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>
56 #include <linux/vserver/debug.h>
58 #include "mach_traps.h"
60 asmlinkage int system_call(void);
62 struct desc_struct default_ldt[] = { { 0, 0 }, { 0, 0 }, { 0, 0 },
65 /* Do we ignore FPU interrupts ? */
66 char ignore_fpu_irq = 0;
69 * The IDT has to be page-aligned to simplify the Pentium
70 * F0 0F bug workaround.. We have a special link segment
73 struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
75 asmlinkage void divide_error(void);
76 asmlinkage void debug(void);
77 asmlinkage void nmi(void);
78 asmlinkage void int3(void);
79 asmlinkage void overflow(void);
80 asmlinkage void bounds(void);
81 asmlinkage void invalid_op(void);
82 asmlinkage void device_not_available(void);
83 asmlinkage void coprocessor_segment_overrun(void);
84 asmlinkage void invalid_TSS(void);
85 asmlinkage void segment_not_present(void);
86 asmlinkage void stack_segment(void);
87 asmlinkage void general_protection(void);
88 asmlinkage void page_fault(void);
89 asmlinkage void coprocessor_error(void);
90 asmlinkage void simd_coprocessor_error(void);
91 asmlinkage void alignment_check(void);
92 asmlinkage void spurious_interrupt_bug(void);
93 asmlinkage void machine_check(void);
95 static int kstack_depth_to_print = 24;
96 struct notifier_block *i386die_chain;
97 static DEFINE_SPINLOCK(die_notifier_lock);
99 int register_die_notifier(struct notifier_block *nb)
103 spin_lock_irqsave(&die_notifier_lock, flags);
104 err = notifier_chain_register(&i386die_chain, nb);
105 spin_unlock_irqrestore(&die_notifier_lock, flags);
108 EXPORT_SYMBOL(register_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;
116 static void print_addr_and_symbol(unsigned long addr, char *log_lvl)
119 printk(" [<%08lx>] ", addr);
120 print_symbol("%s", addr);
124 static inline unsigned long print_context_stack(struct thread_info *tinfo,
125 unsigned long *stack, unsigned long ebp,
130 #ifdef CONFIG_FRAME_POINTER
131 while (valid_stack_ptr(tinfo, (void *)ebp)) {
132 addr = *(unsigned long *)(ebp + 4);
133 print_addr_and_symbol(addr, log_lvl);
134 ebp = *(unsigned long *)ebp;
137 while (valid_stack_ptr(tinfo, stack)) {
139 if (__kernel_text_address(addr))
140 print_addr_and_symbol(addr, log_lvl);
146 static void show_trace_log_lvl(struct task_struct *task,
147 unsigned long *stack, char *log_lvl)
154 if (task == current) {
155 /* Grab ebp right from our regs */
156 asm ("movl %%ebp, %0" : "=r" (ebp) : );
158 /* ebp is the last reg pushed by switch_to */
159 ebp = *(unsigned long *) task->thread.esp;
163 struct thread_info *context;
164 context = (struct thread_info *)
165 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
166 ebp = print_context_stack(context, stack, ebp, log_lvl);
167 stack = (unsigned long*)context->previous_esp;
171 printk(" =======================\n");
175 void show_trace(struct task_struct *task, unsigned long * stack)
177 show_trace_log_lvl(task, stack, "");
180 static void show_stack_log_lvl(struct task_struct *task, unsigned long *esp,
183 unsigned long *stack;
188 esp = (unsigned long*)task->thread.esp;
190 esp = (unsigned long *)&esp;
195 for(i = 0; i < kstack_depth_to_print; i++) {
196 if (kstack_end(stack))
198 if (i && ((i % 8) == 0)) {
203 printk("%08lx ", *stack++);
207 printk("Call Trace:\n");
208 show_trace_log_lvl(task, esp, log_lvl);
211 void show_stack(struct task_struct *task, unsigned long *esp)
213 show_stack_log_lvl(task, esp, "");
217 * The architecture-independent dump_stack generator
219 void dump_stack(void)
223 show_trace(current, &stack);
226 EXPORT_SYMBOL(dump_stack);
228 void show_registers(struct pt_regs *regs)
235 esp = (unsigned long) (®s->esp);
237 if (user_mode(regs)) {
240 ss = regs->xss & 0xffff;
243 printk(KERN_EMERG "CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\n"
244 "EFLAGS: %08lx (%s %.*s) \n",
245 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
246 print_tainted(), regs->eflags, system_utsname.release,
247 (int)strcspn(system_utsname.version, " "),
248 system_utsname.version);
249 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
250 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
251 regs->eax, regs->ebx, regs->ecx, regs->edx);
252 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
253 regs->esi, regs->edi, regs->ebp, esp);
254 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
255 regs->xds & 0xffff, regs->xes & 0xffff, ss);
256 printk(KERN_EMERG "Process %s (pid: %d[#%u], threadinfo=%p task=%p)",
257 current->comm, current->pid, current->xid,
258 current_thread_info(), current);
260 * When in-kernel, we also print out the stack and code at the
261 * time of the fault..
266 printk("\n" KERN_EMERG "Stack: ");
267 show_stack_log_lvl(NULL, (unsigned long *)esp, KERN_EMERG);
269 printk(KERN_EMERG "Code: ");
271 eip = (u8 __user *)regs->eip - 43;
272 for (i = 0; i < 64; i++, eip++) {
275 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
276 printk(" Bad EIP value.");
279 if (eip == (u8 __user *)regs->eip)
280 printk("<%02x> ", c);
288 static void handle_BUG(struct pt_regs *regs)
298 if (eip < PAGE_OFFSET)
300 if (__get_user(ud2, (unsigned short __user *)eip))
304 if (__get_user(line, (unsigned short __user *)(eip + 2)))
306 if (__get_user(file, (char * __user *)(eip + 4)) ||
307 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
308 file = "<bad filename>";
310 printk(KERN_EMERG "------------[ cut here ]------------\n");
311 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
316 /* Here we know it was a BUG but file-n-line is unavailable */
318 printk(KERN_EMERG "Kernel BUG\n");
321 /* This is gone through when something in the kernel
322 * has done something bad and is about to be terminated.
324 void die(const char * str, struct pt_regs * regs, long err)
329 int lock_owner_depth;
331 .lock = SPIN_LOCK_UNLOCKED,
333 .lock_owner_depth = 0
335 static int die_counter;
340 if (die.lock_owner != raw_smp_processor_id()) {
342 spin_lock_irqsave(&die.lock, flags);
343 die.lock_owner = smp_processor_id();
344 die.lock_owner_depth = 0;
348 local_save_flags(flags);
350 if (++die.lock_owner_depth < 3) {
353 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
354 #ifdef CONFIG_PREEMPT
355 printk(KERN_EMERG "PREEMPT ");
364 #ifdef CONFIG_DEBUG_PAGEALLOC
367 printk("DEBUG_PAGEALLOC");
372 notify_die(DIE_OOPS, (char *)str, regs, err, 255, SIGSEGV);
373 show_registers(regs);
376 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
380 spin_unlock_irqrestore(&die.lock, flags);
382 if (kexec_should_crash(current))
386 panic("Fatal exception in interrupt");
389 printk(KERN_EMERG "Fatal exception: panic in 5 seconds\n");
391 panic("Fatal exception");
396 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
398 if (!user_mode_vm(regs))
402 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
403 struct pt_regs * regs, long error_code,
406 struct task_struct *tsk = current;
407 tsk->thread.error_code = error_code;
408 tsk->thread.trap_no = trapnr;
410 if (regs->eflags & VM_MASK) {
416 if (!user_mode(regs))
421 force_sig_info(signr, info, tsk);
423 force_sig(signr, tsk);
428 if (!fixup_exception(regs))
429 die(str, regs, error_code);
434 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
435 if (ret) goto trap_signal;
440 #define DO_ERROR(trapnr, signr, str, name) \
441 fastcall void do_##name(struct pt_regs * regs, long error_code) \
443 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
446 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
449 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
450 fastcall void do_##name(struct pt_regs * regs, long error_code) \
453 info.si_signo = signr; \
455 info.si_code = sicode; \
456 info.si_addr = (void __user *)siaddr; \
457 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
460 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
463 #define DO_VM86_ERROR(trapnr, signr, str, name) \
464 fastcall void do_##name(struct pt_regs * regs, long error_code) \
466 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
469 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
472 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
473 fastcall void do_##name(struct pt_regs * regs, long error_code) \
476 info.si_signo = signr; \
478 info.si_code = sicode; \
479 info.si_addr = (void __user *)siaddr; \
480 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
483 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
486 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
487 #ifndef CONFIG_KPROBES
488 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
490 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
491 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
492 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
493 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
494 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
495 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
496 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
497 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
498 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
500 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
504 struct tss_struct *tss = &per_cpu(init_tss, cpu);
505 struct thread_struct *thread = ¤t->thread;
508 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
509 * invalid offset set (the LAZY one) and the faulting thread has
510 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
511 * and we set the offset field correctly. Then we let the CPU to
512 * restart the faulting instruction.
514 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
515 thread->io_bitmap_ptr) {
516 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
517 thread->io_bitmap_max);
519 * If the previously set map was extending to higher ports
520 * than the current one, pad extra space with 0xff (no access).
522 if (thread->io_bitmap_max < tss->io_bitmap_max)
523 memset((char *) tss->io_bitmap +
524 thread->io_bitmap_max, 0xff,
525 tss->io_bitmap_max - thread->io_bitmap_max);
526 tss->io_bitmap_max = thread->io_bitmap_max;
527 tss->io_bitmap_base = IO_BITMAP_OFFSET;
528 tss->io_bitmap_owner = thread;
534 current->thread.error_code = error_code;
535 current->thread.trap_no = 13;
537 if (regs->eflags & VM_MASK)
540 if (!user_mode(regs))
543 current->thread.error_code = error_code;
544 current->thread.trap_no = 13;
545 force_sig(SIGSEGV, current);
550 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
554 if (!fixup_exception(regs)) {
555 if (notify_die(DIE_GPF, "general protection fault", regs,
556 error_code, 13, SIGSEGV) == NOTIFY_STOP)
558 die("general protection fault", regs, error_code);
562 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
564 printk(KERN_EMERG "Uhhuh. NMI received. Dazed and confused, but trying "
566 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
569 /* Clear and disable the memory parity error line. */
570 clear_mem_error(reason);
573 static void io_check_error(unsigned char reason, struct pt_regs * regs)
577 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
578 show_registers(regs);
580 /* Re-enable the IOCK line, wait for a few seconds */
581 reason = (reason & 0xf) | 8;
584 while (--i) udelay(1000);
589 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
592 /* Might actually be able to figure out what the guilty party
599 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
600 reason, smp_processor_id());
601 printk("Dazed and confused, but trying to continue\n");
602 printk("Do you have a strange power saving mode enabled?\n");
605 static DEFINE_SPINLOCK(nmi_print_lock);
607 void die_nmi (struct pt_regs *regs, const char *msg)
609 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 0, SIGINT) ==
613 spin_lock(&nmi_print_lock);
615 * We are in trouble anyway, lets at least try
616 * to get a message out.
619 printk(KERN_EMERG "%s", msg);
620 printk(" on CPU%d, eip %08lx, registers:\n",
621 smp_processor_id(), regs->eip);
622 show_registers(regs);
623 printk(KERN_EMERG "console shuts up ...\n");
625 spin_unlock(&nmi_print_lock);
628 /* If we are in kernel we are probably nested up pretty bad
629 * and might aswell get out now while we still can.
631 if (!user_mode(regs)) {
632 current->thread.trap_no = 2;
639 static void default_do_nmi(struct pt_regs * regs)
641 unsigned char reason = 0;
643 /* Only the BSP gets external NMIs from the system. */
644 if (!smp_processor_id())
645 reason = get_nmi_reason();
647 if (!(reason & 0xc0)) {
648 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 0, SIGINT)
651 #ifdef CONFIG_X86_LOCAL_APIC
653 * Ok, so this is none of the documented NMI sources,
654 * so it must be the NMI watchdog.
657 nmi_watchdog_tick(regs);
661 unknown_nmi_error(reason, regs);
664 if (notify_die(DIE_NMI, "nmi", regs, reason, 0, SIGINT) == NOTIFY_STOP)
667 mem_parity_error(reason, regs);
669 io_check_error(reason, regs);
671 * Reassert NMI in case it became active meanwhile
672 * as it's edge-triggered.
677 static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
682 static nmi_callback_t nmi_callback = dummy_nmi_callback;
684 fastcall void do_nmi(struct pt_regs * regs, long error_code)
690 cpu = smp_processor_id();
694 if (!rcu_dereference(nmi_callback)(regs, cpu))
695 default_do_nmi(regs);
700 void set_nmi_callback(nmi_callback_t callback)
702 rcu_assign_pointer(nmi_callback, callback);
704 EXPORT_SYMBOL_GPL(set_nmi_callback);
706 void unset_nmi_callback(void)
708 nmi_callback = dummy_nmi_callback;
710 EXPORT_SYMBOL_GPL(unset_nmi_callback);
712 #ifdef CONFIG_KPROBES
713 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
715 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
718 /* This is an interrupt gate, because kprobes wants interrupts
719 disabled. Normal trap handlers don't. */
720 restore_interrupts(regs);
721 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
726 * Our handling of the processor debug registers is non-trivial.
727 * We do not clear them on entry and exit from the kernel. Therefore
728 * it is possible to get a watchpoint trap here from inside the kernel.
729 * However, the code in ./ptrace.c has ensured that the user can
730 * only set watchpoints on userspace addresses. Therefore the in-kernel
731 * watchpoint trap can only occur in code which is reading/writing
732 * from user space. Such code must not hold kernel locks (since it
733 * can equally take a page fault), therefore it is safe to call
734 * force_sig_info even though that claims and releases locks.
736 * Code in ./signal.c ensures that the debug control register
737 * is restored before we deliver any signal, and therefore that
738 * user code runs with the correct debug control register even though
741 * Being careful here means that we don't have to be as careful in a
742 * lot of more complicated places (task switching can be a bit lazy
743 * about restoring all the debug state, and ptrace doesn't have to
744 * find every occurrence of the TF bit that could be saved away even
747 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
749 unsigned int condition;
750 struct task_struct *tsk = current;
752 get_debugreg(condition, 6);
754 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
755 SIGTRAP) == NOTIFY_STOP)
757 /* It's safe to allow irq's after DR6 has been saved */
758 if (regs->eflags & X86_EFLAGS_IF)
761 /* Mask out spurious debug traps due to lazy DR7 setting */
762 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
763 if (!tsk->thread.debugreg[7])
767 if (regs->eflags & VM_MASK)
770 /* Save debug status register where ptrace can see it */
771 tsk->thread.debugreg[6] = condition;
774 * Single-stepping through TF: make sure we ignore any events in
775 * kernel space (but re-enable TF when returning to user mode).
777 if (condition & DR_STEP) {
779 * We already checked v86 mode above, so we can
780 * check for kernel mode by just checking the CPL
783 if (!user_mode(regs))
784 goto clear_TF_reenable;
787 /* Ok, finally something we can handle */
788 send_sigtrap(tsk, regs, error_code);
790 /* Disable additional traps. They'll be re-enabled when
791 * the signal is delivered.
798 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
802 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
803 regs->eflags &= ~TF_MASK;
808 * Note that we play around with the 'TS' bit in an attempt to get
809 * the correct behaviour even in the presence of the asynchronous
812 void math_error(void __user *eip)
814 struct task_struct * task;
816 unsigned short cwd, swd;
819 * Save the info for the exception handler and clear the error.
823 task->thread.trap_no = 16;
824 task->thread.error_code = 0;
825 info.si_signo = SIGFPE;
827 info.si_code = __SI_FAULT;
830 * (~cwd & swd) will mask out exceptions that are not set to unmasked
831 * status. 0x3f is the exception bits in these regs, 0x200 is the
832 * C1 reg you need in case of a stack fault, 0x040 is the stack
833 * fault bit. We should only be taking one exception at a time,
834 * so if this combination doesn't produce any single exception,
835 * then we have a bad program that isn't syncronizing its FPU usage
836 * and it will suffer the consequences since we won't be able to
837 * fully reproduce the context of the exception
839 cwd = get_fpu_cwd(task);
840 swd = get_fpu_swd(task);
841 switch (swd & ~cwd & 0x3f) {
842 case 0x000: /* No unmasked exception */
844 default: /* Multiple exceptions */
846 case 0x001: /* Invalid Op */
848 * swd & 0x240 == 0x040: Stack Underflow
849 * swd & 0x240 == 0x240: Stack Overflow
850 * User must clear the SF bit (0x40) if set
852 info.si_code = FPE_FLTINV;
854 case 0x002: /* Denormalize */
855 case 0x010: /* Underflow */
856 info.si_code = FPE_FLTUND;
858 case 0x004: /* Zero Divide */
859 info.si_code = FPE_FLTDIV;
861 case 0x008: /* Overflow */
862 info.si_code = FPE_FLTOVF;
864 case 0x020: /* Precision */
865 info.si_code = FPE_FLTRES;
868 force_sig_info(SIGFPE, &info, task);
871 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
874 math_error((void __user *)regs->eip);
877 static void simd_math_error(void __user *eip)
879 struct task_struct * task;
881 unsigned short mxcsr;
884 * Save the info for the exception handler and clear the error.
888 task->thread.trap_no = 19;
889 task->thread.error_code = 0;
890 info.si_signo = SIGFPE;
892 info.si_code = __SI_FAULT;
895 * The SIMD FPU exceptions are handled a little differently, as there
896 * is only a single status/control register. Thus, to determine which
897 * unmasked exception was caught we must mask the exception mask bits
898 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
900 mxcsr = get_fpu_mxcsr(task);
901 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
905 case 0x001: /* Invalid Op */
906 info.si_code = FPE_FLTINV;
908 case 0x002: /* Denormalize */
909 case 0x010: /* Underflow */
910 info.si_code = FPE_FLTUND;
912 case 0x004: /* Zero Divide */
913 info.si_code = FPE_FLTDIV;
915 case 0x008: /* Overflow */
916 info.si_code = FPE_FLTOVF;
918 case 0x020: /* Precision */
919 info.si_code = FPE_FLTRES;
922 force_sig_info(SIGFPE, &info, task);
925 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
929 /* Handle SIMD FPU exceptions on PIII+ processors. */
931 simd_math_error((void __user *)regs->eip);
934 * Handle strange cache flush from user space exception
935 * in all other cases. This is undocumented behaviour.
937 if (regs->eflags & VM_MASK) {
938 handle_vm86_fault((struct kernel_vm86_regs *)regs,
942 current->thread.trap_no = 19;
943 current->thread.error_code = error_code;
944 die_if_kernel("cache flush denied", regs, error_code);
945 force_sig(SIGSEGV, current);
949 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
953 /* No need to warn about this any longer. */
954 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
958 fastcall void setup_x86_bogus_stack(unsigned char * stk)
960 unsigned long *switch16_ptr, *switch32_ptr;
961 struct pt_regs *regs;
962 unsigned long stack_top, stack_bot;
963 unsigned short iret_frame16_off;
964 int cpu = smp_processor_id();
965 /* reserve the space on 32bit stack for the magic switch16 pointer */
966 memmove(stk, stk + 8, sizeof(struct pt_regs));
967 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
968 regs = (struct pt_regs *)stk;
969 /* now the switch32 on 16bit stack */
970 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
971 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
972 switch32_ptr = (unsigned long *)(stack_top - 8);
973 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
974 /* copy iret frame on 16bit stack */
975 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
976 /* fill in the switch pointers */
977 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
978 switch16_ptr[1] = __ESPFIX_SS;
979 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
980 8 - CPU_16BIT_STACK_SIZE;
981 switch32_ptr[1] = __KERNEL_DS;
984 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
986 unsigned long *switch32_ptr;
987 unsigned char *stack16, *stack32;
988 unsigned long stack_top, stack_bot;
990 int cpu = smp_processor_id();
991 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
992 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
993 switch32_ptr = (unsigned long *)(stack_top - 8);
994 /* copy the data from 16bit stack to 32bit stack */
995 len = CPU_16BIT_STACK_SIZE - 8 - sp;
996 stack16 = (unsigned char *)(stack_bot + sp);
997 stack32 = (unsigned char *)
998 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
999 memcpy(stack32, stack16, len);
1004 * 'math_state_restore()' saves the current math information in the
1005 * old math state array, and gets the new ones from the current task
1007 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1008 * Don't touch unless you *really* know how it works.
1010 * Must be called with kernel preemption disabled (in this case,
1011 * local interrupts are disabled at the call-site in entry.S).
1013 asmlinkage void math_state_restore(struct pt_regs regs)
1015 struct thread_info *thread = current_thread_info();
1016 struct task_struct *tsk = thread->task;
1018 clts(); /* Allow maths ops (or we recurse) */
1019 if (!tsk_used_math(tsk))
1022 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1025 #ifndef CONFIG_MATH_EMULATION
1027 asmlinkage void math_emulate(long arg)
1029 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1030 printk(KERN_EMERG "killing %s.\n",current->comm);
1031 force_sig(SIGFPE,current);
1035 #endif /* CONFIG_MATH_EMULATION */
1037 #ifdef CONFIG_X86_F00F_BUG
1038 void __init trap_init_f00f_bug(void)
1040 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1043 * Update the IDT descriptor and reload the IDT so that
1044 * it uses the read-only mapped virtual address.
1046 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1047 load_idt(&idt_descr);
1051 #define _set_gate(gate_addr,type,dpl,addr,seg) \
1054 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1055 "movw %4,%%dx\n\t" \
1056 "movl %%eax,%0\n\t" \
1058 :"=m" (*((long *) (gate_addr))), \
1059 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1060 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1061 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1066 * This needs to use 'idt_table' rather than 'idt', and
1067 * thus use the _nonmapped_ version of the IDT, as the
1068 * Pentium F0 0F bugfix can have resulted in the mapped
1069 * IDT being write-protected.
1071 void set_intr_gate(unsigned int n, void *addr)
1073 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1077 * This routine sets up an interrupt gate at directory privilege level 3.
1079 static inline void set_system_intr_gate(unsigned int n, void *addr)
1081 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1084 static void __init set_trap_gate(unsigned int n, void *addr)
1086 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1089 static void __init set_system_gate(unsigned int n, void *addr)
1091 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1094 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1096 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1100 void __init trap_init(void)
1103 void __iomem *p = ioremap(0x0FFFD9, 4);
1104 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1110 #ifdef CONFIG_X86_LOCAL_APIC
1111 init_apic_mappings();
1114 set_trap_gate(0,÷_error);
1115 set_intr_gate(1,&debug);
1116 set_intr_gate(2,&nmi);
1117 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1118 set_system_gate(4,&overflow);
1119 set_trap_gate(5,&bounds);
1120 set_trap_gate(6,&invalid_op);
1121 set_trap_gate(7,&device_not_available);
1122 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1123 set_trap_gate(9,&coprocessor_segment_overrun);
1124 set_trap_gate(10,&invalid_TSS);
1125 set_trap_gate(11,&segment_not_present);
1126 set_trap_gate(12,&stack_segment);
1127 set_trap_gate(13,&general_protection);
1128 set_intr_gate(14,&page_fault);
1129 set_trap_gate(15,&spurious_interrupt_bug);
1130 set_trap_gate(16,&coprocessor_error);
1131 set_trap_gate(17,&alignment_check);
1132 #ifdef CONFIG_X86_MCE
1133 set_trap_gate(18,&machine_check);
1135 set_trap_gate(19,&simd_coprocessor_error);
1139 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1140 * Generates a compile-time "error: zero width for bit-field" if
1141 * the alignment is wrong.
1143 struct fxsrAlignAssert {
1144 int _:!(offsetof(struct task_struct,
1145 thread.i387.fxsave) & 15);
1148 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1149 set_in_cr4(X86_CR4_OSFXSR);
1153 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1155 set_in_cr4(X86_CR4_OSXMMEXCPT);
1159 set_system_gate(SYSCALL_VECTOR,&system_call);
1162 * Should be a barrier for any external CPU state.
1169 static int __init kstack_setup(char *s)
1171 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1174 __setup("kstack=", kstack_setup);