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>
41 #include <linux/mca.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/system.h>
54 #include <asm/traps.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.
128 if (!user_mode(regs))
135 * We want error_code and trap_no 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_no = trapnr;
147 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
148 printk_ratelimit()) {
150 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
151 tsk->comm, tsk->pid, str,
152 regs->ip, regs->sp, error_code);
153 print_vma_addr(" in ", regs->ip);
159 force_sig_info(signr, info, tsk);
161 force_sig(signr, tsk);
165 if (!fixup_exception(regs)) {
166 tsk->thread.error_code = error_code;
167 tsk->thread.trap_no = trapnr;
168 die(str, regs, error_code);
174 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
181 #define DO_ERROR(trapnr, signr, str, name) \
182 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
184 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
187 conditional_sti(regs); \
188 do_trap(trapnr, signr, str, regs, error_code, NULL); \
191 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
192 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
195 info.si_signo = signr; \
197 info.si_code = sicode; \
198 info.si_addr = (void __user *)siaddr; \
199 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
202 conditional_sti(regs); \
203 do_trap(trapnr, signr, str, regs, error_code, &info); \
206 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
207 DO_ERROR(4, SIGSEGV, "overflow", overflow)
208 DO_ERROR(5, SIGSEGV, "bounds", bounds)
209 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
210 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
211 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
212 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
214 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
216 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
219 /* Runs on IST stack */
220 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
222 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
223 12, SIGBUS) == NOTIFY_STOP)
225 preempt_conditional_sti(regs);
226 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
227 preempt_conditional_cli(regs);
230 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
232 static const char str[] = "double fault";
233 struct task_struct *tsk = current;
235 /* Return not checked because double check cannot be ignored */
236 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
238 tsk->thread.error_code = error_code;
239 tsk->thread.trap_no = 8;
242 * This is always a kernel trap and never fixable (and thus must
246 die(str, regs, error_code);
250 dotraplinkage void __kprobes
251 do_general_protection(struct pt_regs *regs, long error_code)
253 struct task_struct *tsk;
255 conditional_sti(regs);
258 if (regs->flags & X86_VM_MASK)
263 if (!user_mode(regs))
266 tsk->thread.error_code = error_code;
267 tsk->thread.trap_no = 13;
269 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
270 printk_ratelimit()) {
272 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
273 tsk->comm, task_pid_nr(tsk),
274 regs->ip, regs->sp, error_code);
275 print_vma_addr(" in ", regs->ip);
279 force_sig(SIGSEGV, tsk);
285 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
290 if (fixup_exception(regs))
293 tsk->thread.error_code = error_code;
294 tsk->thread.trap_no = 13;
295 if (notify_die(DIE_GPF, "general protection fault", regs,
296 error_code, 13, SIGSEGV) == NOTIFY_STOP)
298 die("general protection fault", regs, error_code);
301 /* May run on IST stack. */
302 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
304 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
305 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
308 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
309 #ifdef CONFIG_KPROBES
310 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
314 if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
320 * Let others (NMI) know that the debug stack is in use
321 * as we may switch to the interrupt stack.
323 debug_stack_usage_inc();
324 preempt_conditional_sti(regs);
325 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
326 preempt_conditional_cli(regs);
327 debug_stack_usage_dec();
332 * Help handler running on IST stack to switch back to user stack
333 * for scheduling or signal handling. The actual stack switch is done in
336 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
338 struct pt_regs *regs = eregs;
339 /* Did already sync */
340 if (eregs == (struct pt_regs *)eregs->sp)
342 /* Exception from user space */
343 else if (user_mode(eregs))
344 regs = task_pt_regs(current);
346 * Exception from kernel and interrupts are enabled. Move to
347 * kernel process stack.
349 else if (eregs->flags & X86_EFLAGS_IF)
350 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
358 * Our handling of the processor debug registers is non-trivial.
359 * We do not clear them on entry and exit from the kernel. Therefore
360 * it is possible to get a watchpoint trap here from inside the kernel.
361 * However, the code in ./ptrace.c has ensured that the user can
362 * only set watchpoints on userspace addresses. Therefore the in-kernel
363 * watchpoint trap can only occur in code which is reading/writing
364 * from user space. Such code must not hold kernel locks (since it
365 * can equally take a page fault), therefore it is safe to call
366 * force_sig_info even though that claims and releases locks.
368 * Code in ./signal.c ensures that the debug control register
369 * is restored before we deliver any signal, and therefore that
370 * user code runs with the correct debug control register even though
373 * Being careful here means that we don't have to be as careful in a
374 * lot of more complicated places (task switching can be a bit lazy
375 * about restoring all the debug state, and ptrace doesn't have to
376 * find every occurrence of the TF bit that could be saved away even
379 * May run on IST stack.
381 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
383 struct task_struct *tsk = current;
388 get_debugreg(dr6, 6);
390 /* Filter out all the reserved bits which are preset to 1 */
391 dr6 &= ~DR6_RESERVED;
394 * If dr6 has no reason to give us about the origin of this trap,
395 * then it's very likely the result of an icebp/int01 trap.
396 * User wants a sigtrap for that.
398 if (!dr6 && user_mode(regs))
401 /* Catch kmemcheck conditions first of all! */
402 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
405 /* DR6 may or may not be cleared by the CPU */
409 * The processor cleared BTF, so don't mark that we need it set.
411 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
413 /* Store the virtualized DR6 value */
414 tsk->thread.debugreg6 = dr6;
416 if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
417 SIGTRAP) == NOTIFY_STOP)
421 * Let others (NMI) know that the debug stack is in use
422 * as we may switch to the interrupt stack.
424 debug_stack_usage_inc();
426 /* It's safe to allow irq's after DR6 has been saved */
427 preempt_conditional_sti(regs);
429 if (regs->flags & X86_VM_MASK) {
430 handle_vm86_trap((struct kernel_vm86_regs *) regs,
432 preempt_conditional_cli(regs);
433 debug_stack_usage_dec();
438 * Single-stepping through system calls: ignore any exceptions in
439 * kernel space, but re-enable TF when returning to user mode.
441 * We already checked v86 mode above, so we can check for kernel mode
442 * by just checking the CPL of CS.
444 if ((dr6 & DR_STEP) && !user_mode(regs)) {
445 tsk->thread.debugreg6 &= ~DR_STEP;
446 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
447 regs->flags &= ~X86_EFLAGS_TF;
449 si_code = get_si_code(tsk->thread.debugreg6);
450 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
451 send_sigtrap(tsk, regs, error_code, si_code);
452 preempt_conditional_cli(regs);
453 debug_stack_usage_dec();
459 * Note that we play around with the 'TS' bit in an attempt to get
460 * the correct behaviour even in the presence of the asynchronous
463 void math_error(struct pt_regs *regs, int error_code, int trapnr)
465 struct task_struct *task = current;
468 char *str = (trapnr == 16) ? "fpu exception" : "simd exception";
470 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
472 conditional_sti(regs);
474 if (!user_mode_vm(regs))
476 if (!fixup_exception(regs)) {
477 task->thread.error_code = error_code;
478 task->thread.trap_no = trapnr;
479 die(str, regs, error_code);
485 * Save the info for the exception handler and clear the error.
488 task->thread.trap_no = trapnr;
489 task->thread.error_code = error_code;
490 info.si_signo = SIGFPE;
492 info.si_addr = (void __user *)regs->ip;
494 unsigned short cwd, swd;
496 * (~cwd & swd) will mask out exceptions that are not set to unmasked
497 * status. 0x3f is the exception bits in these regs, 0x200 is the
498 * C1 reg you need in case of a stack fault, 0x040 is the stack
499 * fault bit. We should only be taking one exception at a time,
500 * so if this combination doesn't produce any single exception,
501 * then we have a bad program that isn't synchronizing its FPU usage
502 * and it will suffer the consequences since we won't be able to
503 * fully reproduce the context of the exception
505 cwd = get_fpu_cwd(task);
506 swd = get_fpu_swd(task);
511 * The SIMD FPU exceptions are handled a little differently, as there
512 * is only a single status/control register. Thus, to determine which
513 * unmasked exception was caught we must mask the exception mask bits
514 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
516 unsigned short mxcsr = get_fpu_mxcsr(task);
517 err = ~(mxcsr >> 7) & mxcsr;
520 if (err & 0x001) { /* Invalid op */
522 * swd & 0x240 == 0x040: Stack Underflow
523 * swd & 0x240 == 0x240: Stack Overflow
524 * User must clear the SF bit (0x40) if set
526 info.si_code = FPE_FLTINV;
527 } else if (err & 0x004) { /* Divide by Zero */
528 info.si_code = FPE_FLTDIV;
529 } else if (err & 0x008) { /* Overflow */
530 info.si_code = FPE_FLTOVF;
531 } else if (err & 0x012) { /* Denormal, Underflow */
532 info.si_code = FPE_FLTUND;
533 } else if (err & 0x020) { /* Precision */
534 info.si_code = FPE_FLTRES;
537 * If we're using IRQ 13, or supposedly even some trap 16
538 * implementations, it's possible we get a spurious trap...
540 return; /* Spurious trap, no error */
542 force_sig_info(SIGFPE, &info, task);
545 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
551 math_error(regs, error_code, 16);
555 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
557 math_error(regs, error_code, 19);
561 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
563 conditional_sti(regs);
565 /* No need to warn about this any longer. */
566 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
570 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
574 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
579 * __math_state_restore assumes that cr0.TS is already clear and the
580 * fpu state is all ready for use. Used during context switch.
582 void __math_state_restore(void)
584 struct thread_info *thread = current_thread_info();
585 struct task_struct *tsk = thread->task;
588 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
590 if (unlikely(restore_fpu_checking(tsk))) {
592 force_sig(SIGSEGV, tsk);
596 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
601 * 'math_state_restore()' saves the current math information in the
602 * old math state array, and gets the new ones from the current task
604 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
605 * Don't touch unless you *really* know how it works.
607 * Must be called with kernel preemption disabled (in this case,
608 * local interrupts are disabled at the call-site in entry.S).
610 asmlinkage void math_state_restore(void)
612 struct thread_info *thread = current_thread_info();
613 struct task_struct *tsk = thread->task;
615 if (!tsk_used_math(tsk)) {
618 * does a slab alloc which can sleep
624 do_group_exit(SIGKILL);
630 clts(); /* Allow maths ops (or we recurse) */
632 __math_state_restore();
634 EXPORT_SYMBOL_GPL(math_state_restore);
636 dotraplinkage void __kprobes
637 do_device_not_available(struct pt_regs *regs, long error_code)
639 #ifdef CONFIG_MATH_EMULATION
640 if (read_cr0() & X86_CR0_EM) {
641 struct math_emu_info info = { };
643 conditional_sti(regs);
650 math_state_restore(); /* interrupts still off */
652 conditional_sti(regs);
657 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
662 info.si_signo = SIGILL;
664 info.si_code = ILL_BADSTK;
666 if (notify_die(DIE_TRAP, "iret exception",
667 regs, error_code, 32, SIGILL) == NOTIFY_STOP)
669 do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
673 /* Set of traps needed for early debugging. */
674 void __init early_trap_init(void)
676 set_intr_gate_ist(1, &debug, DEBUG_STACK);
677 /* int3 can be called from all */
678 set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
679 set_intr_gate(14, &page_fault);
680 load_idt(&idt_descr);
683 void __init trap_init(void)
688 void __iomem *p = early_ioremap(0x0FFFD9, 4);
690 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
695 set_intr_gate(0, ÷_error);
696 set_intr_gate_ist(2, &nmi, NMI_STACK);
697 /* int4 can be called from all */
698 set_system_intr_gate(4, &overflow);
699 set_intr_gate(5, &bounds);
700 set_intr_gate(6, &invalid_op);
701 set_intr_gate(7, &device_not_available);
703 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
705 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
707 set_intr_gate(9, &coprocessor_segment_overrun);
708 set_intr_gate(10, &invalid_TSS);
709 set_intr_gate(11, &segment_not_present);
710 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
711 set_intr_gate(13, &general_protection);
712 set_intr_gate(15, &spurious_interrupt_bug);
713 set_intr_gate(16, &coprocessor_error);
714 set_intr_gate(17, &alignment_check);
715 #ifdef CONFIG_X86_MCE
716 set_intr_gate_ist(18, &machine_check, MCE_STACK);
718 set_intr_gate(19, &simd_coprocessor_error);
720 /* Reserve all the builtin and the syscall vector: */
721 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
722 set_bit(i, used_vectors);
724 #ifdef CONFIG_IA32_EMULATION
725 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
726 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
730 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
731 set_bit(SYSCALL_VECTOR, used_vectors);
735 * Should be a barrier for any external CPU state:
739 x86_init.irqs.trap_init();
742 memcpy(&nmi_idt_table, &idt_table, IDT_ENTRIES * 16);
743 set_nmi_gate(1, &debug);
744 set_nmi_gate(3, &int3);