swd;
553 * (~cwd & swd) will mask out exceptions that are not set to unmasked
554 * status. 0x3f is the exception bits in these regs, 0x200 is the
555 * C1 reg you need in case of a stack fault, 0x040 is the stack
556 * fault bit. We should only be taking one exception at a time,
557 * so if this combination doesn't produce any single exception,
558 * then we have a bad program that isn't synchronizing its FPU usage
559 * and it will suffer the consequences since we won't be able to
560 * fully reproduce the context of the exception
562 cwd = get_fpu_cwd(task);
563 swd = get_fpu_swd(task);
568 * The SIMD FPU exceptions are handled a little differently, as there
569 * is only a single status/control register. Thus, to determine which
570 * unmasked exception was caught we must mask the exception mask bits
571 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
573 unsigned short mxcsr = get_fpu_mxcsr(task);
574 err = ~(mxcsr >> 7) & mxcsr;
577 if (err & 0x001) { /* Invalid op */
579 * swd & 0x240 == 0x040: Stack Underflow
580 * swd & 0x240 == 0x240: Stack Overflow
581 * User must clear the SF bit (0x40) if set
583 info.si_code = FPE_FLTINV;
584 } else if (err & 0x004) { /* Divide by Zero */
585 info.si_code = FPE_FLTDIV;
586 } else if (err & 0x008) { /* Overflow */
587 info.si_code = FPE_FLTOVF;
588 } else if (err & 0x012) { /* Denormal, Underflow */
589 info.si_code = FPE_FLTUND;
590 } else if (err & 0x020) { /* Precision */
591 info.si_code = FPE_FLTRES;
594 * If we're using IRQ 13, or supposedly even some trap
595 * X86_TRAP_MF implementations, it's possible
596 * we get a spurious trap, which is not an error.
600 force_sig_info(SIGFPE, &info, task);
603 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
605 enum ctx_state prev_state;
607 prev_state = exception_enter();
608 math_error(regs, error_code, X86_TRAP_MF);
609 exception_exit(prev_state);
613 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
615 enum ctx_state prev_state;
617 prev_state = exception_enter();
618 math_error(regs, error_code, X86_TRAP_XF);
619 exception_exit(prev_state);
623 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
625 conditional_sti(regs);
627 /* No need to warn about this any longer. */
628 pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
632 asmlinkage __visible void __attribute__((weak)) smp_thermal_interrupt(void)
636 asmlinkage __visible void __attribute__((weak)) smp_threshold_interrupt(void)
641 * 'math_state_restore()' saves the current math information in the
642 * old math state array, and gets the new ones from the current task
644 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
645 * Don't touch unless you *really* know how it works.
647 * Must be called with kernel preemption disabled (eg with local
648 * local interrupts as in the case of do_device_not_available).
650 void math_state_restore(void)
652 struct task_struct *tsk = current;
654 if (!tsk_used_math(tsk)) {
657 * does a slab alloc which can sleep
663 do_group_exit(SIGKILL);
669 __thread_fpu_begin(tsk);
672 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
674 if (unlikely(restore_fpu_checking(tsk))) {
676 force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
680 tsk->thread.fpu_counter++;
682 EXPORT_SYMBOL_GPL(math_state_restore);
685 do_device_not_available(struct pt_regs *regs, long error_code)
687 enum ctx_state prev_state;
689 prev_state = exception_enter();
690 BUG_ON(use_eager_fpu());
692 #ifdef CONFIG_MATH_EMULATION
693 if (read_cr0() & X86_CR0_EM) {
694 struct math_emu_info info = { };
696 conditional_sti(regs);
700 exception_exit(prev_state);
704 math_state_restore(); /* interrupts still off */
706 conditional_sti(regs);
708 exception_exit(prev_state);
710 NOKPROBE_SYMBOL(do_device_not_available);
713 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
716 enum ctx_state prev_state;
718 prev_state = exception_enter();
721 info.si_signo = SIGILL;
723 info.si_code = ILL_BADSTK;
725 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
726 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
727 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
730 exception_exit(prev_state);
734 /* Set of traps needed for early debugging. */
735 void __init early_trap_init(void)
737 set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
738 /* int3 can be called from all */
739 set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
741 set_intr_gate(X86_TRAP_PF, page_fault);
743 load_idt(&idt_descr);
746 void __init early_trap_pf_init(void)
749 set_intr_gate(X86_TRAP_PF, page_fault);
753 void __init trap_init(void)
758 void __iomem *p = early_ioremap(0x0FFFD9, 4);
760 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
765 set_intr_gate(X86_TRAP_DE, divide_error);
766 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
767 /* int4 can be called from all */
768 set_system_intr_gate(X86_TRAP_OF, &overflow);
769 set_intr_gate(X86_TRAP_BR, bounds);
770 set_intr_gate(X86_TRAP_UD, invalid_op);
771 set_intr_gate(X86_TRAP_NM, device_not_available);
773 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
775 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
777 set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
778 set_intr_gate(X86_TRAP_TS, invalid_TSS);
779 set_intr_gate(X86_TRAP_NP, segment_not_present);
780 set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
781 set_intr_gate(X86_TRAP_GP, general_protection);
782 set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
783 set_intr_gate(X86_TRAP_MF, coprocessor_error);
784 set_intr_gate(X86_TRAP_AC, alignment_check);
785 #ifdef CONFIG_X86_MCE
786 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
788 set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
790 /* Reserve all the builtin and the syscall vector: */
791 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
792 set_bit(i, used_vectors);
794 #ifdef CONFIG_IA32_EMULATION
795 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
796 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
800 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
801 set_bit(SYSCALL_VECTOR, used_vectors);
805 * Set the IDT descriptor to a fixed read-only location, so that the
806 * "sidt" instruction will not leak the location of the kernel, and
807 * to defend the IDT against arbitrary memory write vulnerabilities.
808 * It will be reloaded in cpu_init() */
809 __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
810 idt_descr.address = fix_to_virt(FIX_RO_IDT);
813 * Should be a barrier for any external CPU state:
817 x86_init.irqs.trap_init();
820 memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
821 set_nmi_gate(X86_TRAP_DB, &debug);
822 set_nmi_gate(X86_TRAP_BP, &int3);