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 void __kprobes
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);
177 #define DO_ERROR(trapnr, signr, str, name) \
178 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
180 enum ctx_state prev_state; \
182 prev_state = exception_enter(); \
183 if (notify_die(DIE_TRAP, str, regs, error_code, \
184 trapnr, signr) == NOTIFY_STOP) { \
185 exception_exit(prev_state); \
188 conditional_sti(regs); \
189 do_trap(trapnr, signr, str, regs, error_code, NULL); \
190 exception_exit(prev_state); \
193 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
194 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
197 enum ctx_state prev_state; \
199 info.si_signo = signr; \
201 info.si_code = sicode; \
202 info.si_addr = (void __user *)siaddr; \
203 prev_state = exception_enter(); \
204 if (notify_die(DIE_TRAP, str, regs, error_code, \
205 trapnr, signr) == NOTIFY_STOP) { \
206 exception_exit(prev_state); \
209 conditional_sti(regs); \
210 do_trap(trapnr, signr, str, regs, error_code, &info); \
211 exception_exit(prev_state); \
214 DO_ERROR_INFO(X86_TRAP_DE, SIGFPE, "divide error", divide_error, FPE_INTDIV,
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,
220 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",
221 coprocessor_segment_overrun)
222 DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
223 DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
225 DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
227 DO_ERROR_INFO(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check,
231 /* Runs on IST stack */
232 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
234 enum ctx_state prev_state;
236 prev_state = exception_enter();
237 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
238 X86_TRAP_SS, SIGBUS) != NOTIFY_STOP) {
239 preempt_conditional_sti(regs);
240 do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
241 preempt_conditional_cli(regs);
243 exception_exit(prev_state);
246 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
248 static const char str[] = "double fault";
249 struct task_struct *tsk = current;
252 /* Return not checked because double check cannot be ignored */
253 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
255 tsk->thread.error_code = error_code;
256 tsk->thread.trap_nr = X86_TRAP_DF;
258 #ifdef CONFIG_DOUBLEFAULT
259 df_debug(regs, error_code);
262 * This is always a kernel trap and never fixable (and thus must
266 die(str, regs, error_code);
270 dotraplinkage void __kprobes
271 do_general_protection(struct pt_regs *regs, long error_code)
273 struct task_struct *tsk;
274 enum ctx_state prev_state;
276 prev_state = exception_enter();
277 conditional_sti(regs);
280 if (regs->flags & X86_VM_MASK) {
282 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
288 if (!user_mode(regs)) {
289 if (fixup_exception(regs))
292 tsk->thread.error_code = error_code;
293 tsk->thread.trap_nr = X86_TRAP_GP;
294 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
295 X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
296 die("general protection fault", regs, error_code);
300 tsk->thread.error_code = error_code;
301 tsk->thread.trap_nr = X86_TRAP_GP;
303 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
304 printk_ratelimit()) {
305 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
306 tsk->comm, task_pid_nr(tsk),
307 regs->ip, regs->sp, error_code);
308 print_vma_addr(" in ", regs->ip);
312 force_sig(SIGSEGV, tsk);
314 exception_exit(prev_state);
317 /* May run on IST stack. */
318 dotraplinkage void __kprobes notrace do_int3(struct pt_regs *regs, long error_code)
320 enum ctx_state prev_state;
322 #ifdef CONFIG_DYNAMIC_FTRACE
324 * ftrace must be first, everything else may cause a recursive crash.
325 * See note by declaration of modifying_ftrace_code in ftrace.c
327 if (unlikely(atomic_read(&modifying_ftrace_code)) &&
328 ftrace_int3_handler(regs))
331 if (poke_int3_handler(regs))
334 prev_state = exception_enter();
335 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
336 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
337 SIGTRAP) == NOTIFY_STOP)
339 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
341 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
342 SIGTRAP) == NOTIFY_STOP)
346 * Let others (NMI) know that the debug stack is in use
347 * as we may switch to the interrupt stack.
349 debug_stack_usage_inc();
350 preempt_conditional_sti(regs);
351 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
352 preempt_conditional_cli(regs);
353 debug_stack_usage_dec();
355 exception_exit(prev_state);
360 * Help handler running on IST stack to switch back to user stack
361 * for scheduling or signal handling. The actual stack switch is done in
364 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
366 struct pt_regs *regs = eregs;
367 /* Did already sync */
368 if (eregs == (struct pt_regs *)eregs->sp)
370 /* Exception from user space */
371 else if (user_mode(eregs))
372 regs = task_pt_regs(current);
374 * Exception from kernel and interrupts are enabled. Move to
375 * kernel process stack.
377 else if (eregs->flags & X86_EFLAGS_IF)
378 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
386 * Our handling of the processor debug registers is non-trivial.
387 * We do not clear them on entry and exit from the kernel. Therefore
388 * it is possible to get a watchpoint trap here from inside the kernel.
389 * However, the code in ./ptrace.c has ensured that the user can
390 * only set watchpoints on userspace addresses. Therefore the in-kernel
391 * watchpoint trap can only occur in code which is reading/writing
392 * from user space. Such code must not hold kernel locks (since it
393 * can equally take a page fault), therefore it is safe to call
394 * force_sig_info even though that claims and releases locks.
396 * Code in ./signal.c ensures that the debug control register
397 * is restored before we deliver any signal, and therefore that
398 * user code runs with the correct debug control register even though
401 * Being careful here means that we don't have to be as careful in a
402 * lot of more complicated places (task switching can be a bit lazy
403 * about restoring all the debug state, and ptrace doesn't have to
404 * find every occurrence of the TF bit that could be saved away even
407 * May run on IST stack.
409 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
411 struct task_struct *tsk = current;
412 enum ctx_state prev_state;
417 prev_state = exception_enter();
419 get_debugreg(dr6, 6);
421 /* Filter out all the reserved bits which are preset to 1 */
422 dr6 &= ~DR6_RESERVED;
425 * If dr6 has no reason to give us about the origin of this trap,
426 * then it's very likely the result of an icebp/int01 trap.
427 * User wants a sigtrap for that.
429 if (!dr6 && user_mode(regs))
432 /* Catch kmemcheck conditions first of all! */
433 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
436 /* DR6 may or may not be cleared by the CPU */
440 * The processor cleared BTF, so don't mark that we need it set.
442 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
444 /* Store the virtualized DR6 value */
445 tsk->thread.debugreg6 = dr6;
447 if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
448 SIGTRAP) == NOTIFY_STOP)
452 * Let others (NMI) know that the debug stack is in use
453 * as we may switch to the interrupt stack.
455 debug_stack_usage_inc();
457 /* It's safe to allow irq's after DR6 has been saved */
458 preempt_conditional_sti(regs);
460 if (regs->flags & X86_VM_MASK) {
461 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
463 preempt_conditional_cli(regs);
464 debug_stack_usage_dec();
469 * Single-stepping through system calls: ignore any exceptions in
470 * kernel space, but re-enable TF when returning to user mode.
472 * We already checked v86 mode above, so we can check for kernel mode
473 * by just checking the CPL of CS.
475 if ((dr6 & DR_STEP) && !user_mode(regs)) {
476 tsk->thread.debugreg6 &= ~DR_STEP;
477 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
478 regs->flags &= ~X86_EFLAGS_TF;
480 si_code = get_si_code(tsk->thread.debugreg6);
481 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
482 send_sigtrap(tsk, regs, error_code, si_code);
483 preempt_conditional_cli(regs);
484 debug_stack_usage_dec();
487 exception_exit(prev_state);
491 * Note that we play around with the 'TS' bit in an attempt to get
492 * the correct behaviour even in the presence of the asynchronous
495 void math_error(struct pt_regs *regs, int error_code, int trapnr)
497 struct task_struct *task = current;
500 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
503 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
505 conditional_sti(regs);
507 if (!user_mode_vm(regs))
509 if (!fixup_exception(regs)) {
510 task->thread.error_code = error_code;
511 task->thread.trap_nr = trapnr;
512 die(str, regs, error_code);
518 * Save the info for the exception handler and clear the error.
521 task->thread.trap_nr = trapnr;
522 task->thread.error_code = error_code;
523 info.si_signo = SIGFPE;
525 info.si_addr = (void __user *)regs->ip;
526 if (trapnr == X86_TRAP_MF) {
527 unsigned short cwd, swd;
529 * (~cwd & swd) will mask out exceptions that are not set to unmasked
530 * status. 0x3f is the exception bits in these regs, 0x200 is the
531 * C1 reg you need in case of a stack fault, 0x040 is the stack
532 * fault bit. We should only be taking one exception at a time,
533 * so if this combination doesn't produce any single exception,
534 * then we have a bad program that isn't synchronizing its FPU usage
535 * and it will suffer the consequences since we won't be able to
536 * fully reproduce the context of the exception
538 cwd = get_fpu_cwd(task);
539 swd = get_fpu_swd(task);
544 * The SIMD FPU exceptions are handled a little differently, as there
545 * is only a single status/control register. Thus, to determine which
546 * unmasked exception was caught we must mask the exception mask bits
547 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
549 unsigned short mxcsr = get_fpu_mxcsr(task);
550 err = ~(mxcsr >> 7) & mxcsr;
553 if (err & 0x001) { /* Invalid op */
555 * swd & 0x240 == 0x040: Stack Underflow
556 * swd & 0x240 == 0x240: Stack Overflow
557 * User must clear the SF bit (0x40) if set
559 info.si_code = FPE_FLTINV;
560 } else if (err & 0x004) { /* Divide by Zero */
561 info.si_code = FPE_FLTDIV;
562 } else if (err & 0x008) { /* Overflow */
563 info.si_code = FPE_FLTOVF;
564 } else if (err & 0x012) { /* Denormal, Underflow */
565 info.si_code = FPE_FLTUND;
566 } else if (err & 0x020) { /* Precision */
567 info.si_code = FPE_FLTRES;
570 * If we're using IRQ 13, or supposedly even some trap
571 * X86_TRAP_MF implementations, it's possible
572 * we get a spurious trap, which is not an error.
576 force_sig_info(SIGFPE, &info, task);
579 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
581 enum ctx_state prev_state;
583 prev_state = exception_enter();
584 math_error(regs, error_code, X86_TRAP_MF);
585 exception_exit(prev_state);
589 do_simd_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_XF);
595 exception_exit(prev_state);
599 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
601 conditional_sti(regs);
603 /* No need to warn about this any longer. */
604 pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
608 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
612 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
617 * 'math_state_restore()' saves the current math information in the
618 * old math state array, and gets the new ones from the current task
620 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
621 * Don't touch unless you *really* know how it works.
623 * Must be called with kernel preemption disabled (eg with local
624 * local interrupts as in the case of do_device_not_available).
626 void math_state_restore(void)
628 struct task_struct *tsk = current;
630 if (!tsk_used_math(tsk)) {
633 * does a slab alloc which can sleep
639 do_group_exit(SIGKILL);
645 __thread_fpu_begin(tsk);
648 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
650 if (unlikely(restore_fpu_checking(tsk))) {
652 force_sig(SIGSEGV, tsk);
658 EXPORT_SYMBOL_GPL(math_state_restore);
660 dotraplinkage void __kprobes
661 do_device_not_available(struct pt_regs *regs, long error_code)
663 enum ctx_state prev_state;
665 prev_state = exception_enter();
666 BUG_ON(use_eager_fpu());
668 #ifdef CONFIG_MATH_EMULATION
669 if (read_cr0() & X86_CR0_EM) {
670 struct math_emu_info info = { };
672 conditional_sti(regs);
676 exception_exit(prev_state);
680 math_state_restore(); /* interrupts still off */
682 conditional_sti(regs);
684 exception_exit(prev_state);
688 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
691 enum ctx_state prev_state;
693 prev_state = exception_enter();
696 info.si_signo = SIGILL;
698 info.si_code = ILL_BADSTK;
700 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
701 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
702 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
705 exception_exit(prev_state);
709 /* Set of traps needed for early debugging. */
710 void __init early_trap_init(void)
712 set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
713 /* int3 can be called from all */
714 set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
716 set_intr_gate(X86_TRAP_PF, &page_fault);
718 load_idt(&idt_descr);
721 void __init early_trap_pf_init(void)
724 set_intr_gate(X86_TRAP_PF, &page_fault);
728 void __init trap_init(void)
733 void __iomem *p = early_ioremap(0x0FFFD9, 4);
735 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
740 set_intr_gate(X86_TRAP_DE, ÷_error);
741 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
742 /* int4 can be called from all */
743 set_system_intr_gate(X86_TRAP_OF, &overflow);
744 set_intr_gate(X86_TRAP_BR, &bounds);
745 set_intr_gate(X86_TRAP_UD, &invalid_op);
746 set_intr_gate(X86_TRAP_NM, &device_not_available);
748 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
750 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
752 set_intr_gate(X86_TRAP_OLD_MF, &coprocessor_segment_overrun);
753 set_intr_gate(X86_TRAP_TS, &invalid_TSS);
754 set_intr_gate(X86_TRAP_NP, &segment_not_present);
755 set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
756 set_intr_gate(X86_TRAP_GP, &general_protection);
757 set_intr_gate(X86_TRAP_SPURIOUS, &spurious_interrupt_bug);
758 set_intr_gate(X86_TRAP_MF, &coprocessor_error);
759 set_intr_gate(X86_TRAP_AC, &alignment_check);
760 #ifdef CONFIG_X86_MCE
761 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
763 set_intr_gate(X86_TRAP_XF, &simd_coprocessor_error);
765 /* Reserve all the builtin and the syscall vector: */
766 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
767 set_bit(i, used_vectors);
769 #ifdef CONFIG_IA32_EMULATION
770 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
771 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
775 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
776 set_bit(SYSCALL_VECTOR, used_vectors);
780 * Set the IDT descriptor to a fixed read-only location, so that the
781 * "sidt" instruction will not leak the location of the kernel, and
782 * to defend the IDT against arbitrary memory write vulnerabilities.
783 * It will be reloaded in cpu_init() */
784 __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
785 idt_descr.address = fix_to_virt(FIX_RO_IDT);
788 * Should be a barrier for any external CPU state:
792 x86_init.irqs.trap_init();
795 memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
796 set_nmi_gate(X86_TRAP_DB, &debug);
797 set_nmi_gate(X86_TRAP_BP, &int3);