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>
31 #include <linux/unwind.h>
34 #include <linux/ioport.h>
35 #include <linux/eisa.h>
39 #include <linux/mca.h>
42 #include <asm/processor.h>
43 #include <asm/system.h>
44 #include <asm/uaccess.h>
46 #include <asm/atomic.h>
47 #include <asm/debugreg.h>
51 #include <asm/unwind.h>
53 #include <asm/arch_hooks.h>
54 #include <asm/kdebug.h>
56 #include <linux/module.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 ATOMIC_NOTIFIER_HEAD(i386die_chain);
98 int register_die_notifier(struct notifier_block *nb)
101 return atomic_notifier_chain_register(&i386die_chain, nb);
103 EXPORT_SYMBOL(register_die_notifier);
105 int unregister_die_notifier(struct notifier_block *nb)
107 return atomic_notifier_chain_unregister(&i386die_chain, nb);
109 EXPORT_SYMBOL(unregister_die_notifier);
111 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
113 return p > (void *)tinfo &&
114 p < (void *)tinfo + THREAD_SIZE - 3;
118 * Print CONFIG_STACK_BACKTRACE_COLS address/symbol entries per line.
120 static inline int print_addr_and_symbol(unsigned long addr, char *log_lvl,
126 #if CONFIG_STACK_BACKTRACE_COLS == 1
127 printk(" [<%08lx>] ", addr);
129 printk(" <%08lx> ", addr);
131 print_symbol("%s", addr);
133 printed = (printed + 1) % CONFIG_STACK_BACKTRACE_COLS;
142 static inline unsigned long print_context_stack(struct thread_info *tinfo,
143 unsigned long *stack, unsigned long ebp,
147 int printed = 0; /* nr of entries already printed on current line */
149 #ifdef CONFIG_FRAME_POINTER
150 while (valid_stack_ptr(tinfo, (void *)ebp)) {
151 addr = *(unsigned long *)(ebp + 4);
152 printed = print_addr_and_symbol(addr, log_lvl, printed);
154 * break out of recursive entries (such as
155 * end_of_stack_stop_unwind_function):
157 if (ebp == *(unsigned long *)ebp)
159 ebp = *(unsigned long *)ebp;
162 while (valid_stack_ptr(tinfo, stack)) {
164 if (__kernel_text_address(addr))
165 printed = print_addr_and_symbol(addr, log_lvl, printed);
174 static asmlinkage void show_trace_unwind(struct unwind_frame_info *info, void *log_lvl)
176 int printed = 0; /* nr of entries already printed on current line */
178 while (unwind(info) == 0 && UNW_PC(info)) {
179 printed = print_addr_and_symbol(UNW_PC(info), log_lvl, printed);
180 if (arch_unw_user_mode(info))
187 static void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
188 unsigned long *stack, char *log_lvl)
191 struct unwind_frame_info info;
197 if (unwind_init_frame_info(&info, task, regs) == 0) {
198 show_trace_unwind(&info, log_lvl);
201 } else if (task == current) {
202 if (unwind_init_running(&info, show_trace_unwind, log_lvl) == 0)
205 if (unwind_init_blocked(&info, task) == 0) {
206 show_trace_unwind(&info, log_lvl);
211 if (task == current) {
212 /* Grab ebp right from our regs */
213 asm ("movl %%ebp, %0" : "=r" (ebp) : );
215 /* ebp is the last reg pushed by switch_to */
216 ebp = *(unsigned long *) task->thread.esp;
220 struct thread_info *context;
221 context = (struct thread_info *)
222 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
223 ebp = print_context_stack(context, stack, ebp, log_lvl);
224 stack = (unsigned long*)context->previous_esp;
227 printk("%s =======================\n", log_lvl);
231 void show_trace(struct task_struct *task, struct pt_regs *regs, unsigned long * stack)
233 show_trace_log_lvl(task, regs, stack, "");
236 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
237 unsigned long *esp, char *log_lvl)
239 unsigned long *stack;
244 esp = (unsigned long*)task->thread.esp;
246 esp = (unsigned long *)&esp;
250 for(i = 0; i < kstack_depth_to_print; i++) {
251 if (kstack_end(stack))
253 if (i && ((i % 8) == 0))
254 printk("\n%s ", log_lvl);
255 printk("%08lx ", *stack++);
257 printk("\n%sCall Trace:\n", log_lvl);
258 show_trace_log_lvl(task, regs, esp, log_lvl);
261 void show_stack(struct task_struct *task, unsigned long *esp)
264 show_stack_log_lvl(task, NULL, esp, "");
268 * The architecture-independent dump_stack generator
270 void dump_stack(void)
274 show_trace(current, NULL, &stack);
277 EXPORT_SYMBOL(dump_stack);
279 void show_registers(struct pt_regs *regs)
286 esp = (unsigned long) (®s->esp);
288 if (user_mode_vm(regs)) {
291 ss = regs->xss & 0xffff;
294 printk(KERN_EMERG "CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\n"
295 "EFLAGS: %08lx (%s %.*s) \n",
296 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
297 print_tainted(), regs->eflags, system_utsname.release,
298 (int)strcspn(system_utsname.version, " "),
299 system_utsname.version);
300 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
301 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
302 regs->eax, regs->ebx, regs->ecx, regs->edx);
303 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
304 regs->esi, regs->edi, regs->ebp, esp);
305 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
306 regs->xds & 0xffff, regs->xes & 0xffff, ss);
307 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
308 TASK_COMM_LEN, current->comm, current->pid,
309 current_thread_info(), current, current->thread_info);
311 * When in-kernel, we also print out the stack and code at the
312 * time of the fault..
317 printk("\n" KERN_EMERG "Stack: ");
318 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
320 printk(KERN_EMERG "Code: ");
322 eip = (u8 __user *)regs->eip - 43;
323 for (i = 0; i < 64; i++, eip++) {
326 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
327 printk(" Bad EIP value.");
330 if (eip == (u8 __user *)regs->eip)
331 printk("<%02x> ", c);
339 static void handle_BUG(struct pt_regs *regs)
349 if (eip < PAGE_OFFSET)
351 if (__get_user(ud2, (unsigned short __user *)eip))
355 if (__get_user(line, (unsigned short __user *)(eip + 2)))
357 if (__get_user(file, (char * __user *)(eip + 4)) ||
358 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
359 file = "<bad filename>";
361 printk(KERN_EMERG "------------[ cut here ]------------\n");
362 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
367 /* Here we know it was a BUG but file-n-line is unavailable */
369 printk(KERN_EMERG "Kernel BUG\n");
372 /* This is gone through when something in the kernel
373 * has done something bad and is about to be terminated.
375 void die(const char * str, struct pt_regs * regs, long err)
380 int lock_owner_depth;
382 .lock = SPIN_LOCK_UNLOCKED,
384 .lock_owner_depth = 0
386 static int die_counter;
391 if (die.lock_owner != raw_smp_processor_id()) {
393 spin_lock_irqsave(&die.lock, flags);
394 die.lock_owner = smp_processor_id();
395 die.lock_owner_depth = 0;
399 local_save_flags(flags);
401 if (++die.lock_owner_depth < 3) {
407 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
408 #ifdef CONFIG_PREEMPT
409 printk(KERN_EMERG "PREEMPT ");
418 #ifdef CONFIG_DEBUG_PAGEALLOC
421 printk("DEBUG_PAGEALLOC");
426 if (notify_die(DIE_OOPS, str, regs, err,
427 current->thread.trap_no, SIGSEGV) !=
429 show_registers(regs);
430 /* Executive summary in case the oops scrolled away */
431 esp = (unsigned long) (®s->esp);
433 if (user_mode(regs)) {
435 ss = regs->xss & 0xffff;
437 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
438 print_symbol("%s", regs->eip);
439 printk(" SS:ESP %04x:%08lx\n", ss, esp);
444 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
448 spin_unlock_irqrestore(&die.lock, flags);
453 if (kexec_should_crash(current))
457 panic("Fatal exception in interrupt");
460 printk(KERN_EMERG "Fatal exception: panic in 5 seconds\n");
462 panic("Fatal exception");
468 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
470 if (!user_mode_vm(regs))
474 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
475 struct pt_regs * regs, long error_code,
478 struct task_struct *tsk = current;
479 tsk->thread.error_code = error_code;
480 tsk->thread.trap_no = trapnr;
482 if (regs->eflags & VM_MASK) {
488 if (!user_mode(regs))
493 force_sig_info(signr, info, tsk);
495 force_sig(signr, tsk);
500 if (!fixup_exception(regs))
501 die(str, regs, error_code);
506 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
507 if (ret) goto trap_signal;
512 #define DO_ERROR(trapnr, signr, str, name) \
513 fastcall void do_##name(struct pt_regs * regs, long error_code) \
515 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
518 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
521 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
522 fastcall void do_##name(struct pt_regs * regs, long error_code) \
525 info.si_signo = signr; \
527 info.si_code = sicode; \
528 info.si_addr = (void __user *)siaddr; \
529 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
532 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
535 #define DO_VM86_ERROR(trapnr, signr, str, name) \
536 fastcall void do_##name(struct pt_regs * regs, long error_code) \
538 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
541 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
544 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
545 fastcall void do_##name(struct pt_regs * regs, long error_code) \
548 info.si_signo = signr; \
550 info.si_code = sicode; \
551 info.si_addr = (void __user *)siaddr; \
552 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
555 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
558 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
559 #ifndef CONFIG_KPROBES
560 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
562 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
563 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
564 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
565 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
566 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
567 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
568 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
569 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
570 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
572 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
576 struct tss_struct *tss = &per_cpu(init_tss, cpu);
577 struct thread_struct *thread = ¤t->thread;
580 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
581 * invalid offset set (the LAZY one) and the faulting thread has
582 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
583 * and we set the offset field correctly. Then we let the CPU to
584 * restart the faulting instruction.
586 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
587 thread->io_bitmap_ptr) {
588 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
589 thread->io_bitmap_max);
591 * If the previously set map was extending to higher ports
592 * than the current one, pad extra space with 0xff (no access).
594 if (thread->io_bitmap_max < tss->io_bitmap_max)
595 memset((char *) tss->io_bitmap +
596 thread->io_bitmap_max, 0xff,
597 tss->io_bitmap_max - thread->io_bitmap_max);
598 tss->io_bitmap_max = thread->io_bitmap_max;
599 tss->io_bitmap_base = IO_BITMAP_OFFSET;
600 tss->io_bitmap_owner = thread;
606 current->thread.error_code = error_code;
607 current->thread.trap_no = 13;
609 if (regs->eflags & VM_MASK)
612 if (!user_mode(regs))
615 current->thread.error_code = error_code;
616 current->thread.trap_no = 13;
617 force_sig(SIGSEGV, current);
622 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
626 if (!fixup_exception(regs)) {
627 if (notify_die(DIE_GPF, "general protection fault", regs,
628 error_code, 13, SIGSEGV) == NOTIFY_STOP)
630 die("general protection fault", regs, error_code);
634 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
636 printk(KERN_EMERG "Uhhuh. NMI received. Dazed and confused, but trying "
638 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
641 /* Clear and disable the memory parity error line. */
642 clear_mem_error(reason);
645 static void io_check_error(unsigned char reason, struct pt_regs * regs)
649 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
650 show_registers(regs);
652 /* Re-enable the IOCK line, wait for a few seconds */
653 reason = (reason & 0xf) | 8;
656 while (--i) udelay(1000);
661 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
664 /* Might actually be able to figure out what the guilty party
671 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
672 reason, smp_processor_id());
673 printk("Dazed and confused, but trying to continue\n");
674 printk("Do you have a strange power saving mode enabled?\n");
677 static DEFINE_SPINLOCK(nmi_print_lock);
679 void die_nmi (struct pt_regs *regs, const char *msg)
681 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
685 spin_lock(&nmi_print_lock);
687 * We are in trouble anyway, lets at least try
688 * to get a message out.
691 printk(KERN_EMERG "%s", msg);
692 printk(" on CPU%d, eip %08lx, registers:\n",
693 smp_processor_id(), regs->eip);
694 show_registers(regs);
695 printk(KERN_EMERG "console shuts up ...\n");
697 spin_unlock(&nmi_print_lock);
700 /* If we are in kernel we are probably nested up pretty bad
701 * and might aswell get out now while we still can.
703 if (!user_mode_vm(regs)) {
704 current->thread.trap_no = 2;
711 static void default_do_nmi(struct pt_regs * regs)
713 unsigned char reason = 0;
715 /* Only the BSP gets external NMIs from the system. */
716 if (!smp_processor_id())
717 reason = get_nmi_reason();
719 if (!(reason & 0xc0)) {
720 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
723 #ifdef CONFIG_X86_LOCAL_APIC
725 * Ok, so this is none of the documented NMI sources,
726 * so it must be the NMI watchdog.
729 nmi_watchdog_tick(regs);
733 unknown_nmi_error(reason, regs);
736 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
739 mem_parity_error(reason, regs);
741 io_check_error(reason, regs);
743 * Reassert NMI in case it became active meanwhile
744 * as it's edge-triggered.
749 static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
754 static nmi_callback_t nmi_callback = dummy_nmi_callback;
756 fastcall void do_nmi(struct pt_regs * regs, long error_code)
762 cpu = smp_processor_id();
766 if (!rcu_dereference(nmi_callback)(regs, cpu))
767 default_do_nmi(regs);
772 void set_nmi_callback(nmi_callback_t callback)
775 rcu_assign_pointer(nmi_callback, callback);
777 EXPORT_SYMBOL_GPL(set_nmi_callback);
779 void unset_nmi_callback(void)
781 nmi_callback = dummy_nmi_callback;
783 EXPORT_SYMBOL_GPL(unset_nmi_callback);
785 #ifdef CONFIG_KPROBES
786 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
788 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
791 /* This is an interrupt gate, because kprobes wants interrupts
792 disabled. Normal trap handlers don't. */
793 restore_interrupts(regs);
794 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
799 * Our handling of the processor debug registers is non-trivial.
800 * We do not clear them on entry and exit from the kernel. Therefore
801 * it is possible to get a watchpoint trap here from inside the kernel.
802 * However, the code in ./ptrace.c has ensured that the user can
803 * only set watchpoints on userspace addresses. Therefore the in-kernel
804 * watchpoint trap can only occur in code which is reading/writing
805 * from user space. Such code must not hold kernel locks (since it
806 * can equally take a page fault), therefore it is safe to call
807 * force_sig_info even though that claims and releases locks.
809 * Code in ./signal.c ensures that the debug control register
810 * is restored before we deliver any signal, and therefore that
811 * user code runs with the correct debug control register even though
814 * Being careful here means that we don't have to be as careful in a
815 * lot of more complicated places (task switching can be a bit lazy
816 * about restoring all the debug state, and ptrace doesn't have to
817 * find every occurrence of the TF bit that could be saved away even
820 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
822 unsigned int condition;
823 struct task_struct *tsk = current;
825 get_debugreg(condition, 6);
827 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
828 SIGTRAP) == NOTIFY_STOP)
830 /* It's safe to allow irq's after DR6 has been saved */
831 if (regs->eflags & X86_EFLAGS_IF)
834 /* Mask out spurious debug traps due to lazy DR7 setting */
835 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
836 if (!tsk->thread.debugreg[7])
840 if (regs->eflags & VM_MASK)
843 /* Save debug status register where ptrace can see it */
844 tsk->thread.debugreg[6] = condition;
847 * Single-stepping through TF: make sure we ignore any events in
848 * kernel space (but re-enable TF when returning to user mode).
850 if (condition & DR_STEP) {
852 * We already checked v86 mode above, so we can
853 * check for kernel mode by just checking the CPL
856 if (!user_mode(regs))
857 goto clear_TF_reenable;
860 /* Ok, finally something we can handle */
861 send_sigtrap(tsk, regs, error_code);
863 /* Disable additional traps. They'll be re-enabled when
864 * the signal is delivered.
871 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
875 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
876 regs->eflags &= ~TF_MASK;
881 * Note that we play around with the 'TS' bit in an attempt to get
882 * the correct behaviour even in the presence of the asynchronous
885 void math_error(void __user *eip)
887 struct task_struct * task;
889 unsigned short cwd, swd;
892 * Save the info for the exception handler and clear the error.
896 task->thread.trap_no = 16;
897 task->thread.error_code = 0;
898 info.si_signo = SIGFPE;
900 info.si_code = __SI_FAULT;
903 * (~cwd & swd) will mask out exceptions that are not set to unmasked
904 * status. 0x3f is the exception bits in these regs, 0x200 is the
905 * C1 reg you need in case of a stack fault, 0x040 is the stack
906 * fault bit. We should only be taking one exception at a time,
907 * so if this combination doesn't produce any single exception,
908 * then we have a bad program that isn't syncronizing its FPU usage
909 * and it will suffer the consequences since we won't be able to
910 * fully reproduce the context of the exception
912 cwd = get_fpu_cwd(task);
913 swd = get_fpu_swd(task);
914 switch (swd & ~cwd & 0x3f) {
915 case 0x000: /* No unmasked exception */
917 default: /* Multiple exceptions */
919 case 0x001: /* Invalid Op */
921 * swd & 0x240 == 0x040: Stack Underflow
922 * swd & 0x240 == 0x240: Stack Overflow
923 * User must clear the SF bit (0x40) if set
925 info.si_code = FPE_FLTINV;
927 case 0x002: /* Denormalize */
928 case 0x010: /* Underflow */
929 info.si_code = FPE_FLTUND;
931 case 0x004: /* Zero Divide */
932 info.si_code = FPE_FLTDIV;
934 case 0x008: /* Overflow */
935 info.si_code = FPE_FLTOVF;
937 case 0x020: /* Precision */
938 info.si_code = FPE_FLTRES;
941 force_sig_info(SIGFPE, &info, task);
944 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
947 math_error((void __user *)regs->eip);
950 static void simd_math_error(void __user *eip)
952 struct task_struct * task;
954 unsigned short mxcsr;
957 * Save the info for the exception handler and clear the error.
961 task->thread.trap_no = 19;
962 task->thread.error_code = 0;
963 info.si_signo = SIGFPE;
965 info.si_code = __SI_FAULT;
968 * The SIMD FPU exceptions are handled a little differently, as there
969 * is only a single status/control register. Thus, to determine which
970 * unmasked exception was caught we must mask the exception mask bits
971 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
973 mxcsr = get_fpu_mxcsr(task);
974 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
978 case 0x001: /* Invalid Op */
979 info.si_code = FPE_FLTINV;
981 case 0x002: /* Denormalize */
982 case 0x010: /* Underflow */
983 info.si_code = FPE_FLTUND;
985 case 0x004: /* Zero Divide */
986 info.si_code = FPE_FLTDIV;
988 case 0x008: /* Overflow */
989 info.si_code = FPE_FLTOVF;
991 case 0x020: /* Precision */
992 info.si_code = FPE_FLTRES;
995 force_sig_info(SIGFPE, &info, task);
998 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1002 /* Handle SIMD FPU exceptions on PIII+ processors. */
1004 simd_math_error((void __user *)regs->eip);
1007 * Handle strange cache flush from user space exception
1008 * in all other cases. This is undocumented behaviour.
1010 if (regs->eflags & VM_MASK) {
1011 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1015 current->thread.trap_no = 19;
1016 current->thread.error_code = error_code;
1017 die_if_kernel("cache flush denied", regs, error_code);
1018 force_sig(SIGSEGV, current);
1022 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1026 /* No need to warn about this any longer. */
1027 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1031 fastcall void setup_x86_bogus_stack(unsigned char * stk)
1033 unsigned long *switch16_ptr, *switch32_ptr;
1034 struct pt_regs *regs;
1035 unsigned long stack_top, stack_bot;
1036 unsigned short iret_frame16_off;
1037 int cpu = smp_processor_id();
1038 /* reserve the space on 32bit stack for the magic switch16 pointer */
1039 memmove(stk, stk + 8, sizeof(struct pt_regs));
1040 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
1041 regs = (struct pt_regs *)stk;
1042 /* now the switch32 on 16bit stack */
1043 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1044 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1045 switch32_ptr = (unsigned long *)(stack_top - 8);
1046 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
1047 /* copy iret frame on 16bit stack */
1048 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
1049 /* fill in the switch pointers */
1050 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
1051 switch16_ptr[1] = __ESPFIX_SS;
1052 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
1053 8 - CPU_16BIT_STACK_SIZE;
1054 switch32_ptr[1] = __KERNEL_DS;
1057 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
1059 unsigned long *switch32_ptr;
1060 unsigned char *stack16, *stack32;
1061 unsigned long stack_top, stack_bot;
1063 int cpu = smp_processor_id();
1064 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1065 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1066 switch32_ptr = (unsigned long *)(stack_top - 8);
1067 /* copy the data from 16bit stack to 32bit stack */
1068 len = CPU_16BIT_STACK_SIZE - 8 - sp;
1069 stack16 = (unsigned char *)(stack_bot + sp);
1070 stack32 = (unsigned char *)
1071 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
1072 memcpy(stack32, stack16, len);
1077 * 'math_state_restore()' saves the current math information in the
1078 * old math state array, and gets the new ones from the current task
1080 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1081 * Don't touch unless you *really* know how it works.
1083 * Must be called with kernel preemption disabled (in this case,
1084 * local interrupts are disabled at the call-site in entry.S).
1086 asmlinkage void math_state_restore(struct pt_regs regs)
1088 struct thread_info *thread = current_thread_info();
1089 struct task_struct *tsk = thread->task;
1091 clts(); /* Allow maths ops (or we recurse) */
1092 if (!tsk_used_math(tsk))
1095 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1098 #ifndef CONFIG_MATH_EMULATION
1100 asmlinkage void math_emulate(long arg)
1102 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1103 printk(KERN_EMERG "killing %s.\n",current->comm);
1104 force_sig(SIGFPE,current);
1108 #endif /* CONFIG_MATH_EMULATION */
1110 #ifdef CONFIG_X86_F00F_BUG
1111 void __init trap_init_f00f_bug(void)
1113 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1116 * Update the IDT descriptor and reload the IDT so that
1117 * it uses the read-only mapped virtual address.
1119 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1120 load_idt(&idt_descr);
1124 #define _set_gate(gate_addr,type,dpl,addr,seg) \
1127 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1128 "movw %4,%%dx\n\t" \
1129 "movl %%eax,%0\n\t" \
1131 :"=m" (*((long *) (gate_addr))), \
1132 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1133 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1134 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1139 * This needs to use 'idt_table' rather than 'idt', and
1140 * thus use the _nonmapped_ version of the IDT, as the
1141 * Pentium F0 0F bugfix can have resulted in the mapped
1142 * IDT being write-protected.
1144 void set_intr_gate(unsigned int n, void *addr)
1146 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1150 * This routine sets up an interrupt gate at directory privilege level 3.
1152 static inline void set_system_intr_gate(unsigned int n, void *addr)
1154 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1157 static void __init set_trap_gate(unsigned int n, void *addr)
1159 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1162 static void __init set_system_gate(unsigned int n, void *addr)
1164 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1167 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1169 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1173 void __init trap_init(void)
1176 void __iomem *p = ioremap(0x0FFFD9, 4);
1177 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1183 #ifdef CONFIG_X86_LOCAL_APIC
1184 init_apic_mappings();
1187 set_trap_gate(0,÷_error);
1188 set_intr_gate(1,&debug);
1189 set_intr_gate(2,&nmi);
1190 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1191 set_system_gate(4,&overflow);
1192 set_trap_gate(5,&bounds);
1193 set_trap_gate(6,&invalid_op);
1194 set_trap_gate(7,&device_not_available);
1195 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1196 set_trap_gate(9,&coprocessor_segment_overrun);
1197 set_trap_gate(10,&invalid_TSS);
1198 set_trap_gate(11,&segment_not_present);
1199 set_trap_gate(12,&stack_segment);
1200 set_trap_gate(13,&general_protection);
1201 set_intr_gate(14,&page_fault);
1202 set_trap_gate(15,&spurious_interrupt_bug);
1203 set_trap_gate(16,&coprocessor_error);
1204 set_trap_gate(17,&alignment_check);
1205 #ifdef CONFIG_X86_MCE
1206 set_trap_gate(18,&machine_check);
1208 set_trap_gate(19,&simd_coprocessor_error);
1212 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1213 * Generates a compile-time "error: zero width for bit-field" if
1214 * the alignment is wrong.
1216 struct fxsrAlignAssert {
1217 int _:!(offsetof(struct task_struct,
1218 thread.i387.fxsave) & 15);
1221 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1222 set_in_cr4(X86_CR4_OSFXSR);
1226 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1228 set_in_cr4(X86_CR4_OSXMMEXCPT);
1232 set_system_gate(SYSCALL_VECTOR,&system_call);
1235 * Should be a barrier for any external CPU state.
1242 static int __init kstack_setup(char *s)
1244 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1247 __setup("kstack=", kstack_setup);