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/mach_traps.h>
62 #include <asm/x86_init.h>
63 #include <asm/pgalloc.h>
64 #include <asm/proto.h>
66 #include <asm/processor-flags.h>
67 #include <asm/setup.h>
69 asmlinkage int system_call(void);
72 * The IDT has to be page-aligned to simplify the Pentium
73 * F0 0F bug workaround.
75 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
78 DECLARE_BITMAP(used_vectors, NR_VECTORS);
79 EXPORT_SYMBOL_GPL(used_vectors);
81 static inline void conditional_sti(struct pt_regs *regs)
83 if (regs->flags & X86_EFLAGS_IF)
87 static inline void preempt_conditional_sti(struct pt_regs *regs)
90 if (regs->flags & X86_EFLAGS_IF)
94 static inline void conditional_cli(struct pt_regs *regs)
96 if (regs->flags & X86_EFLAGS_IF)
100 static inline void preempt_conditional_cli(struct pt_regs *regs)
102 if (regs->flags & X86_EFLAGS_IF)
108 do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
109 struct pt_regs *regs, long error_code)
112 if (regs->flags & X86_VM_MASK) {
114 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
115 * On nmi (interrupt 2), do_trap should not be called.
117 if (trapnr < X86_TRAP_UD) {
118 if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
125 if (!user_mode(regs)) {
126 if (!fixup_exception(regs)) {
127 tsk->thread.error_code = error_code;
128 tsk->thread.trap_nr = trapnr;
129 die(str, regs, error_code);
137 static void __kprobes
138 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
139 long error_code, siginfo_t *info)
141 struct task_struct *tsk = current;
144 if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
147 * We want error_code and trap_nr set for userspace faults and
148 * kernelspace faults which result in die(), but not
149 * kernelspace faults which are fixed up. die() gives the
150 * process no chance to handle the signal and notice the
151 * kernel fault information, so that won't result in polluting
152 * the information about previously queued, but not yet
153 * delivered, faults. See also do_general_protection below.
155 tsk->thread.error_code = error_code;
156 tsk->thread.trap_nr = trapnr;
159 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
160 printk_ratelimit()) {
161 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
162 tsk->comm, tsk->pid, str,
163 regs->ip, regs->sp, error_code);
164 print_vma_addr(" in ", regs->ip);
170 force_sig_info(signr, info, tsk);
172 force_sig(signr, tsk);
175 #define DO_ERROR(trapnr, signr, str, name) \
176 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
178 exception_enter(regs); \
179 if (notify_die(DIE_TRAP, str, regs, error_code, \
180 trapnr, signr) == NOTIFY_STOP) { \
181 exception_exit(regs); \
184 conditional_sti(regs); \
185 do_trap(trapnr, signr, str, regs, error_code, NULL); \
186 exception_exit(regs); \
189 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
190 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
193 info.si_signo = signr; \
195 info.si_code = sicode; \
196 info.si_addr = (void __user *)siaddr; \
197 exception_enter(regs); \
198 if (notify_die(DIE_TRAP, str, regs, error_code, \
199 trapnr, signr) == NOTIFY_STOP) { \
200 exception_exit(regs); \
203 conditional_sti(regs); \
204 do_trap(trapnr, signr, str, regs, error_code, &info); \
205 exception_exit(regs); \
208 DO_ERROR_INFO(X86_TRAP_DE, SIGFPE, "divide error", divide_error, FPE_INTDIV,
210 DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
211 DO_ERROR(X86_TRAP_BR, SIGSEGV, "bounds", bounds)
212 DO_ERROR_INFO(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN,
214 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",
215 coprocessor_segment_overrun)
216 DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
217 DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
219 DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
221 DO_ERROR_INFO(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check,
225 /* Runs on IST stack */
226 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
228 exception_enter(regs);
229 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
230 X86_TRAP_SS, SIGBUS) != NOTIFY_STOP) {
231 preempt_conditional_sti(regs);
232 do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
233 preempt_conditional_cli(regs);
235 exception_exit(regs);
238 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
240 static const char str[] = "double fault";
241 struct task_struct *tsk = current;
243 exception_enter(regs);
244 /* Return not checked because double check cannot be ignored */
245 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
247 tsk->thread.error_code = error_code;
248 tsk->thread.trap_nr = X86_TRAP_DF;
251 * This is always a kernel trap and never fixable (and thus must
255 die(str, regs, error_code);
259 dotraplinkage void __kprobes
260 do_general_protection(struct pt_regs *regs, long error_code)
262 struct task_struct *tsk;
264 exception_enter(regs);
265 conditional_sti(regs);
268 if (regs->flags & X86_VM_MASK) {
270 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
276 if (!user_mode(regs)) {
277 if (fixup_exception(regs))
280 tsk->thread.error_code = error_code;
281 tsk->thread.trap_nr = X86_TRAP_GP;
282 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
283 X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
284 die("general protection fault", regs, error_code);
288 tsk->thread.error_code = error_code;
289 tsk->thread.trap_nr = X86_TRAP_GP;
291 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
292 printk_ratelimit()) {
293 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
294 tsk->comm, task_pid_nr(tsk),
295 regs->ip, regs->sp, error_code);
296 print_vma_addr(" in ", regs->ip);
300 force_sig(SIGSEGV, tsk);
302 exception_exit(regs);
305 /* May run on IST stack. */
306 dotraplinkage void __kprobes notrace do_int3(struct pt_regs *regs, long error_code)
308 #ifdef CONFIG_DYNAMIC_FTRACE
310 * ftrace must be first, everything else may cause a recursive crash.
311 * See note by declaration of modifying_ftrace_code in ftrace.c
313 if (unlikely(atomic_read(&modifying_ftrace_code)) &&
314 ftrace_int3_handler(regs))
317 exception_enter(regs);
318 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
319 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
320 SIGTRAP) == NOTIFY_STOP)
322 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
324 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
325 SIGTRAP) == NOTIFY_STOP)
329 * Let others (NMI) know that the debug stack is in use
330 * as we may switch to the interrupt stack.
332 debug_stack_usage_inc();
333 preempt_conditional_sti(regs);
334 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
335 preempt_conditional_cli(regs);
336 debug_stack_usage_dec();
338 exception_exit(regs);
343 * Help handler running on IST stack to switch back to user stack
344 * for scheduling or signal handling. The actual stack switch is done in
347 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
349 struct pt_regs *regs = eregs;
350 /* Did already sync */
351 if (eregs == (struct pt_regs *)eregs->sp)
353 /* Exception from user space */
354 else if (user_mode(eregs))
355 regs = task_pt_regs(current);
357 * Exception from kernel and interrupts are enabled. Move to
358 * kernel process stack.
360 else if (eregs->flags & X86_EFLAGS_IF)
361 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
369 * Our handling of the processor debug registers is non-trivial.
370 * We do not clear them on entry and exit from the kernel. Therefore
371 * it is possible to get a watchpoint trap here from inside the kernel.
372 * However, the code in ./ptrace.c has ensured that the user can
373 * only set watchpoints on userspace addresses. Therefore the in-kernel
374 * watchpoint trap can only occur in code which is reading/writing
375 * from user space. Such code must not hold kernel locks (since it
376 * can equally take a page fault), therefore it is safe to call
377 * force_sig_info even though that claims and releases locks.
379 * Code in ./signal.c ensures that the debug control register
380 * is restored before we deliver any signal, and therefore that
381 * user code runs with the correct debug control register even though
384 * Being careful here means that we don't have to be as careful in a
385 * lot of more complicated places (task switching can be a bit lazy
386 * about restoring all the debug state, and ptrace doesn't have to
387 * find every occurrence of the TF bit that could be saved away even
390 * May run on IST stack.
392 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
394 struct task_struct *tsk = current;
399 exception_enter(regs);
401 get_debugreg(dr6, 6);
403 /* Filter out all the reserved bits which are preset to 1 */
404 dr6 &= ~DR6_RESERVED;
407 * If dr6 has no reason to give us about the origin of this trap,
408 * then it's very likely the result of an icebp/int01 trap.
409 * User wants a sigtrap for that.
411 if (!dr6 && user_mode(regs))
414 /* Catch kmemcheck conditions first of all! */
415 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
418 /* DR6 may or may not be cleared by the CPU */
422 * The processor cleared BTF, so don't mark that we need it set.
424 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
426 /* Store the virtualized DR6 value */
427 tsk->thread.debugreg6 = dr6;
429 if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
430 SIGTRAP) == NOTIFY_STOP)
434 * Let others (NMI) know that the debug stack is in use
435 * as we may switch to the interrupt stack.
437 debug_stack_usage_inc();
439 /* It's safe to allow irq's after DR6 has been saved */
440 preempt_conditional_sti(regs);
442 if (regs->flags & X86_VM_MASK) {
443 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
445 preempt_conditional_cli(regs);
446 debug_stack_usage_dec();
451 * Single-stepping through system calls: ignore any exceptions in
452 * kernel space, but re-enable TF when returning to user mode.
454 * We already checked v86 mode above, so we can check for kernel mode
455 * by just checking the CPL of CS.
457 if ((dr6 & DR_STEP) && !user_mode(regs)) {
458 tsk->thread.debugreg6 &= ~DR_STEP;
459 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
460 regs->flags &= ~X86_EFLAGS_TF;
462 si_code = get_si_code(tsk->thread.debugreg6);
463 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
464 send_sigtrap(tsk, regs, error_code, si_code);
465 preempt_conditional_cli(regs);
466 debug_stack_usage_dec();
469 exception_exit(regs);
473 * Note that we play around with the 'TS' bit in an attempt to get
474 * the correct behaviour even in the presence of the asynchronous
477 void math_error(struct pt_regs *regs, int error_code, int trapnr)
479 struct task_struct *task = current;
482 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
485 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
487 conditional_sti(regs);
489 if (!user_mode_vm(regs))
491 if (!fixup_exception(regs)) {
492 task->thread.error_code = error_code;
493 task->thread.trap_nr = trapnr;
494 die(str, regs, error_code);
500 * Save the info for the exception handler and clear the error.
503 task->thread.trap_nr = trapnr;
504 task->thread.error_code = error_code;
505 info.si_signo = SIGFPE;
507 info.si_addr = (void __user *)regs->ip;
508 if (trapnr == X86_TRAP_MF) {
509 unsigned short cwd, swd;
511 * (~cwd & swd) will mask out exceptions that are not set to unmasked
512 * status. 0x3f is the exception bits in these regs, 0x200 is the
513 * C1 reg you need in case of a stack fault, 0x040 is the stack
514 * fault bit. We should only be taking one exception at a time,
515 * so if this combination doesn't produce any single exception,
516 * then we have a bad program that isn't synchronizing its FPU usage
517 * and it will suffer the consequences since we won't be able to
518 * fully reproduce the context of the exception
520 cwd = get_fpu_cwd(task);
521 swd = get_fpu_swd(task);
526 * The SIMD FPU exceptions are handled a little differently, as there
527 * is only a single status/control register. Thus, to determine which
528 * unmasked exception was caught we must mask the exception mask bits
529 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
531 unsigned short mxcsr = get_fpu_mxcsr(task);
532 err = ~(mxcsr >> 7) & mxcsr;
535 if (err & 0x001) { /* Invalid op */
537 * swd & 0x240 == 0x040: Stack Underflow
538 * swd & 0x240 == 0x240: Stack Overflow
539 * User must clear the SF bit (0x40) if set
541 info.si_code = FPE_FLTINV;
542 } else if (err & 0x004) { /* Divide by Zero */
543 info.si_code = FPE_FLTDIV;
544 } else if (err & 0x008) { /* Overflow */
545 info.si_code = FPE_FLTOVF;
546 } else if (err & 0x012) { /* Denormal, Underflow */
547 info.si_code = FPE_FLTUND;
548 } else if (err & 0x020) { /* Precision */
549 info.si_code = FPE_FLTRES;
552 * If we're using IRQ 13, or supposedly even some trap
553 * X86_TRAP_MF implementations, it's possible
554 * we get a spurious trap, which is not an error.
558 force_sig_info(SIGFPE, &info, task);
561 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
563 exception_enter(regs);
564 math_error(regs, error_code, X86_TRAP_MF);
565 exception_exit(regs);
569 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
571 exception_enter(regs);
572 math_error(regs, error_code, X86_TRAP_XF);
573 exception_exit(regs);
577 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
579 conditional_sti(regs);
581 /* No need to warn about this any longer. */
582 pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
586 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
590 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
595 * 'math_state_restore()' saves the current math information in the
596 * old math state array, and gets the new ones from the current task
598 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
599 * Don't touch unless you *really* know how it works.
601 * Must be called with kernel preemption disabled (eg with local
602 * local interrupts as in the case of do_device_not_available).
604 void math_state_restore(void)
606 struct task_struct *tsk = current;
608 if (!tsk_used_math(tsk)) {
611 * does a slab alloc which can sleep
617 do_group_exit(SIGKILL);
623 __thread_fpu_begin(tsk);
626 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
628 if (unlikely(restore_fpu_checking(tsk))) {
630 force_sig(SIGSEGV, tsk);
636 EXPORT_SYMBOL_GPL(math_state_restore);
638 dotraplinkage void __kprobes
639 do_device_not_available(struct pt_regs *regs, long error_code)
641 exception_enter(regs);
642 BUG_ON(use_eager_fpu());
644 #ifdef CONFIG_MATH_EMULATION
645 if (read_cr0() & X86_CR0_EM) {
646 struct math_emu_info info = { };
648 conditional_sti(regs);
652 exception_exit(regs);
656 math_state_restore(); /* interrupts still off */
658 conditional_sti(regs);
660 exception_exit(regs);
664 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
668 exception_enter(regs);
671 info.si_signo = SIGILL;
673 info.si_code = ILL_BADSTK;
675 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
676 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
677 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
680 exception_exit(regs);
684 /* Set of traps needed for early debugging. */
685 void __init early_trap_init(void)
687 set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
688 /* int3 can be called from all */
689 set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
691 set_intr_gate(X86_TRAP_PF, &page_fault);
693 load_idt(&idt_descr);
696 void __init early_trap_pf_init(void)
699 set_intr_gate(X86_TRAP_PF, &page_fault);
703 void __init trap_init(void)
708 void __iomem *p = early_ioremap(0x0FFFD9, 4);
710 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
715 set_intr_gate(X86_TRAP_DE, ÷_error);
716 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
717 /* int4 can be called from all */
718 set_system_intr_gate(X86_TRAP_OF, &overflow);
719 set_intr_gate(X86_TRAP_BR, &bounds);
720 set_intr_gate(X86_TRAP_UD, &invalid_op);
721 set_intr_gate(X86_TRAP_NM, &device_not_available);
723 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
725 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
727 set_intr_gate(X86_TRAP_OLD_MF, &coprocessor_segment_overrun);
728 set_intr_gate(X86_TRAP_TS, &invalid_TSS);
729 set_intr_gate(X86_TRAP_NP, &segment_not_present);
730 set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
731 set_intr_gate(X86_TRAP_GP, &general_protection);
732 set_intr_gate(X86_TRAP_SPURIOUS, &spurious_interrupt_bug);
733 set_intr_gate(X86_TRAP_MF, &coprocessor_error);
734 set_intr_gate(X86_TRAP_AC, &alignment_check);
735 #ifdef CONFIG_X86_MCE
736 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
738 set_intr_gate(X86_TRAP_XF, &simd_coprocessor_error);
740 /* Reserve all the builtin and the syscall vector: */
741 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
742 set_bit(i, used_vectors);
744 #ifdef CONFIG_IA32_EMULATION
745 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
746 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
750 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
751 set_bit(SYSCALL_VECTOR, used_vectors);
755 * Should be a barrier for any external CPU state:
759 x86_init.irqs.trap_init();
762 memcpy(&nmi_idt_table, &idt_table, IDT_ENTRIES * 16);
763 set_nmi_gate(X86_TRAP_DB, &debug);
764 set_nmi_gate(X86_TRAP_BP, &int3);