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>
57 #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)
111 static void __kprobes
112 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
113 long error_code, siginfo_t *info)
115 struct task_struct *tsk = current;
118 if (regs->flags & X86_VM_MASK) {
120 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
121 * On nmi (interrupt 2), do_trap should not be called.
129 if (!user_mode(regs))
136 * We want error_code and trap_no set for userspace faults and
137 * kernelspace faults which result in die(), but not
138 * kernelspace faults which are fixed up. die() gives the
139 * process no chance to handle the signal and notice the
140 * kernel fault information, so that won't result in polluting
141 * the information about previously queued, but not yet
142 * delivered, faults. See also do_general_protection below.
144 tsk->thread.error_code = error_code;
145 tsk->thread.trap_no = trapnr;
148 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
149 printk_ratelimit()) {
151 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
152 tsk->comm, tsk->pid, str,
153 regs->ip, regs->sp, error_code);
154 print_vma_addr(" in ", regs->ip);
160 force_sig_info(signr, info, tsk);
162 force_sig(signr, tsk);
166 if (!fixup_exception(regs)) {
167 tsk->thread.error_code = error_code;
168 tsk->thread.trap_no = trapnr;
169 die(str, regs, error_code);
175 if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
182 #define DO_ERROR(trapnr, signr, str, name) \
183 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
185 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
188 conditional_sti(regs); \
189 do_trap(trapnr, signr, str, regs, error_code, NULL); \
192 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
193 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
196 info.si_signo = signr; \
198 info.si_code = sicode; \
199 info.si_addr = (void __user *)siaddr; \
200 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
203 conditional_sti(regs); \
204 do_trap(trapnr, signr, str, regs, error_code, &info); \
207 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
208 DO_ERROR(4, SIGSEGV, "overflow", overflow)
209 DO_ERROR(5, SIGSEGV, "bounds", bounds)
210 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
211 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
212 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
213 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
215 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
217 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
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 12, SIGBUS) == NOTIFY_STOP)
226 preempt_conditional_sti(regs);
227 do_trap(12, 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, 8, SIGSEGV);
239 tsk->thread.error_code = error_code;
240 tsk->thread.trap_no = 8;
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_no = 13;
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_no = 13;
296 if (notify_die(DIE_GPF, "general protection fault", regs,
297 error_code, 13, SIGSEGV) == NOTIFY_STOP)
299 die("general protection fault", regs, error_code);
302 /* May run on IST stack. */
303 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
305 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
306 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
309 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
311 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
316 * Let others (NMI) know that the debug stack is in use
317 * as we may switch to the interrupt stack.
319 debug_stack_usage_inc();
320 preempt_conditional_sti(regs);
321 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
322 preempt_conditional_cli(regs);
323 debug_stack_usage_dec();
328 * Help handler running on IST stack to switch back to user stack
329 * for scheduling or signal handling. The actual stack switch is done in
332 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
334 struct pt_regs *regs = eregs;
335 /* Did already sync */
336 if (eregs == (struct pt_regs *)eregs->sp)
338 /* Exception from user space */
339 else if (user_mode(eregs))
340 regs = task_pt_regs(current);
342 * Exception from kernel and interrupts are enabled. Move to
343 * kernel process stack.
345 else if (eregs->flags & X86_EFLAGS_IF)
346 regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
354 * Our handling of the processor debug registers is non-trivial.
355 * We do not clear them on entry and exit from the kernel. Therefore
356 * it is possible to get a watchpoint trap here from inside the kernel.
357 * However, the code in ./ptrace.c has ensured that the user can
358 * only set watchpoints on userspace addresses. Therefore the in-kernel
359 * watchpoint trap can only occur in code which is reading/writing
360 * from user space. Such code must not hold kernel locks (since it
361 * can equally take a page fault), therefore it is safe to call
362 * force_sig_info even though that claims and releases locks.
364 * Code in ./signal.c ensures that the debug control register
365 * is restored before we deliver any signal, and therefore that
366 * user code runs with the correct debug control register even though
369 * Being careful here means that we don't have to be as careful in a
370 * lot of more complicated places (task switching can be a bit lazy
371 * about restoring all the debug state, and ptrace doesn't have to
372 * find every occurrence of the TF bit that could be saved away even
375 * May run on IST stack.
377 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
379 struct task_struct *tsk = current;
384 get_debugreg(dr6, 6);
386 /* Filter out all the reserved bits which are preset to 1 */
387 dr6 &= ~DR6_RESERVED;
390 * If dr6 has no reason to give us about the origin of this trap,
391 * then it's very likely the result of an icebp/int01 trap.
392 * User wants a sigtrap for that.
394 if (!dr6 && user_mode(regs))
397 /* Catch kmemcheck conditions first of all! */
398 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
401 /* DR6 may or may not be cleared by the CPU */
405 * The processor cleared BTF, so don't mark that we need it set.
407 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
409 /* Store the virtualized DR6 value */
410 tsk->thread.debugreg6 = dr6;
412 if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
413 SIGTRAP) == NOTIFY_STOP)
417 * Let others (NMI) know that the debug stack is in use
418 * as we may switch to the interrupt stack.
420 debug_stack_usage_inc();
422 /* It's safe to allow irq's after DR6 has been saved */
423 preempt_conditional_sti(regs);
425 if (regs->flags & X86_VM_MASK) {
426 handle_vm86_trap((struct kernel_vm86_regs *) regs,
428 preempt_conditional_cli(regs);
429 debug_stack_usage_dec();
434 * Single-stepping through system calls: ignore any exceptions in
435 * kernel space, but re-enable TF when returning to user mode.
437 * We already checked v86 mode above, so we can check for kernel mode
438 * by just checking the CPL of CS.
440 if ((dr6 & DR_STEP) && !user_mode(regs)) {
441 tsk->thread.debugreg6 &= ~DR_STEP;
442 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
443 regs->flags &= ~X86_EFLAGS_TF;
445 si_code = get_si_code(tsk->thread.debugreg6);
446 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
447 send_sigtrap(tsk, regs, error_code, si_code);
448 preempt_conditional_cli(regs);
449 debug_stack_usage_dec();
455 * Note that we play around with the 'TS' bit in an attempt to get
456 * the correct behaviour even in the presence of the asynchronous
459 void math_error(struct pt_regs *regs, int error_code, int trapnr)
461 struct task_struct *task = current;
464 char *str = (trapnr == 16) ? "fpu exception" : "simd exception";
466 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
468 conditional_sti(regs);
470 if (!user_mode_vm(regs))
472 if (!fixup_exception(regs)) {
473 task->thread.error_code = error_code;
474 task->thread.trap_no = trapnr;
475 die(str, regs, error_code);
481 * Save the info for the exception handler and clear the error.
484 task->thread.trap_no = trapnr;
485 task->thread.error_code = error_code;
486 info.si_signo = SIGFPE;
488 info.si_addr = (void __user *)regs->ip;
490 unsigned short cwd, swd;
492 * (~cwd & swd) will mask out exceptions that are not set to unmasked
493 * status. 0x3f is the exception bits in these regs, 0x200 is the
494 * C1 reg you need in case of a stack fault, 0x040 is the stack
495 * fault bit. We should only be taking one exception at a time,
496 * so if this combination doesn't produce any single exception,
497 * then we have a bad program that isn't synchronizing its FPU usage
498 * and it will suffer the consequences since we won't be able to
499 * fully reproduce the context of the exception
501 cwd = get_fpu_cwd(task);
502 swd = get_fpu_swd(task);
507 * The SIMD FPU exceptions are handled a little differently, as there
508 * is only a single status/control register. Thus, to determine which
509 * unmasked exception was caught we must mask the exception mask bits
510 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
512 unsigned short mxcsr = get_fpu_mxcsr(task);
513 err = ~(mxcsr >> 7) & mxcsr;
516 if (err & 0x001) { /* Invalid op */
518 * swd & 0x240 == 0x040: Stack Underflow
519 * swd & 0x240 == 0x240: Stack Overflow
520 * User must clear the SF bit (0x40) if set
522 info.si_code = FPE_FLTINV;
523 } else if (err & 0x004) { /* Divide by Zero */
524 info.si_code = FPE_FLTDIV;
525 } else if (err & 0x008) { /* Overflow */
526 info.si_code = FPE_FLTOVF;
527 } else if (err & 0x012) { /* Denormal, Underflow */
528 info.si_code = FPE_FLTUND;
529 } else if (err & 0x020) { /* Precision */
530 info.si_code = FPE_FLTRES;
533 * If we're using IRQ 13, or supposedly even some trap 16
534 * implementations, it's possible we get a spurious trap...
536 return; /* Spurious trap, no error */
538 force_sig_info(SIGFPE, &info, task);
541 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
547 math_error(regs, error_code, 16);
551 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
553 math_error(regs, error_code, 19);
557 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
559 conditional_sti(regs);
561 /* No need to warn about this any longer. */
562 printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
566 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
570 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
575 * 'math_state_restore()' saves the current math information in the
576 * old math state array, and gets the new ones from the current task
578 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
579 * Don't touch unless you *really* know how it works.
581 * Must be called with kernel preemption disabled (eg with local
582 * local interrupts as in the case of do_device_not_available).
584 void math_state_restore(void)
586 struct task_struct *tsk = current;
588 if (!tsk_used_math(tsk)) {
591 * does a slab alloc which can sleep
597 do_group_exit(SIGKILL);
603 __thread_fpu_begin(tsk);
605 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
607 if (unlikely(restore_fpu_checking(tsk))) {
608 __thread_fpu_end(tsk);
609 force_sig(SIGSEGV, tsk);
615 EXPORT_SYMBOL_GPL(math_state_restore);
617 dotraplinkage void __kprobes
618 do_device_not_available(struct pt_regs *regs, long error_code)
620 #ifdef CONFIG_MATH_EMULATION
621 if (read_cr0() & X86_CR0_EM) {
622 struct math_emu_info info = { };
624 conditional_sti(regs);
631 math_state_restore(); /* interrupts still off */
633 conditional_sti(regs);
638 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
643 info.si_signo = SIGILL;
645 info.si_code = ILL_BADSTK;
647 if (notify_die(DIE_TRAP, "iret exception",
648 regs, error_code, 32, SIGILL) == NOTIFY_STOP)
650 do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
654 /* Set of traps needed for early debugging. */
655 void __init early_trap_init(void)
657 set_intr_gate_ist(1, &debug, DEBUG_STACK);
658 /* int3 can be called from all */
659 set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
660 set_intr_gate(14, &page_fault);
661 load_idt(&idt_descr);
664 void __init trap_init(void)
669 void __iomem *p = early_ioremap(0x0FFFD9, 4);
671 if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
676 set_intr_gate(0, ÷_error);
677 set_intr_gate_ist(2, &nmi, NMI_STACK);
678 /* int4 can be called from all */
679 set_system_intr_gate(4, &overflow);
680 set_intr_gate(5, &bounds);
681 set_intr_gate(6, &invalid_op);
682 set_intr_gate(7, &device_not_available);
684 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
686 set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
688 set_intr_gate(9, &coprocessor_segment_overrun);
689 set_intr_gate(10, &invalid_TSS);
690 set_intr_gate(11, &segment_not_present);
691 set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
692 set_intr_gate(13, &general_protection);
693 set_intr_gate(15, &spurious_interrupt_bug);
694 set_intr_gate(16, &coprocessor_error);
695 set_intr_gate(17, &alignment_check);
696 #ifdef CONFIG_X86_MCE
697 set_intr_gate_ist(18, &machine_check, MCE_STACK);
699 set_intr_gate(19, &simd_coprocessor_error);
701 /* Reserve all the builtin and the syscall vector: */
702 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
703 set_bit(i, used_vectors);
705 #ifdef CONFIG_IA32_EMULATION
706 set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
707 set_bit(IA32_SYSCALL_VECTOR, used_vectors);
711 set_system_trap_gate(SYSCALL_VECTOR, &system_call);
712 set_bit(SYSCALL_VECTOR, used_vectors);
716 * Should be a barrier for any external CPU state:
720 x86_init.irqs.trap_init();
723 memcpy(&nmi_idt_table, &idt_table, IDT_ENTRIES * 16);
724 set_nmi_gate(1, &debug);
725 set_nmi_gate(3, &int3);