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/sched.h>
15 #include <linux/kernel.h>
16 #include <linux/string.h>
17 #include <linux/errno.h>
18 #include <linux/timer.h>
20 #include <linux/init.h>
21 #include <linux/delay.h>
22 #include <linux/spinlock.h>
23 #include <linux/interrupt.h>
24 #include <linux/highmem.h>
25 #include <linux/kallsyms.h>
26 #include <linux/ptrace.h>
27 #include <linux/utsname.h>
28 #include <linux/kprobes.h>
29 #include <linux/kexec.h>
30 #include <linux/unwind.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>
50 #include <asm/unwind.h>
52 #include <asm/arch_hooks.h>
53 #include <asm/kdebug.h>
54 #include <asm/stacktrace.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 #ifdef CONFIG_STACK_UNWIND
97 static int call_trace = 1;
99 #define call_trace (-1)
101 ATOMIC_NOTIFIER_HEAD(i386die_chain);
103 int register_die_notifier(struct notifier_block *nb)
106 return atomic_notifier_chain_register(&i386die_chain, nb);
108 EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
110 int unregister_die_notifier(struct notifier_block *nb)
112 return atomic_notifier_chain_unregister(&i386die_chain, nb);
114 EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
116 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
118 return p > (void *)tinfo &&
119 p < (void *)tinfo + THREAD_SIZE - 3;
122 static inline unsigned long print_context_stack(struct thread_info *tinfo,
123 unsigned long *stack, unsigned long ebp,
124 struct stacktrace_ops *ops, void *data)
128 #ifdef CONFIG_FRAME_POINTER
129 while (valid_stack_ptr(tinfo, (void *)ebp)) {
130 addr = *(unsigned long *)(ebp + 4);
131 ops->address(data, addr);
133 * break out of recursive entries (such as
134 * end_of_stack_stop_unwind_function):
136 if (ebp == *(unsigned long *)ebp)
138 ebp = *(unsigned long *)ebp;
141 while (valid_stack_ptr(tinfo, stack)) {
143 if (__kernel_text_address(addr))
144 ops->address(data, addr);
150 struct ops_and_data {
151 struct stacktrace_ops *ops;
155 static asmlinkage int
156 dump_trace_unwind(struct unwind_frame_info *info, void *data)
158 struct ops_and_data *oad = (struct ops_and_data *)data;
161 while (unwind(info) == 0 && UNW_PC(info)) {
163 oad->ops->address(oad->data, UNW_PC(info));
164 if (arch_unw_user_mode(info))
170 void dump_trace(struct task_struct *task, struct pt_regs *regs,
171 unsigned long *stack,
172 struct stacktrace_ops *ops, void *data)
174 unsigned long ebp = 0;
179 if (call_trace >= 0) {
181 struct unwind_frame_info info;
182 struct ops_and_data oad = { .ops = ops, .data = data };
185 if (unwind_init_frame_info(&info, task, regs) == 0)
186 unw_ret = dump_trace_unwind(&info, &oad);
187 } else if (task == current)
188 unw_ret = unwind_init_running(&info, dump_trace_unwind, &oad);
190 if (unwind_init_blocked(&info, task) == 0)
191 unw_ret = dump_trace_unwind(&info, &oad);
194 if (call_trace == 1 && !arch_unw_user_mode(&info)) {
195 ops->warning_symbol(data, "DWARF2 unwinder stuck at %s\n",
197 if (UNW_SP(&info) >= PAGE_OFFSET) {
198 ops->warning(data, "Leftover inexact backtrace:\n");
199 stack = (void *)UNW_SP(&info);
204 ops->warning(data, "Full inexact backtrace again:\n");
205 } else if (call_trace >= 1)
208 ops->warning(data, "Full inexact backtrace again:\n");
210 ops->warning(data, "Inexact backtrace:\n");
215 if (task && task != current)
216 stack = (unsigned long *)task->thread.esp;
219 #ifdef CONFIG_FRAME_POINTER
221 if (task == current) {
222 /* Grab ebp right from our regs */
223 asm ("movl %%ebp, %0" : "=r" (ebp) : );
225 /* ebp is the last reg pushed by switch_to */
226 ebp = *(unsigned long *) task->thread.esp;
232 struct thread_info *context;
233 context = (struct thread_info *)
234 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
235 ebp = print_context_stack(context, stack, ebp, ops, data);
236 /* Should be after the line below, but somewhere
237 in early boot context comes out corrupted and we
238 can't reference it -AK */
239 if (ops->stack(data, "IRQ") < 0)
241 stack = (unsigned long*)context->previous_esp;
246 EXPORT_SYMBOL(dump_trace);
249 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
252 print_symbol(msg, symbol);
256 static void print_trace_warning(void *data, char *msg)
258 printk("%s%s\n", (char *)data, msg);
261 static int print_trace_stack(void *data, char *name)
267 * Print one address/symbol entries per line.
269 static void print_trace_address(void *data, unsigned long addr)
271 printk("%s [<%08lx>] ", (char *)data, addr);
272 print_symbol("%s\n", addr);
275 static struct stacktrace_ops print_trace_ops = {
276 .warning = print_trace_warning,
277 .warning_symbol = print_trace_warning_symbol,
278 .stack = print_trace_stack,
279 .address = print_trace_address,
283 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
284 unsigned long * stack, char *log_lvl)
286 dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
287 printk("%s =======================\n", log_lvl);
290 void show_trace(struct task_struct *task, struct pt_regs *regs,
291 unsigned long * stack)
293 show_trace_log_lvl(task, regs, stack, "");
296 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
297 unsigned long *esp, char *log_lvl)
299 unsigned long *stack;
304 esp = (unsigned long*)task->thread.esp;
306 esp = (unsigned long *)&esp;
310 for(i = 0; i < kstack_depth_to_print; i++) {
311 if (kstack_end(stack))
313 if (i && ((i % 8) == 0))
314 printk("\n%s ", log_lvl);
315 printk("%08lx ", *stack++);
317 printk("\n%sCall Trace:\n", log_lvl);
318 show_trace_log_lvl(task, regs, esp, log_lvl);
321 void show_stack(struct task_struct *task, unsigned long *esp)
324 show_stack_log_lvl(task, NULL, esp, "");
328 * The architecture-independent dump_stack generator
330 void dump_stack(void)
334 show_trace(current, NULL, &stack);
337 EXPORT_SYMBOL(dump_stack);
339 void show_registers(struct pt_regs *regs)
346 esp = (unsigned long) (®s->esp);
348 if (user_mode_vm(regs)) {
351 ss = regs->xss & 0xffff;
354 printk(KERN_EMERG "CPU: %d\n"
355 KERN_EMERG "EIP: %04x:[<%08lx>] %s VLI\n"
356 KERN_EMERG "EFLAGS: %08lx (%s %.*s)\n",
357 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
358 print_tainted(), regs->eflags, system_utsname.release,
359 (int)strcspn(system_utsname.version, " "),
360 system_utsname.version);
361 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
362 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
363 regs->eax, regs->ebx, regs->ecx, regs->edx);
364 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
365 regs->esi, regs->edi, regs->ebp, esp);
366 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
367 regs->xds & 0xffff, regs->xes & 0xffff, ss);
368 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
369 TASK_COMM_LEN, current->comm, current->pid,
370 current_thread_info(), current, current->thread_info);
372 * When in-kernel, we also print out the stack and code at the
373 * time of the fault..
380 printk("\n" KERN_EMERG "Stack: ");
381 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
383 printk(KERN_EMERG "Code: ");
385 eip = (u8 __user *)regs->eip - 43;
386 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
387 /* try starting at EIP */
388 eip = (u8 __user *)regs->eip;
391 for (i = 0; i < code_bytes; i++, eip++) {
392 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
393 printk(" Bad EIP value.");
396 if (eip == (u8 __user *)regs->eip)
397 printk("<%02x> ", c);
405 static void handle_BUG(struct pt_regs *regs)
407 unsigned long eip = regs->eip;
410 if (eip < PAGE_OFFSET)
412 if (__get_user(ud2, (unsigned short __user *)eip))
417 printk(KERN_EMERG "------------[ cut here ]------------\n");
419 #ifdef CONFIG_DEBUG_BUGVERBOSE
425 if (__get_user(line, (unsigned short __user *)(eip + 2)))
427 if (__get_user(file, (char * __user *)(eip + 4)) ||
428 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
429 file = "<bad filename>";
431 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
435 printk(KERN_EMERG "Kernel BUG at [verbose debug info unavailable]\n");
438 /* This is gone through when something in the kernel
439 * has done something bad and is about to be terminated.
441 void die(const char * str, struct pt_regs * regs, long err)
446 int lock_owner_depth;
448 .lock = SPIN_LOCK_UNLOCKED,
450 .lock_owner_depth = 0
452 static int die_counter;
457 if (die.lock_owner != raw_smp_processor_id()) {
459 spin_lock_irqsave(&die.lock, flags);
460 die.lock_owner = smp_processor_id();
461 die.lock_owner_depth = 0;
465 local_save_flags(flags);
467 if (++die.lock_owner_depth < 3) {
473 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
474 #ifdef CONFIG_PREEMPT
475 printk(KERN_EMERG "PREEMPT ");
484 #ifdef CONFIG_DEBUG_PAGEALLOC
487 printk("DEBUG_PAGEALLOC");
492 if (notify_die(DIE_OOPS, str, regs, err,
493 current->thread.trap_no, SIGSEGV) !=
495 show_registers(regs);
496 /* Executive summary in case the oops scrolled away */
497 esp = (unsigned long) (®s->esp);
499 if (user_mode(regs)) {
501 ss = regs->xss & 0xffff;
503 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
504 print_symbol("%s", regs->eip);
505 printk(" SS:ESP %04x:%08lx\n", ss, esp);
510 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
514 spin_unlock_irqrestore(&die.lock, flags);
519 if (kexec_should_crash(current))
523 panic("Fatal exception in interrupt");
526 panic("Fatal exception");
532 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
534 if (!user_mode_vm(regs))
538 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
539 struct pt_regs * regs, long error_code,
542 struct task_struct *tsk = current;
543 tsk->thread.error_code = error_code;
544 tsk->thread.trap_no = trapnr;
546 if (regs->eflags & VM_MASK) {
552 if (!user_mode(regs))
557 force_sig_info(signr, info, tsk);
559 force_sig(signr, tsk);
564 if (!fixup_exception(regs))
565 die(str, regs, error_code);
570 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
571 if (ret) goto trap_signal;
576 #define DO_ERROR(trapnr, signr, str, name) \
577 fastcall void do_##name(struct pt_regs * regs, long error_code) \
579 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
582 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
585 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
586 fastcall void do_##name(struct pt_regs * regs, long error_code) \
589 info.si_signo = signr; \
591 info.si_code = sicode; \
592 info.si_addr = (void __user *)siaddr; \
593 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
596 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
599 #define DO_VM86_ERROR(trapnr, signr, str, name) \
600 fastcall void do_##name(struct pt_regs * regs, long error_code) \
602 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
605 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
608 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
609 fastcall void do_##name(struct pt_regs * regs, long error_code) \
612 info.si_signo = signr; \
614 info.si_code = sicode; \
615 info.si_addr = (void __user *)siaddr; \
616 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
619 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
622 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
623 #ifndef CONFIG_KPROBES
624 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
626 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
627 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
628 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
629 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
630 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
631 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
632 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
633 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
634 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
636 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
640 struct tss_struct *tss = &per_cpu(init_tss, cpu);
641 struct thread_struct *thread = ¤t->thread;
644 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
645 * invalid offset set (the LAZY one) and the faulting thread has
646 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
647 * and we set the offset field correctly. Then we let the CPU to
648 * restart the faulting instruction.
650 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
651 thread->io_bitmap_ptr) {
652 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
653 thread->io_bitmap_max);
655 * If the previously set map was extending to higher ports
656 * than the current one, pad extra space with 0xff (no access).
658 if (thread->io_bitmap_max < tss->io_bitmap_max)
659 memset((char *) tss->io_bitmap +
660 thread->io_bitmap_max, 0xff,
661 tss->io_bitmap_max - thread->io_bitmap_max);
662 tss->io_bitmap_max = thread->io_bitmap_max;
663 tss->io_bitmap_base = IO_BITMAP_OFFSET;
664 tss->io_bitmap_owner = thread;
670 current->thread.error_code = error_code;
671 current->thread.trap_no = 13;
673 if (regs->eflags & VM_MASK)
676 if (!user_mode(regs))
679 current->thread.error_code = error_code;
680 current->thread.trap_no = 13;
681 force_sig(SIGSEGV, current);
686 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
690 if (!fixup_exception(regs)) {
691 if (notify_die(DIE_GPF, "general protection fault", regs,
692 error_code, 13, SIGSEGV) == NOTIFY_STOP)
694 die("general protection fault", regs, error_code);
698 static __kprobes void
699 mem_parity_error(unsigned char reason, struct pt_regs * regs)
701 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
702 "CPU %d.\n", reason, smp_processor_id());
703 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
705 if (panic_on_unrecovered_nmi)
706 panic("NMI: Not continuing");
708 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
710 /* Clear and disable the memory parity error line. */
711 clear_mem_error(reason);
714 static __kprobes void
715 io_check_error(unsigned char reason, struct pt_regs * regs)
719 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
720 show_registers(regs);
722 /* Re-enable the IOCK line, wait for a few seconds */
723 reason = (reason & 0xf) | 8;
726 while (--i) udelay(1000);
731 static __kprobes void
732 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
735 /* Might actually be able to figure out what the guilty party
742 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
743 "CPU %d.\n", reason, smp_processor_id());
744 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
745 if (panic_on_unrecovered_nmi)
746 panic("NMI: Not continuing");
748 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
751 static DEFINE_SPINLOCK(nmi_print_lock);
753 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
755 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
759 spin_lock(&nmi_print_lock);
761 * We are in trouble anyway, lets at least try
762 * to get a message out.
765 printk(KERN_EMERG "%s", msg);
766 printk(" on CPU%d, eip %08lx, registers:\n",
767 smp_processor_id(), regs->eip);
768 show_registers(regs);
769 printk(KERN_EMERG "console shuts up ...\n");
771 spin_unlock(&nmi_print_lock);
774 /* If we are in kernel we are probably nested up pretty bad
775 * and might aswell get out now while we still can.
777 if (!user_mode_vm(regs)) {
778 current->thread.trap_no = 2;
785 static __kprobes void default_do_nmi(struct pt_regs * regs)
787 unsigned char reason = 0;
789 /* Only the BSP gets external NMIs from the system. */
790 if (!smp_processor_id())
791 reason = get_nmi_reason();
793 if (!(reason & 0xc0)) {
794 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
797 #ifdef CONFIG_X86_LOCAL_APIC
799 * Ok, so this is none of the documented NMI sources,
800 * so it must be the NMI watchdog.
802 if (nmi_watchdog_tick(regs, reason))
804 if (!do_nmi_callback(regs, smp_processor_id()))
806 unknown_nmi_error(reason, regs);
810 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
813 mem_parity_error(reason, regs);
815 io_check_error(reason, regs);
817 * Reassert NMI in case it became active meanwhile
818 * as it's edge-triggered.
823 fastcall __kprobes void do_nmi(struct pt_regs * regs, long error_code)
829 cpu = smp_processor_id();
833 default_do_nmi(regs);
838 #ifdef CONFIG_KPROBES
839 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
841 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
844 /* This is an interrupt gate, because kprobes wants interrupts
845 disabled. Normal trap handlers don't. */
846 restore_interrupts(regs);
847 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
852 * Our handling of the processor debug registers is non-trivial.
853 * We do not clear them on entry and exit from the kernel. Therefore
854 * it is possible to get a watchpoint trap here from inside the kernel.
855 * However, the code in ./ptrace.c has ensured that the user can
856 * only set watchpoints on userspace addresses. Therefore the in-kernel
857 * watchpoint trap can only occur in code which is reading/writing
858 * from user space. Such code must not hold kernel locks (since it
859 * can equally take a page fault), therefore it is safe to call
860 * force_sig_info even though that claims and releases locks.
862 * Code in ./signal.c ensures that the debug control register
863 * is restored before we deliver any signal, and therefore that
864 * user code runs with the correct debug control register even though
867 * Being careful here means that we don't have to be as careful in a
868 * lot of more complicated places (task switching can be a bit lazy
869 * about restoring all the debug state, and ptrace doesn't have to
870 * find every occurrence of the TF bit that could be saved away even
873 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
875 unsigned int condition;
876 struct task_struct *tsk = current;
878 get_debugreg(condition, 6);
880 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
881 SIGTRAP) == NOTIFY_STOP)
883 /* It's safe to allow irq's after DR6 has been saved */
884 if (regs->eflags & X86_EFLAGS_IF)
887 /* Mask out spurious debug traps due to lazy DR7 setting */
888 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
889 if (!tsk->thread.debugreg[7])
893 if (regs->eflags & VM_MASK)
896 /* Save debug status register where ptrace can see it */
897 tsk->thread.debugreg[6] = condition;
900 * Single-stepping through TF: make sure we ignore any events in
901 * kernel space (but re-enable TF when returning to user mode).
903 if (condition & DR_STEP) {
905 * We already checked v86 mode above, so we can
906 * check for kernel mode by just checking the CPL
909 if (!user_mode(regs))
910 goto clear_TF_reenable;
913 /* Ok, finally something we can handle */
914 send_sigtrap(tsk, regs, error_code);
916 /* Disable additional traps. They'll be re-enabled when
917 * the signal is delivered.
924 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
928 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
929 regs->eflags &= ~TF_MASK;
934 * Note that we play around with the 'TS' bit in an attempt to get
935 * the correct behaviour even in the presence of the asynchronous
938 void math_error(void __user *eip)
940 struct task_struct * task;
942 unsigned short cwd, swd;
945 * Save the info for the exception handler and clear the error.
949 task->thread.trap_no = 16;
950 task->thread.error_code = 0;
951 info.si_signo = SIGFPE;
953 info.si_code = __SI_FAULT;
956 * (~cwd & swd) will mask out exceptions that are not set to unmasked
957 * status. 0x3f is the exception bits in these regs, 0x200 is the
958 * C1 reg you need in case of a stack fault, 0x040 is the stack
959 * fault bit. We should only be taking one exception at a time,
960 * so if this combination doesn't produce any single exception,
961 * then we have a bad program that isn't syncronizing its FPU usage
962 * and it will suffer the consequences since we won't be able to
963 * fully reproduce the context of the exception
965 cwd = get_fpu_cwd(task);
966 swd = get_fpu_swd(task);
967 switch (swd & ~cwd & 0x3f) {
968 case 0x000: /* No unmasked exception */
970 default: /* Multiple exceptions */
972 case 0x001: /* Invalid Op */
974 * swd & 0x240 == 0x040: Stack Underflow
975 * swd & 0x240 == 0x240: Stack Overflow
976 * User must clear the SF bit (0x40) if set
978 info.si_code = FPE_FLTINV;
980 case 0x002: /* Denormalize */
981 case 0x010: /* Underflow */
982 info.si_code = FPE_FLTUND;
984 case 0x004: /* Zero Divide */
985 info.si_code = FPE_FLTDIV;
987 case 0x008: /* Overflow */
988 info.si_code = FPE_FLTOVF;
990 case 0x020: /* Precision */
991 info.si_code = FPE_FLTRES;
994 force_sig_info(SIGFPE, &info, task);
997 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
1000 math_error((void __user *)regs->eip);
1003 static void simd_math_error(void __user *eip)
1005 struct task_struct * task;
1007 unsigned short mxcsr;
1010 * Save the info for the exception handler and clear the error.
1013 save_init_fpu(task);
1014 task->thread.trap_no = 19;
1015 task->thread.error_code = 0;
1016 info.si_signo = SIGFPE;
1018 info.si_code = __SI_FAULT;
1021 * The SIMD FPU exceptions are handled a little differently, as there
1022 * is only a single status/control register. Thus, to determine which
1023 * unmasked exception was caught we must mask the exception mask bits
1024 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1026 mxcsr = get_fpu_mxcsr(task);
1027 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1031 case 0x001: /* Invalid Op */
1032 info.si_code = FPE_FLTINV;
1034 case 0x002: /* Denormalize */
1035 case 0x010: /* Underflow */
1036 info.si_code = FPE_FLTUND;
1038 case 0x004: /* Zero Divide */
1039 info.si_code = FPE_FLTDIV;
1041 case 0x008: /* Overflow */
1042 info.si_code = FPE_FLTOVF;
1044 case 0x020: /* Precision */
1045 info.si_code = FPE_FLTRES;
1048 force_sig_info(SIGFPE, &info, task);
1051 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1055 /* Handle SIMD FPU exceptions on PIII+ processors. */
1057 simd_math_error((void __user *)regs->eip);
1060 * Handle strange cache flush from user space exception
1061 * in all other cases. This is undocumented behaviour.
1063 if (regs->eflags & VM_MASK) {
1064 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1068 current->thread.trap_no = 19;
1069 current->thread.error_code = error_code;
1070 die_if_kernel("cache flush denied", regs, error_code);
1071 force_sig(SIGSEGV, current);
1075 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1079 /* No need to warn about this any longer. */
1080 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1084 fastcall void setup_x86_bogus_stack(unsigned char * stk)
1086 unsigned long *switch16_ptr, *switch32_ptr;
1087 struct pt_regs *regs;
1088 unsigned long stack_top, stack_bot;
1089 unsigned short iret_frame16_off;
1090 int cpu = smp_processor_id();
1091 /* reserve the space on 32bit stack for the magic switch16 pointer */
1092 memmove(stk, stk + 8, sizeof(struct pt_regs));
1093 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
1094 regs = (struct pt_regs *)stk;
1095 /* now the switch32 on 16bit stack */
1096 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1097 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1098 switch32_ptr = (unsigned long *)(stack_top - 8);
1099 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
1100 /* copy iret frame on 16bit stack */
1101 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
1102 /* fill in the switch pointers */
1103 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
1104 switch16_ptr[1] = __ESPFIX_SS;
1105 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
1106 8 - CPU_16BIT_STACK_SIZE;
1107 switch32_ptr[1] = __KERNEL_DS;
1110 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
1112 unsigned long *switch32_ptr;
1113 unsigned char *stack16, *stack32;
1114 unsigned long stack_top, stack_bot;
1116 int cpu = smp_processor_id();
1117 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1118 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1119 switch32_ptr = (unsigned long *)(stack_top - 8);
1120 /* copy the data from 16bit stack to 32bit stack */
1121 len = CPU_16BIT_STACK_SIZE - 8 - sp;
1122 stack16 = (unsigned char *)(stack_bot + sp);
1123 stack32 = (unsigned char *)
1124 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
1125 memcpy(stack32, stack16, len);
1130 * 'math_state_restore()' saves the current math information in the
1131 * old math state array, and gets the new ones from the current task
1133 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1134 * Don't touch unless you *really* know how it works.
1136 * Must be called with kernel preemption disabled (in this case,
1137 * local interrupts are disabled at the call-site in entry.S).
1139 asmlinkage void math_state_restore(struct pt_regs regs)
1141 struct thread_info *thread = current_thread_info();
1142 struct task_struct *tsk = thread->task;
1144 clts(); /* Allow maths ops (or we recurse) */
1145 if (!tsk_used_math(tsk))
1148 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1151 #ifndef CONFIG_MATH_EMULATION
1153 asmlinkage void math_emulate(long arg)
1155 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1156 printk(KERN_EMERG "killing %s.\n",current->comm);
1157 force_sig(SIGFPE,current);
1161 #endif /* CONFIG_MATH_EMULATION */
1163 #ifdef CONFIG_X86_F00F_BUG
1164 void __init trap_init_f00f_bug(void)
1166 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1169 * Update the IDT descriptor and reload the IDT so that
1170 * it uses the read-only mapped virtual address.
1172 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1173 load_idt(&idt_descr);
1178 * This needs to use 'idt_table' rather than 'idt', and
1179 * thus use the _nonmapped_ version of the IDT, as the
1180 * Pentium F0 0F bugfix can have resulted in the mapped
1181 * IDT being write-protected.
1183 void set_intr_gate(unsigned int n, void *addr)
1185 _set_gate(n, DESCTYPE_INT, addr, __KERNEL_CS);
1189 * This routine sets up an interrupt gate at directory privilege level 3.
1191 static inline void set_system_intr_gate(unsigned int n, void *addr)
1193 _set_gate(n, DESCTYPE_INT | DESCTYPE_DPL3, addr, __KERNEL_CS);
1196 static void __init set_trap_gate(unsigned int n, void *addr)
1198 _set_gate(n, DESCTYPE_TRAP, addr, __KERNEL_CS);
1201 static void __init set_system_gate(unsigned int n, void *addr)
1203 _set_gate(n, DESCTYPE_TRAP | DESCTYPE_DPL3, addr, __KERNEL_CS);
1206 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1208 _set_gate(n, DESCTYPE_TASK, (void *)0, (gdt_entry<<3));
1212 void __init trap_init(void)
1215 void __iomem *p = ioremap(0x0FFFD9, 4);
1216 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1222 #ifdef CONFIG_X86_LOCAL_APIC
1223 init_apic_mappings();
1226 set_trap_gate(0,÷_error);
1227 set_intr_gate(1,&debug);
1228 set_intr_gate(2,&nmi);
1229 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1230 set_system_gate(4,&overflow);
1231 set_trap_gate(5,&bounds);
1232 set_trap_gate(6,&invalid_op);
1233 set_trap_gate(7,&device_not_available);
1234 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1235 set_trap_gate(9,&coprocessor_segment_overrun);
1236 set_trap_gate(10,&invalid_TSS);
1237 set_trap_gate(11,&segment_not_present);
1238 set_trap_gate(12,&stack_segment);
1239 set_trap_gate(13,&general_protection);
1240 set_intr_gate(14,&page_fault);
1241 set_trap_gate(15,&spurious_interrupt_bug);
1242 set_trap_gate(16,&coprocessor_error);
1243 set_trap_gate(17,&alignment_check);
1244 #ifdef CONFIG_X86_MCE
1245 set_trap_gate(18,&machine_check);
1247 set_trap_gate(19,&simd_coprocessor_error);
1251 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1252 * Generates a compile-time "error: zero width for bit-field" if
1253 * the alignment is wrong.
1255 struct fxsrAlignAssert {
1256 int _:!(offsetof(struct task_struct,
1257 thread.i387.fxsave) & 15);
1260 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1261 set_in_cr4(X86_CR4_OSFXSR);
1265 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1267 set_in_cr4(X86_CR4_OSXMMEXCPT);
1271 set_system_gate(SYSCALL_VECTOR,&system_call);
1274 * Should be a barrier for any external CPU state.
1281 static int __init kstack_setup(char *s)
1283 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1286 __setup("kstack=", kstack_setup);
1288 #ifdef CONFIG_STACK_UNWIND
1289 static int __init call_trace_setup(char *s)
1291 if (strcmp(s, "old") == 0)
1293 else if (strcmp(s, "both") == 0)
1295 else if (strcmp(s, "newfallback") == 0)
1297 else if (strcmp(s, "new") == 2)
1301 __setup("call_trace=", call_trace_setup);