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 struct desc_struct default_ldt[] = { { 0, 0 }, { 0, 0 }, { 0, 0 },
67 /* Do we ignore FPU interrupts ? */
68 char ignore_fpu_irq = 0;
71 * The IDT has to be page-aligned to simplify the Pentium
72 * F0 0F bug workaround.. We have a special link segment
75 struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
77 asmlinkage void divide_error(void);
78 asmlinkage void debug(void);
79 asmlinkage void nmi(void);
80 asmlinkage void int3(void);
81 asmlinkage void overflow(void);
82 asmlinkage void bounds(void);
83 asmlinkage void invalid_op(void);
84 asmlinkage void device_not_available(void);
85 asmlinkage void coprocessor_segment_overrun(void);
86 asmlinkage void invalid_TSS(void);
87 asmlinkage void segment_not_present(void);
88 asmlinkage void stack_segment(void);
89 asmlinkage void general_protection(void);
90 asmlinkage void page_fault(void);
91 asmlinkage void coprocessor_error(void);
92 asmlinkage void simd_coprocessor_error(void);
93 asmlinkage void alignment_check(void);
94 asmlinkage void spurious_interrupt_bug(void);
95 asmlinkage void machine_check(void);
97 static int kstack_depth_to_print = 24;
98 #ifdef CONFIG_STACK_UNWIND
99 static int call_trace = 1;
101 #define call_trace (-1)
103 ATOMIC_NOTIFIER_HEAD(i386die_chain);
105 int register_die_notifier(struct notifier_block *nb)
108 return atomic_notifier_chain_register(&i386die_chain, nb);
110 EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
112 int unregister_die_notifier(struct notifier_block *nb)
114 return atomic_notifier_chain_unregister(&i386die_chain, nb);
116 EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
118 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
120 return p > (void *)tinfo &&
121 p < (void *)tinfo + THREAD_SIZE - 3;
124 static inline unsigned long print_context_stack(struct thread_info *tinfo,
125 unsigned long *stack, unsigned long ebp,
126 struct stacktrace_ops *ops, void *data)
130 #ifdef CONFIG_FRAME_POINTER
131 while (valid_stack_ptr(tinfo, (void *)ebp)) {
132 unsigned long new_ebp;
133 addr = *(unsigned long *)(ebp + 4);
134 ops->address(data, addr);
136 * break out of recursive entries (such as
137 * end_of_stack_stop_unwind_function). Also,
138 * we can never allow a frame pointer to
141 new_ebp = *(unsigned long *)ebp;
147 while (valid_stack_ptr(tinfo, stack)) {
149 if (__kernel_text_address(addr))
150 ops->address(data, addr);
156 struct ops_and_data {
157 struct stacktrace_ops *ops;
161 static asmlinkage int
162 dump_trace_unwind(struct unwind_frame_info *info, void *data)
164 struct ops_and_data *oad = (struct ops_and_data *)data;
167 while (unwind(info) == 0 && UNW_PC(info)) {
169 oad->ops->address(oad->data, UNW_PC(info));
170 if (arch_unw_user_mode(info))
176 void dump_trace(struct task_struct *task, struct pt_regs *regs,
177 unsigned long *stack,
178 struct stacktrace_ops *ops, void *data)
180 unsigned long ebp = 0;
185 if (call_trace >= 0) {
187 struct unwind_frame_info info;
188 struct ops_and_data oad = { .ops = ops, .data = data };
191 if (unwind_init_frame_info(&info, task, regs) == 0)
192 unw_ret = dump_trace_unwind(&info, &oad);
193 } else if (task == current)
194 unw_ret = unwind_init_running(&info, dump_trace_unwind, &oad);
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, "DWARF2 unwinder stuck at %s\n",
203 if (UNW_SP(&info) >= PAGE_OFFSET) {
204 ops->warning(data, "Leftover inexact backtrace:\n");
205 stack = (void *)UNW_SP(&info);
210 ops->warning(data, "Full inexact backtrace again:\n");
211 } else if (call_trace >= 1)
214 ops->warning(data, "Full inexact backtrace again:\n");
216 ops->warning(data, "Inexact backtrace:\n");
221 if (task && task != current)
222 stack = (unsigned long *)task->thread.esp;
225 #ifdef CONFIG_FRAME_POINTER
227 if (task == current) {
228 /* Grab ebp right from our regs */
229 asm ("movl %%ebp, %0" : "=r" (ebp) : );
231 /* ebp is the last reg pushed by switch_to */
232 ebp = *(unsigned long *) task->thread.esp;
238 struct thread_info *context;
239 context = (struct thread_info *)
240 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
241 ebp = print_context_stack(context, stack, ebp, ops, data);
242 /* Should be after the line below, but somewhere
243 in early boot context comes out corrupted and we
244 can't reference it -AK */
245 if (ops->stack(data, "IRQ") < 0)
247 stack = (unsigned long*)context->previous_esp;
252 EXPORT_SYMBOL(dump_trace);
255 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
258 print_symbol(msg, symbol);
262 static void print_trace_warning(void *data, char *msg)
264 printk("%s%s\n", (char *)data, msg);
267 static int print_trace_stack(void *data, char *name)
273 * Print one address/symbol entries per line.
275 static void print_trace_address(void *data, unsigned long addr)
277 printk("%s [<%08lx>] ", (char *)data, addr);
278 print_symbol("%s\n", addr);
281 static struct stacktrace_ops print_trace_ops = {
282 .warning = print_trace_warning,
283 .warning_symbol = print_trace_warning_symbol,
284 .stack = print_trace_stack,
285 .address = print_trace_address,
289 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
290 unsigned long * stack, char *log_lvl)
292 dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
293 printk("%s =======================\n", log_lvl);
296 void show_trace(struct task_struct *task, struct pt_regs *regs,
297 unsigned long * stack)
299 show_trace_log_lvl(task, regs, stack, "");
302 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
303 unsigned long *esp, char *log_lvl)
305 unsigned long *stack;
310 esp = (unsigned long*)task->thread.esp;
312 esp = (unsigned long *)&esp;
316 for(i = 0; i < kstack_depth_to_print; i++) {
317 if (kstack_end(stack))
319 if (i && ((i % 8) == 0))
320 printk("\n%s ", log_lvl);
321 printk("%08lx ", *stack++);
323 printk("\n%sCall Trace:\n", log_lvl);
324 show_trace_log_lvl(task, regs, esp, log_lvl);
327 void show_stack(struct task_struct *task, unsigned long *esp)
330 show_stack_log_lvl(task, NULL, esp, "");
334 * The architecture-independent dump_stack generator
336 void dump_stack(void)
340 show_trace(current, NULL, &stack);
343 EXPORT_SYMBOL(dump_stack);
345 void show_registers(struct pt_regs *regs)
352 esp = (unsigned long) (®s->esp);
354 if (user_mode_vm(regs)) {
357 ss = regs->xss & 0xffff;
360 printk(KERN_EMERG "CPU: %d\n"
361 KERN_EMERG "EIP: %04x:[<%08lx>] %s VLI\n"
362 KERN_EMERG "EFLAGS: %08lx (%s %.*s)\n",
363 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
364 print_tainted(), regs->eflags, init_utsname()->release,
365 (int)strcspn(init_utsname()->version, " "),
366 init_utsname()->version);
367 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
368 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
369 regs->eax, regs->ebx, regs->ecx, regs->edx);
370 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
371 regs->esi, regs->edi, regs->ebp, esp);
372 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
373 regs->xds & 0xffff, regs->xes & 0xffff, ss);
374 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
375 TASK_COMM_LEN, current->comm, current->pid,
376 current_thread_info(), current, current->thread_info);
378 * When in-kernel, we also print out the stack and code at the
379 * time of the fault..
386 printk("\n" KERN_EMERG "Stack: ");
387 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
389 printk(KERN_EMERG "Code: ");
391 eip = (u8 __user *)regs->eip - 43;
392 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
393 /* try starting at EIP */
394 eip = (u8 __user *)regs->eip;
397 for (i = 0; i < code_bytes; i++, eip++) {
398 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
399 printk(" Bad EIP value.");
402 if (eip == (u8 __user *)regs->eip)
403 printk("<%02x> ", c);
411 static void handle_BUG(struct pt_regs *regs)
413 unsigned long eip = regs->eip;
416 if (eip < PAGE_OFFSET)
418 if (probe_kernel_address((unsigned short __user *)eip, ud2))
423 printk(KERN_EMERG "------------[ cut here ]------------\n");
425 #ifdef CONFIG_DEBUG_BUGVERBOSE
431 if (probe_kernel_address((unsigned short __user *)(eip + 2),
434 if (__get_user(file, (char * __user *)(eip + 4)) ||
435 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
436 file = "<bad filename>";
438 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
442 printk(KERN_EMERG "Kernel BUG at [verbose debug info unavailable]\n");
445 /* This is gone through when something in the kernel
446 * has done something bad and is about to be terminated.
448 void die(const char * str, struct pt_regs * regs, long err)
453 int lock_owner_depth;
455 .lock = SPIN_LOCK_UNLOCKED,
457 .lock_owner_depth = 0
459 static int die_counter;
464 if (die.lock_owner != raw_smp_processor_id()) {
466 spin_lock_irqsave(&die.lock, flags);
467 die.lock_owner = smp_processor_id();
468 die.lock_owner_depth = 0;
472 local_save_flags(flags);
474 if (++die.lock_owner_depth < 3) {
480 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
481 #ifdef CONFIG_PREEMPT
482 printk(KERN_EMERG "PREEMPT ");
491 #ifdef CONFIG_DEBUG_PAGEALLOC
494 printk("DEBUG_PAGEALLOC");
499 if (notify_die(DIE_OOPS, str, regs, err,
500 current->thread.trap_no, SIGSEGV) !=
502 show_registers(regs);
503 /* Executive summary in case the oops scrolled away */
504 esp = (unsigned long) (®s->esp);
506 if (user_mode(regs)) {
508 ss = regs->xss & 0xffff;
510 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
511 print_symbol("%s", regs->eip);
512 printk(" SS:ESP %04x:%08lx\n", ss, esp);
517 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
521 spin_unlock_irqrestore(&die.lock, flags);
526 if (kexec_should_crash(current))
530 panic("Fatal exception in interrupt");
533 panic("Fatal exception");
539 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
541 if (!user_mode_vm(regs))
545 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
546 struct pt_regs * regs, long error_code,
549 struct task_struct *tsk = current;
550 tsk->thread.error_code = error_code;
551 tsk->thread.trap_no = trapnr;
553 if (regs->eflags & VM_MASK) {
559 if (!user_mode(regs))
564 force_sig_info(signr, info, tsk);
566 force_sig(signr, tsk);
571 if (!fixup_exception(regs))
572 die(str, regs, error_code);
577 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
578 if (ret) goto trap_signal;
583 #define DO_ERROR(trapnr, signr, str, name) \
584 fastcall void do_##name(struct pt_regs * regs, long error_code) \
586 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
589 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
592 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
593 fastcall void do_##name(struct pt_regs * regs, long error_code) \
596 info.si_signo = signr; \
598 info.si_code = sicode; \
599 info.si_addr = (void __user *)siaddr; \
600 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
603 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
606 #define DO_VM86_ERROR(trapnr, signr, str, name) \
607 fastcall void do_##name(struct pt_regs * regs, long error_code) \
609 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
612 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
615 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
616 fastcall void do_##name(struct pt_regs * regs, long error_code) \
619 info.si_signo = signr; \
621 info.si_code = sicode; \
622 info.si_addr = (void __user *)siaddr; \
623 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
626 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
629 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
630 #ifndef CONFIG_KPROBES
631 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
633 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
634 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
635 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
636 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
637 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
638 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
639 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
640 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
641 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
643 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
647 struct tss_struct *tss = &per_cpu(init_tss, cpu);
648 struct thread_struct *thread = ¤t->thread;
651 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
652 * invalid offset set (the LAZY one) and the faulting thread has
653 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
654 * and we set the offset field correctly. Then we let the CPU to
655 * restart the faulting instruction.
657 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
658 thread->io_bitmap_ptr) {
659 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
660 thread->io_bitmap_max);
662 * If the previously set map was extending to higher ports
663 * than the current one, pad extra space with 0xff (no access).
665 if (thread->io_bitmap_max < tss->io_bitmap_max)
666 memset((char *) tss->io_bitmap +
667 thread->io_bitmap_max, 0xff,
668 tss->io_bitmap_max - thread->io_bitmap_max);
669 tss->io_bitmap_max = thread->io_bitmap_max;
670 tss->io_bitmap_base = IO_BITMAP_OFFSET;
671 tss->io_bitmap_owner = thread;
677 current->thread.error_code = error_code;
678 current->thread.trap_no = 13;
680 if (regs->eflags & VM_MASK)
683 if (!user_mode(regs))
686 current->thread.error_code = error_code;
687 current->thread.trap_no = 13;
688 force_sig(SIGSEGV, current);
693 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
697 if (!fixup_exception(regs)) {
698 if (notify_die(DIE_GPF, "general protection fault", regs,
699 error_code, 13, SIGSEGV) == NOTIFY_STOP)
701 die("general protection fault", regs, error_code);
705 static __kprobes void
706 mem_parity_error(unsigned char reason, struct pt_regs * regs)
708 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
709 "CPU %d.\n", reason, smp_processor_id());
710 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
712 if (panic_on_unrecovered_nmi)
713 panic("NMI: Not continuing");
715 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
717 /* Clear and disable the memory parity error line. */
718 clear_mem_error(reason);
721 static __kprobes void
722 io_check_error(unsigned char reason, struct pt_regs * regs)
726 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
727 show_registers(regs);
729 /* Re-enable the IOCK line, wait for a few seconds */
730 reason = (reason & 0xf) | 8;
733 while (--i) udelay(1000);
738 static __kprobes void
739 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
742 /* Might actually be able to figure out what the guilty party
749 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
750 "CPU %d.\n", reason, smp_processor_id());
751 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
752 if (panic_on_unrecovered_nmi)
753 panic("NMI: Not continuing");
755 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
758 static DEFINE_SPINLOCK(nmi_print_lock);
760 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
762 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
766 spin_lock(&nmi_print_lock);
768 * We are in trouble anyway, lets at least try
769 * to get a message out.
772 printk(KERN_EMERG "%s", msg);
773 printk(" on CPU%d, eip %08lx, registers:\n",
774 smp_processor_id(), regs->eip);
775 show_registers(regs);
776 printk(KERN_EMERG "console shuts up ...\n");
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 void setup_x86_bogus_stack(unsigned char * stk)
1093 unsigned long *switch16_ptr, *switch32_ptr;
1094 struct pt_regs *regs;
1095 unsigned long stack_top, stack_bot;
1096 unsigned short iret_frame16_off;
1097 int cpu = smp_processor_id();
1098 /* reserve the space on 32bit stack for the magic switch16 pointer */
1099 memmove(stk, stk + 8, sizeof(struct pt_regs));
1100 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
1101 regs = (struct pt_regs *)stk;
1102 /* now the switch32 on 16bit stack */
1103 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1104 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1105 switch32_ptr = (unsigned long *)(stack_top - 8);
1106 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
1107 /* copy iret frame on 16bit stack */
1108 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
1109 /* fill in the switch pointers */
1110 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
1111 switch16_ptr[1] = __ESPFIX_SS;
1112 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
1113 8 - CPU_16BIT_STACK_SIZE;
1114 switch32_ptr[1] = __KERNEL_DS;
1117 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
1119 unsigned long *switch32_ptr;
1120 unsigned char *stack16, *stack32;
1121 unsigned long stack_top, stack_bot;
1123 int cpu = smp_processor_id();
1124 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1125 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1126 switch32_ptr = (unsigned long *)(stack_top - 8);
1127 /* copy the data from 16bit stack to 32bit stack */
1128 len = CPU_16BIT_STACK_SIZE - 8 - sp;
1129 stack16 = (unsigned char *)(stack_bot + sp);
1130 stack32 = (unsigned char *)
1131 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
1132 memcpy(stack32, stack16, len);
1137 * 'math_state_restore()' saves the current math information in the
1138 * old math state array, and gets the new ones from the current task
1140 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1141 * Don't touch unless you *really* know how it works.
1143 * Must be called with kernel preemption disabled (in this case,
1144 * local interrupts are disabled at the call-site in entry.S).
1146 asmlinkage void math_state_restore(struct pt_regs regs)
1148 struct thread_info *thread = current_thread_info();
1149 struct task_struct *tsk = thread->task;
1151 clts(); /* Allow maths ops (or we recurse) */
1152 if (!tsk_used_math(tsk))
1155 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1158 #ifndef CONFIG_MATH_EMULATION
1160 asmlinkage void math_emulate(long arg)
1162 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1163 printk(KERN_EMERG "killing %s.\n",current->comm);
1164 force_sig(SIGFPE,current);
1168 #endif /* CONFIG_MATH_EMULATION */
1170 #ifdef CONFIG_X86_F00F_BUG
1171 void __init trap_init_f00f_bug(void)
1173 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1176 * Update the IDT descriptor and reload the IDT so that
1177 * it uses the read-only mapped virtual address.
1179 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1180 load_idt(&idt_descr);
1185 * This needs to use 'idt_table' rather than 'idt', and
1186 * thus use the _nonmapped_ version of the IDT, as the
1187 * Pentium F0 0F bugfix can have resulted in the mapped
1188 * IDT being write-protected.
1190 void set_intr_gate(unsigned int n, void *addr)
1192 _set_gate(n, DESCTYPE_INT, addr, __KERNEL_CS);
1196 * This routine sets up an interrupt gate at directory privilege level 3.
1198 static inline void set_system_intr_gate(unsigned int n, void *addr)
1200 _set_gate(n, DESCTYPE_INT | DESCTYPE_DPL3, addr, __KERNEL_CS);
1203 static void __init set_trap_gate(unsigned int n, void *addr)
1205 _set_gate(n, DESCTYPE_TRAP, addr, __KERNEL_CS);
1208 static void __init set_system_gate(unsigned int n, void *addr)
1210 _set_gate(n, DESCTYPE_TRAP | DESCTYPE_DPL3, addr, __KERNEL_CS);
1213 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1215 _set_gate(n, DESCTYPE_TASK, (void *)0, (gdt_entry<<3));
1219 void __init trap_init(void)
1222 void __iomem *p = ioremap(0x0FFFD9, 4);
1223 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1229 #ifdef CONFIG_X86_LOCAL_APIC
1230 init_apic_mappings();
1233 set_trap_gate(0,÷_error);
1234 set_intr_gate(1,&debug);
1235 set_intr_gate(2,&nmi);
1236 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1237 set_system_gate(4,&overflow);
1238 set_trap_gate(5,&bounds);
1239 set_trap_gate(6,&invalid_op);
1240 set_trap_gate(7,&device_not_available);
1241 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1242 set_trap_gate(9,&coprocessor_segment_overrun);
1243 set_trap_gate(10,&invalid_TSS);
1244 set_trap_gate(11,&segment_not_present);
1245 set_trap_gate(12,&stack_segment);
1246 set_trap_gate(13,&general_protection);
1247 set_intr_gate(14,&page_fault);
1248 set_trap_gate(15,&spurious_interrupt_bug);
1249 set_trap_gate(16,&coprocessor_error);
1250 set_trap_gate(17,&alignment_check);
1251 #ifdef CONFIG_X86_MCE
1252 set_trap_gate(18,&machine_check);
1254 set_trap_gate(19,&simd_coprocessor_error);
1258 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1259 * Generates a compile-time "error: zero width for bit-field" if
1260 * the alignment is wrong.
1262 struct fxsrAlignAssert {
1263 int _:!(offsetof(struct task_struct,
1264 thread.i387.fxsave) & 15);
1267 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1268 set_in_cr4(X86_CR4_OSFXSR);
1272 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1274 set_in_cr4(X86_CR4_OSXMMEXCPT);
1278 set_system_gate(SYSCALL_VECTOR,&system_call);
1281 * Should be a barrier for any external CPU state.
1288 static int __init kstack_setup(char *s)
1290 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1293 __setup("kstack=", kstack_setup);
1295 #ifdef CONFIG_STACK_UNWIND
1296 static int __init call_trace_setup(char *s)
1298 if (strcmp(s, "old") == 0)
1300 else if (strcmp(s, "both") == 0)
1302 else if (strcmp(s, "newfallback") == 0)
1304 else if (strcmp(s, "new") == 2)
1308 __setup("call_trace=", call_trace_setup);