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>
63 #include <asm/x86_init.h>
64 #include <asm/pgalloc.h>
65 #include <asm/proto.h>
67 #include <asm/processor-flags.h>
68 #include <asm/setup.h>
70 asmlinkage int system_call(void);
73 * The IDT has to be page-aligned to simplify the Pentium
74 * F0 0F bug workaround.
76 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
79 DECLARE_BITMAP(used_vectors, NR_VECTORS);
80 EXPORT_SYMBOL_GPL(used_vectors);
82 static inline void conditional_sti(struct pt_regs *regs)
84 if (regs->flags & X86_EFLAGS_IF)
88 static inline void preempt_conditional_sti(struct pt_regs *regs)
91 if (regs->flags & X86_EFLAGS_IF)
95 static inline void conditional_cli(struct pt_regs *regs)
97 if (regs->flags & X86_EFLAGS_IF)
101 static inline void preempt_conditional_cli(struct pt_regs *regs)
103 if (regs->flags & X86_EFLAGS_IF)
109 do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
110 struct pt_regs *regs, long error_code)
113 if (regs->flags & X86_VM_MASK) {
115 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
116 * On nmi (interrupt 2), do_trap should not be called.
118 if (trapnr < X86_TRAP_UD) {
119 if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
126 if (!user_mode(regs)) {
127 if (!fixup_exception(regs)) {
128 tsk->thread.error_code = error_code;
129 tsk->thread.trap_nr = trapnr;
130 die(str, regs, error_code);
138 static void __kprobes
139 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
140 long error_code, siginfo_t *info)
142 struct task_struct *tsk = current;
145 if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
148 * We want error_code and trap_nr set for userspace faults and
149 * kernelspace faults which result in die(), but not
150 * kernelspace faults which are fixed up. die() gives the
151 * process no chance to handle the signal and notice the
152 * kernel fault information, so that won't result in polluting
153 * the information about previously queued, but not yet
154 * delivered, faults. See also do_general_protection below.
156 tsk->thread.error_code = error_code;
157 tsk->thread.trap_nr = trapnr;
160 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
161 printk_ratelimit()) {
162 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
163 tsk->comm, tsk->pid, str,
164 regs->ip, regs->sp, error_code);
165 print_vma_addr(" in ", regs->ip);
171 force_sig_info(signr, info, tsk);
173 force_sig(signr, tsk);
176 #define DO_ERROR(trapnr, signr, str, name) \
177 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
179 enum ctx_state prev_state; \
181 prev_state = exception_enter(); \
182 if (notify_die(DIE_TRAP, str, regs, error_code, \
183 trapnr, signr) == NOTIFY_STOP) { \
184 exception_exit(prev_state); \
187 conditional_sti(regs); \
188 do_trap(trapnr, signr, str, regs, error_code, NULL); \
189 exception_exit(prev_state); \
192 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
193 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
196 enum ctx_state prev_state; \
198 info.si_signo = signr; \
200 info.si_code = sicode; \
201 info.si_addr = (void __user *)siaddr; \
202 prev_state = exception_enter(); \
203 if (notify_die(DIE_TRAP, str, regs, error_code, \
204 trapnr, signr) == NOTIFY_STOP) { \
205 exception_exit(prev_state); \
208 conditional_sti(regs); \
209 do_trap(trapnr, signr, str, regs, error_code, &info); \
210 exception_exit(prev_state); \
213 DO_ERROR_INFO(X86_TRAP_DE, SIGFPE, "divide error", divide_error, FPE_INTDIV,
215 DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
216 DO_ERROR(X86_TRAP_BR, SIGSEGV, "bounds", bounds)
217 DO_ERROR_INFO(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN,
219 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",
220 coprocessor_segment_overrun)
221 DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
222 DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
224 DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
226 DO_ERROR_INFO(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check,
230 /* Runs on IST stack */
231 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
233 enum ctx_state prev_state;
235 prev_state = exception_enter();
236 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
237 X86_TRAP_SS, SIGBUS) != NOTIFY_STOP) {
238 preempt_conditional_sti(regs);
239 do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
240 preempt_conditional_cli(regs);
242 exception_exit(prev_state);
245 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
247 static const char str[] = "double fault";
248 struct task_struct *tsk = current;
251 /* Return not checked because double check cannot be ignored */
252 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
254 tsk->thread.error_code = error_code;
255 tsk->thread.trap_nr = X86_TRAP_DF;
257 #ifdef CONFIG_DOUBLEFAULT
258 df_debug(regs, error_code);
261 * This is always a kernel trap and never fixable (and thus must
265 die(str, regs, error_code);
269 dotraplinkage void __kprobes
270 do_general_protection(struct pt_regs *regs, long error_code)
272 struct task_struct *tsk;
273 enum ctx_state prev_state;
275 prev_state = exception_enter();
276 conditional_sti(regs);
279 if (regs->flags & X86_VM_MASK) {
281 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
287 if (!user_mode(regs)) {
288 if (fixup_exception(regs))
291 tsk->thread.error_code = error_code;
292 tsk->thread.trap_nr = X86_TRAP_GP;
293 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
294 X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
295 die("general protection fault", regs, error_code);
299 tsk->thread.error_code = error_code;
300 tsk->thread.trap_nr = X86_TRAP_GP;
302 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
303 printk_ratelimit()) {
304 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
305 tsk->comm, task_pid_nr(tsk),
306 regs->ip, regs->sp, error_code);
307 print_vma_addr(" in ", regs->ip);
311 force_sig(SIGSEGV, tsk);
313 exception_exit(prev_state);
316 /* May run on IST stack. */
317 dotraplinkage void __kprobes notrace do_int3(struct pt_regs *regs, long error_code)
319 enum ctx_state prev_state;
321 #ifdef CONFIG_DYNAMIC_FTRACE
323 * ftrace must be first, everything else may cause a recursive crash.
324 * See note by declaration of modifying_ftrace_code in ftrace.c
326 if (unlikely(atomic_read(&modifying_ftrace_code)) &&
327 ftrace_int3_handler(regs))
330 prev_state = exception_enter();
331 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
332 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
333 SIGTRAP) == NOTIFY_STOP)
335 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
337 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
338 SIGTRAP) == NOTIFY_STOP)
342 * Let others (NMI) know that the debug stack is in use
343 * as we may switch to the interrupt stack.
345 debug_stack_usage_inc();
346 preempt_conditional_sti(regs);
347 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
348 preempt_conditional_cli(regs);
349 debug_stack_usage_dec();
351 exception_exit(prev_state);
356 * Help handler running on IST stack to switch back to user stack
357 * for scheduling or signal handling. The actual stack switch is done in
360 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
362 struct pt_regs *regs = eregs;
363 /* Did already sync */
364 if (eregs == (struct pt_regs *)eregs->sp)
366 /* Exception from user space */
367 else if (user_mode(eregs))
368 regs = task_pt_regs(current);
370 * Exception from kernel and interrupts are enabled. Move to
371 * kernel process stack.
373 else if (eregs->flags & X86_EFLAGS_IF)
374 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
382 * Our handling of the processor debug registers is non-trivial.
383 * We do not clear them on entry and exit from the kernel. Therefore
384 * it is possible to get a watchpoint trap here from inside the kernel.
385 * However, the code in ./ptrace.c has ensured that the user can
386 * only set watchpoints on userspace addresses. Therefore the in-kernel
387 * watchpoint trap can only occur in code which is reading/writing
388 * from user space. Such code must not hold kernel locks (since it
389 * can equally take a page fault), therefore it is safe to call
390 * force_sig_info even though that claims and releases locks.
392 * Code in ./signal.c ensures that the debug control register
393 * is restored before we deliver any signal, and therefore that
394 * user code runs with the correct debug control register even though
397 * Being careful here means that we don't have to be as careful in a
398 * lot of more complicated places (task switching can be a bit lazy
399 * about restoring all the debug state, and ptrace doesn't have to
400 * find every occurrence of the TF bit that could be saved away even
403 * May run on IST stack.
405 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
407 struct task_struct *tsk = current;
408 enum ctx_state prev_state;
413 prev_state = exception_enter();
415 get_debugreg(dr6, 6);
417 /* Filter out all the reserved bits which are preset to 1 */
418 dr6 &= ~DR6_RESERVED;
421 * If dr6 has no reason to give us about the origin of this trap,
422 * then it's very likely the result of an icebp/int01 trap.
423 * User wants a sigtrap for that.
425 if (!dr6 && user_mode(regs))
428 /* Catch kmemcheck conditions first of all! */
429 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
432 /* DR6 may or may not be cleared by the CPU */
436 * The processor cleared BTF, so don't mark that we need it set.
438 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
440 /* Store the virtualized DR6 value */
441 tsk->thread.debugreg6 = dr6;
443 if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
444 SIGTRAP) == NOTIFY_STOP)
448 * Let others (NMI) know that the debug stack is in use
449 * as we may switch to the interrupt stack.
451 debug_stack_usage_inc();
453 /* It's safe to allow irq's after DR6 has been saved */
454 preempt_conditional_sti(regs);
456 if (regs->flags & X86_VM_MASK) {
457 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
459 preempt_conditional_cli(regs);
460 debug_stack_usage_dec();
465 * Single-stepping through system calls: ignore any exceptions in
466 * kernel space, but re-enable TF when returning to user mode.
468 * We already checked v86 mode above, so we can check for kernel mode
469 * by just checking the CPL of CS.
471 if ((dr6 & DR_STEP) && !user_mode(regs)) {
472 tsk->thread.debugreg6 &= ~DR_STEP;
473 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
474 regs->flags &= ~X86_EFLAGS_TF;
476 si_code = get_si_code(tsk->thread.debugreg6);
477 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
478 send_sigtrap(tsk, regs, error_code, si_code);
479 preempt_conditional_cli(regs);
480 debug_stack_usage_dec();
483 exception_exit(prev_state);
487 * Note that we play around with the 'TS' bit in an attempt to get
488 * the correct behaviour even in the presence of the asynchronous
491 void math_error(struct pt_regs *regs, int error_code, int trapnr)
493 struct task_struct *task = current;
496 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
499 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
501 conditional_sti(regs);
503 if (!user_mode_vm(regs))
505 if (!fixup_exception(regs)) {
506 task->thread.error_code = error_code;
507 task->thread.trap_nr = trapnr;
508 die(str, regs, error_code);
514 * Save the info for the exception handler and clear the error.
517 task->thread.trap_nr = trapnr;
518 task->thread.error_code = error_code;
519 info.si_signo = SIGFPE;
521 info.si_addr = (void __user *)regs->ip;
522 if (trapnr == X86_TRAP_MF) {
523 unsigned short cwd, swd;
525 * (~cwd & swd) will mask out exceptions that are not set to unmasked
526 * status. 0x3f is the exception bits in these regs, 0x200 is the
527 * C1 reg you need in case of a stack fault, 0x040 is the stack
528 * fault bit. We should only be taking one exception at a time,
529 * so if this combination doesn't produce any single exception,
530 * then we have a bad program that isn't synchronizing its FPU usage
531 * and it will suffer the consequences since we won't be able to
532 * fully reproduce the context of the exception
534 cwd = get_fpu_cwd(task);
535 swd = get_fpu_swd(task);
540 * The SIMD FPU exceptions are handled a little differently, as there
541 * is only a single status/control register. Thus, to determine which
542 * unmasked exception was caught we must mask the exception mask bits
543 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
545 unsigned short mxcsr = get_fpu_mxcsr(task);
546 err = ~(mxcsr >> 7) & mxcsr;
549 if (err & 0x001) { /* Invalid op */
551 * swd & 0x240 == 0x040: Stack Underflow
552 * swd & 0x240 == 0x240: Stack Overflow
553 * User must clear the SF bit (0x40) if set
555 info.si_code = FPE_FLTINV;
556 } else if (err & 0x004) { /* Divide by Zero */
557 info.si_code = FPE_FLTDIV;
558 } else if (err & 0x008) { /* Overflow */
559 info.si_code = FPE_FLTOVF;
560 } else if (err & 0x012) { /* Denormal, Underflow */
561 info.si_code = FPE_FLTUND;
562 } else if (err & 0x020) { /* Precision */
563 info.si_code = FPE_FLTRES;
566 * If we're using IRQ 13, or supposedly even some trap
567 * X86_TRAP_MF implementations, it's possible
568 * we get a spurious trap, which is not an error.
572 force_sig_info(SIGFPE, &info, task);
575 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
577 enum ctx_state prev_state;
579 prev_state = exception_enter();
580 math_error(regs, error_code, X86_TRAP_MF);
581 exception_exit(prev_state);
585 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
587 enum ctx_state prev_state;
589 prev_state = exception_enter();
590 math_error(regs, error_code, X86_TRAP_XF);
591 exception_exit(prev_state);
595 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
597 conditional_sti(regs);
599 /* No need to warn about this any longer. */
600 pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
604 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
608 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
613 * 'math_state_restore()' saves the current math information in the
614 * old math state array, and gets the new ones from the current task
616 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
617 * Don't touch unless you *really* know how it works.
619 * Must be called with kernel preemption disabled (eg with local
620 * local interrupts as in the case of do_device_not_available).
622 void math_state_restore(void)
624 struct task_struct *tsk = current;
626 if (!tsk_used_math(tsk)) {
629 * does a slab alloc which can sleep
635 do_group_exit(SIGKILL);
641 __thread_fpu_begin(tsk);
644 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
646 if (unlikely(restore_fpu_checking(tsk))) {
648 force_sig(SIGSEGV, tsk);
654 EXPORT_SYMBOL_GPL(math_state_restore);
656 dotraplinkage void __kprobes
657 do_device_not_available(struct pt_regs *regs, long error_code)
659 enum ctx_state prev_state;
661 prev_state = exception_enter();
662 BUG_ON(use_eager_fpu());
664 #ifdef CONFIG_MATH_EMULATION
665 if (read_cr0() & X86_CR0_EM) {
666 struct math_emu_info info = { };
668 conditional_sti(regs);
672 exception_exit(prev_state);
676 math_state_restore(); /* interrupts still off */
678 conditional_sti(regs);
680 exception_exit(prev_state);
684 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
687 enum ctx_state prev_state;
689 prev_state = exception_enter();
692 info.si_signo = SIGILL;
694 info.si_code = ILL_BADSTK;
696 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
697 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
698 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
701 exception_exit(prev_state);
705 /* Set of traps needed for early debugging. */
706 void __init early_trap_init(void)
708 set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
709 /* int3 can be called from all */
710 set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
712 set_intr_gate(X86_TRAP_PF, &page_fault);
714 load_idt(&idt_descr);
717 void __init early_trap_pf_init(void)
720 set_intr_gate(X86_TRAP_PF, &page_fault);
724 void __init trap_init(void)
729 void __iomem *p = early_ioremap(0x0FFFD9, 4);
731 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
736 set_intr_gate(X86_TRAP_DE, ÷_error);
737 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
738 /* int4 can be called from all */
739 set_system_intr_gate(X86_TRAP_OF, &overflow);
740 set_intr_gate(X86_TRAP_BR, &bounds);
741 set_intr_gate(X86_TRAP_UD, &invalid_op);
742 set_intr_gate(X86_TRAP_NM, &device_not_available);
744 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
746 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
748 set_intr_gate(X86_TRAP_OLD_MF, &coprocessor_segment_overrun);
749 set_intr_gate(X86_TRAP_TS, &invalid_TSS);
750 set_intr_gate(X86_TRAP_NP, &segment_not_present);
751 set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
752 set_intr_gate(X86_TRAP_GP, &general_protection);
753 set_intr_gate(X86_TRAP_SPURIOUS, &spurious_interrupt_bug);
754 set_intr_gate(X86_TRAP_MF, &coprocessor_error);
755 set_intr_gate(X86_TRAP_AC, &alignment_check);
756 #ifdef CONFIG_X86_MCE
757 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
759 set_intr_gate(X86_TRAP_XF, &simd_coprocessor_error);
761 /* Reserve all the builtin and the syscall vector: */
762 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
763 set_bit(i, used_vectors);
765 #ifdef CONFIG_IA32_EMULATION
766 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
767 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
771 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
772 set_bit(SYSCALL_VECTOR, used_vectors);
776 * Set the IDT descriptor to a fixed read-only location, so that the
777 * "sidt" instruction will not leak the location of the kernel, and
778 * to defend the IDT against arbitrary memory write vulnerabilities.
779 * It will be reloaded in cpu_init() */
780 __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
781 idt_descr.address = fix_to_virt(FIX_RO_IDT);
784 * Should be a barrier for any external CPU state:
788 x86_init.irqs.trap_init();
791 memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
792 set_nmi_gate(X86_TRAP_DB, &debug);
793 set_nmi_gate(X86_TRAP_BP, &int3);