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 addr = *(unsigned long *)(ebp + 4);
133 ops->address(data, addr);
135 * break out of recursive entries (such as
136 * end_of_stack_stop_unwind_function):
138 if (ebp == *(unsigned long *)ebp)
140 ebp = *(unsigned long *)ebp;
143 while (valid_stack_ptr(tinfo, stack)) {
145 if (__kernel_text_address(addr))
146 ops->address(data, addr);
152 struct ops_and_data {
153 struct stacktrace_ops *ops;
157 static asmlinkage int
158 dump_trace_unwind(struct unwind_frame_info *info, void *data)
160 struct ops_and_data *oad = (struct ops_and_data *)data;
163 while (unwind(info) == 0 && UNW_PC(info)) {
165 oad->ops->address(oad->data, UNW_PC(info));
166 if (arch_unw_user_mode(info))
172 void dump_trace(struct task_struct *task, struct pt_regs *regs,
173 unsigned long *stack,
174 struct stacktrace_ops *ops, void *data)
176 unsigned long ebp = 0;
181 if (call_trace >= 0) {
183 struct unwind_frame_info info;
184 struct ops_and_data oad = { .ops = ops, .data = data };
187 if (unwind_init_frame_info(&info, task, regs) == 0)
188 unw_ret = dump_trace_unwind(&info, &oad);
189 } else if (task == current)
190 unw_ret = unwind_init_running(&info, dump_trace_unwind, &oad);
192 if (unwind_init_blocked(&info, task) == 0)
193 unw_ret = dump_trace_unwind(&info, &oad);
196 if (call_trace == 1 && !arch_unw_user_mode(&info)) {
197 ops->warning_symbol(data, "DWARF2 unwinder stuck at %s\n",
199 if (UNW_SP(&info) >= PAGE_OFFSET) {
200 ops->warning(data, "Leftover inexact backtrace:\n");
201 stack = (void *)UNW_SP(&info);
206 ops->warning(data, "Full inexact backtrace again:\n");
207 } else if (call_trace >= 1)
210 ops->warning(data, "Full inexact backtrace again:\n");
212 ops->warning(data, "Inexact backtrace:\n");
217 if (task && task != current)
218 stack = (unsigned long *)task->thread.esp;
221 #ifdef CONFIG_FRAME_POINTER
223 if (task == current) {
224 /* Grab ebp right from our regs */
225 asm ("movl %%ebp, %0" : "=r" (ebp) : );
227 /* ebp is the last reg pushed by switch_to */
228 ebp = *(unsigned long *) task->thread.esp;
234 struct thread_info *context;
235 context = (struct thread_info *)
236 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
237 ebp = print_context_stack(context, stack, ebp, ops, data);
238 /* Should be after the line below, but somewhere
239 in early boot context comes out corrupted and we
240 can't reference it -AK */
241 if (ops->stack(data, "IRQ") < 0)
243 stack = (unsigned long*)context->previous_esp;
248 EXPORT_SYMBOL(dump_trace);
251 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
254 print_symbol(msg, symbol);
258 static void print_trace_warning(void *data, char *msg)
260 printk("%s%s\n", (char *)data, msg);
263 static int print_trace_stack(void *data, char *name)
269 * Print one address/symbol entries per line.
271 static void print_trace_address(void *data, unsigned long addr)
273 printk("%s [<%08lx>] ", (char *)data, addr);
274 print_symbol("%s\n", addr);
277 static struct stacktrace_ops print_trace_ops = {
278 .warning = print_trace_warning,
279 .warning_symbol = print_trace_warning_symbol,
280 .stack = print_trace_stack,
281 .address = print_trace_address,
285 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
286 unsigned long * stack, char *log_lvl)
288 dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
289 printk("%s =======================\n", log_lvl);
292 void show_trace(struct task_struct *task, struct pt_regs *regs,
293 unsigned long * stack)
295 show_trace_log_lvl(task, regs, stack, "");
298 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
299 unsigned long *esp, char *log_lvl)
301 unsigned long *stack;
306 esp = (unsigned long*)task->thread.esp;
308 esp = (unsigned long *)&esp;
312 for(i = 0; i < kstack_depth_to_print; i++) {
313 if (kstack_end(stack))
315 if (i && ((i % 8) == 0))
316 printk("\n%s ", log_lvl);
317 printk("%08lx ", *stack++);
319 printk("\n%sCall Trace:\n", log_lvl);
320 show_trace_log_lvl(task, regs, esp, log_lvl);
323 void show_stack(struct task_struct *task, unsigned long *esp)
326 show_stack_log_lvl(task, NULL, esp, "");
330 * The architecture-independent dump_stack generator
332 void dump_stack(void)
336 show_trace(current, NULL, &stack);
339 EXPORT_SYMBOL(dump_stack);
341 void show_registers(struct pt_regs *regs)
348 esp = (unsigned long) (®s->esp);
350 if (user_mode_vm(regs)) {
353 ss = regs->xss & 0xffff;
356 printk(KERN_EMERG "CPU: %d\n"
357 KERN_EMERG "EIP: %04x:[<%08lx>] %s VLI\n"
358 KERN_EMERG "EFLAGS: %08lx (%s %.*s)\n",
359 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
360 print_tainted(), regs->eflags, init_utsname()->release,
361 (int)strcspn(init_utsname()->version, " "),
362 init_utsname()->version);
363 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
364 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
365 regs->eax, regs->ebx, regs->ecx, regs->edx);
366 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
367 regs->esi, regs->edi, regs->ebp, esp);
368 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
369 regs->xds & 0xffff, regs->xes & 0xffff, ss);
370 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
371 TASK_COMM_LEN, current->comm, current->pid,
372 current_thread_info(), current, current->thread_info);
374 * When in-kernel, we also print out the stack and code at the
375 * time of the fault..
382 printk("\n" KERN_EMERG "Stack: ");
383 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
385 printk(KERN_EMERG "Code: ");
387 eip = (u8 __user *)regs->eip - 43;
388 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
389 /* try starting at EIP */
390 eip = (u8 __user *)regs->eip;
393 for (i = 0; i < code_bytes; i++, eip++) {
394 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
395 printk(" Bad EIP value.");
398 if (eip == (u8 __user *)regs->eip)
399 printk("<%02x> ", c);
407 static void handle_BUG(struct pt_regs *regs)
409 unsigned long eip = regs->eip;
412 if (eip < PAGE_OFFSET)
414 if (probe_kernel_address((unsigned short __user *)eip, ud2))
419 printk(KERN_EMERG "------------[ cut here ]------------\n");
421 #ifdef CONFIG_DEBUG_BUGVERBOSE
427 if (probe_kernel_address((unsigned short __user *)(eip + 2),
430 if (__get_user(file, (char * __user *)(eip + 4)) ||
431 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
432 file = "<bad filename>";
434 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
438 printk(KERN_EMERG "Kernel BUG at [verbose debug info unavailable]\n");
441 /* This is gone through when something in the kernel
442 * has done something bad and is about to be terminated.
444 void die(const char * str, struct pt_regs * regs, long err)
449 int lock_owner_depth;
451 .lock = SPIN_LOCK_UNLOCKED,
453 .lock_owner_depth = 0
455 static int die_counter;
460 if (die.lock_owner != raw_smp_processor_id()) {
462 spin_lock_irqsave(&die.lock, flags);
463 die.lock_owner = smp_processor_id();
464 die.lock_owner_depth = 0;
468 local_save_flags(flags);
470 if (++die.lock_owner_depth < 3) {
476 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
477 #ifdef CONFIG_PREEMPT
478 printk(KERN_EMERG "PREEMPT ");
487 #ifdef CONFIG_DEBUG_PAGEALLOC
490 printk("DEBUG_PAGEALLOC");
495 if (notify_die(DIE_OOPS, str, regs, err,
496 current->thread.trap_no, SIGSEGV) !=
498 show_registers(regs);
499 /* Executive summary in case the oops scrolled away */
500 esp = (unsigned long) (®s->esp);
502 if (user_mode(regs)) {
504 ss = regs->xss & 0xffff;
506 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
507 print_symbol("%s", regs->eip);
508 printk(" SS:ESP %04x:%08lx\n", ss, esp);
513 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
517 spin_unlock_irqrestore(&die.lock, flags);
522 if (kexec_should_crash(current))
526 panic("Fatal exception in interrupt");
529 panic("Fatal exception");
535 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
537 if (!user_mode_vm(regs))
541 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
542 struct pt_regs * regs, long error_code,
545 struct task_struct *tsk = current;
546 tsk->thread.error_code = error_code;
547 tsk->thread.trap_no = trapnr;
549 if (regs->eflags & VM_MASK) {
555 if (!user_mode(regs))
560 force_sig_info(signr, info, tsk);
562 force_sig(signr, tsk);
567 if (!fixup_exception(regs))
568 die(str, regs, error_code);
573 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
574 if (ret) goto trap_signal;
579 #define DO_ERROR(trapnr, signr, str, name) \
580 fastcall void do_##name(struct pt_regs * regs, long error_code) \
582 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
585 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
588 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
589 fastcall void do_##name(struct pt_regs * regs, long error_code) \
592 info.si_signo = signr; \
594 info.si_code = sicode; \
595 info.si_addr = (void __user *)siaddr; \
596 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
599 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
602 #define DO_VM86_ERROR(trapnr, signr, str, name) \
603 fastcall void do_##name(struct pt_regs * regs, long error_code) \
605 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
608 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
611 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
612 fastcall void do_##name(struct pt_regs * regs, long error_code) \
615 info.si_signo = signr; \
617 info.si_code = sicode; \
618 info.si_addr = (void __user *)siaddr; \
619 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
622 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
625 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
626 #ifndef CONFIG_KPROBES
627 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
629 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
630 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
631 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
632 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
633 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
634 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
635 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
636 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
637 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
639 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
643 struct tss_struct *tss = &per_cpu(init_tss, cpu);
644 struct thread_struct *thread = ¤t->thread;
647 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
648 * invalid offset set (the LAZY one) and the faulting thread has
649 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
650 * and we set the offset field correctly. Then we let the CPU to
651 * restart the faulting instruction.
653 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
654 thread->io_bitmap_ptr) {
655 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
656 thread->io_bitmap_max);
658 * If the previously set map was extending to higher ports
659 * than the current one, pad extra space with 0xff (no access).
661 if (thread->io_bitmap_max < tss->io_bitmap_max)
662 memset((char *) tss->io_bitmap +
663 thread->io_bitmap_max, 0xff,
664 tss->io_bitmap_max - thread->io_bitmap_max);
665 tss->io_bitmap_max = thread->io_bitmap_max;
666 tss->io_bitmap_base = IO_BITMAP_OFFSET;
667 tss->io_bitmap_owner = thread;
673 current->thread.error_code = error_code;
674 current->thread.trap_no = 13;
676 if (regs->eflags & VM_MASK)
679 if (!user_mode(regs))
682 current->thread.error_code = error_code;
683 current->thread.trap_no = 13;
684 force_sig(SIGSEGV, current);
689 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
693 if (!fixup_exception(regs)) {
694 if (notify_die(DIE_GPF, "general protection fault", regs,
695 error_code, 13, SIGSEGV) == NOTIFY_STOP)
697 die("general protection fault", regs, error_code);
701 static __kprobes void
702 mem_parity_error(unsigned char reason, struct pt_regs * regs)
704 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
705 "CPU %d.\n", reason, smp_processor_id());
706 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
708 if (panic_on_unrecovered_nmi)
709 panic("NMI: Not continuing");
711 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
713 /* Clear and disable the memory parity error line. */
714 clear_mem_error(reason);
717 static __kprobes void
718 io_check_error(unsigned char reason, struct pt_regs * regs)
722 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
723 show_registers(regs);
725 /* Re-enable the IOCK line, wait for a few seconds */
726 reason = (reason & 0xf) | 8;
729 while (--i) udelay(1000);
734 static __kprobes void
735 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
738 /* Might actually be able to figure out what the guilty party
745 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
746 "CPU %d.\n", reason, smp_processor_id());
747 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
748 if (panic_on_unrecovered_nmi)
749 panic("NMI: Not continuing");
751 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
754 static DEFINE_SPINLOCK(nmi_print_lock);
756 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
758 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
762 spin_lock(&nmi_print_lock);
764 * We are in trouble anyway, lets at least try
765 * to get a message out.
768 printk(KERN_EMERG "%s", msg);
769 printk(" on CPU%d, eip %08lx, registers:\n",
770 smp_processor_id(), regs->eip);
771 show_registers(regs);
772 printk(KERN_EMERG "console shuts up ...\n");
774 spin_unlock(&nmi_print_lock);
777 /* If we are in kernel we are probably nested up pretty bad
778 * and might aswell get out now while we still can.
780 if (!user_mode_vm(regs)) {
781 current->thread.trap_no = 2;
788 static __kprobes void default_do_nmi(struct pt_regs * regs)
790 unsigned char reason = 0;
792 /* Only the BSP gets external NMIs from the system. */
793 if (!smp_processor_id())
794 reason = get_nmi_reason();
796 if (!(reason & 0xc0)) {
797 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
800 #ifdef CONFIG_X86_LOCAL_APIC
802 * Ok, so this is none of the documented NMI sources,
803 * so it must be the NMI watchdog.
805 if (nmi_watchdog_tick(regs, reason))
807 if (!do_nmi_callback(regs, smp_processor_id()))
809 unknown_nmi_error(reason, regs);
813 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
816 mem_parity_error(reason, regs);
818 io_check_error(reason, regs);
820 * Reassert NMI in case it became active meanwhile
821 * as it's edge-triggered.
826 fastcall __kprobes void do_nmi(struct pt_regs * regs, long error_code)
832 cpu = smp_processor_id();
836 default_do_nmi(regs);
841 #ifdef CONFIG_KPROBES
842 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
844 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
847 /* This is an interrupt gate, because kprobes wants interrupts
848 disabled. Normal trap handlers don't. */
849 restore_interrupts(regs);
850 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
855 * Our handling of the processor debug registers is non-trivial.
856 * We do not clear them on entry and exit from the kernel. Therefore
857 * it is possible to get a watchpoint trap here from inside the kernel.
858 * However, the code in ./ptrace.c has ensured that the user can
859 * only set watchpoints on userspace addresses. Therefore the in-kernel
860 * watchpoint trap can only occur in code which is reading/writing
861 * from user space. Such code must not hold kernel locks (since it
862 * can equally take a page fault), therefore it is safe to call
863 * force_sig_info even though that claims and releases locks.
865 * Code in ./signal.c ensures that the debug control register
866 * is restored before we deliver any signal, and therefore that
867 * user code runs with the correct debug control register even though
870 * Being careful here means that we don't have to be as careful in a
871 * lot of more complicated places (task switching can be a bit lazy
872 * about restoring all the debug state, and ptrace doesn't have to
873 * find every occurrence of the TF bit that could be saved away even
876 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
878 unsigned int condition;
879 struct task_struct *tsk = current;
881 get_debugreg(condition, 6);
883 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
884 SIGTRAP) == NOTIFY_STOP)
886 /* It's safe to allow irq's after DR6 has been saved */
887 if (regs->eflags & X86_EFLAGS_IF)
890 /* Mask out spurious debug traps due to lazy DR7 setting */
891 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
892 if (!tsk->thread.debugreg[7])
896 if (regs->eflags & VM_MASK)
899 /* Save debug status register where ptrace can see it */
900 tsk->thread.debugreg[6] = condition;
903 * Single-stepping through TF: make sure we ignore any events in
904 * kernel space (but re-enable TF when returning to user mode).
906 if (condition & DR_STEP) {
908 * We already checked v86 mode above, so we can
909 * check for kernel mode by just checking the CPL
912 if (!user_mode(regs))
913 goto clear_TF_reenable;
916 /* Ok, finally something we can handle */
917 send_sigtrap(tsk, regs, error_code);
919 /* Disable additional traps. They'll be re-enabled when
920 * the signal is delivered.
927 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
931 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
932 regs->eflags &= ~TF_MASK;
937 * Note that we play around with the 'TS' bit in an attempt to get
938 * the correct behaviour even in the presence of the asynchronous
941 void math_error(void __user *eip)
943 struct task_struct * task;
945 unsigned short cwd, swd;
948 * Save the info for the exception handler and clear the error.
952 task->thread.trap_no = 16;
953 task->thread.error_code = 0;
954 info.si_signo = SIGFPE;
956 info.si_code = __SI_FAULT;
959 * (~cwd & swd) will mask out exceptions that are not set to unmasked
960 * status. 0x3f is the exception bits in these regs, 0x200 is the
961 * C1 reg you need in case of a stack fault, 0x040 is the stack
962 * fault bit. We should only be taking one exception at a time,
963 * so if this combination doesn't produce any single exception,
964 * then we have a bad program that isn't syncronizing its FPU usage
965 * and it will suffer the consequences since we won't be able to
966 * fully reproduce the context of the exception
968 cwd = get_fpu_cwd(task);
969 swd = get_fpu_swd(task);
970 switch (swd & ~cwd & 0x3f) {
971 case 0x000: /* No unmasked exception */
973 default: /* Multiple exceptions */
975 case 0x001: /* Invalid Op */
977 * swd & 0x240 == 0x040: Stack Underflow
978 * swd & 0x240 == 0x240: Stack Overflow
979 * User must clear the SF bit (0x40) if set
981 info.si_code = FPE_FLTINV;
983 case 0x002: /* Denormalize */
984 case 0x010: /* Underflow */
985 info.si_code = FPE_FLTUND;
987 case 0x004: /* Zero Divide */
988 info.si_code = FPE_FLTDIV;
990 case 0x008: /* Overflow */
991 info.si_code = FPE_FLTOVF;
993 case 0x020: /* Precision */
994 info.si_code = FPE_FLTRES;
997 force_sig_info(SIGFPE, &info, task);
1000 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
1003 math_error((void __user *)regs->eip);
1006 static void simd_math_error(void __user *eip)
1008 struct task_struct * task;
1010 unsigned short mxcsr;
1013 * Save the info for the exception handler and clear the error.
1016 save_init_fpu(task);
1017 task->thread.trap_no = 19;
1018 task->thread.error_code = 0;
1019 info.si_signo = SIGFPE;
1021 info.si_code = __SI_FAULT;
1024 * The SIMD FPU exceptions are handled a little differently, as there
1025 * is only a single status/control register. Thus, to determine which
1026 * unmasked exception was caught we must mask the exception mask bits
1027 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1029 mxcsr = get_fpu_mxcsr(task);
1030 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1034 case 0x001: /* Invalid Op */
1035 info.si_code = FPE_FLTINV;
1037 case 0x002: /* Denormalize */
1038 case 0x010: /* Underflow */
1039 info.si_code = FPE_FLTUND;
1041 case 0x004: /* Zero Divide */
1042 info.si_code = FPE_FLTDIV;
1044 case 0x008: /* Overflow */
1045 info.si_code = FPE_FLTOVF;
1047 case 0x020: /* Precision */
1048 info.si_code = FPE_FLTRES;
1051 force_sig_info(SIGFPE, &info, task);
1054 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1058 /* Handle SIMD FPU exceptions on PIII+ processors. */
1060 simd_math_error((void __user *)regs->eip);
1063 * Handle strange cache flush from user space exception
1064 * in all other cases. This is undocumented behaviour.
1066 if (regs->eflags & VM_MASK) {
1067 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1071 current->thread.trap_no = 19;
1072 current->thread.error_code = error_code;
1073 die_if_kernel("cache flush denied", regs, error_code);
1074 force_sig(SIGSEGV, current);
1078 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1082 /* No need to warn about this any longer. */
1083 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1087 fastcall void setup_x86_bogus_stack(unsigned char * stk)
1089 unsigned long *switch16_ptr, *switch32_ptr;
1090 struct pt_regs *regs;
1091 unsigned long stack_top, stack_bot;
1092 unsigned short iret_frame16_off;
1093 int cpu = smp_processor_id();
1094 /* reserve the space on 32bit stack for the magic switch16 pointer */
1095 memmove(stk, stk + 8, sizeof(struct pt_regs));
1096 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
1097 regs = (struct pt_regs *)stk;
1098 /* now the switch32 on 16bit stack */
1099 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1100 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1101 switch32_ptr = (unsigned long *)(stack_top - 8);
1102 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
1103 /* copy iret frame on 16bit stack */
1104 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
1105 /* fill in the switch pointers */
1106 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
1107 switch16_ptr[1] = __ESPFIX_SS;
1108 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
1109 8 - CPU_16BIT_STACK_SIZE;
1110 switch32_ptr[1] = __KERNEL_DS;
1113 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
1115 unsigned long *switch32_ptr;
1116 unsigned char *stack16, *stack32;
1117 unsigned long stack_top, stack_bot;
1119 int cpu = smp_processor_id();
1120 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1121 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1122 switch32_ptr = (unsigned long *)(stack_top - 8);
1123 /* copy the data from 16bit stack to 32bit stack */
1124 len = CPU_16BIT_STACK_SIZE - 8 - sp;
1125 stack16 = (unsigned char *)(stack_bot + sp);
1126 stack32 = (unsigned char *)
1127 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
1128 memcpy(stack32, stack16, len);
1133 * 'math_state_restore()' saves the current math information in the
1134 * old math state array, and gets the new ones from the current task
1136 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1137 * Don't touch unless you *really* know how it works.
1139 * Must be called with kernel preemption disabled (in this case,
1140 * local interrupts are disabled at the call-site in entry.S).
1142 asmlinkage void math_state_restore(struct pt_regs regs)
1144 struct thread_info *thread = current_thread_info();
1145 struct task_struct *tsk = thread->task;
1147 clts(); /* Allow maths ops (or we recurse) */
1148 if (!tsk_used_math(tsk))
1151 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1154 #ifndef CONFIG_MATH_EMULATION
1156 asmlinkage void math_emulate(long arg)
1158 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1159 printk(KERN_EMERG "killing %s.\n",current->comm);
1160 force_sig(SIGFPE,current);
1164 #endif /* CONFIG_MATH_EMULATION */
1166 #ifdef CONFIG_X86_F00F_BUG
1167 void __init trap_init_f00f_bug(void)
1169 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1172 * Update the IDT descriptor and reload the IDT so that
1173 * it uses the read-only mapped virtual address.
1175 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1176 load_idt(&idt_descr);
1181 * This needs to use 'idt_table' rather than 'idt', and
1182 * thus use the _nonmapped_ version of the IDT, as the
1183 * Pentium F0 0F bugfix can have resulted in the mapped
1184 * IDT being write-protected.
1186 void set_intr_gate(unsigned int n, void *addr)
1188 _set_gate(n, DESCTYPE_INT, addr, __KERNEL_CS);
1192 * This routine sets up an interrupt gate at directory privilege level 3.
1194 static inline void set_system_intr_gate(unsigned int n, void *addr)
1196 _set_gate(n, DESCTYPE_INT | DESCTYPE_DPL3, addr, __KERNEL_CS);
1199 static void __init set_trap_gate(unsigned int n, void *addr)
1201 _set_gate(n, DESCTYPE_TRAP, addr, __KERNEL_CS);
1204 static void __init set_system_gate(unsigned int n, void *addr)
1206 _set_gate(n, DESCTYPE_TRAP | DESCTYPE_DPL3, addr, __KERNEL_CS);
1209 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1211 _set_gate(n, DESCTYPE_TASK, (void *)0, (gdt_entry<<3));
1215 void __init trap_init(void)
1218 void __iomem *p = ioremap(0x0FFFD9, 4);
1219 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1225 #ifdef CONFIG_X86_LOCAL_APIC
1226 init_apic_mappings();
1229 set_trap_gate(0,÷_error);
1230 set_intr_gate(1,&debug);
1231 set_intr_gate(2,&nmi);
1232 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1233 set_system_gate(4,&overflow);
1234 set_trap_gate(5,&bounds);
1235 set_trap_gate(6,&invalid_op);
1236 set_trap_gate(7,&device_not_available);
1237 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1238 set_trap_gate(9,&coprocessor_segment_overrun);
1239 set_trap_gate(10,&invalid_TSS);
1240 set_trap_gate(11,&segment_not_present);
1241 set_trap_gate(12,&stack_segment);
1242 set_trap_gate(13,&general_protection);
1243 set_intr_gate(14,&page_fault);
1244 set_trap_gate(15,&spurious_interrupt_bug);
1245 set_trap_gate(16,&coprocessor_error);
1246 set_trap_gate(17,&alignment_check);
1247 #ifdef CONFIG_X86_MCE
1248 set_trap_gate(18,&machine_check);
1250 set_trap_gate(19,&simd_coprocessor_error);
1254 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1255 * Generates a compile-time "error: zero width for bit-field" if
1256 * the alignment is wrong.
1258 struct fxsrAlignAssert {
1259 int _:!(offsetof(struct task_struct,
1260 thread.i387.fxsave) & 15);
1263 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1264 set_in_cr4(X86_CR4_OSFXSR);
1268 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1270 set_in_cr4(X86_CR4_OSXMMEXCPT);
1274 set_system_gate(SYSCALL_VECTOR,&system_call);
1277 * Should be a barrier for any external CPU state.
1284 static int __init kstack_setup(char *s)
1286 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1289 __setup("kstack=", kstack_setup);
1291 #ifdef CONFIG_STACK_UNWIND
1292 static int __init call_trace_setup(char *s)
1294 if (strcmp(s, "old") == 0)
1296 else if (strcmp(s, "both") == 0)
1298 else if (strcmp(s, "newfallback") == 0)
1300 else if (strcmp(s, "new") == 2)
1304 __setup("call_trace=", call_trace_setup);