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.
12 #include <linux/interrupt.h>
13 #include <linux/kallsyms.h>
14 #include <linux/spinlock.h>
15 #include <linux/kprobes.h>
16 #include <linux/uaccess.h>
17 #include <linux/kdebug.h>
18 #include <linux/kgdb.h>
19 #include <linux/kernel.h>
20 #include <linux/module.h>
21 #include <linux/ptrace.h>
22 #include <linux/string.h>
23 #include <linux/delay.h>
24 #include <linux/errno.h>
25 #include <linux/kexec.h>
26 #include <linux/sched.h>
27 #include <linux/timer.h>
28 #include <linux/init.h>
29 #include <linux/bug.h>
30 #include <linux/nmi.h>
32 #include <linux/smp.h>
36 #include <linux/ioport.h>
37 #include <linux/eisa.h>
40 #if defined(CONFIG_EDAC)
41 #include <linux/edac.h>
44 #include <asm/kmemcheck.h>
45 #include <asm/stacktrace.h>
46 #include <asm/processor.h>
47 #include <asm/debugreg.h>
48 #include <linux/atomic.h>
49 #include <asm/ftrace.h>
50 #include <asm/traps.h>
53 #include <asm/fpu-internal.h>
56 #include <asm/mach_traps.h>
59 #include <asm/x86_init.h>
60 #include <asm/pgalloc.h>
61 #include <asm/proto.h>
63 #include <asm/processor-flags.h>
64 #include <asm/setup.h>
66 asmlinkage int system_call(void);
68 /* Do we ignore FPU interrupts ? */
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)
107 static void __kprobes
108 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
109 long error_code, siginfo_t *info)
111 struct task_struct *tsk = current;
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)
125 if (!user_mode(regs))
132 * We want error_code and trap_nr set for userspace faults and
133 * kernelspace faults which result in die(), but not
134 * kernelspace faults which are fixed up. die() gives the
135 * process no chance to handle the signal and notice the
136 * kernel fault information, so that won't result in polluting
137 * the information about previously queued, but not yet
138 * delivered, faults. See also do_general_protection below.
140 tsk->thread.error_code = error_code;
141 tsk->thread.trap_nr = trapnr;
144 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
145 printk_ratelimit()) {
147 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
148 tsk->comm, tsk->pid, str,
149 regs->ip, regs->sp, error_code);
150 print_vma_addr(" in ", regs->ip);
156 force_sig_info(signr, info, tsk);
158 force_sig(signr, tsk);
162 if (!fixup_exception(regs)) {
163 tsk->thread.error_code = error_code;
164 tsk->thread.trap_nr = trapnr;
165 die(str, regs, error_code);
171 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
178 #define DO_ERROR(trapnr, signr, str, name) \
179 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
181 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
184 conditional_sti(regs); \
185 do_trap(trapnr, signr, str, regs, error_code, NULL); \
188 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
189 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
192 info.si_signo = signr; \
194 info.si_code = sicode; \
195 info.si_addr = (void __user *)siaddr; \
196 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
199 conditional_sti(regs); \
200 do_trap(trapnr, signr, str, regs, error_code, &info); \
203 DO_ERROR_INFO(X86_TRAP_DE, SIGFPE, "divide error", divide_error, FPE_INTDIV,
205 DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
206 DO_ERROR(X86_TRAP_BR, SIGSEGV, "bounds", bounds)
207 DO_ERROR_INFO(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN,
209 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",
210 coprocessor_segment_overrun)
211 DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
212 DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
214 DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
216 DO_ERROR_INFO(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check,
220 /* Runs on IST stack */
221 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
223 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
224 X86_TRAP_SS, SIGBUS) == NOTIFY_STOP)
226 preempt_conditional_sti(regs);
227 do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
228 preempt_conditional_cli(regs);
231 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
233 static const char str[] = "double fault";
234 struct task_struct *tsk = current;
236 /* Return not checked because double check cannot be ignored */
237 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
239 tsk->thread.error_code = error_code;
240 tsk->thread.trap_nr = X86_TRAP_DF;
243 * This is always a kernel trap and never fixable (and thus must
247 die(str, regs, error_code);
251 dotraplinkage void __kprobes
252 do_general_protection(struct pt_regs *regs, long error_code)
254 struct task_struct *tsk;
256 conditional_sti(regs);
259 if (regs->flags & X86_VM_MASK)
264 if (!user_mode(regs))
267 tsk->thread.error_code = error_code;
268 tsk->thread.trap_nr = X86_TRAP_GP;
270 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
271 printk_ratelimit()) {
273 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
274 tsk->comm, task_pid_nr(tsk),
275 regs->ip, regs->sp, error_code);
276 print_vma_addr(" in ", regs->ip);
280 force_sig(SIGSEGV, tsk);
286 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
291 if (fixup_exception(regs))
294 tsk->thread.error_code = error_code;
295 tsk->thread.trap_nr = X86_TRAP_GP;
296 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
297 X86_TRAP_GP, SIGSEGV) == NOTIFY_STOP)
299 die("general protection fault", regs, error_code);
302 /* May run on IST stack. */
303 dotraplinkage void __kprobes notrace do_int3(struct pt_regs *regs, long error_code)
305 #ifdef CONFIG_DYNAMIC_FTRACE
306 /* ftrace must be first, everything else may cause a recursive crash */
307 if (unlikely(modifying_ftrace_code) && ftrace_int3_handler(regs))
310 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
311 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
312 SIGTRAP) == NOTIFY_STOP)
314 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
316 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
317 SIGTRAP) == NOTIFY_STOP)
321 * Let others (NMI) know that the debug stack is in use
322 * as we may switch to the interrupt stack.
324 debug_stack_usage_inc();
325 preempt_conditional_sti(regs);
326 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
327 preempt_conditional_cli(regs);
328 debug_stack_usage_dec();
333 * Help handler running on IST stack to switch back to user stack
334 * for scheduling or signal handling. The actual stack switch is done in
337 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
339 struct pt_regs *regs = eregs;
340 /* Did already sync */
341 if (eregs == (struct pt_regs *)eregs->sp)
343 /* Exception from user space */
344 else if (user_mode(eregs))
345 regs = task_pt_regs(current);
347 * Exception from kernel and interrupts are enabled. Move to
348 * kernel process stack.
350 else if (eregs->flags & X86_EFLAGS_IF)
351 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
359 * Our handling of the processor debug registers is non-trivial.
360 * We do not clear them on entry and exit from the kernel. Therefore
361 * it is possible to get a watchpoint trap here from inside the kernel.
362 * However, the code in ./ptrace.c has ensured that the user can
363 * only set watchpoints on userspace addresses. Therefore the in-kernel
364 * watchpoint trap can only occur in code which is reading/writing
365 * from user space. Such code must not hold kernel locks (since it
366 * can equally take a page fault), therefore it is safe to call
367 * force_sig_info even though that claims and releases locks.
369 * Code in ./signal.c ensures that the debug control register
370 * is restored before we deliver any signal, and therefore that
371 * user code runs with the correct debug control register even though
374 * Being careful here means that we don't have to be as careful in a
375 * lot of more complicated places (task switching can be a bit lazy
376 * about restoring all the debug state, and ptrace doesn't have to
377 * find every occurrence of the TF bit that could be saved away even
380 * May run on IST stack.
382 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
384 struct task_struct *tsk = current;
389 get_debugreg(dr6, 6);
391 /* Filter out all the reserved bits which are preset to 1 */
392 dr6 &= ~DR6_RESERVED;
395 * If dr6 has no reason to give us about the origin of this trap,
396 * then it's very likely the result of an icebp/int01 trap.
397 * User wants a sigtrap for that.
399 if (!dr6 && user_mode(regs))
402 /* Catch kmemcheck conditions first of all! */
403 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
406 /* DR6 may or may not be cleared by the CPU */
410 * The processor cleared BTF, so don't mark that we need it set.
412 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
414 /* Store the virtualized DR6 value */
415 tsk->thread.debugreg6 = dr6;
417 if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
418 SIGTRAP) == NOTIFY_STOP)
422 * Let others (NMI) know that the debug stack is in use
423 * as we may switch to the interrupt stack.
425 debug_stack_usage_inc();
427 /* It's safe to allow irq's after DR6 has been saved */
428 preempt_conditional_sti(regs);
430 if (regs->flags & X86_VM_MASK) {
431 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
433 preempt_conditional_cli(regs);
434 debug_stack_usage_dec();
439 * Single-stepping through system calls: ignore any exceptions in
440 * kernel space, but re-enable TF when returning to user mode.
442 * We already checked v86 mode above, so we can check for kernel mode
443 * by just checking the CPL of CS.
445 if ((dr6 & DR_STEP) && !user_mode(regs)) {
446 tsk->thread.debugreg6 &= ~DR_STEP;
447 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
448 regs->flags &= ~X86_EFLAGS_TF;
450 si_code = get_si_code(tsk->thread.debugreg6);
451 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
452 send_sigtrap(tsk, regs, error_code, si_code);
453 preempt_conditional_cli(regs);
454 debug_stack_usage_dec();
460 * Note that we play around with the 'TS' bit in an attempt to get
461 * the correct behaviour even in the presence of the asynchronous
464 void math_error(struct pt_regs *regs, int error_code, int trapnr)
466 struct task_struct *task = current;
469 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
472 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
474 conditional_sti(regs);
476 if (!user_mode_vm(regs))
478 if (!fixup_exception(regs)) {
479 task->thread.error_code = error_code;
480 task->thread.trap_nr = trapnr;
481 die(str, regs, error_code);
487 * Save the info for the exception handler and clear the error.
490 task->thread.trap_nr = trapnr;
491 task->thread.error_code = error_code;
492 info.si_signo = SIGFPE;
494 info.si_addr = (void __user *)regs->ip;
495 if (trapnr == X86_TRAP_MF) {
496 unsigned short cwd, swd;
498 * (~cwd & swd) will mask out exceptions that are not set to unmasked
499 * status. 0x3f is the exception bits in these regs, 0x200 is the
500 * C1 reg you need in case of a stack fault, 0x040 is the stack
501 * fault bit. We should only be taking one exception at a time,
502 * so if this combination doesn't produce any single exception,
503 * then we have a bad program that isn't synchronizing its FPU usage
504 * and it will suffer the consequences since we won't be able to
505 * fully reproduce the context of the exception
507 cwd = get_fpu_cwd(task);
508 swd = get_fpu_swd(task);
513 * The SIMD FPU exceptions are handled a little differently, as there
514 * is only a single status/control register. Thus, to determine which
515 * unmasked exception was caught we must mask the exception mask bits
516 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
518 unsigned short mxcsr = get_fpu_mxcsr(task);
519 err = ~(mxcsr >> 7) & mxcsr;
522 if (err & 0x001) { /* Invalid op */
524 * swd & 0x240 == 0x040: Stack Underflow
525 * swd & 0x240 == 0x240: Stack Overflow
526 * User must clear the SF bit (0x40) if set
528 info.si_code = FPE_FLTINV;
529 } else if (err & 0x004) { /* Divide by Zero */
530 info.si_code = FPE_FLTDIV;
531 } else if (err & 0x008) { /* Overflow */
532 info.si_code = FPE_FLTOVF;
533 } else if (err & 0x012) { /* Denormal, Underflow */
534 info.si_code = FPE_FLTUND;
535 } else if (err & 0x020) { /* Precision */
536 info.si_code = FPE_FLTRES;
539 * If we're using IRQ 13, or supposedly even some trap
540 * X86_TRAP_MF implementations, it's possible
541 * we get a spurious trap, which is not an error.
545 force_sig_info(SIGFPE, &info, task);
548 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
554 math_error(regs, error_code, X86_TRAP_MF);
558 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
560 math_error(regs, error_code, X86_TRAP_XF);
564 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
566 conditional_sti(regs);
568 /* No need to warn about this any longer. */
569 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
573 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
577 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
582 * 'math_state_restore()' saves the current math information in the
583 * old math state array, and gets the new ones from the current task
585 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
586 * Don't touch unless you *really* know how it works.
588 * Must be called with kernel preemption disabled (eg with local
589 * local interrupts as in the case of do_device_not_available).
591 void math_state_restore(void)
593 struct task_struct *tsk = current;
595 if (!tsk_used_math(tsk)) {
598 * does a slab alloc which can sleep
604 do_group_exit(SIGKILL);
610 __thread_fpu_begin(tsk);
612 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
614 if (unlikely(restore_fpu_checking(tsk))) {
615 __thread_fpu_end(tsk);
616 force_sig(SIGSEGV, tsk);
622 EXPORT_SYMBOL_GPL(math_state_restore);
624 dotraplinkage void __kprobes
625 do_device_not_available(struct pt_regs *regs, long error_code)
627 #ifdef CONFIG_MATH_EMULATION
628 if (read_cr0() & X86_CR0_EM) {
629 struct math_emu_info info = { };
631 conditional_sti(regs);
638 math_state_restore(); /* interrupts still off */
640 conditional_sti(regs);
645 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
650 info.si_signo = SIGILL;
652 info.si_code = ILL_BADSTK;
654 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
655 X86_TRAP_IRET, SIGILL) == NOTIFY_STOP)
657 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
662 /* Set of traps needed for early debugging. */
663 void __init early_trap_init(void)
665 set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
666 /* int3 can be called from all */
667 set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
668 set_intr_gate(X86_TRAP_PF, &page_fault);
669 load_idt(&idt_descr);
672 void __init trap_init(void)
677 void __iomem *p = early_ioremap(0x0FFFD9, 4);
679 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
684 set_intr_gate(X86_TRAP_DE, ÷_error);
685 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
686 /* int4 can be called from all */
687 set_system_intr_gate(X86_TRAP_OF, &overflow);
688 set_intr_gate(X86_TRAP_BR, &bounds);
689 set_intr_gate(X86_TRAP_UD, &invalid_op);
690 set_intr_gate(X86_TRAP_NM, &device_not_available);
692 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
694 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
696 set_intr_gate(X86_TRAP_OLD_MF, &coprocessor_segment_overrun);
697 set_intr_gate(X86_TRAP_TS, &invalid_TSS);
698 set_intr_gate(X86_TRAP_NP, &segment_not_present);
699 set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
700 set_intr_gate(X86_TRAP_GP, &general_protection);
701 set_intr_gate(X86_TRAP_SPURIOUS, &spurious_interrupt_bug);
702 set_intr_gate(X86_TRAP_MF, &coprocessor_error);
703 set_intr_gate(X86_TRAP_AC, &alignment_check);
704 #ifdef CONFIG_X86_MCE
705 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
707 set_intr_gate(X86_TRAP_XF, &simd_coprocessor_error);
709 /* Reserve all the builtin and the syscall vector: */
710 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
711 set_bit(i, used_vectors);
713 #ifdef CONFIG_IA32_EMULATION
714 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
715 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
719 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
720 set_bit(SYSCALL_VECTOR, used_vectors);
724 * Should be a barrier for any external CPU state:
728 x86_init.irqs.trap_init();
731 memcpy(&nmi_idt_table, &idt_table, IDT_ENTRIES * 16);
732 set_nmi_gate(X86_TRAP_DB, &debug);
733 set_nmi_gate(X86_TRAP_BP, &int3);