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 ATOMIC_NOTIFIER_HEAD(i386die_chain);
98 extern char last_sysfs_file[];
100 int register_die_notifier(struct notifier_block *nb)
103 return atomic_notifier_chain_register(&i386die_chain, nb);
105 EXPORT_SYMBOL(register_die_notifier);
107 int unregister_die_notifier(struct notifier_block *nb)
109 return atomic_notifier_chain_unregister(&i386die_chain, nb);
111 EXPORT_SYMBOL(unregister_die_notifier);
113 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
115 return p > (void *)tinfo &&
116 p < (void *)tinfo + THREAD_SIZE - 3;
120 * Print CONFIG_STACK_BACKTRACE_COLS address/symbol entries per line.
122 static inline int print_addr_and_symbol(unsigned long addr, char *log_lvl,
128 #if CONFIG_STACK_BACKTRACE_COLS == 1
129 printk(" [<%08lx>] ", addr);
131 printk(" <%08lx> ", addr);
133 print_symbol("%s", addr);
135 printed = (printed + 1) % CONFIG_STACK_BACKTRACE_COLS;
144 static inline unsigned long print_context_stack(struct thread_info *tinfo,
145 unsigned long *stack, unsigned long ebp,
149 int printed = 0; /* nr of entries already printed on current line */
151 #ifdef CONFIG_FRAME_POINTER
152 while (valid_stack_ptr(tinfo, (void *)ebp)) {
153 addr = *(unsigned long *)(ebp + 4);
154 printed = print_addr_and_symbol(addr, log_lvl, printed);
155 ebp = *(unsigned long *)ebp;
158 while (valid_stack_ptr(tinfo, stack)) {
160 if (__kernel_text_address(addr))
161 printed = print_addr_and_symbol(addr, log_lvl, printed);
170 static void show_trace_log_lvl(struct task_struct *task,
171 unsigned long *stack, char *log_lvl)
178 if (task == current) {
179 /* Grab ebp right from our regs */
180 asm ("movl %%ebp, %0" : "=r" (ebp) : );
182 /* ebp is the last reg pushed by switch_to */
183 ebp = *(unsigned long *) task->thread.esp;
187 struct thread_info *context;
188 context = (struct thread_info *)
189 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
190 ebp = print_context_stack(context, stack, ebp, log_lvl);
191 stack = (unsigned long*)context->previous_esp;
194 printk("%s =======================\n", log_lvl);
198 void show_trace(struct task_struct *task, unsigned long * stack)
200 show_trace_log_lvl(task, stack, "");
203 static void show_stack_log_lvl(struct task_struct *task, unsigned long *esp,
206 unsigned long *stack;
211 esp = (unsigned long*)task->thread.esp;
213 esp = (unsigned long *)&esp;
217 for(i = 0; i < kstack_depth_to_print; i++) {
218 if (kstack_end(stack))
220 if (i && ((i % 8) == 0))
221 printk("\n%s ", log_lvl);
222 printk("%08lx ", *stack++);
224 printk("\n%sCall Trace:\n", log_lvl);
225 show_trace_log_lvl(task, esp, log_lvl);
228 void show_stack(struct task_struct *task, unsigned long *esp)
231 show_stack_log_lvl(task, esp, "");
235 * The architecture-independent dump_stack generator
237 void dump_stack(void)
241 show_trace(current, &stack);
244 EXPORT_SYMBOL(dump_stack);
246 void show_registers(struct pt_regs *regs)
253 esp = (unsigned long) (®s->esp);
255 if (user_mode_vm(regs)) {
258 ss = regs->xss & 0xffff;
261 printk(KERN_EMERG "CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\n"
262 "EFLAGS: %08lx (%s %.*s) \n",
263 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
264 print_tainted(), regs->eflags, system_utsname.release,
265 (int)strcspn(system_utsname.version, " "),
266 system_utsname.version);
267 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
268 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
269 regs->eax, regs->ebx, regs->ecx, regs->edx);
270 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
271 regs->esi, regs->edi, regs->ebp, esp);
272 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
273 regs->xds & 0xffff, regs->xes & 0xffff, ss);
274 printk(KERN_EMERG "Process %s (pid: %d[#%u], threadinfo=%p task=%p)",
275 current->comm, current->pid, current->xid,
276 current_thread_info(), current);
278 * When in-kernel, we also print out the stack and code at the
279 * time of the fault..
284 printk("\n" KERN_EMERG "Stack: ");
285 show_stack_log_lvl(NULL, (unsigned long *)esp, KERN_EMERG);
287 printk(KERN_EMERG "Code: ");
289 eip = (u8 __user *)regs->eip - 43;
290 for (i = 0; i < 64; i++, eip++) {
293 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
294 printk(" Bad EIP value.");
297 if (eip == (u8 __user *)regs->eip)
298 printk("<%02x> ", c);
306 static void handle_BUG(struct pt_regs *regs)
316 if (eip < PAGE_OFFSET)
318 if (__get_user(ud2, (unsigned short __user *)eip))
322 if (__get_user(line, (unsigned short __user *)(eip + 2)))
324 if (__get_user(file, (char * __user *)(eip + 4)) ||
325 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
326 file = "<bad filename>";
328 printk(KERN_EMERG "------------[ cut here ]------------\n");
329 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
334 /* Here we know it was a BUG but file-n-line is unavailable */
336 printk(KERN_EMERG "Kernel BUG\n");
339 /* This is gone through when something in the kernel
340 * has done something bad and is about to be terminated.
342 void die(const char * str, struct pt_regs * regs, long err)
347 int lock_owner_depth;
349 .lock = SPIN_LOCK_UNLOCKED,
351 .lock_owner_depth = 0
353 static int die_counter;
360 if (die.lock_owner != raw_smp_processor_id()) {
362 spin_lock_irqsave(&die.lock, flags);
363 die.lock_owner = smp_processor_id();
364 die.lock_owner_depth = 0;
368 local_save_flags(flags);
370 if (++die.lock_owner_depth < 3) {
376 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
377 #ifdef CONFIG_PREEMPT
378 printk(KERN_EMERG "PREEMPT ");
387 #ifdef CONFIG_DEBUG_PAGEALLOC
390 printk("DEBUG_PAGEALLOC");
396 printk(KERN_ALERT "last sysfs file: %s\n", last_sysfs_file);
398 if (notify_die(DIE_OOPS, str, regs, err,
399 current->thread.trap_no, SIGSEGV) != NOTIFY_STOP) {
400 show_registers(regs);
402 /* Executive summary in case the oops scrolled away */
403 esp = (unsigned long) (®s->esp);
405 if (user_mode(regs)) {
407 ss = regs->xss & 0xffff;
409 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
410 print_symbol("%s", regs->eip);
411 printk(" SS:ESP %04x:%08lx\n", ss, esp);
416 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
420 spin_unlock_irqrestore(&die.lock, flags);
425 if (kexec_should_crash(current))
429 panic("Fatal exception in interrupt");
432 printk(KERN_EMERG "Fatal exception: panic in 5 seconds\n");
434 panic("Fatal exception");
440 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
442 if (!user_mode_vm(regs))
446 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
447 struct pt_regs * regs, long error_code,
450 struct task_struct *tsk = current;
451 tsk->thread.error_code = error_code;
452 tsk->thread.trap_no = trapnr;
454 if (regs->eflags & VM_MASK) {
460 if (!user_mode(regs))
465 force_sig_info(signr, info, tsk);
467 force_sig(signr, tsk);
472 if (!fixup_exception(regs))
473 die(str, regs, error_code);
478 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
479 if (ret) goto trap_signal;
484 #define DO_ERROR(trapnr, signr, str, name) \
485 fastcall void do_##name(struct pt_regs * regs, long error_code) \
487 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
490 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
493 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
494 fastcall void do_##name(struct pt_regs * regs, long error_code) \
497 info.si_signo = signr; \
499 info.si_code = sicode; \
500 info.si_addr = (void __user *)siaddr; \
501 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
504 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
507 #define DO_VM86_ERROR(trapnr, signr, str, name) \
508 fastcall void do_##name(struct pt_regs * regs, long error_code) \
510 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
513 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
516 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
517 fastcall void do_##name(struct pt_regs * regs, long error_code) \
520 info.si_signo = signr; \
522 info.si_code = sicode; \
523 info.si_addr = (void __user *)siaddr; \
524 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
527 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
530 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
531 #ifndef CONFIG_KPROBES
532 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
534 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
535 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
536 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
537 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
538 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
539 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
540 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
541 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
545 * lazy-check for CS validity on exec-shield binaries:
547 * the original non-exec stack patch was written by
548 * Solar Designer <solar at openwall.com>. Thanks!
551 check_lazy_exec_limit(int cpu, struct pt_regs *regs, long error_code)
553 struct desc_struct *desc1, *desc2;
554 struct vm_area_struct *vma;
557 if (current->mm == NULL)
561 if (current->mm->context.exec_limit != -1UL) {
563 spin_lock(¤t->mm->page_table_lock);
564 for (vma = current->mm->mmap; vma; vma = vma->vm_next)
565 if ((vma->vm_flags & VM_EXEC) && (vma->vm_end > limit))
567 spin_unlock(¤t->mm->page_table_lock);
568 if (limit >= TASK_SIZE)
570 current->mm->context.exec_limit = limit;
572 set_user_cs(¤t->mm->context.user_cs, limit);
574 desc1 = ¤t->mm->context.user_cs;
575 desc2 = get_cpu_gdt_table(cpu) + GDT_ENTRY_DEFAULT_USER_CS;
577 if (desc1->a != desc2->a || desc1->b != desc2->b) {
579 * The CS was not in sync - reload it and retry the
580 * instruction. If the instruction still faults then
581 * we won't hit this branch next time around.
583 if (print_fatal_signals >= 2) {
584 printk("#GPF fixup (%ld[seg:%lx]) at %08lx, CPU#%d.\n", error_code, error_code/8, regs->eip, smp_processor_id());
585 printk(" exec_limit: %08lx, user_cs: %08lx/%08lx, CPU_cs: %08lx/%08lx.\n", current->mm->context.exec_limit, desc1->a, desc1->b, desc2->a, desc2->b);
587 load_user_cs_desc(cpu, current->mm);
595 * The fixup code for errors in iret jumps to here (iret_exc). It loses
596 * the original trap number and error code. The bogus trap 32 and error
597 * code 0 are what the vanilla kernel delivers via:
598 * DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
600 * In case of a general protection fault in the iret instruction, we
601 * need to check for a lazy CS update for exec-shield.
603 fastcall void do_iret_error(struct pt_regs *regs, long error_code)
605 int ok = check_lazy_exec_limit(get_cpu(), regs, error_code);
607 if (!ok && notify_die(DIE_TRAP, "iret exception", regs,
608 error_code, 32, SIGSEGV) != NOTIFY_STOP) {
610 info.si_signo = SIGSEGV;
612 info.si_code = ILL_BADSTK;
614 do_trap(32, SIGSEGV, "iret exception", 0, regs, error_code,
619 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
623 struct tss_struct *tss = &per_cpu(init_tss, cpu);
624 struct thread_struct *thread = ¤t->thread;
628 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
629 * invalid offset set (the LAZY one) and the faulting thread has
630 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
631 * and we set the offset field correctly. Then we let the CPU to
632 * restart the faulting instruction.
634 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
635 thread->io_bitmap_ptr) {
636 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
637 thread->io_bitmap_max);
639 * If the previously set map was extending to higher ports
640 * than the current one, pad extra space with 0xff (no access).
642 if (thread->io_bitmap_max < tss->io_bitmap_max)
643 memset((char *) tss->io_bitmap +
644 thread->io_bitmap_max, 0xff,
645 tss->io_bitmap_max - thread->io_bitmap_max);
646 tss->io_bitmap_max = thread->io_bitmap_max;
647 tss->io_bitmap_base = IO_BITMAP_OFFSET;
648 tss->io_bitmap_owner = thread;
653 current->thread.error_code = error_code;
654 current->thread.trap_no = 13;
656 if (regs->eflags & VM_MASK)
659 if (!user_mode(regs))
662 ok = check_lazy_exec_limit(cpu, regs, error_code);
669 if (print_fatal_signals) {
670 printk("#GPF(%ld[seg:%lx]) at %08lx, CPU#%d.\n", error_code, error_code/8, regs->eip, smp_processor_id());
671 printk(" exec_limit: %08lx, user_cs: %08lx/%08lx.\n", current->mm->context.exec_limit, current->mm->context.user_cs.a, current->mm->context.user_cs.b);
674 current->thread.error_code = error_code;
675 current->thread.trap_no = 13;
676 force_sig(SIGSEGV, current);
682 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
687 if (!fixup_exception(regs)) {
688 if (notify_die(DIE_GPF, "general protection fault", regs,
689 error_code, 13, SIGSEGV) == NOTIFY_STOP)
691 die("general protection fault", regs, error_code);
695 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
697 printk(KERN_EMERG "Uhhuh. NMI received. Dazed and confused, but trying "
699 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
702 /* Clear and disable the memory parity error line. */
703 clear_mem_error(reason);
706 static void io_check_error(unsigned char reason, struct pt_regs * regs)
708 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
709 show_registers(regs);
711 /* Re-enable the IOCK line, wait for a few seconds */
712 clear_io_check_error(reason);
715 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
718 /* Might actually be able to figure out what the guilty party
725 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
726 reason, smp_processor_id());
727 printk("Dazed and confused, but trying to continue\n");
728 printk("Do you have a strange power saving mode enabled?\n");
731 static DEFINE_SPINLOCK(nmi_print_lock);
733 void die_nmi (struct pt_regs *regs, const char *msg)
735 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
739 spin_lock(&nmi_print_lock);
741 * We are in trouble anyway, lets at least try
742 * to get a message out.
745 printk(KERN_EMERG "%s", msg);
746 printk(" on CPU%d, eip %08lx, registers:\n",
747 smp_processor_id(), regs->eip);
748 show_registers(regs);
749 printk(KERN_EMERG "console shuts up ...\n");
751 spin_unlock(&nmi_print_lock);
754 /* If we are in kernel we are probably nested up pretty bad
755 * and might aswell get out now while we still can.
757 if (!user_mode_vm(regs)) {
758 current->thread.trap_no = 2;
765 static void default_do_nmi(struct pt_regs * regs)
767 unsigned char reason = 0;
769 /* Only the BSP gets external NMIs from the system. */
770 if (!smp_processor_id())
771 reason = get_nmi_reason();
773 if (!(reason & 0xc0)) {
774 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
777 #ifdef CONFIG_X86_LOCAL_APIC
779 * Ok, so this is none of the documented NMI sources,
780 * so it must be the NMI watchdog.
783 nmi_watchdog_tick(regs);
787 unknown_nmi_error(reason, regs);
790 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
793 mem_parity_error(reason, regs);
795 io_check_error(reason, regs);
797 * Reassert NMI in case it became active meanwhile
798 * as it's edge-triggered.
803 static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
808 static nmi_callback_t nmi_callback = dummy_nmi_callback;
810 fastcall void do_nmi(struct pt_regs * regs, long error_code)
816 cpu = smp_processor_id();
820 if (!rcu_dereference(nmi_callback)(regs, cpu))
821 default_do_nmi(regs);
826 void set_nmi_callback(nmi_callback_t callback)
829 rcu_assign_pointer(nmi_callback, callback);
831 EXPORT_SYMBOL_GPL(set_nmi_callback);
833 void unset_nmi_callback(void)
835 nmi_callback = dummy_nmi_callback;
837 EXPORT_SYMBOL_GPL(unset_nmi_callback);
839 #ifdef CONFIG_KPROBES
840 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
842 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
845 /* This is an interrupt gate, because kprobes wants interrupts
846 disabled. Normal trap handlers don't. */
847 restore_interrupts(regs);
848 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
853 * Our handling of the processor debug registers is non-trivial.
854 * We do not clear them on entry and exit from the kernel. Therefore
855 * it is possible to get a watchpoint trap here from inside the kernel.
856 * However, the code in ./ptrace.c has ensured that the user can
857 * only set watchpoints on userspace addresses. Therefore the in-kernel
858 * watchpoint trap can only occur in code which is reading/writing
859 * from user space. Such code must not hold kernel locks (since it
860 * can equally take a page fault), therefore it is safe to call
861 * force_sig_info even though that claims and releases locks.
863 * Code in ./signal.c ensures that the debug control register
864 * is restored before we deliver any signal, and therefore that
865 * user code runs with the correct debug control register even though
868 * Being careful here means that we don't have to be as careful in a
869 * lot of more complicated places (task switching can be a bit lazy
870 * about restoring all the debug state, and ptrace doesn't have to
871 * find every occurrence of the TF bit that could be saved away even
874 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
876 unsigned int condition;
877 struct task_struct *tsk = current;
879 get_debugreg(condition, 6);
881 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
882 SIGTRAP) == NOTIFY_STOP)
884 /* It's safe to allow irq's after DR6 has been saved */
885 if (regs->eflags & X86_EFLAGS_IF)
888 /* Mask out spurious debug traps due to lazy DR7 setting */
889 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
890 if (!tsk->thread.debugreg[7])
894 if (regs->eflags & VM_MASK)
897 /* Save debug status register where ptrace can see it */
898 tsk->thread.debugreg[6] = condition;
901 * Single-stepping through TF: make sure we ignore any events in
902 * kernel space (but re-enable TF when returning to user mode).
904 if (condition & DR_STEP) {
906 * We already checked v86 mode above, so we can
907 * check for kernel mode by just checking the CPL
910 if (!user_mode(regs))
911 goto clear_TF_reenable;
914 /* Ok, finally something we can handle */
915 send_sigtrap(tsk, regs, error_code);
917 /* Disable additional traps. They'll be re-enabled when
918 * the signal is delivered.
925 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
929 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
930 regs->eflags &= ~TF_MASK;
935 * Note that we play around with the 'TS' bit in an attempt to get
936 * the correct behaviour even in the presence of the asynchronous
939 void math_error(void __user *eip)
941 struct task_struct * task;
943 unsigned short cwd, swd;
946 * Save the info for the exception handler and clear the error.
950 task->thread.trap_no = 16;
951 task->thread.error_code = 0;
952 info.si_signo = SIGFPE;
954 info.si_code = __SI_FAULT;
957 * (~cwd & swd) will mask out exceptions that are not set to unmasked
958 * status. 0x3f is the exception bits in these regs, 0x200 is the
959 * C1 reg you need in case of a stack fault, 0x040 is the stack
960 * fault bit. We should only be taking one exception at a time,
961 * so if this combination doesn't produce any single exception,
962 * then we have a bad program that isn't syncronizing its FPU usage
963 * and it will suffer the consequences since we won't be able to
964 * fully reproduce the context of the exception
966 cwd = get_fpu_cwd(task);
967 swd = get_fpu_swd(task);
968 switch (swd & ~cwd & 0x3f) {
969 case 0x000: /* No unmasked exception */
971 default: /* Multiple exceptions */
973 case 0x001: /* Invalid Op */
975 * swd & 0x240 == 0x040: Stack Underflow
976 * swd & 0x240 == 0x240: Stack Overflow
977 * User must clear the SF bit (0x40) if set
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_coprocessor_error(struct pt_regs * regs, long error_code)
1001 math_error((void __user *)regs->eip);
1004 static void simd_math_error(void __user *eip)
1006 struct task_struct * task;
1008 unsigned short mxcsr;
1011 * Save the info for the exception handler and clear the error.
1014 save_init_fpu(task);
1015 task->thread.trap_no = 19;
1016 task->thread.error_code = 0;
1017 info.si_signo = SIGFPE;
1019 info.si_code = __SI_FAULT;
1022 * The SIMD FPU exceptions are handled a little differently, as there
1023 * is only a single status/control register. Thus, to determine which
1024 * unmasked exception was caught we must mask the exception mask bits
1025 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1027 mxcsr = get_fpu_mxcsr(task);
1028 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1032 case 0x001: /* Invalid Op */
1033 info.si_code = FPE_FLTINV;
1035 case 0x002: /* Denormalize */
1036 case 0x010: /* Underflow */
1037 info.si_code = FPE_FLTUND;
1039 case 0x004: /* Zero Divide */
1040 info.si_code = FPE_FLTDIV;
1042 case 0x008: /* Overflow */
1043 info.si_code = FPE_FLTOVF;
1045 case 0x020: /* Precision */
1046 info.si_code = FPE_FLTRES;
1049 force_sig_info(SIGFPE, &info, task);
1052 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1056 /* Handle SIMD FPU exceptions on PIII+ processors. */
1058 simd_math_error((void __user *)regs->eip);
1061 * Handle strange cache flush from user space exception
1062 * in all other cases. This is undocumented behaviour.
1064 if (regs->eflags & VM_MASK) {
1065 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1069 current->thread.trap_no = 19;
1070 current->thread.error_code = error_code;
1071 die_if_kernel("cache flush denied", regs, error_code);
1072 force_sig(SIGSEGV, current);
1076 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1080 /* No need to warn about this any longer. */
1081 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1085 fastcall void setup_x86_bogus_stack(unsigned char * stk)
1087 unsigned long *switch16_ptr, *switch32_ptr;
1088 struct pt_regs *regs;
1089 unsigned long stack_top, stack_bot;
1090 unsigned short iret_frame16_off;
1091 int cpu = smp_processor_id();
1092 /* reserve the space on 32bit stack for the magic switch16 pointer */
1093 memmove(stk, stk + 8, sizeof(struct pt_regs));
1094 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
1095 regs = (struct pt_regs *)stk;
1096 /* now the switch32 on 16bit stack */
1097 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1098 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1099 switch32_ptr = (unsigned long *)(stack_top - 8);
1100 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
1101 /* copy iret frame on 16bit stack */
1102 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
1103 /* fill in the switch pointers */
1104 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
1105 switch16_ptr[1] = __ESPFIX_SS;
1106 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
1107 8 - CPU_16BIT_STACK_SIZE;
1108 switch32_ptr[1] = __KERNEL_DS;
1111 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
1113 unsigned long *switch32_ptr;
1114 unsigned char *stack16, *stack32;
1115 unsigned long stack_top, stack_bot;
1117 int cpu = smp_processor_id();
1118 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1119 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1120 switch32_ptr = (unsigned long *)(stack_top - 8);
1121 /* copy the data from 16bit stack to 32bit stack */
1122 len = CPU_16BIT_STACK_SIZE - 8 - sp;
1123 stack16 = (unsigned char *)(stack_bot + sp);
1124 stack32 = (unsigned char *)
1125 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
1126 memcpy(stack32, stack16, len);
1131 * 'math_state_restore()' saves the current math information in the
1132 * old math state array, and gets the new ones from the current task
1134 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1135 * Don't touch unless you *really* know how it works.
1137 * Must be called with kernel preemption disabled (in this case,
1138 * local interrupts are disabled at the call-site in entry.S).
1140 asmlinkage void math_state_restore(struct pt_regs regs)
1142 struct thread_info *thread = current_thread_info();
1143 struct task_struct *tsk = thread->task;
1145 clts(); /* Allow maths ops (or we recurse) */
1146 if (!tsk_used_math(tsk))
1149 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1152 #ifndef CONFIG_MATH_EMULATION
1154 asmlinkage void math_emulate(long arg)
1156 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1157 printk(KERN_EMERG "killing %s.\n",current->comm);
1158 force_sig(SIGFPE,current);
1162 #endif /* CONFIG_MATH_EMULATION */
1164 #ifdef CONFIG_X86_F00F_BUG
1165 void __init trap_init_f00f_bug(void)
1167 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1170 * Update the IDT descriptor and reload the IDT so that
1171 * it uses the read-only mapped virtual address.
1173 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1174 load_idt(&idt_descr);
1178 #define _set_gate(gate_addr,type,dpl,addr,seg) \
1181 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1182 "movw %4,%%dx\n\t" \
1183 "movl %%eax,%0\n\t" \
1185 :"=m" (*((long *) (gate_addr))), \
1186 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1187 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1188 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1193 * This needs to use 'idt_table' rather than 'idt', and
1194 * thus use the _nonmapped_ version of the IDT, as the
1195 * Pentium F0 0F bugfix can have resulted in the mapped
1196 * IDT being write-protected.
1198 void set_intr_gate(unsigned int n, void *addr)
1200 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1204 * This routine sets up an interrupt gate at directory privilege level 3.
1206 static inline void set_system_intr_gate(unsigned int n, void *addr)
1208 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1211 static void __init set_trap_gate(unsigned int n, void *addr)
1213 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1216 static void __init set_system_gate(unsigned int n, void *addr)
1218 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1221 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1223 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1227 void __init trap_init(void)
1230 void __iomem *p = ioremap(0x0FFFD9, 4);
1231 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1237 #ifdef CONFIG_X86_LOCAL_APIC
1238 init_apic_mappings();
1241 set_trap_gate(0,÷_error);
1242 set_intr_gate(1,&debug);
1243 set_intr_gate(2,&nmi);
1244 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1245 set_system_gate(4,&overflow);
1246 set_trap_gate(5,&bounds);
1247 set_trap_gate(6,&invalid_op);
1248 set_trap_gate(7,&device_not_available);
1249 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1250 set_trap_gate(9,&coprocessor_segment_overrun);
1251 set_trap_gate(10,&invalid_TSS);
1252 set_trap_gate(11,&segment_not_present);
1253 set_trap_gate(12,&stack_segment);
1254 set_trap_gate(13,&general_protection);
1255 set_intr_gate(14,&page_fault);
1256 set_trap_gate(15,&spurious_interrupt_bug);
1257 set_trap_gate(16,&coprocessor_error);
1258 set_trap_gate(17,&alignment_check);
1259 #ifdef CONFIG_X86_MCE
1260 set_trap_gate(18,&machine_check);
1262 set_trap_gate(19,&simd_coprocessor_error);
1266 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1267 * Generates a compile-time "error: zero width for bit-field" if
1268 * the alignment is wrong.
1270 struct fxsrAlignAssert {
1271 int _:!(offsetof(struct task_struct,
1272 thread.i387.fxsave) & 15);
1275 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1276 set_in_cr4(X86_CR4_OSFXSR);
1280 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1282 set_in_cr4(X86_CR4_OSXMMEXCPT);
1286 set_system_gate(SYSCALL_VECTOR,&system_call);
1289 * Should be a barrier for any external CPU state.
1296 static int __init kstack_setup(char *s)
1298 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1301 __setup("kstack=", kstack_setup);