2 * linux/arch/i386/traps.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Pentium III FXSR, SSE support
7 * Gareth Hughes <gareth@valinux.com>, May 2000
11 * 'Traps.c' handles hardware traps and faults after we have saved some
14 #include <linux/config.h>
15 #include <linux/sched.h>
16 #include <linux/kernel.h>
17 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/timer.h>
21 #include <linux/init.h>
22 #include <linux/delay.h>
23 #include <linux/spinlock.h>
24 #include <linux/interrupt.h>
25 #include <linux/highmem.h>
26 #include <linux/kallsyms.h>
27 #include <linux/ptrace.h>
28 #include <linux/utsname.h>
29 #include <linux/kprobes.h>
30 #include <linux/kexec.h>
33 #include <linux/ioport.h>
34 #include <linux/eisa.h>
38 #include <linux/mca.h>
41 #include <asm/processor.h>
42 #include <asm/system.h>
43 #include <asm/uaccess.h>
45 #include <asm/atomic.h>
46 #include <asm/debugreg.h>
52 #include <asm/arch_hooks.h>
53 #include <asm/kdebug.h>
55 #include <linux/module.h>
57 #include "mach_traps.h"
59 asmlinkage int system_call(void);
61 struct desc_struct default_ldt[] = { { 0, 0 }, { 0, 0 }, { 0, 0 },
64 /* Do we ignore FPU interrupts ? */
65 char ignore_fpu_irq = 0;
68 * The IDT has to be page-aligned to simplify the Pentium
69 * F0 0F bug workaround.. We have a special link segment
72 struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
74 asmlinkage void divide_error(void);
75 asmlinkage void debug(void);
76 asmlinkage void nmi(void);
77 asmlinkage void int3(void);
78 asmlinkage void overflow(void);
79 asmlinkage void bounds(void);
80 asmlinkage void invalid_op(void);
81 asmlinkage void device_not_available(void);
82 asmlinkage void coprocessor_segment_overrun(void);
83 asmlinkage void invalid_TSS(void);
84 asmlinkage void segment_not_present(void);
85 asmlinkage void stack_segment(void);
86 asmlinkage void general_protection(void);
87 asmlinkage void page_fault(void);
88 asmlinkage void coprocessor_error(void);
89 asmlinkage void simd_coprocessor_error(void);
90 asmlinkage void alignment_check(void);
91 asmlinkage void spurious_interrupt_bug(void);
92 asmlinkage void machine_check(void);
94 static int kstack_depth_to_print = 24;
95 ATOMIC_NOTIFIER_HEAD(i386die_chain);
97 int register_die_notifier(struct notifier_block *nb)
100 return atomic_notifier_chain_register(&i386die_chain, nb);
102 EXPORT_SYMBOL(register_die_notifier);
104 int unregister_die_notifier(struct notifier_block *nb)
106 return atomic_notifier_chain_unregister(&i386die_chain, nb);
108 EXPORT_SYMBOL(unregister_die_notifier);
110 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
112 return p > (void *)tinfo &&
113 p < (void *)tinfo + THREAD_SIZE - 3;
117 * Print CONFIG_STACK_BACKTRACE_COLS address/symbol entries per line.
119 static inline int print_addr_and_symbol(unsigned long addr, char *log_lvl,
125 #if CONFIG_STACK_BACKTRACE_COLS == 1
126 printk(" [<%08lx>] ", addr);
128 printk(" <%08lx> ", addr);
130 print_symbol("%s", addr);
132 printed = (printed + 1) % CONFIG_STACK_BACKTRACE_COLS;
142 static inline unsigned long print_context_stack(struct thread_info *tinfo,
143 unsigned long *stack, unsigned long ebp,
147 int printed = 0; /* nr of entries already printed on current line */
149 #ifdef CONFIG_FRAME_POINTER
150 while (valid_stack_ptr(tinfo, (void *)ebp)) {
151 addr = *(unsigned long *)(ebp + 4);
152 printed = print_addr_and_symbol(addr, log_lvl, printed);
153 ebp = *(unsigned long *)ebp;
156 while (valid_stack_ptr(tinfo, stack)) {
158 if (__kernel_text_address(addr))
159 printed = print_addr_and_symbol(addr, log_lvl, printed);
168 static void show_trace_log_lvl(struct task_struct *task,
169 unsigned long *stack, char *log_lvl)
176 if (task == current) {
177 /* Grab ebp right from our regs */
178 asm ("movl %%ebp, %0" : "=r" (ebp) : );
180 /* ebp is the last reg pushed by switch_to */
181 ebp = *(unsigned long *) task->thread.esp;
185 struct thread_info *context;
186 context = (struct thread_info *)
187 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
188 ebp = print_context_stack(context, stack, ebp, log_lvl);
189 stack = (unsigned long*)context->previous_esp;
192 printk("%s =======================\n", log_lvl);
196 void show_trace(struct task_struct *task, unsigned long * stack)
198 show_trace_log_lvl(task, stack, "");
201 static void show_stack_log_lvl(struct task_struct *task, unsigned long *esp,
204 unsigned long *stack;
209 esp = (unsigned long*)task->thread.esp;
211 esp = (unsigned long *)&esp;
216 for(i = 0; i < kstack_depth_to_print; i++) {
217 if (kstack_end(stack))
219 if (i && ((i % 8) == 0))
220 printk("\n%s ", log_lvl);
221 printk("%08lx ", *stack++);
223 printk("\n%sCall Trace:\n", log_lvl);
224 show_trace_log_lvl(task, esp, log_lvl);
227 void show_stack(struct task_struct *task, unsigned long *esp)
230 show_stack_log_lvl(task, esp, "");
234 * The architecture-independent dump_stack generator
236 void dump_stack(void)
240 show_trace(current, &stack);
243 EXPORT_SYMBOL(dump_stack);
245 void show_registers(struct pt_regs *regs)
252 esp = (unsigned long) (®s->esp);
254 if (user_mode_vm(regs)) {
257 ss = regs->xss & 0xffff;
260 printk(KERN_EMERG "CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\n"
261 "EFLAGS: %08lx (%s %.*s) \n",
262 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
263 print_tainted(), regs->eflags, system_utsname.release,
264 (int)strcspn(system_utsname.version, " "),
265 system_utsname.version);
266 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
267 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
268 regs->eax, regs->ebx, regs->ecx, regs->edx);
269 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
270 regs->esi, regs->edi, regs->ebp, esp);
271 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
272 regs->xds & 0xffff, regs->xes & 0xffff, ss);
273 printk(KERN_EMERG "Process %s (pid: %d, threadinfo=%p task=%p)",
274 current->comm, current->pid, current_thread_info(), current);
276 * When in-kernel, we also print out the stack and code at the
277 * time of the fault..
282 printk("\n" KERN_EMERG "Stack: ");
283 show_stack_log_lvl(NULL, (unsigned long *)esp, KERN_EMERG);
285 printk(KERN_EMERG "Code: ");
287 eip = (u8 __user *)regs->eip - 43;
288 for (i = 0; i < 64; i++, eip++) {
291 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
292 printk(" Bad EIP value.");
295 if (eip == (u8 __user *)regs->eip)
296 printk("<%02x> ", c);
304 static void handle_BUG(struct pt_regs *regs)
314 if (eip < PAGE_OFFSET)
316 if (__get_user(ud2, (unsigned short __user *)eip))
320 if (__get_user(line, (unsigned short __user *)(eip + 2)))
322 if (__get_user(file, (char * __user *)(eip + 4)) ||
323 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
324 file = "<bad filename>";
326 printk(KERN_EMERG "------------[ cut here ]------------\n");
327 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
332 /* Here we know it was a BUG but file-n-line is unavailable */
334 printk(KERN_EMERG "Kernel BUG\n");
337 /* This is gone through when something in the kernel
338 * has done something bad and is about to be terminated.
340 void die(const char * str, struct pt_regs * regs, long err)
345 int lock_owner_depth;
347 .lock = SPIN_LOCK_UNLOCKED,
349 .lock_owner_depth = 0
351 static int die_counter;
356 if (die.lock_owner != raw_smp_processor_id()) {
358 spin_lock_irqsave(&die.lock, flags);
359 die.lock_owner = smp_processor_id();
360 die.lock_owner_depth = 0;
364 local_save_flags(flags);
366 if (++die.lock_owner_depth < 3) {
372 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
373 #ifdef CONFIG_PREEMPT
374 printk(KERN_EMERG "PREEMPT ");
383 #ifdef CONFIG_DEBUG_PAGEALLOC
386 printk("DEBUG_PAGEALLOC");
391 if (notify_die(DIE_OOPS, str, regs, err,
392 current->thread.trap_no, SIGSEGV) !=
394 show_registers(regs);
395 /* Executive summary in case the oops scrolled away */
396 esp = (unsigned long) (®s->esp);
398 if (user_mode(regs)) {
400 ss = regs->xss & 0xffff;
402 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
403 print_symbol("%s", regs->eip);
404 printk(" SS:ESP %04x:%08lx\n", ss, esp);
409 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
413 spin_unlock_irqrestore(&die.lock, flags);
418 if (kexec_should_crash(current))
422 panic("Fatal exception in interrupt");
425 printk(KERN_EMERG "Fatal exception: panic in 5 seconds\n");
427 panic("Fatal exception");
433 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
435 if (!user_mode_vm(regs))
439 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
440 struct pt_regs * regs, long error_code,
443 struct task_struct *tsk = current;
444 tsk->thread.error_code = error_code;
445 tsk->thread.trap_no = trapnr;
447 if (regs->eflags & VM_MASK) {
453 if (!user_mode(regs))
458 force_sig_info(signr, info, tsk);
460 force_sig(signr, tsk);
465 if (!fixup_exception(regs))
466 die(str, regs, error_code);
471 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
472 if (ret) goto trap_signal;
477 #define DO_ERROR(trapnr, signr, str, name) \
478 fastcall void do_##name(struct pt_regs * regs, long error_code) \
480 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
483 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
486 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
487 fastcall void do_##name(struct pt_regs * regs, long error_code) \
490 info.si_signo = signr; \
492 info.si_code = sicode; \
493 info.si_addr = (void __user *)siaddr; \
494 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
497 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
500 #define DO_VM86_ERROR(trapnr, signr, str, name) \
501 fastcall void do_##name(struct pt_regs * regs, long error_code) \
503 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
506 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
509 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
510 fastcall void do_##name(struct pt_regs * regs, long error_code) \
513 info.si_signo = signr; \
515 info.si_code = sicode; \
516 info.si_addr = (void __user *)siaddr; \
517 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
520 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
523 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
524 #ifndef CONFIG_KPROBES
525 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
527 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
528 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
529 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
530 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
531 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
532 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
533 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
534 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
535 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
537 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
541 struct tss_struct *tss = &per_cpu(init_tss, cpu);
542 struct thread_struct *thread = ¤t->thread;
545 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
546 * invalid offset set (the LAZY one) and the faulting thread has
547 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
548 * and we set the offset field correctly. Then we let the CPU to
549 * restart the faulting instruction.
551 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
552 thread->io_bitmap_ptr) {
553 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
554 thread->io_bitmap_max);
556 * If the previously set map was extending to higher ports
557 * than the current one, pad extra space with 0xff (no access).
559 if (thread->io_bitmap_max < tss->io_bitmap_max)
560 memset((char *) tss->io_bitmap +
561 thread->io_bitmap_max, 0xff,
562 tss->io_bitmap_max - thread->io_bitmap_max);
563 tss->io_bitmap_max = thread->io_bitmap_max;
564 tss->io_bitmap_base = IO_BITMAP_OFFSET;
565 tss->io_bitmap_owner = thread;
571 current->thread.error_code = error_code;
572 current->thread.trap_no = 13;
574 if (regs->eflags & VM_MASK)
577 if (!user_mode(regs))
580 current->thread.error_code = error_code;
581 current->thread.trap_no = 13;
582 force_sig(SIGSEGV, current);
587 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
591 if (!fixup_exception(regs)) {
592 if (notify_die(DIE_GPF, "general protection fault", regs,
593 error_code, 13, SIGSEGV) == NOTIFY_STOP)
595 die("general protection fault", regs, error_code);
599 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
601 printk(KERN_EMERG "Uhhuh. NMI received. Dazed and confused, but trying "
603 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
606 /* Clear and disable the memory parity error line. */
607 clear_mem_error(reason);
610 static void io_check_error(unsigned char reason, struct pt_regs * regs)
614 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
615 show_registers(regs);
617 /* Re-enable the IOCK line, wait for a few seconds */
618 reason = (reason & 0xf) | 8;
621 while (--i) udelay(1000);
626 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
629 /* Might actually be able to figure out what the guilty party
636 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
637 reason, smp_processor_id());
638 printk("Dazed and confused, but trying to continue\n");
639 printk("Do you have a strange power saving mode enabled?\n");
642 static DEFINE_SPINLOCK(nmi_print_lock);
644 void die_nmi (struct pt_regs *regs, const char *msg)
646 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
650 spin_lock(&nmi_print_lock);
652 * We are in trouble anyway, lets at least try
653 * to get a message out.
656 printk(KERN_EMERG "%s", msg);
657 printk(" on CPU%d, eip %08lx, registers:\n",
658 smp_processor_id(), regs->eip);
659 show_registers(regs);
660 printk(KERN_EMERG "console shuts up ...\n");
662 spin_unlock(&nmi_print_lock);
665 /* If we are in kernel we are probably nested up pretty bad
666 * and might aswell get out now while we still can.
668 if (!user_mode_vm(regs)) {
669 current->thread.trap_no = 2;
676 static void default_do_nmi(struct pt_regs * regs)
678 unsigned char reason = 0;
680 /* Only the BSP gets external NMIs from the system. */
681 if (!smp_processor_id())
682 reason = get_nmi_reason();
684 if (!(reason & 0xc0)) {
685 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
688 #ifdef CONFIG_X86_LOCAL_APIC
690 * Ok, so this is none of the documented NMI sources,
691 * so it must be the NMI watchdog.
694 nmi_watchdog_tick(regs);
698 unknown_nmi_error(reason, regs);
701 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
704 mem_parity_error(reason, regs);
706 io_check_error(reason, regs);
708 * Reassert NMI in case it became active meanwhile
709 * as it's edge-triggered.
714 static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
719 static nmi_callback_t nmi_callback = dummy_nmi_callback;
721 fastcall void do_nmi(struct pt_regs * regs, long error_code)
727 cpu = smp_processor_id();
731 if (!rcu_dereference(nmi_callback)(regs, cpu))
732 default_do_nmi(regs);
737 void set_nmi_callback(nmi_callback_t callback)
740 rcu_assign_pointer(nmi_callback, callback);
742 EXPORT_SYMBOL_GPL(set_nmi_callback);
744 void unset_nmi_callback(void)
746 nmi_callback = dummy_nmi_callback;
748 EXPORT_SYMBOL_GPL(unset_nmi_callback);
750 #ifdef CONFIG_KPROBES
751 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
753 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
756 /* This is an interrupt gate, because kprobes wants interrupts
757 disabled. Normal trap handlers don't. */
758 restore_interrupts(regs);
759 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
764 * Our handling of the processor debug registers is non-trivial.
765 * We do not clear them on entry and exit from the kernel. Therefore
766 * it is possible to get a watchpoint trap here from inside the kernel.
767 * However, the code in ./ptrace.c has ensured that the user can
768 * only set watchpoints on userspace addresses. Therefore the in-kernel
769 * watchpoint trap can only occur in code which is reading/writing
770 * from user space. Such code must not hold kernel locks (since it
771 * can equally take a page fault), therefore it is safe to call
772 * force_sig_info even though that claims and releases locks.
774 * Code in ./signal.c ensures that the debug control register
775 * is restored before we deliver any signal, and therefore that
776 * user code runs with the correct debug control register even though
779 * Being careful here means that we don't have to be as careful in a
780 * lot of more complicated places (task switching can be a bit lazy
781 * about restoring all the debug state, and ptrace doesn't have to
782 * find every occurrence of the TF bit that could be saved away even
785 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
787 unsigned int condition;
788 struct task_struct *tsk = current;
790 get_debugreg(condition, 6);
792 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
793 SIGTRAP) == NOTIFY_STOP)
795 /* It's safe to allow irq's after DR6 has been saved */
796 if (regs->eflags & X86_EFLAGS_IF)
799 /* Mask out spurious debug traps due to lazy DR7 setting */
800 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
801 if (!tsk->thread.debugreg[7])
805 if (regs->eflags & VM_MASK)
808 /* Save debug status register where ptrace can see it */
809 tsk->thread.debugreg[6] = condition;
812 * Single-stepping through TF: make sure we ignore any events in
813 * kernel space (but re-enable TF when returning to user mode).
815 if (condition & DR_STEP) {
817 * We already checked v86 mode above, so we can
818 * check for kernel mode by just checking the CPL
821 if (!user_mode(regs))
822 goto clear_TF_reenable;
825 /* Ok, finally something we can handle */
826 send_sigtrap(tsk, regs, error_code);
828 /* Disable additional traps. They'll be re-enabled when
829 * the signal is delivered.
836 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
840 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
841 regs->eflags &= ~TF_MASK;
846 * Note that we play around with the 'TS' bit in an attempt to get
847 * the correct behaviour even in the presence of the asynchronous
850 void math_error(void __user *eip)
852 struct task_struct * task;
854 unsigned short cwd, swd;
857 * Save the info for the exception handler and clear the error.
861 task->thread.trap_no = 16;
862 task->thread.error_code = 0;
863 info.si_signo = SIGFPE;
865 info.si_code = __SI_FAULT;
868 * (~cwd & swd) will mask out exceptions that are not set to unmasked
869 * status. 0x3f is the exception bits in these regs, 0x200 is the
870 * C1 reg you need in case of a stack fault, 0x040 is the stack
871 * fault bit. We should only be taking one exception at a time,
872 * so if this combination doesn't produce any single exception,
873 * then we have a bad program that isn't syncronizing its FPU usage
874 * and it will suffer the consequences since we won't be able to
875 * fully reproduce the context of the exception
877 cwd = get_fpu_cwd(task);
878 swd = get_fpu_swd(task);
879 switch (swd & ~cwd & 0x3f) {
880 case 0x000: /* No unmasked exception */
882 default: /* Multiple exceptions */
884 case 0x001: /* Invalid Op */
886 * swd & 0x240 == 0x040: Stack Underflow
887 * swd & 0x240 == 0x240: Stack Overflow
888 * User must clear the SF bit (0x40) if set
890 info.si_code = FPE_FLTINV;
892 case 0x002: /* Denormalize */
893 case 0x010: /* Underflow */
894 info.si_code = FPE_FLTUND;
896 case 0x004: /* Zero Divide */
897 info.si_code = FPE_FLTDIV;
899 case 0x008: /* Overflow */
900 info.si_code = FPE_FLTOVF;
902 case 0x020: /* Precision */
903 info.si_code = FPE_FLTRES;
906 force_sig_info(SIGFPE, &info, task);
909 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
912 math_error((void __user *)regs->eip);
915 static void simd_math_error(void __user *eip)
917 struct task_struct * task;
919 unsigned short mxcsr;
922 * Save the info for the exception handler and clear the error.
926 task->thread.trap_no = 19;
927 task->thread.error_code = 0;
928 info.si_signo = SIGFPE;
930 info.si_code = __SI_FAULT;
933 * The SIMD FPU exceptions are handled a little differently, as there
934 * is only a single status/control register. Thus, to determine which
935 * unmasked exception was caught we must mask the exception mask bits
936 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
938 mxcsr = get_fpu_mxcsr(task);
939 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
943 case 0x001: /* Invalid Op */
944 info.si_code = FPE_FLTINV;
946 case 0x002: /* Denormalize */
947 case 0x010: /* Underflow */
948 info.si_code = FPE_FLTUND;
950 case 0x004: /* Zero Divide */
951 info.si_code = FPE_FLTDIV;
953 case 0x008: /* Overflow */
954 info.si_code = FPE_FLTOVF;
956 case 0x020: /* Precision */
957 info.si_code = FPE_FLTRES;
960 force_sig_info(SIGFPE, &info, task);
963 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
967 /* Handle SIMD FPU exceptions on PIII+ processors. */
969 simd_math_error((void __user *)regs->eip);
972 * Handle strange cache flush from user space exception
973 * in all other cases. This is undocumented behaviour.
975 if (regs->eflags & VM_MASK) {
976 handle_vm86_fault((struct kernel_vm86_regs *)regs,
980 current->thread.trap_no = 19;
981 current->thread.error_code = error_code;
982 die_if_kernel("cache flush denied", regs, error_code);
983 force_sig(SIGSEGV, current);
987 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
991 /* No need to warn about this any longer. */
992 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
996 fastcall void setup_x86_bogus_stack(unsigned char * stk)
998 unsigned long *switch16_ptr, *switch32_ptr;
999 struct pt_regs *regs;
1000 unsigned long stack_top, stack_bot;
1001 unsigned short iret_frame16_off;
1002 int cpu = smp_processor_id();
1003 /* reserve the space on 32bit stack for the magic switch16 pointer */
1004 memmove(stk, stk + 8, sizeof(struct pt_regs));
1005 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
1006 regs = (struct pt_regs *)stk;
1007 /* now the switch32 on 16bit stack */
1008 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1009 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1010 switch32_ptr = (unsigned long *)(stack_top - 8);
1011 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
1012 /* copy iret frame on 16bit stack */
1013 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
1014 /* fill in the switch pointers */
1015 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
1016 switch16_ptr[1] = __ESPFIX_SS;
1017 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
1018 8 - CPU_16BIT_STACK_SIZE;
1019 switch32_ptr[1] = __KERNEL_DS;
1022 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
1024 unsigned long *switch32_ptr;
1025 unsigned char *stack16, *stack32;
1026 unsigned long stack_top, stack_bot;
1028 int cpu = smp_processor_id();
1029 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1030 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1031 switch32_ptr = (unsigned long *)(stack_top - 8);
1032 /* copy the data from 16bit stack to 32bit stack */
1033 len = CPU_16BIT_STACK_SIZE - 8 - sp;
1034 stack16 = (unsigned char *)(stack_bot + sp);
1035 stack32 = (unsigned char *)
1036 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
1037 memcpy(stack32, stack16, len);
1042 * 'math_state_restore()' saves the current math information in the
1043 * old math state array, and gets the new ones from the current task
1045 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1046 * Don't touch unless you *really* know how it works.
1048 * Must be called with kernel preemption disabled (in this case,
1049 * local interrupts are disabled at the call-site in entry.S).
1051 asmlinkage void math_state_restore(struct pt_regs regs)
1053 struct thread_info *thread = current_thread_info();
1054 struct task_struct *tsk = thread->task;
1056 clts(); /* Allow maths ops (or we recurse) */
1057 if (!tsk_used_math(tsk))
1060 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1063 #ifndef CONFIG_MATH_EMULATION
1065 asmlinkage void math_emulate(long arg)
1067 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1068 printk(KERN_EMERG "killing %s.\n",current->comm);
1069 force_sig(SIGFPE,current);
1073 #endif /* CONFIG_MATH_EMULATION */
1075 #ifdef CONFIG_X86_F00F_BUG
1076 void __init trap_init_f00f_bug(void)
1078 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1081 * Update the IDT descriptor and reload the IDT so that
1082 * it uses the read-only mapped virtual address.
1084 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1085 load_idt(&idt_descr);
1089 #define _set_gate(gate_addr,type,dpl,addr,seg) \
1092 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1093 "movw %4,%%dx\n\t" \
1094 "movl %%eax,%0\n\t" \
1096 :"=m" (*((long *) (gate_addr))), \
1097 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1098 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1099 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1104 * This needs to use 'idt_table' rather than 'idt', and
1105 * thus use the _nonmapped_ version of the IDT, as the
1106 * Pentium F0 0F bugfix can have resulted in the mapped
1107 * IDT being write-protected.
1109 void set_intr_gate(unsigned int n, void *addr)
1111 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1115 * This routine sets up an interrupt gate at directory privilege level 3.
1117 static inline void set_system_intr_gate(unsigned int n, void *addr)
1119 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1122 static void __init set_trap_gate(unsigned int n, void *addr)
1124 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1127 static void __init set_system_gate(unsigned int n, void *addr)
1129 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1132 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1134 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1138 void __init trap_init(void)
1141 void __iomem *p = ioremap(0x0FFFD9, 4);
1142 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1148 #ifdef CONFIG_X86_LOCAL_APIC
1149 init_apic_mappings();
1152 set_trap_gate(0,÷_error);
1153 set_intr_gate(1,&debug);
1154 set_intr_gate(2,&nmi);
1155 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1156 set_system_gate(4,&overflow);
1157 set_trap_gate(5,&bounds);
1158 set_trap_gate(6,&invalid_op);
1159 set_trap_gate(7,&device_not_available);
1160 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1161 set_trap_gate(9,&coprocessor_segment_overrun);
1162 set_trap_gate(10,&invalid_TSS);
1163 set_trap_gate(11,&segment_not_present);
1164 set_trap_gate(12,&stack_segment);
1165 set_trap_gate(13,&general_protection);
1166 set_intr_gate(14,&page_fault);
1167 set_trap_gate(15,&spurious_interrupt_bug);
1168 set_trap_gate(16,&coprocessor_error);
1169 set_trap_gate(17,&alignment_check);
1170 #ifdef CONFIG_X86_MCE
1171 set_trap_gate(18,&machine_check);
1173 set_trap_gate(19,&simd_coprocessor_error);
1177 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1178 * Generates a compile-time "error: zero width for bit-field" if
1179 * the alignment is wrong.
1181 struct fxsrAlignAssert {
1182 int _:!(offsetof(struct task_struct,
1183 thread.i387.fxsave) & 15);
1186 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1187 set_in_cr4(X86_CR4_OSFXSR);
1191 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1193 set_in_cr4(X86_CR4_OSXMMEXCPT);
1197 set_system_gate(SYSCALL_VECTOR,&system_call);
1200 * Should be a barrier for any external CPU state.
1207 static int __init kstack_setup(char *s)
1209 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1212 __setup("kstack=", kstack_setup);