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)
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);
139 static siginfo_t *fill_trap_info(struct pt_regs *regs, int signr, int trapnr,
142 unsigned long siaddr;
147 return SEND_SIG_PRIV;
163 info->si_signo = signr;
165 info->si_code = sicode;
166 info->si_addr = (void __user *)siaddr;
170 static void __kprobes
171 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
172 long error_code, siginfo_t *info)
174 struct task_struct *tsk = current;
177 if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
180 * We want error_code and trap_nr set for userspace faults and
181 * kernelspace faults which result in die(), but not
182 * kernelspace faults which are fixed up. die() gives the
183 * process no chance to handle the signal and notice the
184 * kernel fault information, so that won't result in polluting
185 * the information about previously queued, but not yet
186 * delivered, faults. See also do_general_protection below.
188 tsk->thread.error_code = error_code;
189 tsk->thread.trap_nr = trapnr;
192 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
193 printk_ratelimit()) {
194 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
195 tsk->comm, tsk->pid, str,
196 regs->ip, regs->sp, error_code);
197 print_vma_addr(" in ", regs->ip);
202 force_sig_info(signr, info ?: SEND_SIG_PRIV, tsk);
205 static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
206 unsigned long trapnr, int signr)
208 enum ctx_state prev_state = exception_enter();
211 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
213 conditional_sti(regs);
214 do_trap(trapnr, signr, str, regs, error_code,
215 fill_trap_info(regs, signr, trapnr, &info));
218 exception_exit(prev_state);
221 #define DO_ERROR(trapnr, signr, str, name) \
222 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
224 do_error_trap(regs, error_code, str, trapnr, signr); \
227 DO_ERROR(X86_TRAP_DE, SIGFPE, "divide error", divide_error)
228 DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
229 DO_ERROR(X86_TRAP_BR, SIGSEGV, "bounds", bounds)
230 DO_ERROR(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op)
231 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",coprocessor_segment_overrun)
232 DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
233 DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
235 DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
237 DO_ERROR(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check)
240 /* Runs on IST stack */
241 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
243 enum ctx_state prev_state;
245 prev_state = exception_enter();
246 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
247 X86_TRAP_SS, SIGBUS) != NOTIFY_STOP) {
248 preempt_conditional_sti(regs);
249 do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
250 preempt_conditional_cli(regs);
252 exception_exit(prev_state);
255 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
257 static const char str[] = "double fault";
258 struct task_struct *tsk = current;
261 /* Return not checked because double check cannot be ignored */
262 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
264 tsk->thread.error_code = error_code;
265 tsk->thread.trap_nr = X86_TRAP_DF;
267 #ifdef CONFIG_DOUBLEFAULT
268 df_debug(regs, error_code);
271 * This is always a kernel trap and never fixable (and thus must
275 die(str, regs, error_code);
279 dotraplinkage void __kprobes
280 do_general_protection(struct pt_regs *regs, long error_code)
282 struct task_struct *tsk;
283 enum ctx_state prev_state;
285 prev_state = exception_enter();
286 conditional_sti(regs);
289 if (regs->flags & X86_VM_MASK) {
291 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
297 if (!user_mode(regs)) {
298 if (fixup_exception(regs))
301 tsk->thread.error_code = error_code;
302 tsk->thread.trap_nr = X86_TRAP_GP;
303 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
304 X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
305 die("general protection fault", regs, error_code);
309 tsk->thread.error_code = error_code;
310 tsk->thread.trap_nr = X86_TRAP_GP;
312 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
313 printk_ratelimit()) {
314 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
315 tsk->comm, task_pid_nr(tsk),
316 regs->ip, regs->sp, error_code);
317 print_vma_addr(" in ", regs->ip);
321 force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
323 exception_exit(prev_state);
326 /* May run on IST stack. */
327 dotraplinkage void __kprobes notrace do_int3(struct pt_regs *regs, long error_code)
329 enum ctx_state prev_state;
331 #ifdef CONFIG_DYNAMIC_FTRACE
333 * ftrace must be first, everything else may cause a recursive crash.
334 * See note by declaration of modifying_ftrace_code in ftrace.c
336 if (unlikely(atomic_read(&modifying_ftrace_code)) &&
337 ftrace_int3_handler(regs))
340 if (poke_int3_handler(regs))
343 prev_state = exception_enter();
344 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
345 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
346 SIGTRAP) == NOTIFY_STOP)
348 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
350 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
351 SIGTRAP) == NOTIFY_STOP)
355 * Let others (NMI) know that the debug stack is in use
356 * as we may switch to the interrupt stack.
358 debug_stack_usage_inc();
359 preempt_conditional_sti(regs);
360 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
361 preempt_conditional_cli(regs);
362 debug_stack_usage_dec();
364 exception_exit(prev_state);
369 * Help handler running on IST stack to switch back to user stack
370 * for scheduling or signal handling. The actual stack switch is done in
373 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
375 struct pt_regs *regs = eregs;
376 /* Did already sync */
377 if (eregs == (struct pt_regs *)eregs->sp)
379 /* Exception from user space */
380 else if (user_mode(eregs))
381 regs = task_pt_regs(current);
383 * Exception from kernel and interrupts are enabled. Move to
384 * kernel process stack.
386 else if (eregs->flags & X86_EFLAGS_IF)
387 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
395 * Our handling of the processor debug registers is non-trivial.
396 * We do not clear them on entry and exit from the kernel. Therefore
397 * it is possible to get a watchpoint trap here from inside the kernel.
398 * However, the code in ./ptrace.c has ensured that the user can
399 * only set watchpoints on userspace addresses. Therefore the in-kernel
400 * watchpoint trap can only occur in code which is reading/writing
401 * from user space. Such code must not hold kernel locks (since it
402 * can equally take a page fault), therefore it is safe to call
403 * force_sig_info even though that claims and releases locks.
405 * Code in ./signal.c ensures that the debug control register
406 * is restored before we deliver any signal, and therefore that
407 * user code runs with the correct debug control register even though
410 * Being careful here means that we don't have to be as careful in a
411 * lot of more complicated places (task switching can be a bit lazy
412 * about restoring all the debug state, and ptrace doesn't have to
413 * find every occurrence of the TF bit that could be saved away even
416 * May run on IST stack.
418 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
420 struct task_struct *tsk = current;
421 enum ctx_state prev_state;
426 prev_state = exception_enter();
428 get_debugreg(dr6, 6);
430 /* Filter out all the reserved bits which are preset to 1 */
431 dr6 &= ~DR6_RESERVED;
434 * If dr6 has no reason to give us about the origin of this trap,
435 * then it's very likely the result of an icebp/int01 trap.
436 * User wants a sigtrap for that.
438 if (!dr6 && user_mode(regs))
441 /* Catch kmemcheck conditions first of all! */
442 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
445 /* DR6 may or may not be cleared by the CPU */
449 * The processor cleared BTF, so don't mark that we need it set.
451 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
453 /* Store the virtualized DR6 value */
454 tsk->thread.debugreg6 = dr6;
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);
500 * Note that we play around with the 'TS' bit in an attempt to get
501 * the correct behaviour even in the presence of the asynchronous
504 static void math_error(struct pt_regs *regs, int error_code, int trapnr)
506 struct task_struct *task = current;
509 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
512 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
514 conditional_sti(regs);
516 if (!user_mode_vm(regs))
518 if (!fixup_exception(regs)) {
519 task->thread.error_code = error_code;
520 task->thread.trap_nr = trapnr;
521 die(str, regs, error_code);
527 * Save the info for the exception handler and clear the error.
530 task->thread.trap_nr = trapnr;
531 task->thread.error_code = error_code;
532 info.si_signo = SIGFPE;
534 info.si_addr = (void __user *)regs->ip;
535 if (trapnr == X86_TRAP_MF) {
536 unsigned short cwd, swd;
538 * (~cwd & swd) will mask out exceptions that are not set to unmasked
539 * status. 0x3f is the exception bits in these regs, 0x200 is the
540 * C1 reg you need in case of a stack fault, 0x040 is the stack
541 * fault bit. We should only be taking one exception at a time,
542 * so if this combination doesn't produce any single exception,
543 * then we have a bad program that isn't synchronizing its FPU usage
544 * and it will suffer the consequences since we won't be able to
545 * fully reproduce the context of the exception
547 cwd = get_fpu_cwd(task);
548 swd = get_fpu_swd(task);
553 * The SIMD FPU exceptions are handled a little differently, as there
554 * is only a single status/control register. Thus, to determine which
555 * unmasked exception was caught we must mask the exception mask bits
556 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
558 unsigned short mxcsr = get_fpu_mxcsr(task);
559 err = ~(mxcsr >> 7) & mxcsr;
562 if (err & 0x001) { /* Invalid op */
564 * swd & 0x240 == 0x040: Stack Underflow
565 * swd & 0x240 == 0x240: Stack Overflow
566 * User must clear the SF bit (0x40) if set
568 info.si_code = FPE_FLTINV;
569 } else if (err & 0x004) { /* Divide by Zero */
570 info.si_code = FPE_FLTDIV;
571 } else if (err & 0x008) { /* Overflow */
572 info.si_code = FPE_FLTOVF;
573 } else if (err & 0x012) { /* Denormal, Underflow */
574 info.si_code = FPE_FLTUND;
575 } else if (err & 0x020) { /* Precision */
576 info.si_code = FPE_FLTRES;
579 * If we're using IRQ 13, or supposedly even some trap
580 * X86_TRAP_MF implementations, it's possible
581 * we get a spurious trap, which is not an error.
585 force_sig_info(SIGFPE, &info, task);
588 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
590 enum ctx_state prev_state;
592 prev_state = exception_enter();
593 math_error(regs, error_code, X86_TRAP_MF);
594 exception_exit(prev_state);
598 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
600 enum ctx_state prev_state;
602 prev_state = exception_enter();
603 math_error(regs, error_code, X86_TRAP_XF);
604 exception_exit(prev_state);
608 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
610 conditional_sti(regs);
612 /* No need to warn about this any longer. */
613 pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
617 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
621 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
626 * 'math_state_restore()' saves the current math information in the
627 * old math state array, and gets the new ones from the current task
629 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
630 * Don't touch unless you *really* know how it works.
632 * Must be called with kernel preemption disabled (eg with local
633 * local interrupts as in the case of do_device_not_available).
635 void math_state_restore(void)
637 struct task_struct *tsk = current;
639 if (!tsk_used_math(tsk)) {
642 * does a slab alloc which can sleep
648 do_group_exit(SIGKILL);
654 __thread_fpu_begin(tsk);
657 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
659 if (unlikely(restore_fpu_checking(tsk))) {
661 force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
665 tsk->thread.fpu_counter++;
667 EXPORT_SYMBOL_GPL(math_state_restore);
669 dotraplinkage void __kprobes
670 do_device_not_available(struct pt_regs *regs, long error_code)
672 enum ctx_state prev_state;
674 prev_state = exception_enter();
675 BUG_ON(use_eager_fpu());
677 #ifdef CONFIG_MATH_EMULATION
678 if (read_cr0() & X86_CR0_EM) {
679 struct math_emu_info info = { };
681 conditional_sti(regs);
685 exception_exit(prev_state);
689 math_state_restore(); /* interrupts still off */
691 conditional_sti(regs);
693 exception_exit(prev_state);
697 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
700 enum ctx_state prev_state;
702 prev_state = exception_enter();
705 info.si_signo = SIGILL;
707 info.si_code = ILL_BADSTK;
709 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
710 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
711 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
714 exception_exit(prev_state);
718 /* Set of traps needed for early debugging. */
719 void __init early_trap_init(void)
721 set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
722 /* int3 can be called from all */
723 set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
725 set_intr_gate(X86_TRAP_PF, page_fault);
727 load_idt(&idt_descr);
730 void __init early_trap_pf_init(void)
733 set_intr_gate(X86_TRAP_PF, page_fault);
737 void __init trap_init(void)
742 void __iomem *p = early_ioremap(0x0FFFD9, 4);
744 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
749 set_intr_gate(X86_TRAP_DE, divide_error);
750 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
751 /* int4 can be called from all */
752 set_system_intr_gate(X86_TRAP_OF, &overflow);
753 set_intr_gate(X86_TRAP_BR, bounds);
754 set_intr_gate(X86_TRAP_UD, invalid_op);
755 set_intr_gate(X86_TRAP_NM, device_not_available);
757 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
759 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
761 set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
762 set_intr_gate(X86_TRAP_TS, invalid_TSS);
763 set_intr_gate(X86_TRAP_NP, segment_not_present);
764 set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
765 set_intr_gate(X86_TRAP_GP, general_protection);
766 set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
767 set_intr_gate(X86_TRAP_MF, coprocessor_error);
768 set_intr_gate(X86_TRAP_AC, alignment_check);
769 #ifdef CONFIG_X86_MCE
770 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
772 set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
774 /* Reserve all the builtin and the syscall vector: */
775 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
776 set_bit(i, used_vectors);
778 #ifdef CONFIG_IA32_EMULATION
779 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
780 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
784 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
785 set_bit(SYSCALL_VECTOR, used_vectors);
789 * Set the IDT descriptor to a fixed read-only location, so that the
790 * "sidt" instruction will not leak the location of the kernel, and
791 * to defend the IDT against arbitrary memory write vulnerabilities.
792 * It will be reloaded in cpu_init() */
793 __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
794 idt_descr.address = fix_to_virt(FIX_RO_IDT);
797 * Should be a barrier for any external CPU state:
801 x86_init.irqs.trap_init();
804 memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
805 set_nmi_gate(X86_TRAP_DB, &debug);
806 set_nmi_gate(X86_TRAP_BP, &int3);