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 struct notifier_block *i386die_chain;
96 static DEFINE_SPINLOCK(die_notifier_lock);
98 int register_die_notifier(struct notifier_block *nb)
102 spin_lock_irqsave(&die_notifier_lock, flags);
103 err = notifier_chain_register(&i386die_chain, nb);
104 spin_unlock_irqrestore(&die_notifier_lock, flags);
107 EXPORT_SYMBOL(register_die_notifier);
109 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
111 return p > (void *)tinfo &&
112 p < (void *)tinfo + THREAD_SIZE - 3;
115 static void print_addr_and_symbol(unsigned long addr, char *log_lvl)
118 printk(" [<%08lx>] ", addr);
119 print_symbol("%s", addr);
123 static inline unsigned long print_context_stack(struct thread_info *tinfo,
124 unsigned long *stack, unsigned long ebp,
129 #ifdef CONFIG_FRAME_POINTER
130 while (valid_stack_ptr(tinfo, (void *)ebp)) {
131 addr = *(unsigned long *)(ebp + 4);
132 print_addr_and_symbol(addr, log_lvl);
133 ebp = *(unsigned long *)ebp;
136 while (valid_stack_ptr(tinfo, stack)) {
138 if (__kernel_text_address(addr))
139 print_addr_and_symbol(addr, log_lvl);
145 static void show_trace_log_lvl(struct task_struct *task,
146 unsigned long *stack, char *log_lvl)
153 if (task == current) {
154 /* Grab ebp right from our regs */
155 asm ("movl %%ebp, %0" : "=r" (ebp) : );
157 /* ebp is the last reg pushed by switch_to */
158 ebp = *(unsigned long *) task->thread.esp;
162 struct thread_info *context;
163 context = (struct thread_info *)
164 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
165 ebp = print_context_stack(context, stack, ebp, log_lvl);
166 stack = (unsigned long*)context->previous_esp;
170 printk(" =======================\n");
174 void show_trace(struct task_struct *task, unsigned long * stack)
176 show_trace_log_lvl(task, stack, "");
179 static void show_stack_log_lvl(struct task_struct *task, unsigned long *esp,
182 unsigned long *stack;
187 esp = (unsigned long*)task->thread.esp;
189 esp = (unsigned long *)&esp;
194 for(i = 0; i < kstack_depth_to_print; i++) {
195 if (kstack_end(stack))
197 if (i && ((i % 8) == 0)) {
202 printk("%08lx ", *stack++);
206 printk("Call Trace:\n");
207 show_trace_log_lvl(task, esp, log_lvl);
210 void show_stack(struct task_struct *task, unsigned long *esp)
212 show_stack_log_lvl(task, esp, "");
216 * The architecture-independent dump_stack generator
218 void dump_stack(void)
222 show_trace(current, &stack);
225 EXPORT_SYMBOL(dump_stack);
227 void show_registers(struct pt_regs *regs)
234 esp = (unsigned long) (®s->esp);
236 if (user_mode(regs)) {
239 ss = regs->xss & 0xffff;
242 printk(KERN_EMERG "CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\n"
243 "EFLAGS: %08lx (%s %.*s) \n",
244 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
245 print_tainted(), regs->eflags, system_utsname.release,
246 (int)strcspn(system_utsname.version, " "),
247 system_utsname.version);
248 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
249 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
250 regs->eax, regs->ebx, regs->ecx, regs->edx);
251 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
252 regs->esi, regs->edi, regs->ebp, esp);
253 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
254 regs->xds & 0xffff, regs->xes & 0xffff, ss);
255 printk(KERN_EMERG "Process %s (pid: %d, threadinfo=%p task=%p)",
256 current->comm, current->pid, current_thread_info(), current);
258 * When in-kernel, we also print out the stack and code at the
259 * time of the fault..
264 printk("\n" KERN_EMERG "Stack: ");
265 show_stack_log_lvl(NULL, (unsigned long *)esp, KERN_EMERG);
267 printk(KERN_EMERG "Code: ");
269 eip = (u8 __user *)regs->eip - 43;
270 for (i = 0; i < 64; i++, eip++) {
273 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
274 printk(" Bad EIP value.");
277 if (eip == (u8 __user *)regs->eip)
278 printk("<%02x> ", c);
286 static void handle_BUG(struct pt_regs *regs)
296 if (eip < PAGE_OFFSET)
298 if (__get_user(ud2, (unsigned short __user *)eip))
302 if (__get_user(line, (unsigned short __user *)(eip + 2)))
304 if (__get_user(file, (char * __user *)(eip + 4)) ||
305 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
306 file = "<bad filename>";
308 printk(KERN_EMERG "------------[ cut here ]------------\n");
309 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
314 /* Here we know it was a BUG but file-n-line is unavailable */
316 printk(KERN_EMERG "Kernel BUG\n");
319 /* This is gone through when something in the kernel
320 * has done something bad and is about to be terminated.
322 void die(const char * str, struct pt_regs * regs, long err)
327 int lock_owner_depth;
329 .lock = SPIN_LOCK_UNLOCKED,
331 .lock_owner_depth = 0
333 static int die_counter;
336 if (die.lock_owner != raw_smp_processor_id()) {
338 spin_lock_irqsave(&die.lock, flags);
339 die.lock_owner = smp_processor_id();
340 die.lock_owner_depth = 0;
344 local_save_flags(flags);
346 if (++die.lock_owner_depth < 3) {
349 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
350 #ifdef CONFIG_PREEMPT
351 printk(KERN_EMERG "PREEMPT ");
360 #ifdef CONFIG_DEBUG_PAGEALLOC
363 printk("DEBUG_PAGEALLOC");
368 notify_die(DIE_OOPS, (char *)str, regs, err, 255, SIGSEGV);
369 show_registers(regs);
371 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
375 spin_unlock_irqrestore(&die.lock, flags);
377 if (kexec_should_crash(current))
381 panic("Fatal exception in interrupt");
384 printk(KERN_EMERG "Fatal exception: panic in 5 seconds\n");
386 panic("Fatal exception");
391 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
393 if (!user_mode_vm(regs))
397 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
398 struct pt_regs * regs, long error_code,
401 struct task_struct *tsk = current;
402 tsk->thread.error_code = error_code;
403 tsk->thread.trap_no = trapnr;
405 if (regs->eflags & VM_MASK) {
411 if (!user_mode(regs))
416 force_sig_info(signr, info, tsk);
418 force_sig(signr, tsk);
423 if (!fixup_exception(regs))
424 die(str, regs, error_code);
429 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
430 if (ret) goto trap_signal;
435 #define DO_ERROR(trapnr, signr, str, name) \
436 fastcall void do_##name(struct pt_regs * regs, long error_code) \
438 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
441 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
444 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
445 fastcall void do_##name(struct pt_regs * regs, long error_code) \
448 info.si_signo = signr; \
450 info.si_code = sicode; \
451 info.si_addr = (void __user *)siaddr; \
452 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
455 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
458 #define DO_VM86_ERROR(trapnr, signr, str, name) \
459 fastcall void do_##name(struct pt_regs * regs, long error_code) \
461 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
464 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
467 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
468 fastcall void do_##name(struct pt_regs * regs, long error_code) \
471 info.si_signo = signr; \
473 info.si_code = sicode; \
474 info.si_addr = (void __user *)siaddr; \
475 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
478 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
481 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
482 #ifndef CONFIG_KPROBES
483 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
485 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
486 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
487 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
488 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
489 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
490 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
491 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
492 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
493 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
495 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
499 struct tss_struct *tss = &per_cpu(init_tss, cpu);
500 struct thread_struct *thread = ¤t->thread;
503 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
504 * invalid offset set (the LAZY one) and the faulting thread has
505 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
506 * and we set the offset field correctly. Then we let the CPU to
507 * restart the faulting instruction.
509 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
510 thread->io_bitmap_ptr) {
511 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
512 thread->io_bitmap_max);
514 * If the previously set map was extending to higher ports
515 * than the current one, pad extra space with 0xff (no access).
517 if (thread->io_bitmap_max < tss->io_bitmap_max)
518 memset((char *) tss->io_bitmap +
519 thread->io_bitmap_max, 0xff,
520 tss->io_bitmap_max - thread->io_bitmap_max);
521 tss->io_bitmap_max = thread->io_bitmap_max;
522 tss->io_bitmap_base = IO_BITMAP_OFFSET;
523 tss->io_bitmap_owner = thread;
529 current->thread.error_code = error_code;
530 current->thread.trap_no = 13;
532 if (regs->eflags & VM_MASK)
535 if (!user_mode(regs))
538 current->thread.error_code = error_code;
539 current->thread.trap_no = 13;
540 force_sig(SIGSEGV, current);
545 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
549 if (!fixup_exception(regs)) {
550 if (notify_die(DIE_GPF, "general protection fault", regs,
551 error_code, 13, SIGSEGV) == NOTIFY_STOP)
553 die("general protection fault", regs, error_code);
557 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
559 printk(KERN_EMERG "Uhhuh. NMI received. Dazed and confused, but trying "
561 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
564 /* Clear and disable the memory parity error line. */
565 clear_mem_error(reason);
568 static void io_check_error(unsigned char reason, struct pt_regs * regs)
572 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
573 show_registers(regs);
575 /* Re-enable the IOCK line, wait for a few seconds */
576 reason = (reason & 0xf) | 8;
579 while (--i) udelay(1000);
584 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
587 /* Might actually be able to figure out what the guilty party
594 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
595 reason, smp_processor_id());
596 printk("Dazed and confused, but trying to continue\n");
597 printk("Do you have a strange power saving mode enabled?\n");
600 static DEFINE_SPINLOCK(nmi_print_lock);
602 void die_nmi (struct pt_regs *regs, const char *msg)
604 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 0, SIGINT) ==
608 spin_lock(&nmi_print_lock);
610 * We are in trouble anyway, lets at least try
611 * to get a message out.
614 printk(KERN_EMERG "%s", msg);
615 printk(" on CPU%d, eip %08lx, registers:\n",
616 smp_processor_id(), regs->eip);
617 show_registers(regs);
618 printk(KERN_EMERG "console shuts up ...\n");
620 spin_unlock(&nmi_print_lock);
623 /* If we are in kernel we are probably nested up pretty bad
624 * and might aswell get out now while we still can.
626 if (!user_mode(regs)) {
627 current->thread.trap_no = 2;
634 static void default_do_nmi(struct pt_regs * regs)
636 unsigned char reason = 0;
638 /* Only the BSP gets external NMIs from the system. */
639 if (!smp_processor_id())
640 reason = get_nmi_reason();
642 if (!(reason & 0xc0)) {
643 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 0, SIGINT)
646 #ifdef CONFIG_X86_LOCAL_APIC
648 * Ok, so this is none of the documented NMI sources,
649 * so it must be the NMI watchdog.
652 nmi_watchdog_tick(regs);
656 unknown_nmi_error(reason, regs);
659 if (notify_die(DIE_NMI, "nmi", regs, reason, 0, SIGINT) == NOTIFY_STOP)
662 mem_parity_error(reason, regs);
664 io_check_error(reason, regs);
666 * Reassert NMI in case it became active meanwhile
667 * as it's edge-triggered.
672 static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
677 static nmi_callback_t nmi_callback = dummy_nmi_callback;
679 fastcall void do_nmi(struct pt_regs * regs, long error_code)
685 cpu = smp_processor_id();
689 if (!rcu_dereference(nmi_callback)(regs, cpu))
690 default_do_nmi(regs);
695 void set_nmi_callback(nmi_callback_t callback)
697 rcu_assign_pointer(nmi_callback, callback);
699 EXPORT_SYMBOL_GPL(set_nmi_callback);
701 void unset_nmi_callback(void)
703 nmi_callback = dummy_nmi_callback;
705 EXPORT_SYMBOL_GPL(unset_nmi_callback);
707 #ifdef CONFIG_KPROBES
708 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
710 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
713 /* This is an interrupt gate, because kprobes wants interrupts
714 disabled. Normal trap handlers don't. */
715 restore_interrupts(regs);
716 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
721 * Our handling of the processor debug registers is non-trivial.
722 * We do not clear them on entry and exit from the kernel. Therefore
723 * it is possible to get a watchpoint trap here from inside the kernel.
724 * However, the code in ./ptrace.c has ensured that the user can
725 * only set watchpoints on userspace addresses. Therefore the in-kernel
726 * watchpoint trap can only occur in code which is reading/writing
727 * from user space. Such code must not hold kernel locks (since it
728 * can equally take a page fault), therefore it is safe to call
729 * force_sig_info even though that claims and releases locks.
731 * Code in ./signal.c ensures that the debug control register
732 * is restored before we deliver any signal, and therefore that
733 * user code runs with the correct debug control register even though
736 * Being careful here means that we don't have to be as careful in a
737 * lot of more complicated places (task switching can be a bit lazy
738 * about restoring all the debug state, and ptrace doesn't have to
739 * find every occurrence of the TF bit that could be saved away even
742 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
744 unsigned int condition;
745 struct task_struct *tsk = current;
747 get_debugreg(condition, 6);
749 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
750 SIGTRAP) == NOTIFY_STOP)
752 /* It's safe to allow irq's after DR6 has been saved */
753 if (regs->eflags & X86_EFLAGS_IF)
756 /* Mask out spurious debug traps due to lazy DR7 setting */
757 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
758 if (!tsk->thread.debugreg[7])
762 if (regs->eflags & VM_MASK)
765 /* Save debug status register where ptrace can see it */
766 tsk->thread.debugreg[6] = condition;
769 * Single-stepping through TF: make sure we ignore any events in
770 * kernel space (but re-enable TF when returning to user mode).
772 if (condition & DR_STEP) {
774 * We already checked v86 mode above, so we can
775 * check for kernel mode by just checking the CPL
778 if (!user_mode(regs))
779 goto clear_TF_reenable;
782 /* Ok, finally something we can handle */
783 send_sigtrap(tsk, regs, error_code);
785 /* Disable additional traps. They'll be re-enabled when
786 * the signal is delivered.
793 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
797 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
798 regs->eflags &= ~TF_MASK;
803 * Note that we play around with the 'TS' bit in an attempt to get
804 * the correct behaviour even in the presence of the asynchronous
807 void math_error(void __user *eip)
809 struct task_struct * task;
811 unsigned short cwd, swd;
814 * Save the info for the exception handler and clear the error.
818 task->thread.trap_no = 16;
819 task->thread.error_code = 0;
820 info.si_signo = SIGFPE;
822 info.si_code = __SI_FAULT;
825 * (~cwd & swd) will mask out exceptions that are not set to unmasked
826 * status. 0x3f is the exception bits in these regs, 0x200 is the
827 * C1 reg you need in case of a stack fault, 0x040 is the stack
828 * fault bit. We should only be taking one exception at a time,
829 * so if this combination doesn't produce any single exception,
830 * then we have a bad program that isn't syncronizing its FPU usage
831 * and it will suffer the consequences since we won't be able to
832 * fully reproduce the context of the exception
834 cwd = get_fpu_cwd(task);
835 swd = get_fpu_swd(task);
836 switch (swd & ~cwd & 0x3f) {
837 case 0x000: /* No unmasked exception */
839 default: /* Multiple exceptions */
841 case 0x001: /* Invalid Op */
843 * swd & 0x240 == 0x040: Stack Underflow
844 * swd & 0x240 == 0x240: Stack Overflow
845 * User must clear the SF bit (0x40) if set
847 info.si_code = FPE_FLTINV;
849 case 0x002: /* Denormalize */
850 case 0x010: /* Underflow */
851 info.si_code = FPE_FLTUND;
853 case 0x004: /* Zero Divide */
854 info.si_code = FPE_FLTDIV;
856 case 0x008: /* Overflow */
857 info.si_code = FPE_FLTOVF;
859 case 0x020: /* Precision */
860 info.si_code = FPE_FLTRES;
863 force_sig_info(SIGFPE, &info, task);
866 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
869 math_error((void __user *)regs->eip);
872 static void simd_math_error(void __user *eip)
874 struct task_struct * task;
876 unsigned short mxcsr;
879 * Save the info for the exception handler and clear the error.
883 task->thread.trap_no = 19;
884 task->thread.error_code = 0;
885 info.si_signo = SIGFPE;
887 info.si_code = __SI_FAULT;
890 * The SIMD FPU exceptions are handled a little differently, as there
891 * is only a single status/control register. Thus, to determine which
892 * unmasked exception was caught we must mask the exception mask bits
893 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
895 mxcsr = get_fpu_mxcsr(task);
896 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
900 case 0x001: /* Invalid Op */
901 info.si_code = FPE_FLTINV;
903 case 0x002: /* Denormalize */
904 case 0x010: /* Underflow */
905 info.si_code = FPE_FLTUND;
907 case 0x004: /* Zero Divide */
908 info.si_code = FPE_FLTDIV;
910 case 0x008: /* Overflow */
911 info.si_code = FPE_FLTOVF;
913 case 0x020: /* Precision */
914 info.si_code = FPE_FLTRES;
917 force_sig_info(SIGFPE, &info, task);
920 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
924 /* Handle SIMD FPU exceptions on PIII+ processors. */
926 simd_math_error((void __user *)regs->eip);
929 * Handle strange cache flush from user space exception
930 * in all other cases. This is undocumented behaviour.
932 if (regs->eflags & VM_MASK) {
933 handle_vm86_fault((struct kernel_vm86_regs *)regs,
937 current->thread.trap_no = 19;
938 current->thread.error_code = error_code;
939 die_if_kernel("cache flush denied", regs, error_code);
940 force_sig(SIGSEGV, current);
944 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
948 /* No need to warn about this any longer. */
949 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
953 fastcall void setup_x86_bogus_stack(unsigned char * stk)
955 unsigned long *switch16_ptr, *switch32_ptr;
956 struct pt_regs *regs;
957 unsigned long stack_top, stack_bot;
958 unsigned short iret_frame16_off;
959 int cpu = smp_processor_id();
960 /* reserve the space on 32bit stack for the magic switch16 pointer */
961 memmove(stk, stk + 8, sizeof(struct pt_regs));
962 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
963 regs = (struct pt_regs *)stk;
964 /* now the switch32 on 16bit stack */
965 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
966 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
967 switch32_ptr = (unsigned long *)(stack_top - 8);
968 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
969 /* copy iret frame on 16bit stack */
970 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
971 /* fill in the switch pointers */
972 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
973 switch16_ptr[1] = __ESPFIX_SS;
974 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
975 8 - CPU_16BIT_STACK_SIZE;
976 switch32_ptr[1] = __KERNEL_DS;
979 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
981 unsigned long *switch32_ptr;
982 unsigned char *stack16, *stack32;
983 unsigned long stack_top, stack_bot;
985 int cpu = smp_processor_id();
986 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
987 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
988 switch32_ptr = (unsigned long *)(stack_top - 8);
989 /* copy the data from 16bit stack to 32bit stack */
990 len = CPU_16BIT_STACK_SIZE - 8 - sp;
991 stack16 = (unsigned char *)(stack_bot + sp);
992 stack32 = (unsigned char *)
993 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
994 memcpy(stack32, stack16, len);
999 * 'math_state_restore()' saves the current math information in the
1000 * old math state array, and gets the new ones from the current task
1002 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1003 * Don't touch unless you *really* know how it works.
1005 * Must be called with kernel preemption disabled (in this case,
1006 * local interrupts are disabled at the call-site in entry.S).
1008 asmlinkage void math_state_restore(struct pt_regs regs)
1010 struct thread_info *thread = current_thread_info();
1011 struct task_struct *tsk = thread->task;
1013 clts(); /* Allow maths ops (or we recurse) */
1014 if (!tsk_used_math(tsk))
1017 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1020 #ifndef CONFIG_MATH_EMULATION
1022 asmlinkage void math_emulate(long arg)
1024 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1025 printk(KERN_EMERG "killing %s.\n",current->comm);
1026 force_sig(SIGFPE,current);
1030 #endif /* CONFIG_MATH_EMULATION */
1032 #ifdef CONFIG_X86_F00F_BUG
1033 void __init trap_init_f00f_bug(void)
1035 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1038 * Update the IDT descriptor and reload the IDT so that
1039 * it uses the read-only mapped virtual address.
1041 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1042 load_idt(&idt_descr);
1046 #define _set_gate(gate_addr,type,dpl,addr,seg) \
1049 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1050 "movw %4,%%dx\n\t" \
1051 "movl %%eax,%0\n\t" \
1053 :"=m" (*((long *) (gate_addr))), \
1054 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1055 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1056 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1061 * This needs to use 'idt_table' rather than 'idt', and
1062 * thus use the _nonmapped_ version of the IDT, as the
1063 * Pentium F0 0F bugfix can have resulted in the mapped
1064 * IDT being write-protected.
1066 void set_intr_gate(unsigned int n, void *addr)
1068 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1072 * This routine sets up an interrupt gate at directory privilege level 3.
1074 static inline void set_system_intr_gate(unsigned int n, void *addr)
1076 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1079 static void __init set_trap_gate(unsigned int n, void *addr)
1081 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1084 static void __init set_system_gate(unsigned int n, void *addr)
1086 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1089 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1091 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1095 void __init trap_init(void)
1098 void __iomem *p = ioremap(0x0FFFD9, 4);
1099 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1105 #ifdef CONFIG_X86_LOCAL_APIC
1106 init_apic_mappings();
1109 set_trap_gate(0,÷_error);
1110 set_intr_gate(1,&debug);
1111 set_intr_gate(2,&nmi);
1112 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1113 set_system_gate(4,&overflow);
1114 set_trap_gate(5,&bounds);
1115 set_trap_gate(6,&invalid_op);
1116 set_trap_gate(7,&device_not_available);
1117 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1118 set_trap_gate(9,&coprocessor_segment_overrun);
1119 set_trap_gate(10,&invalid_TSS);
1120 set_trap_gate(11,&segment_not_present);
1121 set_trap_gate(12,&stack_segment);
1122 set_trap_gate(13,&general_protection);
1123 set_intr_gate(14,&page_fault);
1124 set_trap_gate(15,&spurious_interrupt_bug);
1125 set_trap_gate(16,&coprocessor_error);
1126 set_trap_gate(17,&alignment_check);
1127 #ifdef CONFIG_X86_MCE
1128 set_trap_gate(18,&machine_check);
1130 set_trap_gate(19,&simd_coprocessor_error);
1134 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1135 * Generates a compile-time "error: zero width for bit-field" if
1136 * the alignment is wrong.
1138 struct fxsrAlignAssert {
1139 int _:!(offsetof(struct task_struct,
1140 thread.i387.fxsave) & 15);
1143 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1144 set_in_cr4(X86_CR4_OSFXSR);
1148 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1150 set_in_cr4(X86_CR4_OSXMMEXCPT);
1154 set_system_gate(SYSCALL_VECTOR,&system_call);
1157 * Should be a barrier for any external CPU state.
1164 static int __init kstack_setup(char *s)
1166 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1169 __setup("kstack=", kstack_setup);