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)
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);
203 ops->warning(data, "Full inexact backtrace again:\n");
204 } else if (call_trace >= 1)
207 ops->warning(data, "Full inexact backtrace again:\n");
209 ops->warning(data, "Inexact backtrace:\n");
213 if (task && task != current)
214 stack = (unsigned long *)task->thread.esp;
217 if (task == current) {
218 /* Grab ebp right from our regs */
219 asm ("movl %%ebp, %0" : "=r" (ebp) : );
221 /* ebp is the last reg pushed by switch_to */
222 ebp = *(unsigned long *) task->thread.esp;
226 struct thread_info *context;
227 context = (struct thread_info *)
228 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
229 ebp = print_context_stack(context, stack, ebp, ops, data);
230 /* Should be after the line below, but somewhere
231 in early boot context comes out corrupted and we
232 can't reference it -AK */
233 if (ops->stack(data, "IRQ") < 0)
235 stack = (unsigned long*)context->previous_esp;
240 EXPORT_SYMBOL(dump_trace);
243 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
246 print_symbol(msg, symbol);
250 static void print_trace_warning(void *data, char *msg)
252 printk("%s%s\n", (char *)data, msg);
255 static int print_trace_stack(void *data, char *name)
261 * Print one address/symbol entries per line.
263 static void print_trace_address(void *data, unsigned long addr)
265 printk("%s [<%08lx>] ", (char *)data, addr);
266 print_symbol("%s\n", addr);
269 static struct stacktrace_ops print_trace_ops = {
270 .warning = print_trace_warning,
271 .warning_symbol = print_trace_warning_symbol,
272 .stack = print_trace_stack,
273 .address = print_trace_address,
277 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
278 unsigned long * stack, char *log_lvl)
280 dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
281 printk("%s =======================\n", log_lvl);
284 void show_trace(struct task_struct *task, struct pt_regs *regs,
285 unsigned long * stack)
287 show_trace_log_lvl(task, regs, stack, "");
290 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
291 unsigned long *esp, char *log_lvl)
293 unsigned long *stack;
298 esp = (unsigned long*)task->thread.esp;
300 esp = (unsigned long *)&esp;
304 for(i = 0; i < kstack_depth_to_print; i++) {
305 if (kstack_end(stack))
307 if (i && ((i % 8) == 0))
308 printk("\n%s ", log_lvl);
309 printk("%08lx ", *stack++);
311 printk("\n%sCall Trace:\n", log_lvl);
312 show_trace_log_lvl(task, regs, esp, log_lvl);
315 void show_stack(struct task_struct *task, unsigned long *esp)
318 show_stack_log_lvl(task, NULL, esp, "");
322 * The architecture-independent dump_stack generator
324 void dump_stack(void)
328 show_trace(current, NULL, &stack);
331 EXPORT_SYMBOL(dump_stack);
333 void show_registers(struct pt_regs *regs)
340 esp = (unsigned long) (®s->esp);
342 if (user_mode_vm(regs)) {
345 ss = regs->xss & 0xffff;
348 printk(KERN_EMERG "CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\n"
349 "EFLAGS: %08lx (%s %.*s) \n",
350 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
351 print_tainted(), regs->eflags, system_utsname.release,
352 (int)strcspn(system_utsname.version, " "),
353 system_utsname.version);
354 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
355 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
356 regs->eax, regs->ebx, regs->ecx, regs->edx);
357 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
358 regs->esi, regs->edi, regs->ebp, esp);
359 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
360 regs->xds & 0xffff, regs->xes & 0xffff, ss);
361 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
362 TASK_COMM_LEN, current->comm, current->pid,
363 current_thread_info(), current, current->thread_info);
365 * When in-kernel, we also print out the stack and code at the
366 * time of the fault..
371 printk("\n" KERN_EMERG "Stack: ");
372 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
374 printk(KERN_EMERG "Code: ");
376 eip = (u8 __user *)regs->eip - 43;
377 for (i = 0; i < 64; i++, eip++) {
380 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
381 printk(" Bad EIP value.");
384 if (eip == (u8 __user *)regs->eip)
385 printk("<%02x> ", c);
393 static void handle_BUG(struct pt_regs *regs)
395 unsigned long eip = regs->eip;
398 if (eip < PAGE_OFFSET)
400 if (__get_user(ud2, (unsigned short __user *)eip))
405 printk(KERN_EMERG "------------[ cut here ]------------\n");
407 #ifdef CONFIG_DEBUG_BUGVERBOSE
413 if (__get_user(line, (unsigned short __user *)(eip + 2)))
415 if (__get_user(file, (char * __user *)(eip + 4)) ||
416 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
417 file = "<bad filename>";
419 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
423 printk(KERN_EMERG "Kernel BUG at [verbose debug info unavailable]\n");
426 /* This is gone through when something in the kernel
427 * has done something bad and is about to be terminated.
429 void die(const char * str, struct pt_regs * regs, long err)
434 int lock_owner_depth;
436 .lock = SPIN_LOCK_UNLOCKED,
438 .lock_owner_depth = 0
440 static int die_counter;
445 if (die.lock_owner != raw_smp_processor_id()) {
447 spin_lock_irqsave(&die.lock, flags);
448 die.lock_owner = smp_processor_id();
449 die.lock_owner_depth = 0;
453 local_save_flags(flags);
455 if (++die.lock_owner_depth < 3) {
461 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
462 #ifdef CONFIG_PREEMPT
463 printk(KERN_EMERG "PREEMPT ");
472 #ifdef CONFIG_DEBUG_PAGEALLOC
475 printk("DEBUG_PAGEALLOC");
480 if (notify_die(DIE_OOPS, str, regs, err,
481 current->thread.trap_no, SIGSEGV) !=
483 show_registers(regs);
484 /* Executive summary in case the oops scrolled away */
485 esp = (unsigned long) (®s->esp);
487 if (user_mode(regs)) {
489 ss = regs->xss & 0xffff;
491 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
492 print_symbol("%s", regs->eip);
493 printk(" SS:ESP %04x:%08lx\n", ss, esp);
498 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
502 spin_unlock_irqrestore(&die.lock, flags);
507 if (kexec_should_crash(current))
511 panic("Fatal exception in interrupt");
514 panic("Fatal exception");
520 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
522 if (!user_mode_vm(regs))
526 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
527 struct pt_regs * regs, long error_code,
530 struct task_struct *tsk = current;
531 tsk->thread.error_code = error_code;
532 tsk->thread.trap_no = trapnr;
534 if (regs->eflags & VM_MASK) {
540 if (!user_mode(regs))
545 force_sig_info(signr, info, tsk);
547 force_sig(signr, tsk);
552 if (!fixup_exception(regs))
553 die(str, regs, error_code);
558 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
559 if (ret) goto trap_signal;
564 #define DO_ERROR(trapnr, signr, str, name) \
565 fastcall void do_##name(struct pt_regs * regs, long error_code) \
567 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
570 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
573 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
574 fastcall void do_##name(struct pt_regs * regs, long error_code) \
577 info.si_signo = signr; \
579 info.si_code = sicode; \
580 info.si_addr = (void __user *)siaddr; \
581 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
584 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
587 #define DO_VM86_ERROR(trapnr, signr, str, name) \
588 fastcall void do_##name(struct pt_regs * regs, long error_code) \
590 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
593 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
596 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
597 fastcall void do_##name(struct pt_regs * regs, long error_code) \
600 info.si_signo = signr; \
602 info.si_code = sicode; \
603 info.si_addr = (void __user *)siaddr; \
604 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
607 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
610 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
611 #ifndef CONFIG_KPROBES
612 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
614 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
615 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
616 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
617 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
618 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
619 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
620 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
621 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
622 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
624 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
628 struct tss_struct *tss = &per_cpu(init_tss, cpu);
629 struct thread_struct *thread = ¤t->thread;
632 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
633 * invalid offset set (the LAZY one) and the faulting thread has
634 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
635 * and we set the offset field correctly. Then we let the CPU to
636 * restart the faulting instruction.
638 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
639 thread->io_bitmap_ptr) {
640 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
641 thread->io_bitmap_max);
643 * If the previously set map was extending to higher ports
644 * than the current one, pad extra space with 0xff (no access).
646 if (thread->io_bitmap_max < tss->io_bitmap_max)
647 memset((char *) tss->io_bitmap +
648 thread->io_bitmap_max, 0xff,
649 tss->io_bitmap_max - thread->io_bitmap_max);
650 tss->io_bitmap_max = thread->io_bitmap_max;
651 tss->io_bitmap_base = IO_BITMAP_OFFSET;
652 tss->io_bitmap_owner = thread;
658 current->thread.error_code = error_code;
659 current->thread.trap_no = 13;
661 if (regs->eflags & VM_MASK)
664 if (!user_mode(regs))
667 current->thread.error_code = error_code;
668 current->thread.trap_no = 13;
669 force_sig(SIGSEGV, current);
674 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
678 if (!fixup_exception(regs)) {
679 if (notify_die(DIE_GPF, "general protection fault", regs,
680 error_code, 13, SIGSEGV) == NOTIFY_STOP)
682 die("general protection fault", regs, error_code);
686 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
688 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
689 "CPU %d.\n", reason, smp_processor_id());
690 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
692 if (panic_on_unrecovered_nmi)
693 panic("NMI: Not continuing");
695 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
697 /* Clear and disable the memory parity error line. */
698 clear_mem_error(reason);
701 static void io_check_error(unsigned char reason, struct pt_regs * regs)
705 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
706 show_registers(regs);
708 /* Re-enable the IOCK line, wait for a few seconds */
709 reason = (reason & 0xf) | 8;
712 while (--i) udelay(1000);
717 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
720 /* Might actually be able to figure out what the guilty party
727 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
728 "CPU %d.\n", reason, smp_processor_id());
729 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
730 if (panic_on_unrecovered_nmi)
731 panic("NMI: Not continuing");
733 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
736 static DEFINE_SPINLOCK(nmi_print_lock);
738 void die_nmi (struct pt_regs *regs, const char *msg)
740 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
744 spin_lock(&nmi_print_lock);
746 * We are in trouble anyway, lets at least try
747 * to get a message out.
750 printk(KERN_EMERG "%s", msg);
751 printk(" on CPU%d, eip %08lx, registers:\n",
752 smp_processor_id(), regs->eip);
753 show_registers(regs);
754 printk(KERN_EMERG "console shuts up ...\n");
756 spin_unlock(&nmi_print_lock);
759 /* If we are in kernel we are probably nested up pretty bad
760 * and might aswell get out now while we still can.
762 if (!user_mode_vm(regs)) {
763 current->thread.trap_no = 2;
770 static void default_do_nmi(struct pt_regs * regs)
772 unsigned char reason = 0;
774 /* Only the BSP gets external NMIs from the system. */
775 if (!smp_processor_id())
776 reason = get_nmi_reason();
778 if (!(reason & 0xc0)) {
779 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
782 #ifdef CONFIG_X86_LOCAL_APIC
784 * Ok, so this is none of the documented NMI sources,
785 * so it must be the NMI watchdog.
787 if (nmi_watchdog_tick(regs, reason))
789 if (!do_nmi_callback(regs, smp_processor_id()))
791 unknown_nmi_error(reason, regs);
795 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
798 mem_parity_error(reason, regs);
800 io_check_error(reason, regs);
802 * Reassert NMI in case it became active meanwhile
803 * as it's edge-triggered.
808 fastcall void do_nmi(struct pt_regs * regs, long error_code)
814 cpu = smp_processor_id();
818 default_do_nmi(regs);
823 #ifdef CONFIG_KPROBES
824 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
826 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
829 /* This is an interrupt gate, because kprobes wants interrupts
830 disabled. Normal trap handlers don't. */
831 restore_interrupts(regs);
832 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
837 * Our handling of the processor debug registers is non-trivial.
838 * We do not clear them on entry and exit from the kernel. Therefore
839 * it is possible to get a watchpoint trap here from inside the kernel.
840 * However, the code in ./ptrace.c has ensured that the user can
841 * only set watchpoints on userspace addresses. Therefore the in-kernel
842 * watchpoint trap can only occur in code which is reading/writing
843 * from user space. Such code must not hold kernel locks (since it
844 * can equally take a page fault), therefore it is safe to call
845 * force_sig_info even though that claims and releases locks.
847 * Code in ./signal.c ensures that the debug control register
848 * is restored before we deliver any signal, and therefore that
849 * user code runs with the correct debug control register even though
852 * Being careful here means that we don't have to be as careful in a
853 * lot of more complicated places (task switching can be a bit lazy
854 * about restoring all the debug state, and ptrace doesn't have to
855 * find every occurrence of the TF bit that could be saved away even
858 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
860 unsigned int condition;
861 struct task_struct *tsk = current;
863 get_debugreg(condition, 6);
865 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
866 SIGTRAP) == NOTIFY_STOP)
868 /* It's safe to allow irq's after DR6 has been saved */
869 if (regs->eflags & X86_EFLAGS_IF)
872 /* Mask out spurious debug traps due to lazy DR7 setting */
873 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
874 if (!tsk->thread.debugreg[7])
878 if (regs->eflags & VM_MASK)
881 /* Save debug status register where ptrace can see it */
882 tsk->thread.debugreg[6] = condition;
885 * Single-stepping through TF: make sure we ignore any events in
886 * kernel space (but re-enable TF when returning to user mode).
888 if (condition & DR_STEP) {
890 * We already checked v86 mode above, so we can
891 * check for kernel mode by just checking the CPL
894 if (!user_mode(regs))
895 goto clear_TF_reenable;
898 /* Ok, finally something we can handle */
899 send_sigtrap(tsk, regs, error_code);
901 /* Disable additional traps. They'll be re-enabled when
902 * the signal is delivered.
909 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
913 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
914 regs->eflags &= ~TF_MASK;
919 * Note that we play around with the 'TS' bit in an attempt to get
920 * the correct behaviour even in the presence of the asynchronous
923 void math_error(void __user *eip)
925 struct task_struct * task;
927 unsigned short cwd, swd;
930 * Save the info for the exception handler and clear the error.
934 task->thread.trap_no = 16;
935 task->thread.error_code = 0;
936 info.si_signo = SIGFPE;
938 info.si_code = __SI_FAULT;
941 * (~cwd & swd) will mask out exceptions that are not set to unmasked
942 * status. 0x3f is the exception bits in these regs, 0x200 is the
943 * C1 reg you need in case of a stack fault, 0x040 is the stack
944 * fault bit. We should only be taking one exception at a time,
945 * so if this combination doesn't produce any single exception,
946 * then we have a bad program that isn't syncronizing its FPU usage
947 * and it will suffer the consequences since we won't be able to
948 * fully reproduce the context of the exception
950 cwd = get_fpu_cwd(task);
951 swd = get_fpu_swd(task);
952 switch (swd & ~cwd & 0x3f) {
953 case 0x000: /* No unmasked exception */
955 default: /* Multiple exceptions */
957 case 0x001: /* Invalid Op */
959 * swd & 0x240 == 0x040: Stack Underflow
960 * swd & 0x240 == 0x240: Stack Overflow
961 * User must clear the SF bit (0x40) if set
963 info.si_code = FPE_FLTINV;
965 case 0x002: /* Denormalize */
966 case 0x010: /* Underflow */
967 info.si_code = FPE_FLTUND;
969 case 0x004: /* Zero Divide */
970 info.si_code = FPE_FLTDIV;
972 case 0x008: /* Overflow */
973 info.si_code = FPE_FLTOVF;
975 case 0x020: /* Precision */
976 info.si_code = FPE_FLTRES;
979 force_sig_info(SIGFPE, &info, task);
982 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
985 math_error((void __user *)regs->eip);
988 static void simd_math_error(void __user *eip)
990 struct task_struct * task;
992 unsigned short mxcsr;
995 * Save the info for the exception handler and clear the error.
999 task->thread.trap_no = 19;
1000 task->thread.error_code = 0;
1001 info.si_signo = SIGFPE;
1003 info.si_code = __SI_FAULT;
1006 * The SIMD FPU exceptions are handled a little differently, as there
1007 * is only a single status/control register. Thus, to determine which
1008 * unmasked exception was caught we must mask the exception mask bits
1009 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1011 mxcsr = get_fpu_mxcsr(task);
1012 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1016 case 0x001: /* Invalid Op */
1017 info.si_code = FPE_FLTINV;
1019 case 0x002: /* Denormalize */
1020 case 0x010: /* Underflow */
1021 info.si_code = FPE_FLTUND;
1023 case 0x004: /* Zero Divide */
1024 info.si_code = FPE_FLTDIV;
1026 case 0x008: /* Overflow */
1027 info.si_code = FPE_FLTOVF;
1029 case 0x020: /* Precision */
1030 info.si_code = FPE_FLTRES;
1033 force_sig_info(SIGFPE, &info, task);
1036 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1040 /* Handle SIMD FPU exceptions on PIII+ processors. */
1042 simd_math_error((void __user *)regs->eip);
1045 * Handle strange cache flush from user space exception
1046 * in all other cases. This is undocumented behaviour.
1048 if (regs->eflags & VM_MASK) {
1049 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1053 current->thread.trap_no = 19;
1054 current->thread.error_code = error_code;
1055 die_if_kernel("cache flush denied", regs, error_code);
1056 force_sig(SIGSEGV, current);
1060 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1064 /* No need to warn about this any longer. */
1065 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1069 fastcall void setup_x86_bogus_stack(unsigned char * stk)
1071 unsigned long *switch16_ptr, *switch32_ptr;
1072 struct pt_regs *regs;
1073 unsigned long stack_top, stack_bot;
1074 unsigned short iret_frame16_off;
1075 int cpu = smp_processor_id();
1076 /* reserve the space on 32bit stack for the magic switch16 pointer */
1077 memmove(stk, stk + 8, sizeof(struct pt_regs));
1078 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
1079 regs = (struct pt_regs *)stk;
1080 /* now the switch32 on 16bit stack */
1081 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1082 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1083 switch32_ptr = (unsigned long *)(stack_top - 8);
1084 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
1085 /* copy iret frame on 16bit stack */
1086 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
1087 /* fill in the switch pointers */
1088 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
1089 switch16_ptr[1] = __ESPFIX_SS;
1090 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
1091 8 - CPU_16BIT_STACK_SIZE;
1092 switch32_ptr[1] = __KERNEL_DS;
1095 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
1097 unsigned long *switch32_ptr;
1098 unsigned char *stack16, *stack32;
1099 unsigned long stack_top, stack_bot;
1101 int cpu = smp_processor_id();
1102 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1103 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1104 switch32_ptr = (unsigned long *)(stack_top - 8);
1105 /* copy the data from 16bit stack to 32bit stack */
1106 len = CPU_16BIT_STACK_SIZE - 8 - sp;
1107 stack16 = (unsigned char *)(stack_bot + sp);
1108 stack32 = (unsigned char *)
1109 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
1110 memcpy(stack32, stack16, len);
1115 * 'math_state_restore()' saves the current math information in the
1116 * old math state array, and gets the new ones from the current task
1118 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1119 * Don't touch unless you *really* know how it works.
1121 * Must be called with kernel preemption disabled (in this case,
1122 * local interrupts are disabled at the call-site in entry.S).
1124 asmlinkage void math_state_restore(struct pt_regs regs)
1126 struct thread_info *thread = current_thread_info();
1127 struct task_struct *tsk = thread->task;
1129 clts(); /* Allow maths ops (or we recurse) */
1130 if (!tsk_used_math(tsk))
1133 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1136 #ifndef CONFIG_MATH_EMULATION
1138 asmlinkage void math_emulate(long arg)
1140 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1141 printk(KERN_EMERG "killing %s.\n",current->comm);
1142 force_sig(SIGFPE,current);
1146 #endif /* CONFIG_MATH_EMULATION */
1148 #ifdef CONFIG_X86_F00F_BUG
1149 void __init trap_init_f00f_bug(void)
1151 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1154 * Update the IDT descriptor and reload the IDT so that
1155 * it uses the read-only mapped virtual address.
1157 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1158 load_idt(&idt_descr);
1162 #define _set_gate(gate_addr,type,dpl,addr,seg) \
1165 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1166 "movw %4,%%dx\n\t" \
1167 "movl %%eax,%0\n\t" \
1169 :"=m" (*((long *) (gate_addr))), \
1170 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1171 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1172 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1177 * This needs to use 'idt_table' rather than 'idt', and
1178 * thus use the _nonmapped_ version of the IDT, as the
1179 * Pentium F0 0F bugfix can have resulted in the mapped
1180 * IDT being write-protected.
1182 void set_intr_gate(unsigned int n, void *addr)
1184 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1188 * This routine sets up an interrupt gate at directory privilege level 3.
1190 static inline void set_system_intr_gate(unsigned int n, void *addr)
1192 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1195 static void __init set_trap_gate(unsigned int n, void *addr)
1197 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1200 static void __init set_system_gate(unsigned int n, void *addr)
1202 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1205 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1207 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1211 void __init trap_init(void)
1214 void __iomem *p = ioremap(0x0FFFD9, 4);
1215 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1221 #ifdef CONFIG_X86_LOCAL_APIC
1222 init_apic_mappings();
1225 set_trap_gate(0,÷_error);
1226 set_intr_gate(1,&debug);
1227 set_intr_gate(2,&nmi);
1228 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1229 set_system_gate(4,&overflow);
1230 set_trap_gate(5,&bounds);
1231 set_trap_gate(6,&invalid_op);
1232 set_trap_gate(7,&device_not_available);
1233 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1234 set_trap_gate(9,&coprocessor_segment_overrun);
1235 set_trap_gate(10,&invalid_TSS);
1236 set_trap_gate(11,&segment_not_present);
1237 set_trap_gate(12,&stack_segment);
1238 set_trap_gate(13,&general_protection);
1239 set_intr_gate(14,&page_fault);
1240 set_trap_gate(15,&spurious_interrupt_bug);
1241 set_trap_gate(16,&coprocessor_error);
1242 set_trap_gate(17,&alignment_check);
1243 #ifdef CONFIG_X86_MCE
1244 set_trap_gate(18,&machine_check);
1246 set_trap_gate(19,&simd_coprocessor_error);
1250 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1251 * Generates a compile-time "error: zero width for bit-field" if
1252 * the alignment is wrong.
1254 struct fxsrAlignAssert {
1255 int _:!(offsetof(struct task_struct,
1256 thread.i387.fxsave) & 15);
1259 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1260 set_in_cr4(X86_CR4_OSFXSR);
1264 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1266 set_in_cr4(X86_CR4_OSXMMEXCPT);
1270 set_system_gate(SYSCALL_VECTOR,&system_call);
1273 * Should be a barrier for any external CPU state.
1280 static int __init kstack_setup(char *s)
1282 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1285 __setup("kstack=", kstack_setup);
1287 #ifdef CONFIG_STACK_UNWIND
1288 static int __init call_trace_setup(char *s)
1290 if (strcmp(s, "old") == 0)
1292 else if (strcmp(s, "both") == 0)
1294 else if (strcmp(s, "newfallback") == 0)
1296 else if (strcmp(s, "new") == 2)
1300 __setup("call_trace=", call_trace_setup);