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 inline unsigned long print_context_stack(struct thread_info *tinfo,
116 unsigned long *stack, unsigned long ebp)
120 #ifdef CONFIG_FRAME_POINTER
121 while (valid_stack_ptr(tinfo, (void *)ebp)) {
122 addr = *(unsigned long *)(ebp + 4);
123 printk(" [<%08lx>] ", addr);
124 print_symbol("%s", addr);
126 ebp = *(unsigned long *)ebp;
129 while (valid_stack_ptr(tinfo, stack)) {
131 if (__kernel_text_address(addr)) {
132 printk(" [<%08lx>]", addr);
133 print_symbol(" %s", addr);
141 void show_trace(struct task_struct *task, unsigned long * stack)
148 if (task == current) {
149 /* Grab ebp right from our regs */
150 asm ("movl %%ebp, %0" : "=r" (ebp) : );
152 /* ebp is the last reg pushed by switch_to */
153 ebp = *(unsigned long *) task->thread.esp;
157 struct thread_info *context;
158 context = (struct thread_info *)
159 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
160 ebp = print_context_stack(context, stack, ebp);
161 stack = (unsigned long*)context->previous_esp;
164 printk(" =======================\n");
168 void show_stack(struct task_struct *task, unsigned long *esp)
170 unsigned long *stack;
175 esp = (unsigned long*)task->thread.esp;
177 esp = (unsigned long *)&esp;
181 for(i = 0; i < kstack_depth_to_print; i++) {
182 if (kstack_end(stack))
184 if (i && ((i % 8) == 0))
186 printk("%08lx ", *stack++);
188 printk("\nCall Trace:\n");
189 show_trace(task, esp);
193 * The architecture-independent dump_stack generator
195 void dump_stack(void)
199 show_trace(current, &stack);
202 EXPORT_SYMBOL(dump_stack);
204 void show_registers(struct pt_regs *regs)
211 esp = (unsigned long) (®s->esp);
213 if (user_mode(regs)) {
216 ss = regs->xss & 0xffff;
219 printk("CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\nEFLAGS: %08lx"
221 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
222 print_tainted(), regs->eflags, system_utsname.release);
223 print_symbol("EIP is at %s\n", regs->eip);
224 printk("eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
225 regs->eax, regs->ebx, regs->ecx, regs->edx);
226 printk("esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
227 regs->esi, regs->edi, regs->ebp, esp);
228 printk("ds: %04x es: %04x ss: %04x\n",
229 regs->xds & 0xffff, regs->xes & 0xffff, ss);
230 printk("Process %s (pid: %d, threadinfo=%p task=%p)",
231 current->comm, current->pid, current_thread_info(), current);
233 * When in-kernel, we also print out the stack and code at the
234 * time of the fault..
240 show_stack(NULL, (unsigned long*)esp);
244 eip = (u8 __user *)regs->eip - 43;
245 for (i = 0; i < 64; i++, eip++) {
248 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
249 printk(" Bad EIP value.");
252 if (eip == (u8 __user *)regs->eip)
253 printk("<%02x> ", c);
261 static void handle_BUG(struct pt_regs *regs)
271 if (eip < PAGE_OFFSET)
273 if (__get_user(ud2, (unsigned short __user *)eip))
277 if (__get_user(line, (unsigned short __user *)(eip + 2)))
279 if (__get_user(file, (char * __user *)(eip + 4)) ||
280 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
281 file = "<bad filename>";
283 printk("------------[ cut here ]------------\n");
284 printk(KERN_ALERT "kernel BUG at %s:%d!\n", file, line);
289 /* Here we know it was a BUG but file-n-line is unavailable */
291 printk("Kernel BUG\n");
294 /* This is gone through when something in the kernel
295 * has done something bad and is about to be terminated.
297 void die(const char * str, struct pt_regs * regs, long err)
302 int lock_owner_depth;
304 .lock = SPIN_LOCK_UNLOCKED,
306 .lock_owner_depth = 0
308 static int die_counter;
311 if (die.lock_owner != raw_smp_processor_id()) {
313 spin_lock_irqsave(&die.lock, flags);
314 die.lock_owner = smp_processor_id();
315 die.lock_owner_depth = 0;
319 local_save_flags(flags);
321 if (++die.lock_owner_depth < 3) {
324 printk(KERN_ALERT "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
325 #ifdef CONFIG_PREEMPT
333 #ifdef CONFIG_DEBUG_PAGEALLOC
334 printk("DEBUG_PAGEALLOC");
339 notify_die(DIE_OOPS, (char *)str, regs, err, 255, SIGSEGV);
340 show_registers(regs);
342 printk(KERN_ERR "Recursive die() failure, output suppressed\n");
346 spin_unlock_irqrestore(&die.lock, flags);
348 if (kexec_should_crash(current))
352 panic("Fatal exception in interrupt");
355 printk(KERN_EMERG "Fatal exception: panic in 5 seconds\n");
357 panic("Fatal exception");
362 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
364 if (!user_mode_vm(regs))
368 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
369 struct pt_regs * regs, long error_code,
372 struct task_struct *tsk = current;
373 tsk->thread.error_code = error_code;
374 tsk->thread.trap_no = trapnr;
376 if (regs->eflags & VM_MASK) {
382 if (!user_mode(regs))
387 force_sig_info(signr, info, tsk);
389 force_sig(signr, tsk);
394 if (!fixup_exception(regs))
395 die(str, regs, error_code);
400 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
401 if (ret) goto trap_signal;
406 #define DO_ERROR(trapnr, signr, str, name) \
407 fastcall void do_##name(struct pt_regs * regs, long error_code) \
409 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
412 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
415 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
416 fastcall void do_##name(struct pt_regs * regs, long error_code) \
419 info.si_signo = signr; \
421 info.si_code = sicode; \
422 info.si_addr = (void __user *)siaddr; \
423 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
426 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
429 #define DO_VM86_ERROR(trapnr, signr, str, name) \
430 fastcall void do_##name(struct pt_regs * regs, long error_code) \
432 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
435 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
438 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
439 fastcall void do_##name(struct pt_regs * regs, long error_code) \
442 info.si_signo = signr; \
444 info.si_code = sicode; \
445 info.si_addr = (void __user *)siaddr; \
446 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
449 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
452 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
453 #ifndef CONFIG_KPROBES
454 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
456 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
457 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
458 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
459 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
460 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
461 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
462 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
463 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
464 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
466 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
470 struct tss_struct *tss = &per_cpu(init_tss, cpu);
471 struct thread_struct *thread = ¤t->thread;
474 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
475 * invalid offset set (the LAZY one) and the faulting thread has
476 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
477 * and we set the offset field correctly. Then we let the CPU to
478 * restart the faulting instruction.
480 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
481 thread->io_bitmap_ptr) {
482 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
483 thread->io_bitmap_max);
485 * If the previously set map was extending to higher ports
486 * than the current one, pad extra space with 0xff (no access).
488 if (thread->io_bitmap_max < tss->io_bitmap_max)
489 memset((char *) tss->io_bitmap +
490 thread->io_bitmap_max, 0xff,
491 tss->io_bitmap_max - thread->io_bitmap_max);
492 tss->io_bitmap_max = thread->io_bitmap_max;
493 tss->io_bitmap_base = IO_BITMAP_OFFSET;
494 tss->io_bitmap_owner = thread;
500 current->thread.error_code = error_code;
501 current->thread.trap_no = 13;
503 if (regs->eflags & VM_MASK)
506 if (!user_mode(regs))
509 current->thread.error_code = error_code;
510 current->thread.trap_no = 13;
511 force_sig(SIGSEGV, current);
516 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
520 if (!fixup_exception(regs)) {
521 if (notify_die(DIE_GPF, "general protection fault", regs,
522 error_code, 13, SIGSEGV) == NOTIFY_STOP)
524 die("general protection fault", regs, error_code);
528 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
530 printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
531 printk("You probably have a hardware problem with your RAM chips\n");
533 /* Clear and disable the memory parity error line. */
534 clear_mem_error(reason);
537 static void io_check_error(unsigned char reason, struct pt_regs * regs)
541 printk("NMI: IOCK error (debug interrupt?)\n");
542 show_registers(regs);
544 /* Re-enable the IOCK line, wait for a few seconds */
545 reason = (reason & 0xf) | 8;
548 while (--i) udelay(1000);
553 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
556 /* Might actually be able to figure out what the guilty party
563 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
564 reason, smp_processor_id());
565 printk("Dazed and confused, but trying to continue\n");
566 printk("Do you have a strange power saving mode enabled?\n");
569 static DEFINE_SPINLOCK(nmi_print_lock);
571 void die_nmi (struct pt_regs *regs, const char *msg)
573 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 0, SIGINT) ==
577 spin_lock(&nmi_print_lock);
579 * We are in trouble anyway, lets at least try
580 * to get a message out.
584 printk(" on CPU%d, eip %08lx, registers:\n",
585 smp_processor_id(), regs->eip);
586 show_registers(regs);
587 printk("console shuts up ...\n");
589 spin_unlock(&nmi_print_lock);
592 /* If we are in kernel we are probably nested up pretty bad
593 * and might aswell get out now while we still can.
595 if (!user_mode(regs)) {
596 current->thread.trap_no = 2;
603 static void default_do_nmi(struct pt_regs * regs)
605 unsigned char reason = 0;
607 /* Only the BSP gets external NMIs from the system. */
608 if (!smp_processor_id())
609 reason = get_nmi_reason();
611 if (!(reason & 0xc0)) {
612 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 0, SIGINT)
615 #ifdef CONFIG_X86_LOCAL_APIC
617 * Ok, so this is none of the documented NMI sources,
618 * so it must be the NMI watchdog.
621 nmi_watchdog_tick(regs);
625 unknown_nmi_error(reason, regs);
628 if (notify_die(DIE_NMI, "nmi", regs, reason, 0, SIGINT) == NOTIFY_STOP)
631 mem_parity_error(reason, regs);
633 io_check_error(reason, regs);
635 * Reassert NMI in case it became active meanwhile
636 * as it's edge-triggered.
641 static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
646 static nmi_callback_t nmi_callback = dummy_nmi_callback;
648 fastcall void do_nmi(struct pt_regs * regs, long error_code)
654 cpu = smp_processor_id();
658 if (!rcu_dereference(nmi_callback)(regs, cpu))
659 default_do_nmi(regs);
664 void set_nmi_callback(nmi_callback_t callback)
666 rcu_assign_pointer(nmi_callback, callback);
668 EXPORT_SYMBOL_GPL(set_nmi_callback);
670 void unset_nmi_callback(void)
672 nmi_callback = dummy_nmi_callback;
674 EXPORT_SYMBOL_GPL(unset_nmi_callback);
676 #ifdef CONFIG_KPROBES
677 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
679 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
682 /* This is an interrupt gate, because kprobes wants interrupts
683 disabled. Normal trap handlers don't. */
684 restore_interrupts(regs);
685 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
690 * Our handling of the processor debug registers is non-trivial.
691 * We do not clear them on entry and exit from the kernel. Therefore
692 * it is possible to get a watchpoint trap here from inside the kernel.
693 * However, the code in ./ptrace.c has ensured that the user can
694 * only set watchpoints on userspace addresses. Therefore the in-kernel
695 * watchpoint trap can only occur in code which is reading/writing
696 * from user space. Such code must not hold kernel locks (since it
697 * can equally take a page fault), therefore it is safe to call
698 * force_sig_info even though that claims and releases locks.
700 * Code in ./signal.c ensures that the debug control register
701 * is restored before we deliver any signal, and therefore that
702 * user code runs with the correct debug control register even though
705 * Being careful here means that we don't have to be as careful in a
706 * lot of more complicated places (task switching can be a bit lazy
707 * about restoring all the debug state, and ptrace doesn't have to
708 * find every occurrence of the TF bit that could be saved away even
711 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
713 unsigned int condition;
714 struct task_struct *tsk = current;
716 get_debugreg(condition, 6);
718 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
719 SIGTRAP) == NOTIFY_STOP)
721 /* It's safe to allow irq's after DR6 has been saved */
722 if (regs->eflags & X86_EFLAGS_IF)
725 /* Mask out spurious debug traps due to lazy DR7 setting */
726 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
727 if (!tsk->thread.debugreg[7])
731 if (regs->eflags & VM_MASK)
734 /* Save debug status register where ptrace can see it */
735 tsk->thread.debugreg[6] = condition;
738 * Single-stepping through TF: make sure we ignore any events in
739 * kernel space (but re-enable TF when returning to user mode).
741 if (condition & DR_STEP) {
743 * We already checked v86 mode above, so we can
744 * check for kernel mode by just checking the CPL
747 if (!user_mode(regs))
748 goto clear_TF_reenable;
751 /* Ok, finally something we can handle */
752 send_sigtrap(tsk, regs, error_code);
754 /* Disable additional traps. They'll be re-enabled when
755 * the signal is delivered.
762 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
766 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
767 regs->eflags &= ~TF_MASK;
772 * Note that we play around with the 'TS' bit in an attempt to get
773 * the correct behaviour even in the presence of the asynchronous
776 void math_error(void __user *eip)
778 struct task_struct * task;
780 unsigned short cwd, swd;
783 * Save the info for the exception handler and clear the error.
787 task->thread.trap_no = 16;
788 task->thread.error_code = 0;
789 info.si_signo = SIGFPE;
791 info.si_code = __SI_FAULT;
794 * (~cwd & swd) will mask out exceptions that are not set to unmasked
795 * status. 0x3f is the exception bits in these regs, 0x200 is the
796 * C1 reg you need in case of a stack fault, 0x040 is the stack
797 * fault bit. We should only be taking one exception at a time,
798 * so if this combination doesn't produce any single exception,
799 * then we have a bad program that isn't syncronizing its FPU usage
800 * and it will suffer the consequences since we won't be able to
801 * fully reproduce the context of the exception
803 cwd = get_fpu_cwd(task);
804 swd = get_fpu_swd(task);
805 switch (swd & ~cwd & 0x3f) {
806 case 0x000: /* No unmasked exception */
808 default: /* Multiple exceptions */
810 case 0x001: /* Invalid Op */
812 * swd & 0x240 == 0x040: Stack Underflow
813 * swd & 0x240 == 0x240: Stack Overflow
814 * User must clear the SF bit (0x40) if set
816 info.si_code = FPE_FLTINV;
818 case 0x002: /* Denormalize */
819 case 0x010: /* Underflow */
820 info.si_code = FPE_FLTUND;
822 case 0x004: /* Zero Divide */
823 info.si_code = FPE_FLTDIV;
825 case 0x008: /* Overflow */
826 info.si_code = FPE_FLTOVF;
828 case 0x020: /* Precision */
829 info.si_code = FPE_FLTRES;
832 force_sig_info(SIGFPE, &info, task);
835 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
838 math_error((void __user *)regs->eip);
841 static void simd_math_error(void __user *eip)
843 struct task_struct * task;
845 unsigned short mxcsr;
848 * Save the info for the exception handler and clear the error.
852 task->thread.trap_no = 19;
853 task->thread.error_code = 0;
854 info.si_signo = SIGFPE;
856 info.si_code = __SI_FAULT;
859 * The SIMD FPU exceptions are handled a little differently, as there
860 * is only a single status/control register. Thus, to determine which
861 * unmasked exception was caught we must mask the exception mask bits
862 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
864 mxcsr = get_fpu_mxcsr(task);
865 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
869 case 0x001: /* Invalid Op */
870 info.si_code = FPE_FLTINV;
872 case 0x002: /* Denormalize */
873 case 0x010: /* Underflow */
874 info.si_code = FPE_FLTUND;
876 case 0x004: /* Zero Divide */
877 info.si_code = FPE_FLTDIV;
879 case 0x008: /* Overflow */
880 info.si_code = FPE_FLTOVF;
882 case 0x020: /* Precision */
883 info.si_code = FPE_FLTRES;
886 force_sig_info(SIGFPE, &info, task);
889 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
893 /* Handle SIMD FPU exceptions on PIII+ processors. */
895 simd_math_error((void __user *)regs->eip);
898 * Handle strange cache flush from user space exception
899 * in all other cases. This is undocumented behaviour.
901 if (regs->eflags & VM_MASK) {
902 handle_vm86_fault((struct kernel_vm86_regs *)regs,
906 current->thread.trap_no = 19;
907 current->thread.error_code = error_code;
908 die_if_kernel("cache flush denied", regs, error_code);
909 force_sig(SIGSEGV, current);
913 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
917 /* No need to warn about this any longer. */
918 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
922 fastcall void setup_x86_bogus_stack(unsigned char * stk)
924 unsigned long *switch16_ptr, *switch32_ptr;
925 struct pt_regs *regs;
926 unsigned long stack_top, stack_bot;
927 unsigned short iret_frame16_off;
928 int cpu = smp_processor_id();
929 /* reserve the space on 32bit stack for the magic switch16 pointer */
930 memmove(stk, stk + 8, sizeof(struct pt_regs));
931 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
932 regs = (struct pt_regs *)stk;
933 /* now the switch32 on 16bit stack */
934 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
935 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
936 switch32_ptr = (unsigned long *)(stack_top - 8);
937 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
938 /* copy iret frame on 16bit stack */
939 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
940 /* fill in the switch pointers */
941 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
942 switch16_ptr[1] = __ESPFIX_SS;
943 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
944 8 - CPU_16BIT_STACK_SIZE;
945 switch32_ptr[1] = __KERNEL_DS;
948 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
950 unsigned long *switch32_ptr;
951 unsigned char *stack16, *stack32;
952 unsigned long stack_top, stack_bot;
954 int cpu = smp_processor_id();
955 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
956 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
957 switch32_ptr = (unsigned long *)(stack_top - 8);
958 /* copy the data from 16bit stack to 32bit stack */
959 len = CPU_16BIT_STACK_SIZE - 8 - sp;
960 stack16 = (unsigned char *)(stack_bot + sp);
961 stack32 = (unsigned char *)
962 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
963 memcpy(stack32, stack16, len);
968 * 'math_state_restore()' saves the current math information in the
969 * old math state array, and gets the new ones from the current task
971 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
972 * Don't touch unless you *really* know how it works.
974 * Must be called with kernel preemption disabled (in this case,
975 * local interrupts are disabled at the call-site in entry.S).
977 asmlinkage void math_state_restore(struct pt_regs regs)
979 struct thread_info *thread = current_thread_info();
980 struct task_struct *tsk = thread->task;
982 clts(); /* Allow maths ops (or we recurse) */
983 if (!tsk_used_math(tsk))
986 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
989 #ifndef CONFIG_MATH_EMULATION
991 asmlinkage void math_emulate(long arg)
993 printk("math-emulation not enabled and no coprocessor found.\n");
994 printk("killing %s.\n",current->comm);
995 force_sig(SIGFPE,current);
999 #endif /* CONFIG_MATH_EMULATION */
1001 #ifdef CONFIG_X86_F00F_BUG
1002 void __init trap_init_f00f_bug(void)
1004 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1007 * Update the IDT descriptor and reload the IDT so that
1008 * it uses the read-only mapped virtual address.
1010 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1011 load_idt(&idt_descr);
1015 #define _set_gate(gate_addr,type,dpl,addr,seg) \
1018 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1019 "movw %4,%%dx\n\t" \
1020 "movl %%eax,%0\n\t" \
1022 :"=m" (*((long *) (gate_addr))), \
1023 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1024 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1025 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1030 * This needs to use 'idt_table' rather than 'idt', and
1031 * thus use the _nonmapped_ version of the IDT, as the
1032 * Pentium F0 0F bugfix can have resulted in the mapped
1033 * IDT being write-protected.
1035 void set_intr_gate(unsigned int n, void *addr)
1037 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1041 * This routine sets up an interrupt gate at directory privilege level 3.
1043 static inline void set_system_intr_gate(unsigned int n, void *addr)
1045 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1048 static void __init set_trap_gate(unsigned int n, void *addr)
1050 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1053 static void __init set_system_gate(unsigned int n, void *addr)
1055 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1058 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1060 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1064 void __init trap_init(void)
1067 void __iomem *p = ioremap(0x0FFFD9, 4);
1068 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1074 #ifdef CONFIG_X86_LOCAL_APIC
1075 init_apic_mappings();
1078 set_trap_gate(0,÷_error);
1079 set_intr_gate(1,&debug);
1080 set_intr_gate(2,&nmi);
1081 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1082 set_system_gate(4,&overflow);
1083 set_trap_gate(5,&bounds);
1084 set_trap_gate(6,&invalid_op);
1085 set_trap_gate(7,&device_not_available);
1086 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1087 set_trap_gate(9,&coprocessor_segment_overrun);
1088 set_trap_gate(10,&invalid_TSS);
1089 set_trap_gate(11,&segment_not_present);
1090 set_trap_gate(12,&stack_segment);
1091 set_trap_gate(13,&general_protection);
1092 set_intr_gate(14,&page_fault);
1093 set_trap_gate(15,&spurious_interrupt_bug);
1094 set_trap_gate(16,&coprocessor_error);
1095 set_trap_gate(17,&alignment_check);
1096 #ifdef CONFIG_X86_MCE
1097 set_trap_gate(18,&machine_check);
1099 set_trap_gate(19,&simd_coprocessor_error);
1103 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1104 * Generates a compile-time "error: zero width for bit-field" if
1105 * the alignment is wrong.
1107 struct fxsrAlignAssert {
1108 int _:!(offsetof(struct task_struct,
1109 thread.i387.fxsave) & 15);
1112 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1113 set_in_cr4(X86_CR4_OSFXSR);
1117 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1119 set_in_cr4(X86_CR4_OSXMMEXCPT);
1123 set_system_gate(SYSCALL_VECTOR,&system_call);
1126 * Should be a barrier for any external CPU state.
1133 static int __init kstack_setup(char *s)
1135 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1138 __setup("kstack=", kstack_setup);