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/interrupt.h>
16 #include <linux/kallsyms.h>
17 #include <linux/spinlock.h>
18 #include <linux/kprobes.h>
19 #include <linux/uaccess.h>
20 #include <linux/kdebug.h>
21 #include <linux/kgdb.h>
22 #include <linux/kernel.h>
23 #include <linux/module.h>
24 #include <linux/ptrace.h>
25 #include <linux/string.h>
26 #include <linux/delay.h>
27 #include <linux/errno.h>
28 #include <linux/kexec.h>
29 #include <linux/sched.h>
30 #include <linux/timer.h>
31 #include <linux/init.h>
32 #include <linux/bug.h>
33 #include <linux/nmi.h>
35 #include <linux/smp.h>
39 #include <linux/ioport.h>
40 #include <linux/eisa.h>
43 #if defined(CONFIG_EDAC)
44 #include <linux/edac.h>
47 #include <asm/kmemcheck.h>
48 #include <asm/stacktrace.h>
49 #include <asm/processor.h>
50 #include <asm/debugreg.h>
51 #include <linux/atomic.h>
52 #include <asm/ftrace.h>
53 #include <asm/traps.h>
56 #include <asm/fpu-internal.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);
72 /* Do we ignore FPU interrupts ? */
76 * The IDT has to be page-aligned to simplify the Pentium
77 * F0 0F bug workaround.
79 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
82 DECLARE_BITMAP(used_vectors, NR_VECTORS);
83 EXPORT_SYMBOL_GPL(used_vectors);
85 static inline void conditional_sti(struct pt_regs *regs)
87 if (regs->flags & X86_EFLAGS_IF)
91 static inline void preempt_conditional_sti(struct pt_regs *regs)
94 if (regs->flags & X86_EFLAGS_IF)
98 static inline void conditional_cli(struct pt_regs *regs)
100 if (regs->flags & X86_EFLAGS_IF)
104 static inline void preempt_conditional_cli(struct pt_regs *regs)
106 if (regs->flags & X86_EFLAGS_IF)
112 do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
113 struct pt_regs *regs, long error_code)
116 if (regs->flags & X86_VM_MASK) {
118 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
119 * On nmi (interrupt 2), do_trap should not be called.
121 if (trapnr < X86_TRAP_UD) {
122 if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
129 if (!user_mode(regs)) {
130 if (!fixup_exception(regs)) {
131 tsk->thread.error_code = error_code;
132 tsk->thread.trap_nr = trapnr;
133 die(str, regs, error_code);
141 static void __kprobes
142 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
143 long error_code, siginfo_t *info)
145 struct task_struct *tsk = current;
148 if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
151 * We want error_code and trap_nr set for userspace faults and
152 * kernelspace faults which result in die(), but not
153 * kernelspace faults which are fixed up. die() gives the
154 * process no chance to handle the signal and notice the
155 * kernel fault information, so that won't result in polluting
156 * the information about previously queued, but not yet
157 * delivered, faults. See also do_general_protection below.
159 tsk->thread.error_code = error_code;
160 tsk->thread.trap_nr = trapnr;
163 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
164 printk_ratelimit()) {
165 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
166 tsk->comm, tsk->pid, str,
167 regs->ip, regs->sp, error_code);
168 print_vma_addr(" in ", regs->ip);
174 force_sig_info(signr, info, tsk);
176 force_sig(signr, tsk);
179 #define DO_ERROR(trapnr, signr, str, name) \
180 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
182 exception_enter(regs); \
183 if (notify_die(DIE_TRAP, str, regs, error_code, \
184 trapnr, signr) == NOTIFY_STOP) { \
185 exception_exit(regs); \
188 conditional_sti(regs); \
189 do_trap(trapnr, signr, str, regs, error_code, NULL); \
190 exception_exit(regs); \
193 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
194 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
197 info.si_signo = signr; \
199 info.si_code = sicode; \
200 info.si_addr = (void __user *)siaddr; \
201 exception_enter(regs); \
202 if (notify_die(DIE_TRAP, str, regs, error_code, \
203 trapnr, signr) == NOTIFY_STOP) { \
204 exception_exit(regs); \
207 conditional_sti(regs); \
208 do_trap(trapnr, signr, str, regs, error_code, &info); \
209 exception_exit(regs); \
212 DO_ERROR_INFO(X86_TRAP_DE, SIGFPE, "divide error", divide_error, FPE_INTDIV,
214 DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
215 DO_ERROR(X86_TRAP_BR, SIGSEGV, "bounds", bounds)
216 DO_ERROR_INFO(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN,
218 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",
219 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,
229 /* Runs on IST stack */
230 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
232 exception_enter(regs);
233 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
234 X86_TRAP_SS, SIGBUS) != NOTIFY_STOP) {
235 preempt_conditional_sti(regs);
236 do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
237 preempt_conditional_cli(regs);
239 exception_exit(regs);
242 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
244 static const char str[] = "double fault";
245 struct task_struct *tsk = current;
247 exception_enter(regs);
248 /* Return not checked because double check cannot be ignored */
249 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
251 tsk->thread.error_code = error_code;
252 tsk->thread.trap_nr = X86_TRAP_DF;
255 * This is always a kernel trap and never fixable (and thus must
259 die(str, regs, error_code);
263 dotraplinkage void __kprobes
264 do_general_protection(struct pt_regs *regs, long error_code)
266 struct task_struct *tsk;
268 exception_enter(regs);
269 conditional_sti(regs);
272 if (regs->flags & X86_VM_MASK) {
274 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
280 if (!user_mode(regs)) {
281 if (fixup_exception(regs))
284 tsk->thread.error_code = error_code;
285 tsk->thread.trap_nr = X86_TRAP_GP;
286 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
287 X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
288 die("general protection fault", regs, error_code);
292 tsk->thread.error_code = error_code;
293 tsk->thread.trap_nr = X86_TRAP_GP;
295 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
296 printk_ratelimit()) {
297 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
298 tsk->comm, task_pid_nr(tsk),
299 regs->ip, regs->sp, error_code);
300 print_vma_addr(" in ", regs->ip);
304 force_sig(SIGSEGV, tsk);
306 exception_exit(regs);
309 /* May run on IST stack. */
310 dotraplinkage void __kprobes notrace do_int3(struct pt_regs *regs, long error_code)
312 #ifdef CONFIG_DYNAMIC_FTRACE
314 * ftrace must be first, everything else may cause a recursive crash.
315 * See note by declaration of modifying_ftrace_code in ftrace.c
317 if (unlikely(atomic_read(&modifying_ftrace_code)) &&
318 ftrace_int3_handler(regs))
321 exception_enter(regs);
322 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
323 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
324 SIGTRAP) == NOTIFY_STOP)
326 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
328 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
329 SIGTRAP) == NOTIFY_STOP)
333 * Let others (NMI) know that the debug stack is in use
334 * as we may switch to the interrupt stack.
336 debug_stack_usage_inc();
337 preempt_conditional_sti(regs);
338 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
339 preempt_conditional_cli(regs);
340 debug_stack_usage_dec();
342 exception_exit(regs);
347 * Help handler running on IST stack to switch back to user stack
348 * for scheduling or signal handling. The actual stack switch is done in
351 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
353 struct pt_regs *regs = eregs;
354 /* Did already sync */
355 if (eregs == (struct pt_regs *)eregs->sp)
357 /* Exception from user space */
358 else if (user_mode(eregs))
359 regs = task_pt_regs(current);
361 * Exception from kernel and interrupts are enabled. Move to
362 * kernel process stack.
364 else if (eregs->flags & X86_EFLAGS_IF)
365 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
373 * Our handling of the processor debug registers is non-trivial.
374 * We do not clear them on entry and exit from the kernel. Therefore
375 * it is possible to get a watchpoint trap here from inside the kernel.
376 * However, the code in ./ptrace.c has ensured that the user can
377 * only set watchpoints on userspace addresses. Therefore the in-kernel
378 * watchpoint trap can only occur in code which is reading/writing
379 * from user space. Such code must not hold kernel locks (since it
380 * can equally take a page fault), therefore it is safe to call
381 * force_sig_info even though that claims and releases locks.
383 * Code in ./signal.c ensures that the debug control register
384 * is restored before we deliver any signal, and therefore that
385 * user code runs with the correct debug control register even though
388 * Being careful here means that we don't have to be as careful in a
389 * lot of more complicated places (task switching can be a bit lazy
390 * about restoring all the debug state, and ptrace doesn't have to
391 * find every occurrence of the TF bit that could be saved away even
394 * May run on IST stack.
396 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
398 struct task_struct *tsk = current;
403 exception_enter(regs);
405 get_debugreg(dr6, 6);
407 /* Filter out all the reserved bits which are preset to 1 */
408 dr6 &= ~DR6_RESERVED;
411 * If dr6 has no reason to give us about the origin of this trap,
412 * then it's very likely the result of an icebp/int01 trap.
413 * User wants a sigtrap for that.
415 if (!dr6 && user_mode(regs))
418 /* Catch kmemcheck conditions first of all! */
419 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
422 /* DR6 may or may not be cleared by the CPU */
426 * The processor cleared BTF, so don't mark that we need it set.
428 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
430 /* Store the virtualized DR6 value */
431 tsk->thread.debugreg6 = dr6;
433 if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
434 SIGTRAP) == NOTIFY_STOP)
438 * Let others (NMI) know that the debug stack is in use
439 * as we may switch to the interrupt stack.
441 debug_stack_usage_inc();
443 /* It's safe to allow irq's after DR6 has been saved */
444 preempt_conditional_sti(regs);
446 if (regs->flags & X86_VM_MASK) {
447 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
449 preempt_conditional_cli(regs);
450 debug_stack_usage_dec();
455 * Single-stepping through system calls: ignore any exceptions in
456 * kernel space, but re-enable TF when returning to user mode.
458 * We already checked v86 mode above, so we can check for kernel mode
459 * by just checking the CPL of CS.
461 if ((dr6 & DR_STEP) && !user_mode(regs)) {
462 tsk->thread.debugreg6 &= ~DR_STEP;
463 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
464 regs->flags &= ~X86_EFLAGS_TF;
466 si_code = get_si_code(tsk->thread.debugreg6);
467 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
468 send_sigtrap(tsk, regs, error_code, si_code);
469 preempt_conditional_cli(regs);
470 debug_stack_usage_dec();
473 exception_exit(regs);
477 * Note that we play around with the 'TS' bit in an attempt to get
478 * the correct behaviour even in the presence of the asynchronous
481 void math_error(struct pt_regs *regs, int error_code, int trapnr)
483 struct task_struct *task = current;
486 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
489 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
491 conditional_sti(regs);
493 if (!user_mode_vm(regs))
495 if (!fixup_exception(regs)) {
496 task->thread.error_code = error_code;
497 task->thread.trap_nr = trapnr;
498 die(str, regs, error_code);
504 * Save the info for the exception handler and clear the error.
507 task->thread.trap_nr = trapnr;
508 task->thread.error_code = error_code;
509 info.si_signo = SIGFPE;
511 info.si_addr = (void __user *)regs->ip;
512 if (trapnr == X86_TRAP_MF) {
513 unsigned short cwd, swd;
515 * (~cwd & swd) will mask out exceptions that are not set to unmasked
516 * status. 0x3f is the exception bits in these regs, 0x200 is the
517 * C1 reg you need in case of a stack fault, 0x040 is the stack
518 * fault bit. We should only be taking one exception at a time,
519 * so if this combination doesn't produce any single exception,
520 * then we have a bad program that isn't synchronizing its FPU usage
521 * and it will suffer the consequences since we won't be able to
522 * fully reproduce the context of the exception
524 cwd = get_fpu_cwd(task);
525 swd = get_fpu_swd(task);
530 * The SIMD FPU exceptions are handled a little differently, as there
531 * is only a single status/control register. Thus, to determine which
532 * unmasked exception was caught we must mask the exception mask bits
533 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
535 unsigned short mxcsr = get_fpu_mxcsr(task);
536 err = ~(mxcsr >> 7) & mxcsr;
539 if (err & 0x001) { /* Invalid op */
541 * swd & 0x240 == 0x040: Stack Underflow
542 * swd & 0x240 == 0x240: Stack Overflow
543 * User must clear the SF bit (0x40) if set
545 info.si_code = FPE_FLTINV;
546 } else if (err & 0x004) { /* Divide by Zero */
547 info.si_code = FPE_FLTDIV;
548 } else if (err & 0x008) { /* Overflow */
549 info.si_code = FPE_FLTOVF;
550 } else if (err & 0x012) { /* Denormal, Underflow */
551 info.si_code = FPE_FLTUND;
552 } else if (err & 0x020) { /* Precision */
553 info.si_code = FPE_FLTRES;
556 * If we're using IRQ 13, or supposedly even some trap
557 * X86_TRAP_MF implementations, it's possible
558 * we get a spurious trap, which is not an error.
562 force_sig_info(SIGFPE, &info, task);
565 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
570 exception_enter(regs);
571 math_error(regs, error_code, X86_TRAP_MF);
572 exception_exit(regs);
576 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
578 exception_enter(regs);
579 math_error(regs, error_code, X86_TRAP_XF);
580 exception_exit(regs);
584 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
586 conditional_sti(regs);
588 /* No need to warn about this any longer. */
589 pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
593 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
597 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
602 * 'math_state_restore()' saves the current math information in the
603 * old math state array, and gets the new ones from the current task
605 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
606 * Don't touch unless you *really* know how it works.
608 * Must be called with kernel preemption disabled (eg with local
609 * local interrupts as in the case of do_device_not_available).
611 void math_state_restore(void)
613 struct task_struct *tsk = current;
615 if (!tsk_used_math(tsk)) {
618 * does a slab alloc which can sleep
624 do_group_exit(SIGKILL);
630 __thread_fpu_begin(tsk);
633 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
635 if (unlikely(restore_fpu_checking(tsk))) {
637 force_sig(SIGSEGV, tsk);
643 EXPORT_SYMBOL_GPL(math_state_restore);
645 dotraplinkage void __kprobes
646 do_device_not_available(struct pt_regs *regs, long error_code)
648 exception_enter(regs);
649 BUG_ON(use_eager_fpu());
651 #ifdef CONFIG_MATH_EMULATION
652 if (read_cr0() & X86_CR0_EM) {
653 struct math_emu_info info = { };
655 conditional_sti(regs);
659 exception_exit(regs);
663 math_state_restore(); /* interrupts still off */
665 conditional_sti(regs);
667 exception_exit(regs);
671 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
675 exception_enter(regs);
678 info.si_signo = SIGILL;
680 info.si_code = ILL_BADSTK;
682 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
683 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
684 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
687 exception_exit(regs);
691 /* Set of traps needed for early debugging. */
692 void __init early_trap_init(void)
694 set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
695 /* int3 can be called from all */
696 set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
697 set_intr_gate(X86_TRAP_PF, &page_fault);
698 load_idt(&idt_descr);
701 void __init trap_init(void)
706 void __iomem *p = early_ioremap(0x0FFFD9, 4);
708 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
713 set_intr_gate(X86_TRAP_DE, ÷_error);
714 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
715 /* int4 can be called from all */
716 set_system_intr_gate(X86_TRAP_OF, &overflow);
717 set_intr_gate(X86_TRAP_BR, &bounds);
718 set_intr_gate(X86_TRAP_UD, &invalid_op);
719 set_intr_gate(X86_TRAP_NM, &device_not_available);
721 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
723 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
725 set_intr_gate(X86_TRAP_OLD_MF, &coprocessor_segment_overrun);
726 set_intr_gate(X86_TRAP_TS, &invalid_TSS);
727 set_intr_gate(X86_TRAP_NP, &segment_not_present);
728 set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
729 set_intr_gate(X86_TRAP_GP, &general_protection);
730 set_intr_gate(X86_TRAP_SPURIOUS, &spurious_interrupt_bug);
731 set_intr_gate(X86_TRAP_MF, &coprocessor_error);
732 set_intr_gate(X86_TRAP_AC, &alignment_check);
733 #ifdef CONFIG_X86_MCE
734 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
736 set_intr_gate(X86_TRAP_XF, &simd_coprocessor_error);
738 /* Reserve all the builtin and the syscall vector: */
739 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
740 set_bit(i, used_vectors);
742 #ifdef CONFIG_IA32_EMULATION
743 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
744 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
748 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
749 set_bit(SYSCALL_VECTOR, used_vectors);
753 * Should be a barrier for any external CPU state:
757 x86_init.irqs.trap_init();
760 memcpy(&nmi_idt_table, &idt_table, IDT_ENTRIES * 16);
761 set_nmi_gate(X86_TRAP_DB, &debug);
762 set_nmi_gate(X86_TRAP_BP, &int3);