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>
31 #include <linux/uaccess.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>
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 int panic_on_unrecovered_nmi;
62 asmlinkage int system_call(void);
64 /* Do we ignore FPU interrupts ? */
65 char ignore_fpu_irq = 0;
68 * The IDT has to be page-aligned to simplify the Pentium
69 * F0 0F bug workaround.. We have a special link segment
72 struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
74 asmlinkage void divide_error(void);
75 asmlinkage void debug(void);
76 asmlinkage void nmi(void);
77 asmlinkage void int3(void);
78 asmlinkage void overflow(void);
79 asmlinkage void bounds(void);
80 asmlinkage void invalid_op(void);
81 asmlinkage void device_not_available(void);
82 asmlinkage void coprocessor_segment_overrun(void);
83 asmlinkage void invalid_TSS(void);
84 asmlinkage void segment_not_present(void);
85 asmlinkage void stack_segment(void);
86 asmlinkage void general_protection(void);
87 asmlinkage void page_fault(void);
88 asmlinkage void coprocessor_error(void);
89 asmlinkage void simd_coprocessor_error(void);
90 asmlinkage void alignment_check(void);
91 asmlinkage void spurious_interrupt_bug(void);
92 asmlinkage void machine_check(void);
94 static int kstack_depth_to_print = 24;
95 #ifdef CONFIG_STACK_UNWIND
96 static int call_trace = 1;
98 #define call_trace (-1)
100 ATOMIC_NOTIFIER_HEAD(i386die_chain);
102 int register_die_notifier(struct notifier_block *nb)
105 return atomic_notifier_chain_register(&i386die_chain, nb);
107 EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
109 int unregister_die_notifier(struct notifier_block *nb)
111 return atomic_notifier_chain_unregister(&i386die_chain, nb);
113 EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
115 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
117 return p > (void *)tinfo &&
118 p < (void *)tinfo + THREAD_SIZE - 3;
121 static inline unsigned long print_context_stack(struct thread_info *tinfo,
122 unsigned long *stack, unsigned long ebp,
123 struct stacktrace_ops *ops, void *data)
127 #ifdef CONFIG_FRAME_POINTER
128 while (valid_stack_ptr(tinfo, (void *)ebp)) {
129 unsigned long new_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). Also,
135 * we can never allow a frame pointer to
138 new_ebp = *(unsigned long *)ebp;
144 while (valid_stack_ptr(tinfo, stack)) {
146 if (__kernel_text_address(addr))
147 ops->address(data, addr);
153 struct ops_and_data {
154 struct stacktrace_ops *ops;
158 static asmlinkage int
159 dump_trace_unwind(struct unwind_frame_info *info, void *data)
161 struct ops_and_data *oad = (struct ops_and_data *)data;
164 while (unwind(info) == 0 && UNW_PC(info)) {
166 oad->ops->address(oad->data, UNW_PC(info));
167 if (arch_unw_user_mode(info))
173 #define MSG(msg) ops->warning(data, msg)
175 void dump_trace(struct task_struct *task, struct pt_regs *regs,
176 unsigned long *stack,
177 struct stacktrace_ops *ops, void *data)
179 unsigned long ebp = 0;
184 if (call_trace >= 0) {
186 struct unwind_frame_info info;
187 struct ops_and_data oad = { .ops = ops, .data = data };
190 if (unwind_init_frame_info(&info, task, regs) == 0)
191 unw_ret = dump_trace_unwind(&info, &oad);
192 } else if (task == current)
193 unw_ret = unwind_init_running(&info, dump_trace_unwind,
196 if (unwind_init_blocked(&info, task) == 0)
197 unw_ret = dump_trace_unwind(&info, &oad);
200 if (call_trace == 1 && !arch_unw_user_mode(&info)) {
201 ops->warning_symbol(data,
202 "DWARF2 unwinder stuck at %s",
204 if (UNW_SP(&info) >= PAGE_OFFSET) {
205 MSG("Leftover inexact backtrace:");
206 stack = (void *)UNW_SP(&info);
211 MSG("Full inexact backtrace again:");
212 } else if (call_trace >= 1)
215 MSG("Full inexact backtrace again:");
217 MSG("Inexact backtrace:");
222 if (task && task != current)
223 stack = (unsigned long *)task->thread.esp;
226 #ifdef CONFIG_FRAME_POINTER
228 if (task == current) {
229 /* Grab ebp right from our regs */
230 asm ("movl %%ebp, %0" : "=r" (ebp) : );
232 /* ebp is the last reg pushed by switch_to */
233 ebp = *(unsigned long *) task->thread.esp;
239 struct thread_info *context;
240 context = (struct thread_info *)
241 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
242 ebp = print_context_stack(context, stack, ebp, ops, data);
243 /* Should be after the line below, but somewhere
244 in early boot context comes out corrupted and we
245 can't reference it -AK */
246 if (ops->stack(data, "IRQ") < 0)
248 stack = (unsigned long*)context->previous_esp;
253 EXPORT_SYMBOL(dump_trace);
256 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
259 print_symbol(msg, symbol);
263 static void print_trace_warning(void *data, char *msg)
265 printk("%s%s\n", (char *)data, msg);
268 static int print_trace_stack(void *data, char *name)
274 * Print one address/symbol entries per line.
276 static void print_trace_address(void *data, unsigned long addr)
278 printk("%s [<%08lx>] ", (char *)data, addr);
279 print_symbol("%s\n", addr);
282 static struct stacktrace_ops print_trace_ops = {
283 .warning = print_trace_warning,
284 .warning_symbol = print_trace_warning_symbol,
285 .stack = print_trace_stack,
286 .address = print_trace_address,
290 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
291 unsigned long * stack, char *log_lvl)
293 dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
294 printk("%s =======================\n", log_lvl);
297 void show_trace(struct task_struct *task, struct pt_regs *regs,
298 unsigned long * stack)
300 show_trace_log_lvl(task, regs, stack, "");
303 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
304 unsigned long *esp, char *log_lvl)
306 unsigned long *stack;
311 esp = (unsigned long*)task->thread.esp;
313 esp = (unsigned long *)&esp;
317 for(i = 0; i < kstack_depth_to_print; i++) {
318 if (kstack_end(stack))
320 if (i && ((i % 8) == 0))
321 printk("\n%s ", log_lvl);
322 printk("%08lx ", *stack++);
324 printk("\n%sCall Trace:\n", log_lvl);
325 show_trace_log_lvl(task, regs, esp, log_lvl);
328 void show_stack(struct task_struct *task, unsigned long *esp)
331 show_stack_log_lvl(task, NULL, esp, "");
335 * The architecture-independent dump_stack generator
337 void dump_stack(void)
341 show_trace(current, NULL, &stack);
344 EXPORT_SYMBOL(dump_stack);
346 void show_registers(struct pt_regs *regs)
353 esp = (unsigned long) (®s->esp);
355 if (user_mode_vm(regs)) {
358 ss = regs->xss & 0xffff;
361 printk(KERN_EMERG "CPU: %d\n"
362 KERN_EMERG "EIP: %04x:[<%08lx>] %s VLI\n"
363 KERN_EMERG "EFLAGS: %08lx (%s %.*s)\n",
364 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
365 print_tainted(), regs->eflags, init_utsname()->release,
366 (int)strcspn(init_utsname()->version, " "),
367 init_utsname()->version);
368 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
369 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
370 regs->eax, regs->ebx, regs->ecx, regs->edx);
371 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
372 regs->esi, regs->edi, regs->ebp, esp);
373 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
374 regs->xds & 0xffff, regs->xes & 0xffff, ss);
375 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
376 TASK_COMM_LEN, current->comm, current->pid,
377 current_thread_info(), current, current->thread_info);
379 * When in-kernel, we also print out the stack and code at the
380 * time of the fault..
387 printk("\n" KERN_EMERG "Stack: ");
388 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
390 printk(KERN_EMERG "Code: ");
392 eip = (u8 __user *)regs->eip - 43;
393 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
394 /* try starting at EIP */
395 eip = (u8 __user *)regs->eip;
398 for (i = 0; i < code_bytes; i++, eip++) {
399 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
400 printk(" Bad EIP value.");
403 if (eip == (u8 __user *)regs->eip)
404 printk("<%02x> ", c);
412 static void handle_BUG(struct pt_regs *regs)
414 unsigned long eip = regs->eip;
417 if (eip < PAGE_OFFSET)
419 if (probe_kernel_address((unsigned short __user *)eip, ud2))
424 printk(KERN_EMERG "------------[ cut here ]------------\n");
426 #ifdef CONFIG_DEBUG_BUGVERBOSE
432 if (probe_kernel_address((unsigned short __user *)(eip + 2),
435 if (__get_user(file, (char * __user *)(eip + 4)) ||
436 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
437 file = "<bad filename>";
439 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
443 printk(KERN_EMERG "Kernel BUG at [verbose debug info unavailable]\n");
446 /* This is gone through when something in the kernel
447 * has done something bad and is about to be terminated.
449 void die(const char * str, struct pt_regs * regs, long err)
454 int lock_owner_depth;
456 .lock = SPIN_LOCK_UNLOCKED,
458 .lock_owner_depth = 0
460 static int die_counter;
465 if (die.lock_owner != raw_smp_processor_id()) {
467 spin_lock_irqsave(&die.lock, flags);
468 die.lock_owner = smp_processor_id();
469 die.lock_owner_depth = 0;
473 local_save_flags(flags);
475 if (++die.lock_owner_depth < 3) {
481 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
482 #ifdef CONFIG_PREEMPT
483 printk(KERN_EMERG "PREEMPT ");
492 #ifdef CONFIG_DEBUG_PAGEALLOC
495 printk("DEBUG_PAGEALLOC");
500 if (notify_die(DIE_OOPS, str, regs, err,
501 current->thread.trap_no, SIGSEGV) !=
503 show_registers(regs);
504 /* Executive summary in case the oops scrolled away */
505 esp = (unsigned long) (®s->esp);
507 if (user_mode(regs)) {
509 ss = regs->xss & 0xffff;
511 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
512 print_symbol("%s", regs->eip);
513 printk(" SS:ESP %04x:%08lx\n", ss, esp);
518 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
522 spin_unlock_irqrestore(&die.lock, flags);
527 if (kexec_should_crash(current))
531 panic("Fatal exception in interrupt");
534 panic("Fatal exception");
540 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
542 if (!user_mode_vm(regs))
546 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
547 struct pt_regs * regs, long error_code,
550 struct task_struct *tsk = current;
551 tsk->thread.error_code = error_code;
552 tsk->thread.trap_no = trapnr;
554 if (regs->eflags & VM_MASK) {
560 if (!user_mode(regs))
565 force_sig_info(signr, info, tsk);
567 force_sig(signr, tsk);
572 if (!fixup_exception(regs))
573 die(str, regs, error_code);
578 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
579 if (ret) goto trap_signal;
584 #define DO_ERROR(trapnr, signr, str, name) \
585 fastcall void do_##name(struct pt_regs * regs, long error_code) \
587 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
590 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
593 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
594 fastcall void do_##name(struct pt_regs * regs, long error_code) \
597 info.si_signo = signr; \
599 info.si_code = sicode; \
600 info.si_addr = (void __user *)siaddr; \
601 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
604 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
607 #define DO_VM86_ERROR(trapnr, signr, str, name) \
608 fastcall void do_##name(struct pt_regs * regs, long error_code) \
610 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
613 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
616 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
617 fastcall void do_##name(struct pt_regs * regs, long error_code) \
620 info.si_signo = signr; \
622 info.si_code = sicode; \
623 info.si_addr = (void __user *)siaddr; \
624 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
627 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
630 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
631 #ifndef CONFIG_KPROBES
632 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
634 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
635 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
636 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
637 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
638 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
639 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
640 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
641 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
642 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
644 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
648 struct tss_struct *tss = &per_cpu(init_tss, cpu);
649 struct thread_struct *thread = ¤t->thread;
652 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
653 * invalid offset set (the LAZY one) and the faulting thread has
654 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
655 * and we set the offset field correctly. Then we let the CPU to
656 * restart the faulting instruction.
658 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
659 thread->io_bitmap_ptr) {
660 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
661 thread->io_bitmap_max);
663 * If the previously set map was extending to higher ports
664 * than the current one, pad extra space with 0xff (no access).
666 if (thread->io_bitmap_max < tss->io_bitmap_max)
667 memset((char *) tss->io_bitmap +
668 thread->io_bitmap_max, 0xff,
669 tss->io_bitmap_max - thread->io_bitmap_max);
670 tss->io_bitmap_max = thread->io_bitmap_max;
671 tss->io_bitmap_base = IO_BITMAP_OFFSET;
672 tss->io_bitmap_owner = thread;
678 current->thread.error_code = error_code;
679 current->thread.trap_no = 13;
681 if (regs->eflags & VM_MASK)
684 if (!user_mode(regs))
687 current->thread.error_code = error_code;
688 current->thread.trap_no = 13;
689 force_sig(SIGSEGV, current);
694 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
698 if (!fixup_exception(regs)) {
699 if (notify_die(DIE_GPF, "general protection fault", regs,
700 error_code, 13, SIGSEGV) == NOTIFY_STOP)
702 die("general protection fault", regs, error_code);
706 static __kprobes void
707 mem_parity_error(unsigned char reason, struct pt_regs * regs)
709 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
710 "CPU %d.\n", reason, smp_processor_id());
711 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
713 if (panic_on_unrecovered_nmi)
714 panic("NMI: Not continuing");
716 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
718 /* Clear and disable the memory parity error line. */
719 clear_mem_error(reason);
722 static __kprobes void
723 io_check_error(unsigned char reason, struct pt_regs * regs)
727 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
728 show_registers(regs);
730 /* Re-enable the IOCK line, wait for a few seconds */
731 reason = (reason & 0xf) | 8;
734 while (--i) udelay(1000);
739 static __kprobes void
740 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
743 /* Might actually be able to figure out what the guilty party
750 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
751 "CPU %d.\n", reason, smp_processor_id());
752 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
753 if (panic_on_unrecovered_nmi)
754 panic("NMI: Not continuing");
756 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
759 static DEFINE_SPINLOCK(nmi_print_lock);
761 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
763 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
767 spin_lock(&nmi_print_lock);
769 * We are in trouble anyway, lets at least try
770 * to get a message out.
773 printk(KERN_EMERG "%s", msg);
774 printk(" on CPU%d, eip %08lx, registers:\n",
775 smp_processor_id(), regs->eip);
776 show_registers(regs);
778 spin_unlock(&nmi_print_lock);
781 /* If we are in kernel we are probably nested up pretty bad
782 * and might aswell get out now while we still can.
784 if (!user_mode_vm(regs)) {
785 current->thread.trap_no = 2;
792 static __kprobes void default_do_nmi(struct pt_regs * regs)
794 unsigned char reason = 0;
796 /* Only the BSP gets external NMIs from the system. */
797 if (!smp_processor_id())
798 reason = get_nmi_reason();
800 if (!(reason & 0xc0)) {
801 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
804 #ifdef CONFIG_X86_LOCAL_APIC
806 * Ok, so this is none of the documented NMI sources,
807 * so it must be the NMI watchdog.
809 if (nmi_watchdog_tick(regs, reason))
811 if (!do_nmi_callback(regs, smp_processor_id()))
813 unknown_nmi_error(reason, regs);
817 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
820 mem_parity_error(reason, regs);
822 io_check_error(reason, regs);
824 * Reassert NMI in case it became active meanwhile
825 * as it's edge-triggered.
830 fastcall __kprobes void do_nmi(struct pt_regs * regs, long error_code)
836 cpu = smp_processor_id();
840 default_do_nmi(regs);
845 #ifdef CONFIG_KPROBES
846 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
848 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
851 /* This is an interrupt gate, because kprobes wants interrupts
852 disabled. Normal trap handlers don't. */
853 restore_interrupts(regs);
854 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
859 * Our handling of the processor debug registers is non-trivial.
860 * We do not clear them on entry and exit from the kernel. Therefore
861 * it is possible to get a watchpoint trap here from inside the kernel.
862 * However, the code in ./ptrace.c has ensured that the user can
863 * only set watchpoints on userspace addresses. Therefore the in-kernel
864 * watchpoint trap can only occur in code which is reading/writing
865 * from user space. Such code must not hold kernel locks (since it
866 * can equally take a page fault), therefore it is safe to call
867 * force_sig_info even though that claims and releases locks.
869 * Code in ./signal.c ensures that the debug control register
870 * is restored before we deliver any signal, and therefore that
871 * user code runs with the correct debug control register even though
874 * Being careful here means that we don't have to be as careful in a
875 * lot of more complicated places (task switching can be a bit lazy
876 * about restoring all the debug state, and ptrace doesn't have to
877 * find every occurrence of the TF bit that could be saved away even
880 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
882 unsigned int condition;
883 struct task_struct *tsk = current;
885 get_debugreg(condition, 6);
887 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
888 SIGTRAP) == NOTIFY_STOP)
890 /* It's safe to allow irq's after DR6 has been saved */
891 if (regs->eflags & X86_EFLAGS_IF)
894 /* Mask out spurious debug traps due to lazy DR7 setting */
895 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
896 if (!tsk->thread.debugreg[7])
900 if (regs->eflags & VM_MASK)
903 /* Save debug status register where ptrace can see it */
904 tsk->thread.debugreg[6] = condition;
907 * Single-stepping through TF: make sure we ignore any events in
908 * kernel space (but re-enable TF when returning to user mode).
910 if (condition & DR_STEP) {
912 * We already checked v86 mode above, so we can
913 * check for kernel mode by just checking the CPL
916 if (!user_mode(regs))
917 goto clear_TF_reenable;
920 /* Ok, finally something we can handle */
921 send_sigtrap(tsk, regs, error_code);
923 /* Disable additional traps. They'll be re-enabled when
924 * the signal is delivered.
931 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
935 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
936 regs->eflags &= ~TF_MASK;
941 * Note that we play around with the 'TS' bit in an attempt to get
942 * the correct behaviour even in the presence of the asynchronous
945 void math_error(void __user *eip)
947 struct task_struct * task;
949 unsigned short cwd, swd;
952 * Save the info for the exception handler and clear the error.
956 task->thread.trap_no = 16;
957 task->thread.error_code = 0;
958 info.si_signo = SIGFPE;
960 info.si_code = __SI_FAULT;
963 * (~cwd & swd) will mask out exceptions that are not set to unmasked
964 * status. 0x3f is the exception bits in these regs, 0x200 is the
965 * C1 reg you need in case of a stack fault, 0x040 is the stack
966 * fault bit. We should only be taking one exception at a time,
967 * so if this combination doesn't produce any single exception,
968 * then we have a bad program that isn't syncronizing its FPU usage
969 * and it will suffer the consequences since we won't be able to
970 * fully reproduce the context of the exception
972 cwd = get_fpu_cwd(task);
973 swd = get_fpu_swd(task);
974 switch (swd & ~cwd & 0x3f) {
975 case 0x000: /* No unmasked exception */
977 default: /* Multiple exceptions */
979 case 0x001: /* Invalid Op */
981 * swd & 0x240 == 0x040: Stack Underflow
982 * swd & 0x240 == 0x240: Stack Overflow
983 * User must clear the SF bit (0x40) if set
985 info.si_code = FPE_FLTINV;
987 case 0x002: /* Denormalize */
988 case 0x010: /* Underflow */
989 info.si_code = FPE_FLTUND;
991 case 0x004: /* Zero Divide */
992 info.si_code = FPE_FLTDIV;
994 case 0x008: /* Overflow */
995 info.si_code = FPE_FLTOVF;
997 case 0x020: /* Precision */
998 info.si_code = FPE_FLTRES;
1001 force_sig_info(SIGFPE, &info, task);
1004 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
1007 math_error((void __user *)regs->eip);
1010 static void simd_math_error(void __user *eip)
1012 struct task_struct * task;
1014 unsigned short mxcsr;
1017 * Save the info for the exception handler and clear the error.
1020 save_init_fpu(task);
1021 task->thread.trap_no = 19;
1022 task->thread.error_code = 0;
1023 info.si_signo = SIGFPE;
1025 info.si_code = __SI_FAULT;
1028 * The SIMD FPU exceptions are handled a little differently, as there
1029 * is only a single status/control register. Thus, to determine which
1030 * unmasked exception was caught we must mask the exception mask bits
1031 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1033 mxcsr = get_fpu_mxcsr(task);
1034 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1038 case 0x001: /* Invalid Op */
1039 info.si_code = FPE_FLTINV;
1041 case 0x002: /* Denormalize */
1042 case 0x010: /* Underflow */
1043 info.si_code = FPE_FLTUND;
1045 case 0x004: /* Zero Divide */
1046 info.si_code = FPE_FLTDIV;
1048 case 0x008: /* Overflow */
1049 info.si_code = FPE_FLTOVF;
1051 case 0x020: /* Precision */
1052 info.si_code = FPE_FLTRES;
1055 force_sig_info(SIGFPE, &info, task);
1058 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1062 /* Handle SIMD FPU exceptions on PIII+ processors. */
1064 simd_math_error((void __user *)regs->eip);
1067 * Handle strange cache flush from user space exception
1068 * in all other cases. This is undocumented behaviour.
1070 if (regs->eflags & VM_MASK) {
1071 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1075 current->thread.trap_no = 19;
1076 current->thread.error_code = error_code;
1077 die_if_kernel("cache flush denied", regs, error_code);
1078 force_sig(SIGSEGV, current);
1082 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1086 /* No need to warn about this any longer. */
1087 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1091 fastcall unsigned long patch_espfix_desc(unsigned long uesp,
1094 int cpu = smp_processor_id();
1095 struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu);
1096 struct desc_struct *gdt = (struct desc_struct *)cpu_gdt_descr->address;
1097 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1098 unsigned long new_kesp = kesp - base;
1099 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1100 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1101 /* Set up base for espfix segment */
1102 desc &= 0x00f0ff0000000000ULL;
1103 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1104 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1105 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1106 (lim_pages & 0xffff);
1107 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1112 * 'math_state_restore()' saves the current math information in the
1113 * old math state array, and gets the new ones from the current task
1115 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1116 * Don't touch unless you *really* know how it works.
1118 * Must be called with kernel preemption disabled (in this case,
1119 * local interrupts are disabled at the call-site in entry.S).
1121 asmlinkage void math_state_restore(void)
1123 struct thread_info *thread = current_thread_info();
1124 struct task_struct *tsk = thread->task;
1126 clts(); /* Allow maths ops (or we recurse) */
1127 if (!tsk_used_math(tsk))
1130 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1134 #ifndef CONFIG_MATH_EMULATION
1136 asmlinkage void math_emulate(long arg)
1138 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1139 printk(KERN_EMERG "killing %s.\n",current->comm);
1140 force_sig(SIGFPE,current);
1144 #endif /* CONFIG_MATH_EMULATION */
1146 #ifdef CONFIG_X86_F00F_BUG
1147 void __init trap_init_f00f_bug(void)
1149 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1152 * Update the IDT descriptor and reload the IDT so that
1153 * it uses the read-only mapped virtual address.
1155 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1156 load_idt(&idt_descr);
1161 * This needs to use 'idt_table' rather than 'idt', and
1162 * thus use the _nonmapped_ version of the IDT, as the
1163 * Pentium F0 0F bugfix can have resulted in the mapped
1164 * IDT being write-protected.
1166 void set_intr_gate(unsigned int n, void *addr)
1168 _set_gate(n, DESCTYPE_INT, addr, __KERNEL_CS);
1172 * This routine sets up an interrupt gate at directory privilege level 3.
1174 static inline void set_system_intr_gate(unsigned int n, void *addr)
1176 _set_gate(n, DESCTYPE_INT | DESCTYPE_DPL3, addr, __KERNEL_CS);
1179 static void __init set_trap_gate(unsigned int n, void *addr)
1181 _set_gate(n, DESCTYPE_TRAP, addr, __KERNEL_CS);
1184 static void __init set_system_gate(unsigned int n, void *addr)
1186 _set_gate(n, DESCTYPE_TRAP | DESCTYPE_DPL3, addr, __KERNEL_CS);
1189 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1191 _set_gate(n, DESCTYPE_TASK, (void *)0, (gdt_entry<<3));
1195 void __init trap_init(void)
1198 void __iomem *p = ioremap(0x0FFFD9, 4);
1199 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1205 #ifdef CONFIG_X86_LOCAL_APIC
1206 init_apic_mappings();
1209 set_trap_gate(0,÷_error);
1210 set_intr_gate(1,&debug);
1211 set_intr_gate(2,&nmi);
1212 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1213 set_system_gate(4,&overflow);
1214 set_trap_gate(5,&bounds);
1215 set_trap_gate(6,&invalid_op);
1216 set_trap_gate(7,&device_not_available);
1217 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1218 set_trap_gate(9,&coprocessor_segment_overrun);
1219 set_trap_gate(10,&invalid_TSS);
1220 set_trap_gate(11,&segment_not_present);
1221 set_trap_gate(12,&stack_segment);
1222 set_trap_gate(13,&general_protection);
1223 set_intr_gate(14,&page_fault);
1224 set_trap_gate(15,&spurious_interrupt_bug);
1225 set_trap_gate(16,&coprocessor_error);
1226 set_trap_gate(17,&alignment_check);
1227 #ifdef CONFIG_X86_MCE
1228 set_trap_gate(18,&machine_check);
1230 set_trap_gate(19,&simd_coprocessor_error);
1234 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1235 * Generates a compile-time "error: zero width for bit-field" if
1236 * the alignment is wrong.
1238 struct fxsrAlignAssert {
1239 int _:!(offsetof(struct task_struct,
1240 thread.i387.fxsave) & 15);
1243 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1244 set_in_cr4(X86_CR4_OSFXSR);
1248 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1250 set_in_cr4(X86_CR4_OSXMMEXCPT);
1254 set_system_gate(SYSCALL_VECTOR,&system_call);
1257 * Should be a barrier for any external CPU state.
1264 static int __init kstack_setup(char *s)
1266 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1269 __setup("kstack=", kstack_setup);
1271 #ifdef CONFIG_STACK_UNWIND
1272 static int __init call_trace_setup(char *s)
1274 if (strcmp(s, "old") == 0)
1276 else if (strcmp(s, "both") == 0)
1278 else if (strcmp(s, "newfallback") == 0)
1280 else if (strcmp(s, "new") == 2)
1284 __setup("call_trace=", call_trace_setup);