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>
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);
71 /* Do we ignore FPU interrupts ? */
75 * The IDT has to be page-aligned to simplify the Pentium
76 * F0 0F bug workaround.
78 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
81 DECLARE_BITMAP(used_vectors, NR_VECTORS);
82 EXPORT_SYMBOL_GPL(used_vectors);
84 static inline void conditional_sti(struct pt_regs *regs)
86 if (regs->flags & X86_EFLAGS_IF)
90 static inline void preempt_conditional_sti(struct pt_regs *regs)
93 if (regs->flags & X86_EFLAGS_IF)
97 static inline void conditional_cli(struct pt_regs *regs)
99 if (regs->flags & X86_EFLAGS_IF)
103 static inline void preempt_conditional_cli(struct pt_regs *regs)
105 if (regs->flags & X86_EFLAGS_IF)
110 static void __kprobes
111 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
112 long error_code, siginfo_t *info)
114 struct task_struct *tsk = current;
117 if (regs->flags & X86_VM_MASK) {
119 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
120 * On nmi (interrupt 2), do_trap should not be called.
122 if (trapnr < X86_TRAP_UD)
128 if (!user_mode(regs))
135 * We want error_code and trap_nr set for userspace faults and
136 * kernelspace faults which result in die(), but not
137 * kernelspace faults which are fixed up. die() gives the
138 * process no chance to handle the signal and notice the
139 * kernel fault information, so that won't result in polluting
140 * the information about previously queued, but not yet
141 * delivered, faults. See also do_general_protection below.
143 tsk->thread.error_code = error_code;
144 tsk->thread.trap_nr = trapnr;
147 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
148 printk_ratelimit()) {
149 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
150 tsk->comm, tsk->pid, str,
151 regs->ip, regs->sp, error_code);
152 print_vma_addr(" in ", regs->ip);
158 force_sig_info(signr, info, tsk);
160 force_sig(signr, tsk);
164 if (!fixup_exception(regs)) {
165 tsk->thread.error_code = error_code;
166 tsk->thread.trap_nr = trapnr;
167 die(str, regs, error_code);
173 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
180 #define DO_ERROR(trapnr, signr, str, name) \
181 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
183 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
186 conditional_sti(regs); \
187 do_trap(trapnr, signr, str, regs, error_code, NULL); \
190 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
191 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
194 info.si_signo = signr; \
196 info.si_code = sicode; \
197 info.si_addr = (void __user *)siaddr; \
198 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
201 conditional_sti(regs); \
202 do_trap(trapnr, signr, str, regs, error_code, &info); \
205 DO_ERROR_INFO(X86_TRAP_DE, SIGFPE, "divide error", divide_error, FPE_INTDIV,
207 DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
208 DO_ERROR(X86_TRAP_BR, SIGSEGV, "bounds", bounds)
209 DO_ERROR_INFO(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN,
211 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",
212 coprocessor_segment_overrun)
213 DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
214 DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
216 DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
218 DO_ERROR_INFO(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check,
222 /* Runs on IST stack */
223 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
225 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
226 X86_TRAP_SS, SIGBUS) == NOTIFY_STOP)
228 preempt_conditional_sti(regs);
229 do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
230 preempt_conditional_cli(regs);
233 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
235 static const char str[] = "double fault";
236 struct task_struct *tsk = current;
238 /* Return not checked because double check cannot be ignored */
239 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
241 tsk->thread.error_code = error_code;
242 tsk->thread.trap_nr = X86_TRAP_DF;
245 * This is always a kernel trap and never fixable (and thus must
249 die(str, regs, error_code);
253 dotraplinkage void __kprobes
254 do_general_protection(struct pt_regs *regs, long error_code)
256 struct task_struct *tsk;
258 conditional_sti(regs);
261 if (regs->flags & X86_VM_MASK)
266 if (!user_mode(regs))
269 tsk->thread.error_code = error_code;
270 tsk->thread.trap_nr = X86_TRAP_GP;
272 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
273 printk_ratelimit()) {
274 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
275 tsk->comm, task_pid_nr(tsk),
276 regs->ip, regs->sp, error_code);
277 print_vma_addr(" in ", regs->ip);
281 force_sig(SIGSEGV, tsk);
287 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
292 if (fixup_exception(regs))
295 tsk->thread.error_code = error_code;
296 tsk->thread.trap_nr = X86_TRAP_GP;
297 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
298 X86_TRAP_GP, SIGSEGV) == NOTIFY_STOP)
300 die("general protection fault", regs, error_code);
303 /* May run on IST stack. */
304 dotraplinkage void __kprobes notrace do_int3(struct pt_regs *regs, long error_code)
306 #ifdef CONFIG_DYNAMIC_FTRACE
308 * ftrace must be first, everything else may cause a recursive crash.
309 * See note by declaration of modifying_ftrace_code in ftrace.c
311 if (unlikely(atomic_read(&modifying_ftrace_code)) &&
312 ftrace_int3_handler(regs))
315 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
316 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
317 SIGTRAP) == NOTIFY_STOP)
319 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
321 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
322 SIGTRAP) == NOTIFY_STOP)
326 * Let others (NMI) know that the debug stack is in use
327 * as we may switch to the interrupt stack.
329 debug_stack_usage_inc();
330 preempt_conditional_sti(regs);
331 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
332 preempt_conditional_cli(regs);
333 debug_stack_usage_dec();
338 * Help handler running on IST stack to switch back to user stack
339 * for scheduling or signal handling. The actual stack switch is done in
342 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
344 struct pt_regs *regs = eregs;
345 /* Did already sync */
346 if (eregs == (struct pt_regs *)eregs->sp)
348 /* Exception from user space */
349 else if (user_mode(eregs))
350 regs = task_pt_regs(current);
352 * Exception from kernel and interrupts are enabled. Move to
353 * kernel process stack.
355 else if (eregs->flags & X86_EFLAGS_IF)
356 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
364 * Our handling of the processor debug registers is non-trivial.
365 * We do not clear them on entry and exit from the kernel. Therefore
366 * it is possible to get a watchpoint trap here from inside the kernel.
367 * However, the code in ./ptrace.c has ensured that the user can
368 * only set watchpoints on userspace addresses. Therefore the in-kernel
369 * watchpoint trap can only occur in code which is reading/writing
370 * from user space. Such code must not hold kernel locks (since it
371 * can equally take a page fault), therefore it is safe to call
372 * force_sig_info even though that claims and releases locks.
374 * Code in ./signal.c ensures that the debug control register
375 * is restored before we deliver any signal, and therefore that
376 * user code runs with the correct debug control register even though
379 * Being careful here means that we don't have to be as careful in a
380 * lot of more complicated places (task switching can be a bit lazy
381 * about restoring all the debug state, and ptrace doesn't have to
382 * find every occurrence of the TF bit that could be saved away even
385 * May run on IST stack.
387 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
389 struct task_struct *tsk = current;
394 get_debugreg(dr6, 6);
396 /* Filter out all the reserved bits which are preset to 1 */
397 dr6 &= ~DR6_RESERVED;
400 * If dr6 has no reason to give us about the origin of this trap,
401 * then it's very likely the result of an icebp/int01 trap.
402 * User wants a sigtrap for that.
404 if (!dr6 && user_mode(regs))
407 /* Catch kmemcheck conditions first of all! */
408 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
411 /* DR6 may or may not be cleared by the CPU */
415 * The processor cleared BTF, so don't mark that we need it set.
417 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
419 /* Store the virtualized DR6 value */
420 tsk->thread.debugreg6 = dr6;
422 if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
423 SIGTRAP) == NOTIFY_STOP)
427 * Let others (NMI) know that the debug stack is in use
428 * as we may switch to the interrupt stack.
430 debug_stack_usage_inc();
432 /* It's safe to allow irq's after DR6 has been saved */
433 preempt_conditional_sti(regs);
435 if (regs->flags & X86_VM_MASK) {
436 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
438 preempt_conditional_cli(regs);
439 debug_stack_usage_dec();
444 * Single-stepping through system calls: ignore any exceptions in
445 * kernel space, but re-enable TF when returning to user mode.
447 * We already checked v86 mode above, so we can check for kernel mode
448 * by just checking the CPL of CS.
450 if ((dr6 & DR_STEP) && !user_mode(regs)) {
451 tsk->thread.debugreg6 &= ~DR_STEP;
452 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
453 regs->flags &= ~X86_EFLAGS_TF;
455 si_code = get_si_code(tsk->thread.debugreg6);
456 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
457 send_sigtrap(tsk, regs, error_code, si_code);
458 preempt_conditional_cli(regs);
459 debug_stack_usage_dec();
465 * Note that we play around with the 'TS' bit in an attempt to get
466 * the correct behaviour even in the presence of the asynchronous
469 void math_error(struct pt_regs *regs, int error_code, int trapnr)
471 struct task_struct *task = current;
474 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
477 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
479 conditional_sti(regs);
481 if (!user_mode_vm(regs))
483 if (!fixup_exception(regs)) {
484 task->thread.error_code = error_code;
485 task->thread.trap_nr = trapnr;
486 die(str, regs, error_code);
492 * Save the info for the exception handler and clear the error.
495 task->thread.trap_nr = trapnr;
496 task->thread.error_code = error_code;
497 info.si_signo = SIGFPE;
499 info.si_addr = (void __user *)regs->ip;
500 if (trapnr == X86_TRAP_MF) {
501 unsigned short cwd, swd;
503 * (~cwd & swd) will mask out exceptions that are not set to unmasked
504 * status. 0x3f is the exception bits in these regs, 0x200 is the
505 * C1 reg you need in case of a stack fault, 0x040 is the stack
506 * fault bit. We should only be taking one exception at a time,
507 * so if this combination doesn't produce any single exception,
508 * then we have a bad program that isn't synchronizing its FPU usage
509 * and it will suffer the consequences since we won't be able to
510 * fully reproduce the context of the exception
512 cwd = get_fpu_cwd(task);
513 swd = get_fpu_swd(task);
518 * The SIMD FPU exceptions are handled a little differently, as there
519 * is only a single status/control register. Thus, to determine which
520 * unmasked exception was caught we must mask the exception mask bits
521 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
523 unsigned short mxcsr = get_fpu_mxcsr(task);
524 err = ~(mxcsr >> 7) & mxcsr;
527 if (err & 0x001) { /* Invalid op */
529 * swd & 0x240 == 0x040: Stack Underflow
530 * swd & 0x240 == 0x240: Stack Overflow
531 * User must clear the SF bit (0x40) if set
533 info.si_code = FPE_FLTINV;
534 } else if (err & 0x004) { /* Divide by Zero */
535 info.si_code = FPE_FLTDIV;
536 } else if (err & 0x008) { /* Overflow */
537 info.si_code = FPE_FLTOVF;
538 } else if (err & 0x012) { /* Denormal, Underflow */
539 info.si_code = FPE_FLTUND;
540 } else if (err & 0x020) { /* Precision */
541 info.si_code = FPE_FLTRES;
544 * If we're using IRQ 13, or supposedly even some trap
545 * X86_TRAP_MF implementations, it's possible
546 * we get a spurious trap, which is not an error.
550 force_sig_info(SIGFPE, &info, task);
553 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
559 math_error(regs, error_code, X86_TRAP_MF);
563 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
565 math_error(regs, error_code, X86_TRAP_XF);
569 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
571 conditional_sti(regs);
573 /* No need to warn about this any longer. */
574 pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
578 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
582 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
587 * 'math_state_restore()' saves the current math information in the
588 * old math state array, and gets the new ones from the current task
590 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
591 * Don't touch unless you *really* know how it works.
593 * Must be called with kernel preemption disabled (eg with local
594 * local interrupts as in the case of do_device_not_available).
596 void math_state_restore(void)
598 struct task_struct *tsk = current;
600 if (!tsk_used_math(tsk)) {
603 * does a slab alloc which can sleep
609 do_group_exit(SIGKILL);
615 __thread_fpu_begin(tsk);
618 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
620 if (unlikely(restore_fpu_checking(tsk))) {
622 force_sig(SIGSEGV, tsk);
628 EXPORT_SYMBOL_GPL(math_state_restore);
630 dotraplinkage void __kprobes
631 do_device_not_available(struct pt_regs *regs, long error_code)
633 BUG_ON(use_eager_fpu());
635 #ifdef CONFIG_MATH_EMULATION
636 if (read_cr0() & X86_CR0_EM) {
637 struct math_emu_info info = { };
639 conditional_sti(regs);
646 math_state_restore(); /* interrupts still off */
648 conditional_sti(regs);
653 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
658 info.si_signo = SIGILL;
660 info.si_code = ILL_BADSTK;
662 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
663 X86_TRAP_IRET, SIGILL) == NOTIFY_STOP)
665 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
670 /* Set of traps needed for early debugging. */
671 void __init early_trap_init(void)
673 set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
674 /* int3 can be called from all */
675 set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
676 set_intr_gate(X86_TRAP_PF, &page_fault);
677 load_idt(&idt_descr);
680 void __init trap_init(void)
685 void __iomem *p = early_ioremap(0x0FFFD9, 4);
687 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
692 set_intr_gate(X86_TRAP_DE, ÷_error);
693 set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
694 /* int4 can be called from all */
695 set_system_intr_gate(X86_TRAP_OF, &overflow);
696 set_intr_gate(X86_TRAP_BR, &bounds);
697 set_intr_gate(X86_TRAP_UD, &invalid_op);
698 set_intr_gate(X86_TRAP_NM, &device_not_available);
700 set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
702 set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
704 set_intr_gate(X86_TRAP_OLD_MF, &coprocessor_segment_overrun);
705 set_intr_gate(X86_TRAP_TS, &invalid_TSS);
706 set_intr_gate(X86_TRAP_NP, &segment_not_present);
707 set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
708 set_intr_gate(X86_TRAP_GP, &general_protection);
709 set_intr_gate(X86_TRAP_SPURIOUS, &spurious_interrupt_bug);
710 set_intr_gate(X86_TRAP_MF, &coprocessor_error);
711 set_intr_gate(X86_TRAP_AC, &alignment_check);
712 #ifdef CONFIG_X86_MCE
713 set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
715 set_intr_gate(X86_TRAP_XF, &simd_coprocessor_error);
717 /* Reserve all the builtin and the syscall vector: */
718 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
719 set_bit(i, used_vectors);
721 #ifdef CONFIG_IA32_EMULATION
722 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
723 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
727 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
728 set_bit(SYSCALL_VECTOR, used_vectors);
732 * Should be a barrier for any external CPU state:
736 x86_init.irqs.trap_init();
739 memcpy(&nmi_idt_table, &idt_table, IDT_ENTRIES * 16);
740 set_nmi_gate(X86_TRAP_DB, &debug);
741 set_nmi_gate(X86_TRAP_BP, &int3);