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/config.h>
15 #include <linux/sched.h>
16 #include <linux/kernel.h>
17 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/timer.h>
21 #include <linux/init.h>
22 #include <linux/delay.h>
23 #include <linux/spinlock.h>
24 #include <linux/interrupt.h>
25 #include <linux/highmem.h>
26 #include <linux/kallsyms.h>
27 #include <linux/ptrace.h>
28 #include <linux/utsname.h>
29 #include <linux/kprobes.h>
30 #include <linux/kexec.h>
31 #include <linux/unwind.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>
44 #include <asm/uaccess.h>
46 #include <asm/atomic.h>
47 #include <asm/debugreg.h>
51 #include <asm/unwind.h>
53 #include <asm/arch_hooks.h>
54 #include <asm/kdebug.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 static int call_trace = 1;
97 ATOMIC_NOTIFIER_HEAD(i386die_chain);
99 int register_die_notifier(struct notifier_block *nb)
102 return atomic_notifier_chain_register(&i386die_chain, nb);
104 EXPORT_SYMBOL(register_die_notifier);
106 int unregister_die_notifier(struct notifier_block *nb)
108 return atomic_notifier_chain_unregister(&i386die_chain, nb);
110 EXPORT_SYMBOL(unregister_die_notifier);
112 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
114 return p > (void *)tinfo &&
115 p < (void *)tinfo + THREAD_SIZE - 3;
119 * Print CONFIG_STACK_BACKTRACE_COLS address/symbol entries per line.
121 static inline int print_addr_and_symbol(unsigned long addr, char *log_lvl,
127 #if CONFIG_STACK_BACKTRACE_COLS == 1
128 printk(" [<%08lx>] ", addr);
130 printk(" <%08lx> ", addr);
132 print_symbol("%s", addr);
134 printed = (printed + 1) % CONFIG_STACK_BACKTRACE_COLS;
143 static inline unsigned long print_context_stack(struct thread_info *tinfo,
144 unsigned long *stack, unsigned long ebp,
148 int printed = 0; /* nr of entries already printed on current line */
150 #ifdef CONFIG_FRAME_POINTER
151 while (valid_stack_ptr(tinfo, (void *)ebp)) {
152 addr = *(unsigned long *)(ebp + 4);
153 printed = print_addr_and_symbol(addr, log_lvl, printed);
155 * break out of recursive entries (such as
156 * end_of_stack_stop_unwind_function):
158 if (ebp == *(unsigned long *)ebp)
160 ebp = *(unsigned long *)ebp;
163 while (valid_stack_ptr(tinfo, stack)) {
165 if (__kernel_text_address(addr))
166 printed = print_addr_and_symbol(addr, log_lvl, printed);
175 static asmlinkage int show_trace_unwind(struct unwind_frame_info *info, void *log_lvl)
178 int printed = 0; /* nr of entries already printed on current line */
180 while (unwind(info) == 0 && UNW_PC(info)) {
182 printed = print_addr_and_symbol(UNW_PC(info), log_lvl, printed);
183 if (arch_unw_user_mode(info))
191 static void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
192 unsigned long *stack, char *log_lvl)
199 if (call_trace >= 0) {
201 struct unwind_frame_info info;
204 if (unwind_init_frame_info(&info, task, regs) == 0)
205 unw_ret = show_trace_unwind(&info, log_lvl);
206 } else if (task == current)
207 unw_ret = unwind_init_running(&info, show_trace_unwind, log_lvl);
209 if (unwind_init_blocked(&info, task) == 0)
210 unw_ret = show_trace_unwind(&info, log_lvl);
215 printk("%sLegacy call trace:\n", log_lvl);
219 if (task == current) {
220 /* Grab ebp right from our regs */
221 asm ("movl %%ebp, %0" : "=r" (ebp) : );
223 /* ebp is the last reg pushed by switch_to */
224 ebp = *(unsigned long *) task->thread.esp;
228 struct thread_info *context;
229 context = (struct thread_info *)
230 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
231 ebp = print_context_stack(context, stack, ebp, log_lvl);
232 stack = (unsigned long*)context->previous_esp;
235 printk("%s =======================\n", log_lvl);
239 void show_trace(struct task_struct *task, struct pt_regs *regs, unsigned long * stack)
241 show_trace_log_lvl(task, regs, stack, "");
244 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
245 unsigned long *esp, char *log_lvl)
247 unsigned long *stack;
252 esp = (unsigned long*)task->thread.esp;
254 esp = (unsigned long *)&esp;
258 for(i = 0; i < kstack_depth_to_print; i++) {
259 if (kstack_end(stack))
261 if (i && ((i % 8) == 0))
262 printk("\n%s ", log_lvl);
263 printk("%08lx ", *stack++);
265 printk("\n%sCall Trace:\n", log_lvl);
266 show_trace_log_lvl(task, regs, esp, log_lvl);
269 void show_stack(struct task_struct *task, unsigned long *esp)
272 show_stack_log_lvl(task, NULL, esp, "");
276 * The architecture-independent dump_stack generator
278 void dump_stack(void)
282 show_trace(current, NULL, &stack);
285 EXPORT_SYMBOL(dump_stack);
287 void show_registers(struct pt_regs *regs)
294 esp = (unsigned long) (®s->esp);
296 if (user_mode_vm(regs)) {
299 ss = regs->xss & 0xffff;
302 printk(KERN_EMERG "CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\n"
303 "EFLAGS: %08lx (%s %.*s) \n",
304 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
305 print_tainted(), regs->eflags, system_utsname.release,
306 (int)strcspn(system_utsname.version, " "),
307 system_utsname.version);
308 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
309 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
310 regs->eax, regs->ebx, regs->ecx, regs->edx);
311 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
312 regs->esi, regs->edi, regs->ebp, esp);
313 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
314 regs->xds & 0xffff, regs->xes & 0xffff, ss);
315 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
316 TASK_COMM_LEN, current->comm, current->pid,
317 current_thread_info(), current, current->thread_info);
319 * When in-kernel, we also print out the stack and code at the
320 * time of the fault..
325 printk("\n" KERN_EMERG "Stack: ");
326 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
328 printk(KERN_EMERG "Code: ");
330 eip = (u8 __user *)regs->eip - 43;
331 for (i = 0; i < 64; i++, eip++) {
334 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
335 printk(" Bad EIP value.");
338 if (eip == (u8 __user *)regs->eip)
339 printk("<%02x> ", c);
347 static void handle_BUG(struct pt_regs *regs)
357 if (eip < PAGE_OFFSET)
359 if (__get_user(ud2, (unsigned short __user *)eip))
363 if (__get_user(line, (unsigned short __user *)(eip + 2)))
365 if (__get_user(file, (char * __user *)(eip + 4)) ||
366 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
367 file = "<bad filename>";
369 printk(KERN_EMERG "------------[ cut here ]------------\n");
370 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
375 /* Here we know it was a BUG but file-n-line is unavailable */
377 printk(KERN_EMERG "Kernel BUG\n");
380 /* This is gone through when something in the kernel
381 * has done something bad and is about to be terminated.
383 void die(const char * str, struct pt_regs * regs, long err)
388 int lock_owner_depth;
390 .lock = SPIN_LOCK_UNLOCKED,
392 .lock_owner_depth = 0
394 static int die_counter;
399 if (die.lock_owner != raw_smp_processor_id()) {
401 spin_lock_irqsave(&die.lock, flags);
402 die.lock_owner = smp_processor_id();
403 die.lock_owner_depth = 0;
407 local_save_flags(flags);
409 if (++die.lock_owner_depth < 3) {
415 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
416 #ifdef CONFIG_PREEMPT
417 printk(KERN_EMERG "PREEMPT ");
426 #ifdef CONFIG_DEBUG_PAGEALLOC
429 printk("DEBUG_PAGEALLOC");
434 if (notify_die(DIE_OOPS, str, regs, err,
435 current->thread.trap_no, SIGSEGV) !=
437 show_registers(regs);
438 /* Executive summary in case the oops scrolled away */
439 esp = (unsigned long) (®s->esp);
441 if (user_mode(regs)) {
443 ss = regs->xss & 0xffff;
445 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
446 print_symbol("%s", regs->eip);
447 printk(" SS:ESP %04x:%08lx\n", ss, esp);
452 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
456 spin_unlock_irqrestore(&die.lock, flags);
461 if (kexec_should_crash(current))
465 panic("Fatal exception in interrupt");
468 printk(KERN_EMERG "Fatal exception: panic in 5 seconds\n");
470 panic("Fatal exception");
476 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
478 if (!user_mode_vm(regs))
482 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
483 struct pt_regs * regs, long error_code,
486 struct task_struct *tsk = current;
487 tsk->thread.error_code = error_code;
488 tsk->thread.trap_no = trapnr;
490 if (regs->eflags & VM_MASK) {
496 if (!user_mode(regs))
501 force_sig_info(signr, info, tsk);
503 force_sig(signr, tsk);
508 if (!fixup_exception(regs))
509 die(str, regs, error_code);
514 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
515 if (ret) goto trap_signal;
520 #define DO_ERROR(trapnr, signr, str, name) \
521 fastcall void do_##name(struct pt_regs * regs, long error_code) \
523 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
526 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
529 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
530 fastcall void do_##name(struct pt_regs * regs, long error_code) \
533 info.si_signo = signr; \
535 info.si_code = sicode; \
536 info.si_addr = (void __user *)siaddr; \
537 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
540 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
543 #define DO_VM86_ERROR(trapnr, signr, str, name) \
544 fastcall void do_##name(struct pt_regs * regs, long error_code) \
546 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
549 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
552 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
553 fastcall void do_##name(struct pt_regs * regs, long error_code) \
556 info.si_signo = signr; \
558 info.si_code = sicode; \
559 info.si_addr = (void __user *)siaddr; \
560 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
563 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
566 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
567 #ifndef CONFIG_KPROBES
568 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
570 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
571 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
572 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
573 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
574 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
575 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
576 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
577 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
578 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
580 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
584 struct tss_struct *tss = &per_cpu(init_tss, cpu);
585 struct thread_struct *thread = ¤t->thread;
588 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
589 * invalid offset set (the LAZY one) and the faulting thread has
590 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
591 * and we set the offset field correctly. Then we let the CPU to
592 * restart the faulting instruction.
594 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
595 thread->io_bitmap_ptr) {
596 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
597 thread->io_bitmap_max);
599 * If the previously set map was extending to higher ports
600 * than the current one, pad extra space with 0xff (no access).
602 if (thread->io_bitmap_max < tss->io_bitmap_max)
603 memset((char *) tss->io_bitmap +
604 thread->io_bitmap_max, 0xff,
605 tss->io_bitmap_max - thread->io_bitmap_max);
606 tss->io_bitmap_max = thread->io_bitmap_max;
607 tss->io_bitmap_base = IO_BITMAP_OFFSET;
608 tss->io_bitmap_owner = thread;
614 current->thread.error_code = error_code;
615 current->thread.trap_no = 13;
617 if (regs->eflags & VM_MASK)
620 if (!user_mode(regs))
623 current->thread.error_code = error_code;
624 current->thread.trap_no = 13;
625 force_sig(SIGSEGV, current);
630 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
634 if (!fixup_exception(regs)) {
635 if (notify_die(DIE_GPF, "general protection fault", regs,
636 error_code, 13, SIGSEGV) == NOTIFY_STOP)
638 die("general protection fault", regs, error_code);
642 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
644 printk(KERN_EMERG "Uhhuh. NMI received. Dazed and confused, but trying "
646 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
649 /* Clear and disable the memory parity error line. */
650 clear_mem_error(reason);
653 static void io_check_error(unsigned char reason, struct pt_regs * regs)
657 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
658 show_registers(regs);
660 /* Re-enable the IOCK line, wait for a few seconds */
661 reason = (reason & 0xf) | 8;
664 while (--i) udelay(1000);
669 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
672 /* Might actually be able to figure out what the guilty party
679 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
680 reason, smp_processor_id());
681 printk("Dazed and confused, but trying to continue\n");
682 printk("Do you have a strange power saving mode enabled?\n");
685 static DEFINE_SPINLOCK(nmi_print_lock);
687 void die_nmi (struct pt_regs *regs, const char *msg)
689 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
693 spin_lock(&nmi_print_lock);
695 * We are in trouble anyway, lets at least try
696 * to get a message out.
699 printk(KERN_EMERG "%s", msg);
700 printk(" on CPU%d, eip %08lx, registers:\n",
701 smp_processor_id(), regs->eip);
702 show_registers(regs);
703 printk(KERN_EMERG "console shuts up ...\n");
705 spin_unlock(&nmi_print_lock);
708 /* If we are in kernel we are probably nested up pretty bad
709 * and might aswell get out now while we still can.
711 if (!user_mode_vm(regs)) {
712 current->thread.trap_no = 2;
719 static void default_do_nmi(struct pt_regs * regs)
721 unsigned char reason = 0;
723 /* Only the BSP gets external NMIs from the system. */
724 if (!smp_processor_id())
725 reason = get_nmi_reason();
727 if (!(reason & 0xc0)) {
728 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
731 #ifdef CONFIG_X86_LOCAL_APIC
733 * Ok, so this is none of the documented NMI sources,
734 * so it must be the NMI watchdog.
737 nmi_watchdog_tick(regs);
741 unknown_nmi_error(reason, regs);
744 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
747 mem_parity_error(reason, regs);
749 io_check_error(reason, regs);
751 * Reassert NMI in case it became active meanwhile
752 * as it's edge-triggered.
757 static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
762 static nmi_callback_t nmi_callback = dummy_nmi_callback;
764 fastcall void do_nmi(struct pt_regs * regs, long error_code)
770 cpu = smp_processor_id();
774 if (!rcu_dereference(nmi_callback)(regs, cpu))
775 default_do_nmi(regs);
780 void set_nmi_callback(nmi_callback_t callback)
783 rcu_assign_pointer(nmi_callback, callback);
785 EXPORT_SYMBOL_GPL(set_nmi_callback);
787 void unset_nmi_callback(void)
789 nmi_callback = dummy_nmi_callback;
791 EXPORT_SYMBOL_GPL(unset_nmi_callback);
793 #ifdef CONFIG_KPROBES
794 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
796 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
799 /* This is an interrupt gate, because kprobes wants interrupts
800 disabled. Normal trap handlers don't. */
801 restore_interrupts(regs);
802 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
807 * Our handling of the processor debug registers is non-trivial.
808 * We do not clear them on entry and exit from the kernel. Therefore
809 * it is possible to get a watchpoint trap here from inside the kernel.
810 * However, the code in ./ptrace.c has ensured that the user can
811 * only set watchpoints on userspace addresses. Therefore the in-kernel
812 * watchpoint trap can only occur in code which is reading/writing
813 * from user space. Such code must not hold kernel locks (since it
814 * can equally take a page fault), therefore it is safe to call
815 * force_sig_info even though that claims and releases locks.
817 * Code in ./signal.c ensures that the debug control register
818 * is restored before we deliver any signal, and therefore that
819 * user code runs with the correct debug control register even though
822 * Being careful here means that we don't have to be as careful in a
823 * lot of more complicated places (task switching can be a bit lazy
824 * about restoring all the debug state, and ptrace doesn't have to
825 * find every occurrence of the TF bit that could be saved away even
828 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
830 unsigned int condition;
831 struct task_struct *tsk = current;
833 get_debugreg(condition, 6);
835 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
836 SIGTRAP) == NOTIFY_STOP)
838 /* It's safe to allow irq's after DR6 has been saved */
839 if (regs->eflags & X86_EFLAGS_IF)
842 /* Mask out spurious debug traps due to lazy DR7 setting */
843 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
844 if (!tsk->thread.debugreg[7])
848 if (regs->eflags & VM_MASK)
851 /* Save debug status register where ptrace can see it */
852 tsk->thread.debugreg[6] = condition;
855 * Single-stepping through TF: make sure we ignore any events in
856 * kernel space (but re-enable TF when returning to user mode).
858 if (condition & DR_STEP) {
860 * We already checked v86 mode above, so we can
861 * check for kernel mode by just checking the CPL
864 if (!user_mode(regs))
865 goto clear_TF_reenable;
868 /* Ok, finally something we can handle */
869 send_sigtrap(tsk, regs, error_code);
871 /* Disable additional traps. They'll be re-enabled when
872 * the signal is delivered.
879 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
883 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
884 regs->eflags &= ~TF_MASK;
889 * Note that we play around with the 'TS' bit in an attempt to get
890 * the correct behaviour even in the presence of the asynchronous
893 void math_error(void __user *eip)
895 struct task_struct * task;
897 unsigned short cwd, swd;
900 * Save the info for the exception handler and clear the error.
904 task->thread.trap_no = 16;
905 task->thread.error_code = 0;
906 info.si_signo = SIGFPE;
908 info.si_code = __SI_FAULT;
911 * (~cwd & swd) will mask out exceptions that are not set to unmasked
912 * status. 0x3f is the exception bits in these regs, 0x200 is the
913 * C1 reg you need in case of a stack fault, 0x040 is the stack
914 * fault bit. We should only be taking one exception at a time,
915 * so if this combination doesn't produce any single exception,
916 * then we have a bad program that isn't syncronizing its FPU usage
917 * and it will suffer the consequences since we won't be able to
918 * fully reproduce the context of the exception
920 cwd = get_fpu_cwd(task);
921 swd = get_fpu_swd(task);
922 switch (swd & ~cwd & 0x3f) {
923 case 0x000: /* No unmasked exception */
925 default: /* Multiple exceptions */
927 case 0x001: /* Invalid Op */
929 * swd & 0x240 == 0x040: Stack Underflow
930 * swd & 0x240 == 0x240: Stack Overflow
931 * User must clear the SF bit (0x40) if set
933 info.si_code = FPE_FLTINV;
935 case 0x002: /* Denormalize */
936 case 0x010: /* Underflow */
937 info.si_code = FPE_FLTUND;
939 case 0x004: /* Zero Divide */
940 info.si_code = FPE_FLTDIV;
942 case 0x008: /* Overflow */
943 info.si_code = FPE_FLTOVF;
945 case 0x020: /* Precision */
946 info.si_code = FPE_FLTRES;
949 force_sig_info(SIGFPE, &info, task);
952 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
955 math_error((void __user *)regs->eip);
958 static void simd_math_error(void __user *eip)
960 struct task_struct * task;
962 unsigned short mxcsr;
965 * Save the info for the exception handler and clear the error.
969 task->thread.trap_no = 19;
970 task->thread.error_code = 0;
971 info.si_signo = SIGFPE;
973 info.si_code = __SI_FAULT;
976 * The SIMD FPU exceptions are handled a little differently, as there
977 * is only a single status/control register. Thus, to determine which
978 * unmasked exception was caught we must mask the exception mask bits
979 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
981 mxcsr = get_fpu_mxcsr(task);
982 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
986 case 0x001: /* Invalid Op */
987 info.si_code = FPE_FLTINV;
989 case 0x002: /* Denormalize */
990 case 0x010: /* Underflow */
991 info.si_code = FPE_FLTUND;
993 case 0x004: /* Zero Divide */
994 info.si_code = FPE_FLTDIV;
996 case 0x008: /* Overflow */
997 info.si_code = FPE_FLTOVF;
999 case 0x020: /* Precision */
1000 info.si_code = FPE_FLTRES;
1003 force_sig_info(SIGFPE, &info, task);
1006 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1010 /* Handle SIMD FPU exceptions on PIII+ processors. */
1012 simd_math_error((void __user *)regs->eip);
1015 * Handle strange cache flush from user space exception
1016 * in all other cases. This is undocumented behaviour.
1018 if (regs->eflags & VM_MASK) {
1019 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1023 current->thread.trap_no = 19;
1024 current->thread.error_code = error_code;
1025 die_if_kernel("cache flush denied", regs, error_code);
1026 force_sig(SIGSEGV, current);
1030 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1034 /* No need to warn about this any longer. */
1035 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1039 fastcall void setup_x86_bogus_stack(unsigned char * stk)
1041 unsigned long *switch16_ptr, *switch32_ptr;
1042 struct pt_regs *regs;
1043 unsigned long stack_top, stack_bot;
1044 unsigned short iret_frame16_off;
1045 int cpu = smp_processor_id();
1046 /* reserve the space on 32bit stack for the magic switch16 pointer */
1047 memmove(stk, stk + 8, sizeof(struct pt_regs));
1048 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
1049 regs = (struct pt_regs *)stk;
1050 /* now the switch32 on 16bit stack */
1051 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1052 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1053 switch32_ptr = (unsigned long *)(stack_top - 8);
1054 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
1055 /* copy iret frame on 16bit stack */
1056 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
1057 /* fill in the switch pointers */
1058 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
1059 switch16_ptr[1] = __ESPFIX_SS;
1060 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
1061 8 - CPU_16BIT_STACK_SIZE;
1062 switch32_ptr[1] = __KERNEL_DS;
1065 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
1067 unsigned long *switch32_ptr;
1068 unsigned char *stack16, *stack32;
1069 unsigned long stack_top, stack_bot;
1071 int cpu = smp_processor_id();
1072 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1073 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1074 switch32_ptr = (unsigned long *)(stack_top - 8);
1075 /* copy the data from 16bit stack to 32bit stack */
1076 len = CPU_16BIT_STACK_SIZE - 8 - sp;
1077 stack16 = (unsigned char *)(stack_bot + sp);
1078 stack32 = (unsigned char *)
1079 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
1080 memcpy(stack32, stack16, len);
1085 * 'math_state_restore()' saves the current math information in the
1086 * old math state array, and gets the new ones from the current task
1088 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1089 * Don't touch unless you *really* know how it works.
1091 * Must be called with kernel preemption disabled (in this case,
1092 * local interrupts are disabled at the call-site in entry.S).
1094 asmlinkage void math_state_restore(struct pt_regs regs)
1096 struct thread_info *thread = current_thread_info();
1097 struct task_struct *tsk = thread->task;
1099 clts(); /* Allow maths ops (or we recurse) */
1100 if (!tsk_used_math(tsk))
1103 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1106 #ifndef CONFIG_MATH_EMULATION
1108 asmlinkage void math_emulate(long arg)
1110 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1111 printk(KERN_EMERG "killing %s.\n",current->comm);
1112 force_sig(SIGFPE,current);
1116 #endif /* CONFIG_MATH_EMULATION */
1118 #ifdef CONFIG_X86_F00F_BUG
1119 void __init trap_init_f00f_bug(void)
1121 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1124 * Update the IDT descriptor and reload the IDT so that
1125 * it uses the read-only mapped virtual address.
1127 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1128 load_idt(&idt_descr);
1132 #define _set_gate(gate_addr,type,dpl,addr,seg) \
1135 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1136 "movw %4,%%dx\n\t" \
1137 "movl %%eax,%0\n\t" \
1139 :"=m" (*((long *) (gate_addr))), \
1140 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1141 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1142 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1147 * This needs to use 'idt_table' rather than 'idt', and
1148 * thus use the _nonmapped_ version of the IDT, as the
1149 * Pentium F0 0F bugfix can have resulted in the mapped
1150 * IDT being write-protected.
1152 void set_intr_gate(unsigned int n, void *addr)
1154 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1158 * This routine sets up an interrupt gate at directory privilege level 3.
1160 static inline void set_system_intr_gate(unsigned int n, void *addr)
1162 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1165 static void __init set_trap_gate(unsigned int n, void *addr)
1167 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1170 static void __init set_system_gate(unsigned int n, void *addr)
1172 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1175 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1177 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1181 void __init trap_init(void)
1184 void __iomem *p = ioremap(0x0FFFD9, 4);
1185 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1191 #ifdef CONFIG_X86_LOCAL_APIC
1192 init_apic_mappings();
1195 set_trap_gate(0,÷_error);
1196 set_intr_gate(1,&debug);
1197 set_intr_gate(2,&nmi);
1198 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1199 set_system_gate(4,&overflow);
1200 set_trap_gate(5,&bounds);
1201 set_trap_gate(6,&invalid_op);
1202 set_trap_gate(7,&device_not_available);
1203 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1204 set_trap_gate(9,&coprocessor_segment_overrun);
1205 set_trap_gate(10,&invalid_TSS);
1206 set_trap_gate(11,&segment_not_present);
1207 set_trap_gate(12,&stack_segment);
1208 set_trap_gate(13,&general_protection);
1209 set_intr_gate(14,&page_fault);
1210 set_trap_gate(15,&spurious_interrupt_bug);
1211 set_trap_gate(16,&coprocessor_error);
1212 set_trap_gate(17,&alignment_check);
1213 #ifdef CONFIG_X86_MCE
1214 set_trap_gate(18,&machine_check);
1216 set_trap_gate(19,&simd_coprocessor_error);
1220 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1221 * Generates a compile-time "error: zero width for bit-field" if
1222 * the alignment is wrong.
1224 struct fxsrAlignAssert {
1225 int _:!(offsetof(struct task_struct,
1226 thread.i387.fxsave) & 15);
1229 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1230 set_in_cr4(X86_CR4_OSFXSR);
1234 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1236 set_in_cr4(X86_CR4_OSXMMEXCPT);
1240 set_system_gate(SYSCALL_VECTOR,&system_call);
1243 * Should be a barrier for any external CPU state.
1250 static int __init kstack_setup(char *s)
1252 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1255 __setup("kstack=", kstack_setup);
1257 static int __init call_trace_setup(char *s)
1259 if (strcmp(s, "old") == 0)
1261 else if (strcmp(s, "both") == 0)
1263 else if (strcmp(s, "new") == 0)
1267 __setup("call_trace=", call_trace_setup);