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>
32 #include <linux/nmi.h>
33 #include <linux/bug.h>
36 #include <linux/ioport.h>
37 #include <linux/eisa.h>
41 #include <linux/mca.h>
44 #include <asm/processor.h>
45 #include <asm/system.h>
47 #include <asm/atomic.h>
48 #include <asm/debugreg.h>
52 #include <asm/unwind.h>
54 #include <asm/arch_hooks.h>
55 #include <asm/kdebug.h>
56 #include <asm/stacktrace.h>
58 #include <linux/module.h>
60 #include "mach_traps.h"
62 int panic_on_unrecovered_nmi;
64 asmlinkage int system_call(void);
66 /* Do we ignore FPU interrupts ? */
67 char ignore_fpu_irq = 0;
70 * The IDT has to be page-aligned to simplify the Pentium
71 * F0 0F bug workaround.. We have a special link segment
74 struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
76 asmlinkage void divide_error(void);
77 asmlinkage void debug(void);
78 asmlinkage void nmi(void);
79 asmlinkage void int3(void);
80 asmlinkage void overflow(void);
81 asmlinkage void bounds(void);
82 asmlinkage void invalid_op(void);
83 asmlinkage void device_not_available(void);
84 asmlinkage void coprocessor_segment_overrun(void);
85 asmlinkage void invalid_TSS(void);
86 asmlinkage void segment_not_present(void);
87 asmlinkage void stack_segment(void);
88 asmlinkage void general_protection(void);
89 asmlinkage void page_fault(void);
90 asmlinkage void coprocessor_error(void);
91 asmlinkage void simd_coprocessor_error(void);
92 asmlinkage void alignment_check(void);
93 asmlinkage void spurious_interrupt_bug(void);
94 asmlinkage void machine_check(void);
96 int kstack_depth_to_print = 24;
97 #ifdef CONFIG_STACK_UNWIND
98 static int call_trace = 1;
100 #define call_trace (-1)
102 ATOMIC_NOTIFIER_HEAD(i386die_chain);
104 int register_die_notifier(struct notifier_block *nb)
107 return atomic_notifier_chain_register(&i386die_chain, nb);
109 EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
111 int unregister_die_notifier(struct notifier_block *nb)
113 return atomic_notifier_chain_unregister(&i386die_chain, nb);
115 EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
117 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
119 return p > (void *)tinfo &&
120 p < (void *)tinfo + THREAD_SIZE - 3;
123 static inline unsigned long print_context_stack(struct thread_info *tinfo,
124 unsigned long *stack, unsigned long ebp,
125 struct stacktrace_ops *ops, void *data)
129 #ifdef CONFIG_FRAME_POINTER
130 while (valid_stack_ptr(tinfo, (void *)ebp)) {
131 unsigned long new_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). Also,
137 * we can never allow a frame pointer to
140 new_ebp = *(unsigned long *)ebp;
146 while (valid_stack_ptr(tinfo, stack)) {
148 if (__kernel_text_address(addr))
149 ops->address(data, addr);
155 struct ops_and_data {
156 struct stacktrace_ops *ops;
160 static asmlinkage int
161 dump_trace_unwind(struct unwind_frame_info *info, void *data)
163 struct ops_and_data *oad = (struct ops_and_data *)data;
165 unsigned long sp = UNW_SP(info);
167 if (arch_unw_user_mode(info))
169 while (unwind(info) == 0 && UNW_PC(info)) {
171 oad->ops->address(oad->data, UNW_PC(info));
172 if (arch_unw_user_mode(info))
174 if ((sp & ~(PAGE_SIZE - 1)) == (UNW_SP(info) & ~(PAGE_SIZE - 1))
175 && sp > UNW_SP(info))
182 #define MSG(msg) ops->warning(data, msg)
184 void dump_trace(struct task_struct *task, struct pt_regs *regs,
185 unsigned long *stack,
186 struct stacktrace_ops *ops, void *data)
188 unsigned long ebp = 0;
193 if (call_trace >= 0) {
195 struct unwind_frame_info info;
196 struct ops_and_data oad = { .ops = ops, .data = data };
199 if (unwind_init_frame_info(&info, task, regs) == 0)
200 unw_ret = dump_trace_unwind(&info, &oad);
201 } else if (task == current)
202 unw_ret = unwind_init_running(&info, dump_trace_unwind,
205 if (unwind_init_blocked(&info, task) == 0)
206 unw_ret = dump_trace_unwind(&info, &oad);
209 if (call_trace == 1 && !arch_unw_user_mode(&info)) {
210 ops->warning_symbol(data,
211 "DWARF2 unwinder stuck at %s",
213 if (UNW_SP(&info) >= PAGE_OFFSET) {
214 MSG("Leftover inexact backtrace:");
215 stack = (void *)UNW_SP(&info);
220 MSG("Full inexact backtrace again:");
221 } else if (call_trace >= 1)
224 MSG("Full inexact backtrace again:");
226 MSG("Inexact backtrace:");
231 if (task && task != current)
232 stack = (unsigned long *)task->thread.esp;
235 #ifdef CONFIG_FRAME_POINTER
237 if (task == current) {
238 /* Grab ebp right from our regs */
239 asm ("movl %%ebp, %0" : "=r" (ebp) : );
241 /* ebp is the last reg pushed by switch_to */
242 ebp = *(unsigned long *) task->thread.esp;
248 struct thread_info *context;
249 context = (struct thread_info *)
250 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
251 ebp = print_context_stack(context, stack, ebp, ops, data);
252 /* Should be after the line below, but somewhere
253 in early boot context comes out corrupted and we
254 can't reference it -AK */
255 if (ops->stack(data, "IRQ") < 0)
257 stack = (unsigned long*)context->previous_esp;
260 touch_nmi_watchdog();
263 EXPORT_SYMBOL(dump_trace);
266 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
269 print_symbol(msg, symbol);
273 static void print_trace_warning(void *data, char *msg)
275 printk("%s%s\n", (char *)data, msg);
278 static int print_trace_stack(void *data, char *name)
284 * Print one address/symbol entries per line.
286 static void print_trace_address(void *data, unsigned long addr)
288 printk("%s [<%08lx>] ", (char *)data, addr);
289 print_symbol("%s\n", addr);
292 static struct stacktrace_ops print_trace_ops = {
293 .warning = print_trace_warning,
294 .warning_symbol = print_trace_warning_symbol,
295 .stack = print_trace_stack,
296 .address = print_trace_address,
300 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
301 unsigned long * stack, char *log_lvl)
303 dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
304 printk("%s =======================\n", log_lvl);
307 void show_trace(struct task_struct *task, struct pt_regs *regs,
308 unsigned long * stack)
310 show_trace_log_lvl(task, regs, stack, "");
313 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
314 unsigned long *esp, char *log_lvl)
316 unsigned long *stack;
321 esp = (unsigned long*)task->thread.esp;
323 esp = (unsigned long *)&esp;
327 for(i = 0; i < kstack_depth_to_print; i++) {
328 if (kstack_end(stack))
330 if (i && ((i % 8) == 0))
331 printk("\n%s ", log_lvl);
332 printk("%08lx ", *stack++);
334 printk("\n%sCall Trace:\n", log_lvl);
335 show_trace_log_lvl(task, regs, esp, log_lvl);
338 void show_stack(struct task_struct *task, unsigned long *esp)
341 show_stack_log_lvl(task, NULL, esp, "");
345 * The architecture-independent dump_stack generator
347 void dump_stack(void)
351 show_trace(current, NULL, &stack);
354 EXPORT_SYMBOL(dump_stack);
356 void show_registers(struct pt_regs *regs)
363 esp = (unsigned long) (®s->esp);
365 if (user_mode_vm(regs)) {
368 ss = regs->xss & 0xffff;
371 printk(KERN_EMERG "CPU: %d\n"
372 KERN_EMERG "EIP: %04x:[<%08lx>] %s VLI\n"
373 KERN_EMERG "EFLAGS: %08lx (%s %.*s)\n",
374 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
375 print_tainted(), regs->eflags, init_utsname()->release,
376 (int)strcspn(init_utsname()->version, " "),
377 init_utsname()->version);
378 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
379 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
380 regs->eax, regs->ebx, regs->ecx, regs->edx);
381 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
382 regs->esi, regs->edi, regs->ebp, esp);
383 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
384 regs->xds & 0xffff, regs->xes & 0xffff, ss);
385 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
386 TASK_COMM_LEN, current->comm, current->pid,
387 current_thread_info(), current, current->thread_info);
389 * When in-kernel, we also print out the stack and code at the
390 * time of the fault..
397 printk("\n" KERN_EMERG "Stack: ");
398 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
400 printk(KERN_EMERG "Code: ");
402 eip = (u8 *)regs->eip - 43;
403 if (eip < (u8 *)PAGE_OFFSET ||
404 probe_kernel_address(eip, c)) {
405 /* try starting at EIP */
406 eip = (u8 *)regs->eip;
409 for (i = 0; i < code_bytes; i++, eip++) {
410 if (eip < (u8 *)PAGE_OFFSET ||
411 probe_kernel_address(eip, c)) {
412 printk(" Bad EIP value.");
415 if (eip == (u8 *)regs->eip)
416 printk("<%02x> ", c);
424 int is_valid_bugaddr(unsigned long eip)
428 if (eip < PAGE_OFFSET)
430 if (probe_kernel_address((unsigned short *)eip, ud2))
433 return ud2 == 0x0b0f;
437 * This is gone through when something in the kernel has done something bad and
438 * is about to be terminated.
440 void die(const char * str, struct pt_regs * regs, long err)
445 int lock_owner_depth;
447 .lock = __SPIN_LOCK_UNLOCKED(die.lock),
449 .lock_owner_depth = 0
451 static int die_counter;
456 if (die.lock_owner != raw_smp_processor_id()) {
458 spin_lock_irqsave(&die.lock, flags);
459 die.lock_owner = smp_processor_id();
460 die.lock_owner_depth = 0;
464 local_save_flags(flags);
466 if (++die.lock_owner_depth < 3) {
471 report_bug(regs->eip);
473 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
474 #ifdef CONFIG_PREEMPT
475 printk(KERN_EMERG "PREEMPT ");
484 #ifdef CONFIG_DEBUG_PAGEALLOC
487 printk("DEBUG_PAGEALLOC");
492 if (notify_die(DIE_OOPS, str, regs, err,
493 current->thread.trap_no, SIGSEGV) !=
495 show_registers(regs);
496 /* Executive summary in case the oops scrolled away */
497 esp = (unsigned long) (®s->esp);
499 if (user_mode(regs)) {
501 ss = regs->xss & 0xffff;
503 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
504 print_symbol("%s", regs->eip);
505 printk(" SS:ESP %04x:%08lx\n", ss, esp);
510 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
514 spin_unlock_irqrestore(&die.lock, flags);
519 if (kexec_should_crash(current))
523 panic("Fatal exception in interrupt");
526 panic("Fatal exception");
532 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
534 if (!user_mode_vm(regs))
538 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
539 struct pt_regs * regs, long error_code,
542 struct task_struct *tsk = current;
543 tsk->thread.error_code = error_code;
544 tsk->thread.trap_no = trapnr;
546 if (regs->eflags & VM_MASK) {
552 if (!user_mode(regs))
557 force_sig_info(signr, info, tsk);
559 force_sig(signr, tsk);
564 if (!fixup_exception(regs))
565 die(str, regs, error_code);
570 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
571 if (ret) goto trap_signal;
576 #define DO_ERROR(trapnr, signr, str, name) \
577 fastcall void do_##name(struct pt_regs * regs, long error_code) \
579 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
582 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
585 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
586 fastcall void do_##name(struct pt_regs * regs, long error_code) \
589 info.si_signo = signr; \
591 info.si_code = sicode; \
592 info.si_addr = (void __user *)siaddr; \
593 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
596 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
599 #define DO_VM86_ERROR(trapnr, signr, str, name) \
600 fastcall void do_##name(struct pt_regs * regs, long error_code) \
602 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
605 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
608 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
609 fastcall void do_##name(struct pt_regs * regs, long error_code) \
612 info.si_signo = signr; \
614 info.si_code = sicode; \
615 info.si_addr = (void __user *)siaddr; \
616 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
619 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
622 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
623 #ifndef CONFIG_KPROBES
624 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
626 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
627 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
628 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
629 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
630 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
631 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
632 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
633 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
634 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
636 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
640 struct tss_struct *tss = &per_cpu(init_tss, cpu);
641 struct thread_struct *thread = ¤t->thread;
644 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
645 * invalid offset set (the LAZY one) and the faulting thread has
646 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
647 * and we set the offset field correctly. Then we let the CPU to
648 * restart the faulting instruction.
650 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
651 thread->io_bitmap_ptr) {
652 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
653 thread->io_bitmap_max);
655 * If the previously set map was extending to higher ports
656 * than the current one, pad extra space with 0xff (no access).
658 if (thread->io_bitmap_max < tss->io_bitmap_max)
659 memset((char *) tss->io_bitmap +
660 thread->io_bitmap_max, 0xff,
661 tss->io_bitmap_max - thread->io_bitmap_max);
662 tss->io_bitmap_max = thread->io_bitmap_max;
663 tss->io_bitmap_base = IO_BITMAP_OFFSET;
664 tss->io_bitmap_owner = thread;
670 current->thread.error_code = error_code;
671 current->thread.trap_no = 13;
673 if (regs->eflags & VM_MASK)
676 if (!user_mode(regs))
679 current->thread.error_code = error_code;
680 current->thread.trap_no = 13;
681 force_sig(SIGSEGV, current);
686 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
690 if (!fixup_exception(regs)) {
691 if (notify_die(DIE_GPF, "general protection fault", regs,
692 error_code, 13, SIGSEGV) == NOTIFY_STOP)
694 die("general protection fault", regs, error_code);
698 static __kprobes void
699 mem_parity_error(unsigned char reason, struct pt_regs * regs)
701 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
702 "CPU %d.\n", reason, smp_processor_id());
703 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
704 if (panic_on_unrecovered_nmi)
705 panic("NMI: Not continuing");
707 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
709 /* Clear and disable the memory parity error line. */
710 clear_mem_error(reason);
713 static __kprobes void
714 io_check_error(unsigned char reason, struct pt_regs * regs)
718 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
719 show_registers(regs);
721 /* Re-enable the IOCK line, wait for a few seconds */
722 reason = (reason & 0xf) | 8;
725 while (--i) udelay(1000);
730 static __kprobes void
731 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
734 /* Might actually be able to figure out what the guilty party
741 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
742 "CPU %d.\n", reason, smp_processor_id());
743 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
744 if (panic_on_unrecovered_nmi)
745 panic("NMI: Not continuing");
747 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
750 static DEFINE_SPINLOCK(nmi_print_lock);
752 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
754 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
758 spin_lock(&nmi_print_lock);
760 * We are in trouble anyway, lets at least try
761 * to get a message out.
764 printk(KERN_EMERG "%s", msg);
765 printk(" on CPU%d, eip %08lx, registers:\n",
766 smp_processor_id(), regs->eip);
767 show_registers(regs);
769 spin_unlock(&nmi_print_lock);
772 /* If we are in kernel we are probably nested up pretty bad
773 * and might aswell get out now while we still can.
775 if (!user_mode_vm(regs)) {
776 current->thread.trap_no = 2;
783 static __kprobes void default_do_nmi(struct pt_regs * regs)
785 unsigned char reason = 0;
787 /* Only the BSP gets external NMIs from the system. */
788 if (!smp_processor_id())
789 reason = get_nmi_reason();
791 if (!(reason & 0xc0)) {
792 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
795 #ifdef CONFIG_X86_LOCAL_APIC
797 * Ok, so this is none of the documented NMI sources,
798 * so it must be the NMI watchdog.
800 if (nmi_watchdog_tick(regs, reason))
802 if (!do_nmi_callback(regs, smp_processor_id()))
804 unknown_nmi_error(reason, regs);
808 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
811 mem_parity_error(reason, regs);
813 io_check_error(reason, regs);
815 * Reassert NMI in case it became active meanwhile
816 * as it's edge-triggered.
821 fastcall __kprobes void do_nmi(struct pt_regs * regs, long error_code)
827 cpu = smp_processor_id();
831 default_do_nmi(regs);
836 #ifdef CONFIG_KPROBES
837 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
839 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
842 /* This is an interrupt gate, because kprobes wants interrupts
843 disabled. Normal trap handlers don't. */
844 restore_interrupts(regs);
845 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
850 * Our handling of the processor debug registers is non-trivial.
851 * We do not clear them on entry and exit from the kernel. Therefore
852 * it is possible to get a watchpoint trap here from inside the kernel.
853 * However, the code in ./ptrace.c has ensured that the user can
854 * only set watchpoints on userspace addresses. Therefore the in-kernel
855 * watchpoint trap can only occur in code which is reading/writing
856 * from user space. Such code must not hold kernel locks (since it
857 * can equally take a page fault), therefore it is safe to call
858 * force_sig_info even though that claims and releases locks.
860 * Code in ./signal.c ensures that the debug control register
861 * is restored before we deliver any signal, and therefore that
862 * user code runs with the correct debug control register even though
865 * Being careful here means that we don't have to be as careful in a
866 * lot of more complicated places (task switching can be a bit lazy
867 * about restoring all the debug state, and ptrace doesn't have to
868 * find every occurrence of the TF bit that could be saved away even
871 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
873 unsigned int condition;
874 struct task_struct *tsk = current;
876 get_debugreg(condition, 6);
878 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
879 SIGTRAP) == NOTIFY_STOP)
881 /* It's safe to allow irq's after DR6 has been saved */
882 if (regs->eflags & X86_EFLAGS_IF)
885 /* Mask out spurious debug traps due to lazy DR7 setting */
886 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
887 if (!tsk->thread.debugreg[7])
891 if (regs->eflags & VM_MASK)
894 /* Save debug status register where ptrace can see it */
895 tsk->thread.debugreg[6] = condition;
898 * Single-stepping through TF: make sure we ignore any events in
899 * kernel space (but re-enable TF when returning to user mode).
901 if (condition & DR_STEP) {
903 * We already checked v86 mode above, so we can
904 * check for kernel mode by just checking the CPL
907 if (!user_mode(regs))
908 goto clear_TF_reenable;
911 /* Ok, finally something we can handle */
912 send_sigtrap(tsk, regs, error_code);
914 /* Disable additional traps. They'll be re-enabled when
915 * the signal is delivered.
922 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
926 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
927 regs->eflags &= ~TF_MASK;
932 * Note that we play around with the 'TS' bit in an attempt to get
933 * the correct behaviour even in the presence of the asynchronous
936 void math_error(void __user *eip)
938 struct task_struct * task;
940 unsigned short cwd, swd;
943 * Save the info for the exception handler and clear the error.
947 task->thread.trap_no = 16;
948 task->thread.error_code = 0;
949 info.si_signo = SIGFPE;
951 info.si_code = __SI_FAULT;
954 * (~cwd & swd) will mask out exceptions that are not set to unmasked
955 * status. 0x3f is the exception bits in these regs, 0x200 is the
956 * C1 reg you need in case of a stack fault, 0x040 is the stack
957 * fault bit. We should only be taking one exception at a time,
958 * so if this combination doesn't produce any single exception,
959 * then we have a bad program that isn't syncronizing its FPU usage
960 * and it will suffer the consequences since we won't be able to
961 * fully reproduce the context of the exception
963 cwd = get_fpu_cwd(task);
964 swd = get_fpu_swd(task);
965 switch (swd & ~cwd & 0x3f) {
966 case 0x000: /* No unmasked exception */
968 default: /* Multiple exceptions */
970 case 0x001: /* Invalid Op */
972 * swd & 0x240 == 0x040: Stack Underflow
973 * swd & 0x240 == 0x240: Stack Overflow
974 * User must clear the SF bit (0x40) if set
976 info.si_code = FPE_FLTINV;
978 case 0x002: /* Denormalize */
979 case 0x010: /* Underflow */
980 info.si_code = FPE_FLTUND;
982 case 0x004: /* Zero Divide */
983 info.si_code = FPE_FLTDIV;
985 case 0x008: /* Overflow */
986 info.si_code = FPE_FLTOVF;
988 case 0x020: /* Precision */
989 info.si_code = FPE_FLTRES;
992 force_sig_info(SIGFPE, &info, task);
995 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
998 math_error((void __user *)regs->eip);
1001 static void simd_math_error(void __user *eip)
1003 struct task_struct * task;
1005 unsigned short mxcsr;
1008 * Save the info for the exception handler and clear the error.
1011 save_init_fpu(task);
1012 task->thread.trap_no = 19;
1013 task->thread.error_code = 0;
1014 info.si_signo = SIGFPE;
1016 info.si_code = __SI_FAULT;
1019 * The SIMD FPU exceptions are handled a little differently, as there
1020 * is only a single status/control register. Thus, to determine which
1021 * unmasked exception was caught we must mask the exception mask bits
1022 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1024 mxcsr = get_fpu_mxcsr(task);
1025 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1029 case 0x001: /* Invalid Op */
1030 info.si_code = FPE_FLTINV;
1032 case 0x002: /* Denormalize */
1033 case 0x010: /* Underflow */
1034 info.si_code = FPE_FLTUND;
1036 case 0x004: /* Zero Divide */
1037 info.si_code = FPE_FLTDIV;
1039 case 0x008: /* Overflow */
1040 info.si_code = FPE_FLTOVF;
1042 case 0x020: /* Precision */
1043 info.si_code = FPE_FLTRES;
1046 force_sig_info(SIGFPE, &info, task);
1049 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1053 /* Handle SIMD FPU exceptions on PIII+ processors. */
1055 simd_math_error((void __user *)regs->eip);
1058 * Handle strange cache flush from user space exception
1059 * in all other cases. This is undocumented behaviour.
1061 if (regs->eflags & VM_MASK) {
1062 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1066 current->thread.trap_no = 19;
1067 current->thread.error_code = error_code;
1068 die_if_kernel("cache flush denied", regs, error_code);
1069 force_sig(SIGSEGV, current);
1073 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1077 /* No need to warn about this any longer. */
1078 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1082 fastcall unsigned long patch_espfix_desc(unsigned long uesp,
1085 int cpu = smp_processor_id();
1086 struct Xgt_desc_struct *cpu_gdt_descr = &per_cpu(cpu_gdt_descr, cpu);
1087 struct desc_struct *gdt = (struct desc_struct *)cpu_gdt_descr->address;
1088 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1089 unsigned long new_kesp = kesp - base;
1090 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1091 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1092 /* Set up base for espfix segment */
1093 desc &= 0x00f0ff0000000000ULL;
1094 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1095 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1096 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1097 (lim_pages & 0xffff);
1098 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1103 * 'math_state_restore()' saves the current math information in the
1104 * old math state array, and gets the new ones from the current task
1106 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1107 * Don't touch unless you *really* know how it works.
1109 * Must be called with kernel preemption disabled (in this case,
1110 * local interrupts are disabled at the call-site in entry.S).
1112 asmlinkage void math_state_restore(void)
1114 struct thread_info *thread = current_thread_info();
1115 struct task_struct *tsk = thread->task;
1117 clts(); /* Allow maths ops (or we recurse) */
1118 if (!tsk_used_math(tsk))
1121 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1125 #ifndef CONFIG_MATH_EMULATION
1127 asmlinkage void math_emulate(long arg)
1129 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1130 printk(KERN_EMERG "killing %s.\n",current->comm);
1131 force_sig(SIGFPE,current);
1135 #endif /* CONFIG_MATH_EMULATION */
1137 #ifdef CONFIG_X86_F00F_BUG
1138 void __init trap_init_f00f_bug(void)
1140 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1143 * Update the IDT descriptor and reload the IDT so that
1144 * it uses the read-only mapped virtual address.
1146 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1147 load_idt(&idt_descr);
1152 * This needs to use 'idt_table' rather than 'idt', and
1153 * thus use the _nonmapped_ version of the IDT, as the
1154 * Pentium F0 0F bugfix can have resulted in the mapped
1155 * IDT being write-protected.
1157 void set_intr_gate(unsigned int n, void *addr)
1159 _set_gate(n, DESCTYPE_INT, addr, __KERNEL_CS);
1163 * This routine sets up an interrupt gate at directory privilege level 3.
1165 static inline void set_system_intr_gate(unsigned int n, void *addr)
1167 _set_gate(n, DESCTYPE_INT | DESCTYPE_DPL3, addr, __KERNEL_CS);
1170 static void __init set_trap_gate(unsigned int n, void *addr)
1172 _set_gate(n, DESCTYPE_TRAP, addr, __KERNEL_CS);
1175 static void __init set_system_gate(unsigned int n, void *addr)
1177 _set_gate(n, DESCTYPE_TRAP | DESCTYPE_DPL3, addr, __KERNEL_CS);
1180 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1182 _set_gate(n, DESCTYPE_TASK, (void *)0, (gdt_entry<<3));
1186 void __init trap_init(void)
1189 void __iomem *p = ioremap(0x0FFFD9, 4);
1190 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1196 #ifdef CONFIG_X86_LOCAL_APIC
1197 init_apic_mappings();
1200 set_trap_gate(0,÷_error);
1201 set_intr_gate(1,&debug);
1202 set_intr_gate(2,&nmi);
1203 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1204 set_system_gate(4,&overflow);
1205 set_trap_gate(5,&bounds);
1206 set_trap_gate(6,&invalid_op);
1207 set_trap_gate(7,&device_not_available);
1208 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1209 set_trap_gate(9,&coprocessor_segment_overrun);
1210 set_trap_gate(10,&invalid_TSS);
1211 set_trap_gate(11,&segment_not_present);
1212 set_trap_gate(12,&stack_segment);
1213 set_trap_gate(13,&general_protection);
1214 set_intr_gate(14,&page_fault);
1215 set_trap_gate(15,&spurious_interrupt_bug);
1216 set_trap_gate(16,&coprocessor_error);
1217 set_trap_gate(17,&alignment_check);
1218 #ifdef CONFIG_X86_MCE
1219 set_trap_gate(18,&machine_check);
1221 set_trap_gate(19,&simd_coprocessor_error);
1225 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1226 * Generates a compile-time "error: zero width for bit-field" if
1227 * the alignment is wrong.
1229 struct fxsrAlignAssert {
1230 int _:!(offsetof(struct task_struct,
1231 thread.i387.fxsave) & 15);
1234 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1235 set_in_cr4(X86_CR4_OSFXSR);
1239 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1241 set_in_cr4(X86_CR4_OSXMMEXCPT);
1245 set_system_gate(SYSCALL_VECTOR,&system_call);
1248 * Should be a barrier for any external CPU state.
1255 static int __init kstack_setup(char *s)
1257 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1260 __setup("kstack=", kstack_setup);
1262 #ifdef CONFIG_STACK_UNWIND
1263 static int __init call_trace_setup(char *s)
1265 if (strcmp(s, "old") == 0)
1267 else if (strcmp(s, "both") == 0)
1269 else if (strcmp(s, "newfallback") == 0)
1271 else if (strcmp(s, "new") == 2)
1275 __setup("call_trace=", call_trace_setup);