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;
141 static inline unsigned long print_context_stack(struct thread_info *tinfo,
142 unsigned long *stack, unsigned long ebp,
146 int printed = 0; /* nr of entries already printed on current line */
148 #ifdef CONFIG_FRAME_POINTER
149 while (valid_stack_ptr(tinfo, (void *)ebp)) {
150 addr = *(unsigned long *)(ebp + 4);
151 printed = print_addr_and_symbol(addr, log_lvl, printed);
152 ebp = *(unsigned long *)ebp;
155 while (valid_stack_ptr(tinfo, stack)) {
157 if (__kernel_text_address(addr))
158 printed = print_addr_and_symbol(addr, log_lvl, printed);
167 static void show_trace_log_lvl(struct task_struct *task,
168 unsigned long *stack, char *log_lvl)
175 if (task == current) {
176 /* Grab ebp right from our regs */
177 asm ("movl %%ebp, %0" : "=r" (ebp) : );
179 /* ebp is the last reg pushed by switch_to */
180 ebp = *(unsigned long *) task->thread.esp;
184 struct thread_info *context;
185 context = (struct thread_info *)
186 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
187 ebp = print_context_stack(context, stack, ebp, log_lvl);
188 stack = (unsigned long*)context->previous_esp;
191 printk("%s =======================\n", log_lvl);
195 void show_trace(struct task_struct *task, unsigned long * stack)
197 show_trace_log_lvl(task, stack, "");
200 static void show_stack_log_lvl(struct task_struct *task, unsigned long *esp,
203 unsigned long *stack;
208 esp = (unsigned long*)task->thread.esp;
210 esp = (unsigned long *)&esp;
214 for(i = 0; i < kstack_depth_to_print; i++) {
215 if (kstack_end(stack))
217 if (i && ((i % 8) == 0))
218 printk("\n%s ", log_lvl);
219 printk("%08lx ", *stack++);
221 printk("\n%sCall Trace:\n", log_lvl);
222 show_trace_log_lvl(task, esp, log_lvl);
225 void show_stack(struct task_struct *task, unsigned long *esp)
228 show_stack_log_lvl(task, esp, "");
232 * The architecture-independent dump_stack generator
234 void dump_stack(void)
238 show_trace(current, &stack);
241 EXPORT_SYMBOL(dump_stack);
243 void show_registers(struct pt_regs *regs)
250 esp = (unsigned long) (®s->esp);
252 if (user_mode_vm(regs)) {
255 ss = regs->xss & 0xffff;
258 printk(KERN_EMERG "CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\n"
259 "EFLAGS: %08lx (%s %.*s) \n",
260 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
261 print_tainted(), regs->eflags, system_utsname.release,
262 (int)strcspn(system_utsname.version, " "),
263 system_utsname.version);
264 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
265 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
266 regs->eax, regs->ebx, regs->ecx, regs->edx);
267 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
268 regs->esi, regs->edi, regs->ebp, esp);
269 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
270 regs->xds & 0xffff, regs->xes & 0xffff, ss);
271 printk(KERN_EMERG "Process %s (pid: %d, threadinfo=%p task=%p)",
272 current->comm, current->pid, current_thread_info(), current);
274 * When in-kernel, we also print out the stack and code at the
275 * time of the fault..
280 printk("\n" KERN_EMERG "Stack: ");
281 show_stack_log_lvl(NULL, (unsigned long *)esp, KERN_EMERG);
283 printk(KERN_EMERG "Code: ");
285 eip = (u8 __user *)regs->eip - 43;
286 for (i = 0; i < 64; i++, eip++) {
289 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
290 printk(" Bad EIP value.");
293 if (eip == (u8 __user *)regs->eip)
294 printk("<%02x> ", c);
302 static void handle_BUG(struct pt_regs *regs)
312 if (eip < PAGE_OFFSET)
314 if (__get_user(ud2, (unsigned short __user *)eip))
318 if (__get_user(line, (unsigned short __user *)(eip + 2)))
320 if (__get_user(file, (char * __user *)(eip + 4)) ||
321 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
322 file = "<bad filename>";
324 printk(KERN_EMERG "------------[ cut here ]------------\n");
325 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
330 /* Here we know it was a BUG but file-n-line is unavailable */
332 printk(KERN_EMERG "Kernel BUG\n");
335 /* This is gone through when something in the kernel
336 * has done something bad and is about to be terminated.
338 void die(const char * str, struct pt_regs * regs, long err)
343 int lock_owner_depth;
345 .lock = SPIN_LOCK_UNLOCKED,
347 .lock_owner_depth = 0
349 static int die_counter;
354 if (die.lock_owner != raw_smp_processor_id()) {
356 spin_lock_irqsave(&die.lock, flags);
357 die.lock_owner = smp_processor_id();
358 die.lock_owner_depth = 0;
362 local_save_flags(flags);
364 if (++die.lock_owner_depth < 3) {
370 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
371 #ifdef CONFIG_PREEMPT
372 printk(KERN_EMERG "PREEMPT ");
381 #ifdef CONFIG_DEBUG_PAGEALLOC
384 printk("DEBUG_PAGEALLOC");
389 if (notify_die(DIE_OOPS, str, regs, err,
390 current->thread.trap_no, SIGSEGV) !=
392 show_registers(regs);
393 /* Executive summary in case the oops scrolled away */
394 esp = (unsigned long) (®s->esp);
396 if (user_mode(regs)) {
398 ss = regs->xss & 0xffff;
400 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
401 print_symbol("%s", regs->eip);
402 printk(" SS:ESP %04x:%08lx\n", ss, esp);
407 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
411 spin_unlock_irqrestore(&die.lock, flags);
416 if (kexec_should_crash(current))
420 panic("Fatal exception in interrupt");
423 printk(KERN_EMERG "Fatal exception: panic in 5 seconds\n");
425 panic("Fatal exception");
431 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
433 if (!user_mode_vm(regs))
437 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
438 struct pt_regs * regs, long error_code,
441 struct task_struct *tsk = current;
442 tsk->thread.error_code = error_code;
443 tsk->thread.trap_no = trapnr;
445 if (regs->eflags & VM_MASK) {
451 if (!user_mode(regs))
456 force_sig_info(signr, info, tsk);
458 force_sig(signr, tsk);
463 if (!fixup_exception(regs))
464 die(str, regs, error_code);
469 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
470 if (ret) goto trap_signal;
475 #define DO_ERROR(trapnr, signr, str, name) \
476 fastcall void do_##name(struct pt_regs * regs, long error_code) \
478 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
481 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
484 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
485 fastcall void do_##name(struct pt_regs * regs, long error_code) \
488 info.si_signo = signr; \
490 info.si_code = sicode; \
491 info.si_addr = (void __user *)siaddr; \
492 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
495 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
498 #define DO_VM86_ERROR(trapnr, signr, str, name) \
499 fastcall void do_##name(struct pt_regs * regs, long error_code) \
501 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
504 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
507 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
508 fastcall void do_##name(struct pt_regs * regs, long error_code) \
511 info.si_signo = signr; \
513 info.si_code = sicode; \
514 info.si_addr = (void __user *)siaddr; \
515 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
518 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
521 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
522 #ifndef CONFIG_KPROBES
523 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
525 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
526 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
527 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
528 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
529 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
530 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
531 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
532 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
533 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
535 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
539 struct tss_struct *tss = &per_cpu(init_tss, cpu);
540 struct thread_struct *thread = ¤t->thread;
543 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
544 * invalid offset set (the LAZY one) and the faulting thread has
545 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
546 * and we set the offset field correctly. Then we let the CPU to
547 * restart the faulting instruction.
549 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
550 thread->io_bitmap_ptr) {
551 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
552 thread->io_bitmap_max);
554 * If the previously set map was extending to higher ports
555 * than the current one, pad extra space with 0xff (no access).
557 if (thread->io_bitmap_max < tss->io_bitmap_max)
558 memset((char *) tss->io_bitmap +
559 thread->io_bitmap_max, 0xff,
560 tss->io_bitmap_max - thread->io_bitmap_max);
561 tss->io_bitmap_max = thread->io_bitmap_max;
562 tss->io_bitmap_base = IO_BITMAP_OFFSET;
563 tss->io_bitmap_owner = thread;
569 current->thread.error_code = error_code;
570 current->thread.trap_no = 13;
572 if (regs->eflags & VM_MASK)
575 if (!user_mode(regs))
578 current->thread.error_code = error_code;
579 current->thread.trap_no = 13;
580 force_sig(SIGSEGV, current);
585 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
589 if (!fixup_exception(regs)) {
590 if (notify_die(DIE_GPF, "general protection fault", regs,
591 error_code, 13, SIGSEGV) == NOTIFY_STOP)
593 die("general protection fault", regs, error_code);
597 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
599 printk(KERN_EMERG "Uhhuh. NMI received. Dazed and confused, but trying "
601 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
604 /* Clear and disable the memory parity error line. */
605 clear_mem_error(reason);
608 static void io_check_error(unsigned char reason, struct pt_regs * regs)
612 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
613 show_registers(regs);
615 /* Re-enable the IOCK line, wait for a few seconds */
616 reason = (reason & 0xf) | 8;
619 while (--i) udelay(1000);
624 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
627 /* Might actually be able to figure out what the guilty party
634 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
635 reason, smp_processor_id());
636 printk("Dazed and confused, but trying to continue\n");
637 printk("Do you have a strange power saving mode enabled?\n");
640 static DEFINE_SPINLOCK(nmi_print_lock);
642 void die_nmi (struct pt_regs *regs, const char *msg)
644 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
648 spin_lock(&nmi_print_lock);
650 * We are in trouble anyway, lets at least try
651 * to get a message out.
654 printk(KERN_EMERG "%s", msg);
655 printk(" on CPU%d, eip %08lx, registers:\n",
656 smp_processor_id(), regs->eip);
657 show_registers(regs);
658 printk(KERN_EMERG "console shuts up ...\n");
660 spin_unlock(&nmi_print_lock);
663 /* If we are in kernel we are probably nested up pretty bad
664 * and might aswell get out now while we still can.
666 if (!user_mode_vm(regs)) {
667 current->thread.trap_no = 2;
674 static void default_do_nmi(struct pt_regs * regs)
676 unsigned char reason = 0;
678 /* Only the BSP gets external NMIs from the system. */
679 if (!smp_processor_id())
680 reason = get_nmi_reason();
682 if (!(reason & 0xc0)) {
683 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
686 #ifdef CONFIG_X86_LOCAL_APIC
688 * Ok, so this is none of the documented NMI sources,
689 * so it must be the NMI watchdog.
692 nmi_watchdog_tick(regs);
696 unknown_nmi_error(reason, regs);
699 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
702 mem_parity_error(reason, regs);
704 io_check_error(reason, regs);
706 * Reassert NMI in case it became active meanwhile
707 * as it's edge-triggered.
712 static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
717 static nmi_callback_t nmi_callback = dummy_nmi_callback;
719 fastcall void do_nmi(struct pt_regs * regs, long error_code)
725 cpu = smp_processor_id();
729 if (!rcu_dereference(nmi_callback)(regs, cpu))
730 default_do_nmi(regs);
735 void set_nmi_callback(nmi_callback_t callback)
738 rcu_assign_pointer(nmi_callback, callback);
740 EXPORT_SYMBOL_GPL(set_nmi_callback);
742 void unset_nmi_callback(void)
744 nmi_callback = dummy_nmi_callback;
746 EXPORT_SYMBOL_GPL(unset_nmi_callback);
748 #ifdef CONFIG_KPROBES
749 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
751 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
754 /* This is an interrupt gate, because kprobes wants interrupts
755 disabled. Normal trap handlers don't. */
756 restore_interrupts(regs);
757 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
762 * Our handling of the processor debug registers is non-trivial.
763 * We do not clear them on entry and exit from the kernel. Therefore
764 * it is possible to get a watchpoint trap here from inside the kernel.
765 * However, the code in ./ptrace.c has ensured that the user can
766 * only set watchpoints on userspace addresses. Therefore the in-kernel
767 * watchpoint trap can only occur in code which is reading/writing
768 * from user space. Such code must not hold kernel locks (since it
769 * can equally take a page fault), therefore it is safe to call
770 * force_sig_info even though that claims and releases locks.
772 * Code in ./signal.c ensures that the debug control register
773 * is restored before we deliver any signal, and therefore that
774 * user code runs with the correct debug control register even though
777 * Being careful here means that we don't have to be as careful in a
778 * lot of more complicated places (task switching can be a bit lazy
779 * about restoring all the debug state, and ptrace doesn't have to
780 * find every occurrence of the TF bit that could be saved away even
783 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
785 unsigned int condition;
786 struct task_struct *tsk = current;
788 get_debugreg(condition, 6);
790 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
791 SIGTRAP) == NOTIFY_STOP)
793 /* It's safe to allow irq's after DR6 has been saved */
794 if (regs->eflags & X86_EFLAGS_IF)
797 /* Mask out spurious debug traps due to lazy DR7 setting */
798 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
799 if (!tsk->thread.debugreg[7])
803 if (regs->eflags & VM_MASK)
806 /* Save debug status register where ptrace can see it */
807 tsk->thread.debugreg[6] = condition;
810 * Single-stepping through TF: make sure we ignore any events in
811 * kernel space (but re-enable TF when returning to user mode).
813 if (condition & DR_STEP) {
815 * We already checked v86 mode above, so we can
816 * check for kernel mode by just checking the CPL
819 if (!user_mode(regs))
820 goto clear_TF_reenable;
823 /* Ok, finally something we can handle */
824 send_sigtrap(tsk, regs, error_code);
826 /* Disable additional traps. They'll be re-enabled when
827 * the signal is delivered.
834 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
838 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
839 regs->eflags &= ~TF_MASK;
844 * Note that we play around with the 'TS' bit in an attempt to get
845 * the correct behaviour even in the presence of the asynchronous
848 void math_error(void __user *eip)
850 struct task_struct * task;
852 unsigned short cwd, swd;
855 * Save the info for the exception handler and clear the error.
859 task->thread.trap_no = 16;
860 task->thread.error_code = 0;
861 info.si_signo = SIGFPE;
863 info.si_code = __SI_FAULT;
866 * (~cwd & swd) will mask out exceptions that are not set to unmasked
867 * status. 0x3f is the exception bits in these regs, 0x200 is the
868 * C1 reg you need in case of a stack fault, 0x040 is the stack
869 * fault bit. We should only be taking one exception at a time,
870 * so if this combination doesn't produce any single exception,
871 * then we have a bad program that isn't syncronizing its FPU usage
872 * and it will suffer the consequences since we won't be able to
873 * fully reproduce the context of the exception
875 cwd = get_fpu_cwd(task);
876 swd = get_fpu_swd(task);
877 switch (swd & ~cwd & 0x3f) {
878 case 0x000: /* No unmasked exception */
880 default: /* Multiple exceptions */
882 case 0x001: /* Invalid Op */
884 * swd & 0x240 == 0x040: Stack Underflow
885 * swd & 0x240 == 0x240: Stack Overflow
886 * User must clear the SF bit (0x40) if set
888 info.si_code = FPE_FLTINV;
890 case 0x002: /* Denormalize */
891 case 0x010: /* Underflow */
892 info.si_code = FPE_FLTUND;
894 case 0x004: /* Zero Divide */
895 info.si_code = FPE_FLTDIV;
897 case 0x008: /* Overflow */
898 info.si_code = FPE_FLTOVF;
900 case 0x020: /* Precision */
901 info.si_code = FPE_FLTRES;
904 force_sig_info(SIGFPE, &info, task);
907 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
910 math_error((void __user *)regs->eip);
913 static void simd_math_error(void __user *eip)
915 struct task_struct * task;
917 unsigned short mxcsr;
920 * Save the info for the exception handler and clear the error.
924 task->thread.trap_no = 19;
925 task->thread.error_code = 0;
926 info.si_signo = SIGFPE;
928 info.si_code = __SI_FAULT;
931 * The SIMD FPU exceptions are handled a little differently, as there
932 * is only a single status/control register. Thus, to determine which
933 * unmasked exception was caught we must mask the exception mask bits
934 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
936 mxcsr = get_fpu_mxcsr(task);
937 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
941 case 0x001: /* Invalid Op */
942 info.si_code = FPE_FLTINV;
944 case 0x002: /* Denormalize */
945 case 0x010: /* Underflow */
946 info.si_code = FPE_FLTUND;
948 case 0x004: /* Zero Divide */
949 info.si_code = FPE_FLTDIV;
951 case 0x008: /* Overflow */
952 info.si_code = FPE_FLTOVF;
954 case 0x020: /* Precision */
955 info.si_code = FPE_FLTRES;
958 force_sig_info(SIGFPE, &info, task);
961 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
965 /* Handle SIMD FPU exceptions on PIII+ processors. */
967 simd_math_error((void __user *)regs->eip);
970 * Handle strange cache flush from user space exception
971 * in all other cases. This is undocumented behaviour.
973 if (regs->eflags & VM_MASK) {
974 handle_vm86_fault((struct kernel_vm86_regs *)regs,
978 current->thread.trap_no = 19;
979 current->thread.error_code = error_code;
980 die_if_kernel("cache flush denied", regs, error_code);
981 force_sig(SIGSEGV, current);
985 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
989 /* No need to warn about this any longer. */
990 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
994 fastcall void setup_x86_bogus_stack(unsigned char * stk)
996 unsigned long *switch16_ptr, *switch32_ptr;
997 struct pt_regs *regs;
998 unsigned long stack_top, stack_bot;
999 unsigned short iret_frame16_off;
1000 int cpu = smp_processor_id();
1001 /* reserve the space on 32bit stack for the magic switch16 pointer */
1002 memmove(stk, stk + 8, sizeof(struct pt_regs));
1003 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
1004 regs = (struct pt_regs *)stk;
1005 /* now the switch32 on 16bit stack */
1006 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1007 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1008 switch32_ptr = (unsigned long *)(stack_top - 8);
1009 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
1010 /* copy iret frame on 16bit stack */
1011 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
1012 /* fill in the switch pointers */
1013 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
1014 switch16_ptr[1] = __ESPFIX_SS;
1015 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
1016 8 - CPU_16BIT_STACK_SIZE;
1017 switch32_ptr[1] = __KERNEL_DS;
1020 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
1022 unsigned long *switch32_ptr;
1023 unsigned char *stack16, *stack32;
1024 unsigned long stack_top, stack_bot;
1026 int cpu = smp_processor_id();
1027 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1028 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1029 switch32_ptr = (unsigned long *)(stack_top - 8);
1030 /* copy the data from 16bit stack to 32bit stack */
1031 len = CPU_16BIT_STACK_SIZE - 8 - sp;
1032 stack16 = (unsigned char *)(stack_bot + sp);
1033 stack32 = (unsigned char *)
1034 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
1035 memcpy(stack32, stack16, len);
1040 * 'math_state_restore()' saves the current math information in the
1041 * old math state array, and gets the new ones from the current task
1043 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1044 * Don't touch unless you *really* know how it works.
1046 * Must be called with kernel preemption disabled (in this case,
1047 * local interrupts are disabled at the call-site in entry.S).
1049 asmlinkage void math_state_restore(struct pt_regs regs)
1051 struct thread_info *thread = current_thread_info();
1052 struct task_struct *tsk = thread->task;
1054 clts(); /* Allow maths ops (or we recurse) */
1055 if (!tsk_used_math(tsk))
1058 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1061 #ifndef CONFIG_MATH_EMULATION
1063 asmlinkage void math_emulate(long arg)
1065 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1066 printk(KERN_EMERG "killing %s.\n",current->comm);
1067 force_sig(SIGFPE,current);
1071 #endif /* CONFIG_MATH_EMULATION */
1073 #ifdef CONFIG_X86_F00F_BUG
1074 void __init trap_init_f00f_bug(void)
1076 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1079 * Update the IDT descriptor and reload the IDT so that
1080 * it uses the read-only mapped virtual address.
1082 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1083 load_idt(&idt_descr);
1087 #define _set_gate(gate_addr,type,dpl,addr,seg) \
1090 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1091 "movw %4,%%dx\n\t" \
1092 "movl %%eax,%0\n\t" \
1094 :"=m" (*((long *) (gate_addr))), \
1095 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1096 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1097 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1102 * This needs to use 'idt_table' rather than 'idt', and
1103 * thus use the _nonmapped_ version of the IDT, as the
1104 * Pentium F0 0F bugfix can have resulted in the mapped
1105 * IDT being write-protected.
1107 void set_intr_gate(unsigned int n, void *addr)
1109 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1113 * This routine sets up an interrupt gate at directory privilege level 3.
1115 static inline void set_system_intr_gate(unsigned int n, void *addr)
1117 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1120 static void __init set_trap_gate(unsigned int n, void *addr)
1122 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1125 static void __init set_system_gate(unsigned int n, void *addr)
1127 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1130 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1132 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1136 void __init trap_init(void)
1139 void __iomem *p = ioremap(0x0FFFD9, 4);
1140 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1146 #ifdef CONFIG_X86_LOCAL_APIC
1147 init_apic_mappings();
1150 set_trap_gate(0,÷_error);
1151 set_intr_gate(1,&debug);
1152 set_intr_gate(2,&nmi);
1153 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1154 set_system_gate(4,&overflow);
1155 set_trap_gate(5,&bounds);
1156 set_trap_gate(6,&invalid_op);
1157 set_trap_gate(7,&device_not_available);
1158 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1159 set_trap_gate(9,&coprocessor_segment_overrun);
1160 set_trap_gate(10,&invalid_TSS);
1161 set_trap_gate(11,&segment_not_present);
1162 set_trap_gate(12,&stack_segment);
1163 set_trap_gate(13,&general_protection);
1164 set_intr_gate(14,&page_fault);
1165 set_trap_gate(15,&spurious_interrupt_bug);
1166 set_trap_gate(16,&coprocessor_error);
1167 set_trap_gate(17,&alignment_check);
1168 #ifdef CONFIG_X86_MCE
1169 set_trap_gate(18,&machine_check);
1171 set_trap_gate(19,&simd_coprocessor_error);
1175 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1176 * Generates a compile-time "error: zero width for bit-field" if
1177 * the alignment is wrong.
1179 struct fxsrAlignAssert {
1180 int _:!(offsetof(struct task_struct,
1181 thread.i387.fxsave) & 15);
1184 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1185 set_in_cr4(X86_CR4_OSFXSR);
1189 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1191 set_in_cr4(X86_CR4_OSXMMEXCPT);
1195 set_system_gate(SYSCALL_VECTOR,&system_call);
1198 * Should be a barrier for any external CPU state.
1205 static int __init kstack_setup(char *s)
1207 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1210 __setup("kstack=", kstack_setup);