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);
153 * break out of recursive entries (such as
154 * end_of_stack_stop_unwind_function):
156 if (ebp == *(unsigned long *)ebp)
158 ebp = *(unsigned long *)ebp;
161 while (valid_stack_ptr(tinfo, stack)) {
163 if (__kernel_text_address(addr))
164 printed = print_addr_and_symbol(addr, log_lvl, printed);
173 static void show_trace_log_lvl(struct task_struct *task,
174 unsigned long *stack, char *log_lvl)
181 if (task == current) {
182 /* Grab ebp right from our regs */
183 asm ("movl %%ebp, %0" : "=r" (ebp) : );
185 /* ebp is the last reg pushed by switch_to */
186 ebp = *(unsigned long *) task->thread.esp;
190 struct thread_info *context;
191 context = (struct thread_info *)
192 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
193 ebp = print_context_stack(context, stack, ebp, log_lvl);
194 stack = (unsigned long*)context->previous_esp;
197 printk("%s =======================\n", log_lvl);
201 void show_trace(struct task_struct *task, unsigned long * stack)
203 show_trace_log_lvl(task, stack, "");
206 static void show_stack_log_lvl(struct task_struct *task, unsigned long *esp,
209 unsigned long *stack;
214 esp = (unsigned long*)task->thread.esp;
216 esp = (unsigned long *)&esp;
220 for(i = 0; i < kstack_depth_to_print; i++) {
221 if (kstack_end(stack))
223 if (i && ((i % 8) == 0))
224 printk("\n%s ", log_lvl);
225 printk("%08lx ", *stack++);
227 printk("\n%sCall Trace:\n", log_lvl);
228 show_trace_log_lvl(task, esp, log_lvl);
231 void show_stack(struct task_struct *task, unsigned long *esp)
234 show_stack_log_lvl(task, esp, "");
238 * The architecture-independent dump_stack generator
240 void dump_stack(void)
244 show_trace(current, &stack);
247 EXPORT_SYMBOL(dump_stack);
249 void show_registers(struct pt_regs *regs)
256 esp = (unsigned long) (®s->esp);
258 if (user_mode_vm(regs)) {
261 ss = regs->xss & 0xffff;
264 printk(KERN_EMERG "CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\n"
265 "EFLAGS: %08lx (%s %.*s) \n",
266 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
267 print_tainted(), regs->eflags, system_utsname.release,
268 (int)strcspn(system_utsname.version, " "),
269 system_utsname.version);
270 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
271 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
272 regs->eax, regs->ebx, regs->ecx, regs->edx);
273 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
274 regs->esi, regs->edi, regs->ebp, esp);
275 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
276 regs->xds & 0xffff, regs->xes & 0xffff, ss);
277 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
278 TASK_COMM_LEN, current->comm, current->pid,
279 current_thread_info(), current, current->thread_info);
281 * When in-kernel, we also print out the stack and code at the
282 * time of the fault..
287 printk("\n" KERN_EMERG "Stack: ");
288 show_stack_log_lvl(NULL, (unsigned long *)esp, KERN_EMERG);
290 printk(KERN_EMERG "Code: ");
292 eip = (u8 __user *)regs->eip - 43;
293 for (i = 0; i < 64; i++, eip++) {
296 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
297 printk(" Bad EIP value.");
300 if (eip == (u8 __user *)regs->eip)
301 printk("<%02x> ", c);
309 static void handle_BUG(struct pt_regs *regs)
319 if (eip < PAGE_OFFSET)
321 if (__get_user(ud2, (unsigned short __user *)eip))
325 if (__get_user(line, (unsigned short __user *)(eip + 2)))
327 if (__get_user(file, (char * __user *)(eip + 4)) ||
328 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
329 file = "<bad filename>";
331 printk(KERN_EMERG "------------[ cut here ]------------\n");
332 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
337 /* Here we know it was a BUG but file-n-line is unavailable */
339 printk(KERN_EMERG "Kernel BUG\n");
342 /* This is gone through when something in the kernel
343 * has done something bad and is about to be terminated.
345 void die(const char * str, struct pt_regs * regs, long err)
350 int lock_owner_depth;
352 .lock = SPIN_LOCK_UNLOCKED,
354 .lock_owner_depth = 0
356 static int die_counter;
361 if (die.lock_owner != raw_smp_processor_id()) {
363 spin_lock_irqsave(&die.lock, flags);
364 die.lock_owner = smp_processor_id();
365 die.lock_owner_depth = 0;
369 local_save_flags(flags);
371 if (++die.lock_owner_depth < 3) {
377 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
378 #ifdef CONFIG_PREEMPT
379 printk(KERN_EMERG "PREEMPT ");
388 #ifdef CONFIG_DEBUG_PAGEALLOC
391 printk("DEBUG_PAGEALLOC");
396 if (notify_die(DIE_OOPS, str, regs, err,
397 current->thread.trap_no, SIGSEGV) !=
399 show_registers(regs);
400 /* Executive summary in case the oops scrolled away */
401 esp = (unsigned long) (®s->esp);
403 if (user_mode(regs)) {
405 ss = regs->xss & 0xffff;
407 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
408 print_symbol("%s", regs->eip);
409 printk(" SS:ESP %04x:%08lx\n", ss, esp);
414 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
418 spin_unlock_irqrestore(&die.lock, flags);
423 if (kexec_should_crash(current))
427 panic("Fatal exception in interrupt");
430 printk(KERN_EMERG "Fatal exception: panic in 5 seconds\n");
432 panic("Fatal exception");
438 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
440 if (!user_mode_vm(regs))
444 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
445 struct pt_regs * regs, long error_code,
448 struct task_struct *tsk = current;
449 tsk->thread.error_code = error_code;
450 tsk->thread.trap_no = trapnr;
452 if (regs->eflags & VM_MASK) {
458 if (!user_mode(regs))
463 force_sig_info(signr, info, tsk);
465 force_sig(signr, tsk);
470 if (!fixup_exception(regs))
471 die(str, regs, error_code);
476 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
477 if (ret) goto trap_signal;
482 #define DO_ERROR(trapnr, signr, str, name) \
483 fastcall void do_##name(struct pt_regs * regs, long error_code) \
485 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
488 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
491 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
492 fastcall void do_##name(struct pt_regs * regs, long error_code) \
495 info.si_signo = signr; \
497 info.si_code = sicode; \
498 info.si_addr = (void __user *)siaddr; \
499 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
502 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
505 #define DO_VM86_ERROR(trapnr, signr, str, name) \
506 fastcall void do_##name(struct pt_regs * regs, long error_code) \
508 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
511 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
514 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
515 fastcall void do_##name(struct pt_regs * regs, long error_code) \
518 info.si_signo = signr; \
520 info.si_code = sicode; \
521 info.si_addr = (void __user *)siaddr; \
522 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
525 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
528 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
529 #ifndef CONFIG_KPROBES
530 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
532 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
533 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
534 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
535 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
536 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
537 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
538 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
539 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
540 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
542 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
546 struct tss_struct *tss = &per_cpu(init_tss, cpu);
547 struct thread_struct *thread = ¤t->thread;
550 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
551 * invalid offset set (the LAZY one) and the faulting thread has
552 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
553 * and we set the offset field correctly. Then we let the CPU to
554 * restart the faulting instruction.
556 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
557 thread->io_bitmap_ptr) {
558 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
559 thread->io_bitmap_max);
561 * If the previously set map was extending to higher ports
562 * than the current one, pad extra space with 0xff (no access).
564 if (thread->io_bitmap_max < tss->io_bitmap_max)
565 memset((char *) tss->io_bitmap +
566 thread->io_bitmap_max, 0xff,
567 tss->io_bitmap_max - thread->io_bitmap_max);
568 tss->io_bitmap_max = thread->io_bitmap_max;
569 tss->io_bitmap_base = IO_BITMAP_OFFSET;
570 tss->io_bitmap_owner = thread;
576 current->thread.error_code = error_code;
577 current->thread.trap_no = 13;
579 if (regs->eflags & VM_MASK)
582 if (!user_mode(regs))
585 current->thread.error_code = error_code;
586 current->thread.trap_no = 13;
587 force_sig(SIGSEGV, current);
592 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
596 if (!fixup_exception(regs)) {
597 if (notify_die(DIE_GPF, "general protection fault", regs,
598 error_code, 13, SIGSEGV) == NOTIFY_STOP)
600 die("general protection fault", regs, error_code);
604 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
606 printk(KERN_EMERG "Uhhuh. NMI received. Dazed and confused, but trying "
608 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
611 /* Clear and disable the memory parity error line. */
612 clear_mem_error(reason);
615 static void io_check_error(unsigned char reason, struct pt_regs * regs)
619 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
620 show_registers(regs);
622 /* Re-enable the IOCK line, wait for a few seconds */
623 reason = (reason & 0xf) | 8;
626 while (--i) udelay(1000);
631 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
634 /* Might actually be able to figure out what the guilty party
641 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
642 reason, smp_processor_id());
643 printk("Dazed and confused, but trying to continue\n");
644 printk("Do you have a strange power saving mode enabled?\n");
647 static DEFINE_SPINLOCK(nmi_print_lock);
649 void die_nmi (struct pt_regs *regs, const char *msg)
651 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
655 spin_lock(&nmi_print_lock);
657 * We are in trouble anyway, lets at least try
658 * to get a message out.
661 printk(KERN_EMERG "%s", msg);
662 printk(" on CPU%d, eip %08lx, registers:\n",
663 smp_processor_id(), regs->eip);
664 show_registers(regs);
665 printk(KERN_EMERG "console shuts up ...\n");
667 spin_unlock(&nmi_print_lock);
670 /* If we are in kernel we are probably nested up pretty bad
671 * and might aswell get out now while we still can.
673 if (!user_mode_vm(regs)) {
674 current->thread.trap_no = 2;
681 static void default_do_nmi(struct pt_regs * regs)
683 unsigned char reason = 0;
685 /* Only the BSP gets external NMIs from the system. */
686 if (!smp_processor_id())
687 reason = get_nmi_reason();
689 if (!(reason & 0xc0)) {
690 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
693 #ifdef CONFIG_X86_LOCAL_APIC
695 * Ok, so this is none of the documented NMI sources,
696 * so it must be the NMI watchdog.
699 nmi_watchdog_tick(regs);
703 unknown_nmi_error(reason, regs);
706 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
709 mem_parity_error(reason, regs);
711 io_check_error(reason, regs);
713 * Reassert NMI in case it became active meanwhile
714 * as it's edge-triggered.
719 static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
724 static nmi_callback_t nmi_callback = dummy_nmi_callback;
726 fastcall void do_nmi(struct pt_regs * regs, long error_code)
732 cpu = smp_processor_id();
736 if (!rcu_dereference(nmi_callback)(regs, cpu))
737 default_do_nmi(regs);
742 void set_nmi_callback(nmi_callback_t callback)
745 rcu_assign_pointer(nmi_callback, callback);
747 EXPORT_SYMBOL_GPL(set_nmi_callback);
749 void unset_nmi_callback(void)
751 nmi_callback = dummy_nmi_callback;
753 EXPORT_SYMBOL_GPL(unset_nmi_callback);
755 #ifdef CONFIG_KPROBES
756 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
758 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
761 /* This is an interrupt gate, because kprobes wants interrupts
762 disabled. Normal trap handlers don't. */
763 restore_interrupts(regs);
764 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
769 * Our handling of the processor debug registers is non-trivial.
770 * We do not clear them on entry and exit from the kernel. Therefore
771 * it is possible to get a watchpoint trap here from inside the kernel.
772 * However, the code in ./ptrace.c has ensured that the user can
773 * only set watchpoints on userspace addresses. Therefore the in-kernel
774 * watchpoint trap can only occur in code which is reading/writing
775 * from user space. Such code must not hold kernel locks (since it
776 * can equally take a page fault), therefore it is safe to call
777 * force_sig_info even though that claims and releases locks.
779 * Code in ./signal.c ensures that the debug control register
780 * is restored before we deliver any signal, and therefore that
781 * user code runs with the correct debug control register even though
784 * Being careful here means that we don't have to be as careful in a
785 * lot of more complicated places (task switching can be a bit lazy
786 * about restoring all the debug state, and ptrace doesn't have to
787 * find every occurrence of the TF bit that could be saved away even
790 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
792 unsigned int condition;
793 struct task_struct *tsk = current;
795 get_debugreg(condition, 6);
797 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
798 SIGTRAP) == NOTIFY_STOP)
800 /* It's safe to allow irq's after DR6 has been saved */
801 if (regs->eflags & X86_EFLAGS_IF)
804 /* Mask out spurious debug traps due to lazy DR7 setting */
805 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
806 if (!tsk->thread.debugreg[7])
810 if (regs->eflags & VM_MASK)
813 /* Save debug status register where ptrace can see it */
814 tsk->thread.debugreg[6] = condition;
817 * Single-stepping through TF: make sure we ignore any events in
818 * kernel space (but re-enable TF when returning to user mode).
820 if (condition & DR_STEP) {
822 * We already checked v86 mode above, so we can
823 * check for kernel mode by just checking the CPL
826 if (!user_mode(regs))
827 goto clear_TF_reenable;
830 /* Ok, finally something we can handle */
831 send_sigtrap(tsk, regs, error_code);
833 /* Disable additional traps. They'll be re-enabled when
834 * the signal is delivered.
841 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
845 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
846 regs->eflags &= ~TF_MASK;
851 * Note that we play around with the 'TS' bit in an attempt to get
852 * the correct behaviour even in the presence of the asynchronous
855 void math_error(void __user *eip)
857 struct task_struct * task;
859 unsigned short cwd, swd;
862 * Save the info for the exception handler and clear the error.
866 task->thread.trap_no = 16;
867 task->thread.error_code = 0;
868 info.si_signo = SIGFPE;
870 info.si_code = __SI_FAULT;
873 * (~cwd & swd) will mask out exceptions that are not set to unmasked
874 * status. 0x3f is the exception bits in these regs, 0x200 is the
875 * C1 reg you need in case of a stack fault, 0x040 is the stack
876 * fault bit. We should only be taking one exception at a time,
877 * so if this combination doesn't produce any single exception,
878 * then we have a bad program that isn't syncronizing its FPU usage
879 * and it will suffer the consequences since we won't be able to
880 * fully reproduce the context of the exception
882 cwd = get_fpu_cwd(task);
883 swd = get_fpu_swd(task);
884 switch (swd & ~cwd & 0x3f) {
885 case 0x000: /* No unmasked exception */
887 default: /* Multiple exceptions */
889 case 0x001: /* Invalid Op */
891 * swd & 0x240 == 0x040: Stack Underflow
892 * swd & 0x240 == 0x240: Stack Overflow
893 * User must clear the SF bit (0x40) if set
895 info.si_code = FPE_FLTINV;
897 case 0x002: /* Denormalize */
898 case 0x010: /* Underflow */
899 info.si_code = FPE_FLTUND;
901 case 0x004: /* Zero Divide */
902 info.si_code = FPE_FLTDIV;
904 case 0x008: /* Overflow */
905 info.si_code = FPE_FLTOVF;
907 case 0x020: /* Precision */
908 info.si_code = FPE_FLTRES;
911 force_sig_info(SIGFPE, &info, task);
914 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
917 math_error((void __user *)regs->eip);
920 static void simd_math_error(void __user *eip)
922 struct task_struct * task;
924 unsigned short mxcsr;
927 * Save the info for the exception handler and clear the error.
931 task->thread.trap_no = 19;
932 task->thread.error_code = 0;
933 info.si_signo = SIGFPE;
935 info.si_code = __SI_FAULT;
938 * The SIMD FPU exceptions are handled a little differently, as there
939 * is only a single status/control register. Thus, to determine which
940 * unmasked exception was caught we must mask the exception mask bits
941 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
943 mxcsr = get_fpu_mxcsr(task);
944 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
948 case 0x001: /* Invalid Op */
949 info.si_code = FPE_FLTINV;
951 case 0x002: /* Denormalize */
952 case 0x010: /* Underflow */
953 info.si_code = FPE_FLTUND;
955 case 0x004: /* Zero Divide */
956 info.si_code = FPE_FLTDIV;
958 case 0x008: /* Overflow */
959 info.si_code = FPE_FLTOVF;
961 case 0x020: /* Precision */
962 info.si_code = FPE_FLTRES;
965 force_sig_info(SIGFPE, &info, task);
968 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
972 /* Handle SIMD FPU exceptions on PIII+ processors. */
974 simd_math_error((void __user *)regs->eip);
977 * Handle strange cache flush from user space exception
978 * in all other cases. This is undocumented behaviour.
980 if (regs->eflags & VM_MASK) {
981 handle_vm86_fault((struct kernel_vm86_regs *)regs,
985 current->thread.trap_no = 19;
986 current->thread.error_code = error_code;
987 die_if_kernel("cache flush denied", regs, error_code);
988 force_sig(SIGSEGV, current);
992 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
996 /* No need to warn about this any longer. */
997 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1001 fastcall void setup_x86_bogus_stack(unsigned char * stk)
1003 unsigned long *switch16_ptr, *switch32_ptr;
1004 struct pt_regs *regs;
1005 unsigned long stack_top, stack_bot;
1006 unsigned short iret_frame16_off;
1007 int cpu = smp_processor_id();
1008 /* reserve the space on 32bit stack for the magic switch16 pointer */
1009 memmove(stk, stk + 8, sizeof(struct pt_regs));
1010 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
1011 regs = (struct pt_regs *)stk;
1012 /* now the switch32 on 16bit stack */
1013 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1014 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1015 switch32_ptr = (unsigned long *)(stack_top - 8);
1016 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
1017 /* copy iret frame on 16bit stack */
1018 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
1019 /* fill in the switch pointers */
1020 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
1021 switch16_ptr[1] = __ESPFIX_SS;
1022 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
1023 8 - CPU_16BIT_STACK_SIZE;
1024 switch32_ptr[1] = __KERNEL_DS;
1027 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
1029 unsigned long *switch32_ptr;
1030 unsigned char *stack16, *stack32;
1031 unsigned long stack_top, stack_bot;
1033 int cpu = smp_processor_id();
1034 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1035 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1036 switch32_ptr = (unsigned long *)(stack_top - 8);
1037 /* copy the data from 16bit stack to 32bit stack */
1038 len = CPU_16BIT_STACK_SIZE - 8 - sp;
1039 stack16 = (unsigned char *)(stack_bot + sp);
1040 stack32 = (unsigned char *)
1041 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
1042 memcpy(stack32, stack16, len);
1047 * 'math_state_restore()' saves the current math information in the
1048 * old math state array, and gets the new ones from the current task
1050 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1051 * Don't touch unless you *really* know how it works.
1053 * Must be called with kernel preemption disabled (in this case,
1054 * local interrupts are disabled at the call-site in entry.S).
1056 asmlinkage void math_state_restore(struct pt_regs regs)
1058 struct thread_info *thread = current_thread_info();
1059 struct task_struct *tsk = thread->task;
1061 clts(); /* Allow maths ops (or we recurse) */
1062 if (!tsk_used_math(tsk))
1065 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1068 #ifndef CONFIG_MATH_EMULATION
1070 asmlinkage void math_emulate(long arg)
1072 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1073 printk(KERN_EMERG "killing %s.\n",current->comm);
1074 force_sig(SIGFPE,current);
1078 #endif /* CONFIG_MATH_EMULATION */
1080 #ifdef CONFIG_X86_F00F_BUG
1081 void __init trap_init_f00f_bug(void)
1083 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1086 * Update the IDT descriptor and reload the IDT so that
1087 * it uses the read-only mapped virtual address.
1089 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1090 load_idt(&idt_descr);
1094 #define _set_gate(gate_addr,type,dpl,addr,seg) \
1097 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1098 "movw %4,%%dx\n\t" \
1099 "movl %%eax,%0\n\t" \
1101 :"=m" (*((long *) (gate_addr))), \
1102 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1103 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1104 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1109 * This needs to use 'idt_table' rather than 'idt', and
1110 * thus use the _nonmapped_ version of the IDT, as the
1111 * Pentium F0 0F bugfix can have resulted in the mapped
1112 * IDT being write-protected.
1114 void set_intr_gate(unsigned int n, void *addr)
1116 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1120 * This routine sets up an interrupt gate at directory privilege level 3.
1122 static inline void set_system_intr_gate(unsigned int n, void *addr)
1124 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1127 static void __init set_trap_gate(unsigned int n, void *addr)
1129 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1132 static void __init set_system_gate(unsigned int n, void *addr)
1134 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1137 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1139 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1143 void __init trap_init(void)
1146 void __iomem *p = ioremap(0x0FFFD9, 4);
1147 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1153 #ifdef CONFIG_X86_LOCAL_APIC
1154 init_apic_mappings();
1157 set_trap_gate(0,÷_error);
1158 set_intr_gate(1,&debug);
1159 set_intr_gate(2,&nmi);
1160 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1161 set_system_gate(4,&overflow);
1162 set_trap_gate(5,&bounds);
1163 set_trap_gate(6,&invalid_op);
1164 set_trap_gate(7,&device_not_available);
1165 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1166 set_trap_gate(9,&coprocessor_segment_overrun);
1167 set_trap_gate(10,&invalid_TSS);
1168 set_trap_gate(11,&segment_not_present);
1169 set_trap_gate(12,&stack_segment);
1170 set_trap_gate(13,&general_protection);
1171 set_intr_gate(14,&page_fault);
1172 set_trap_gate(15,&spurious_interrupt_bug);
1173 set_trap_gate(16,&coprocessor_error);
1174 set_trap_gate(17,&alignment_check);
1175 #ifdef CONFIG_X86_MCE
1176 set_trap_gate(18,&machine_check);
1178 set_trap_gate(19,&simd_coprocessor_error);
1182 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1183 * Generates a compile-time "error: zero width for bit-field" if
1184 * the alignment is wrong.
1186 struct fxsrAlignAssert {
1187 int _:!(offsetof(struct task_struct,
1188 thread.i387.fxsave) & 15);
1191 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1192 set_in_cr4(X86_CR4_OSFXSR);
1196 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1198 set_in_cr4(X86_CR4_OSXMMEXCPT);
1202 set_system_gate(SYSCALL_VECTOR,&system_call);
1205 * Should be a barrier for any external CPU state.
1212 static int __init kstack_setup(char *s)
1214 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1217 __setup("kstack=", kstack_setup);