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>
55 #include <linux/module.h>
57 #include "mach_traps.h"
59 asmlinkage int system_call(void);
61 struct desc_struct default_ldt[] = { { 0, 0 }, { 0, 0 }, { 0, 0 },
64 /* Do we ignore FPU interrupts ? */
65 char ignore_fpu_irq = 0;
68 * The IDT has to be page-aligned to simplify the Pentium
69 * F0 0F bug workaround.. We have a special link segment
72 struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
74 asmlinkage void divide_error(void);
75 asmlinkage void debug(void);
76 asmlinkage void nmi(void);
77 asmlinkage void int3(void);
78 asmlinkage void overflow(void);
79 asmlinkage void bounds(void);
80 asmlinkage void invalid_op(void);
81 asmlinkage void device_not_available(void);
82 asmlinkage void coprocessor_segment_overrun(void);
83 asmlinkage void invalid_TSS(void);
84 asmlinkage void segment_not_present(void);
85 asmlinkage void stack_segment(void);
86 asmlinkage void general_protection(void);
87 asmlinkage void page_fault(void);
88 asmlinkage void coprocessor_error(void);
89 asmlinkage void simd_coprocessor_error(void);
90 asmlinkage void alignment_check(void);
91 asmlinkage void spurious_interrupt_bug(void);
92 asmlinkage void machine_check(void);
94 static int kstack_depth_to_print = 24;
95 static int call_trace = 1;
96 ATOMIC_NOTIFIER_HEAD(i386die_chain);
98 int register_die_notifier(struct notifier_block *nb)
101 return atomic_notifier_chain_register(&i386die_chain, nb);
103 EXPORT_SYMBOL(register_die_notifier);
105 int unregister_die_notifier(struct notifier_block *nb)
107 return atomic_notifier_chain_unregister(&i386die_chain, nb);
109 EXPORT_SYMBOL(unregister_die_notifier);
111 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
113 return p > (void *)tinfo &&
114 p < (void *)tinfo + THREAD_SIZE - 3;
118 * Print CONFIG_STACK_BACKTRACE_COLS address/symbol entries per line.
120 static inline int print_addr_and_symbol(unsigned long addr, char *log_lvl,
126 #if CONFIG_STACK_BACKTRACE_COLS == 1
127 printk(" [<%08lx>] ", addr);
129 printk(" <%08lx> ", addr);
131 print_symbol("%s", addr);
133 printed = (printed + 1) % CONFIG_STACK_BACKTRACE_COLS;
142 static inline unsigned long print_context_stack(struct thread_info *tinfo,
143 unsigned long *stack, unsigned long ebp,
147 int printed = 0; /* nr of entries already printed on current line */
149 #ifdef CONFIG_FRAME_POINTER
150 while (valid_stack_ptr(tinfo, (void *)ebp)) {
151 addr = *(unsigned long *)(ebp + 4);
152 printed = print_addr_and_symbol(addr, log_lvl, printed);
154 * break out of recursive entries (such as
155 * end_of_stack_stop_unwind_function):
157 if (ebp == *(unsigned long *)ebp)
159 ebp = *(unsigned long *)ebp;
162 while (valid_stack_ptr(tinfo, stack)) {
164 if (__kernel_text_address(addr))
165 printed = print_addr_and_symbol(addr, log_lvl, printed);
174 static asmlinkage int show_trace_unwind(struct unwind_frame_info *info, void *log_lvl)
177 int printed = 0; /* nr of entries already printed on current line */
179 while (unwind(info) == 0 && UNW_PC(info)) {
181 printed = print_addr_and_symbol(UNW_PC(info), log_lvl, printed);
182 if (arch_unw_user_mode(info))
190 static void show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
191 unsigned long *stack, char *log_lvl)
198 if (call_trace >= 0) {
200 struct unwind_frame_info info;
203 if (unwind_init_frame_info(&info, task, regs) == 0)
204 unw_ret = show_trace_unwind(&info, log_lvl);
205 } else if (task == current)
206 unw_ret = unwind_init_running(&info, show_trace_unwind, log_lvl);
208 if (unwind_init_blocked(&info, task) == 0)
209 unw_ret = show_trace_unwind(&info, log_lvl);
214 printk("%sLegacy call trace:\n", log_lvl);
218 if (task == current) {
219 /* Grab ebp right from our regs */
220 asm ("movl %%ebp, %0" : "=r" (ebp) : );
222 /* ebp is the last reg pushed by switch_to */
223 ebp = *(unsigned long *) task->thread.esp;
227 struct thread_info *context;
228 context = (struct thread_info *)
229 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
230 ebp = print_context_stack(context, stack, ebp, log_lvl);
231 stack = (unsigned long*)context->previous_esp;
234 printk("%s =======================\n", log_lvl);
238 void show_trace(struct task_struct *task, struct pt_regs *regs, unsigned long * stack)
240 show_trace_log_lvl(task, regs, stack, "");
243 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
244 unsigned long *esp, char *log_lvl)
246 unsigned long *stack;
251 esp = (unsigned long*)task->thread.esp;
253 esp = (unsigned long *)&esp;
257 for(i = 0; i < kstack_depth_to_print; i++) {
258 if (kstack_end(stack))
260 if (i && ((i % 8) == 0))
261 printk("\n%s ", log_lvl);
262 printk("%08lx ", *stack++);
264 printk("\n%sCall Trace:\n", log_lvl);
265 show_trace_log_lvl(task, regs, esp, log_lvl);
268 void show_stack(struct task_struct *task, unsigned long *esp)
271 show_stack_log_lvl(task, NULL, esp, "");
275 * The architecture-independent dump_stack generator
277 void dump_stack(void)
281 show_trace(current, NULL, &stack);
284 EXPORT_SYMBOL(dump_stack);
286 void show_registers(struct pt_regs *regs)
293 esp = (unsigned long) (®s->esp);
295 if (user_mode_vm(regs)) {
298 ss = regs->xss & 0xffff;
301 printk(KERN_EMERG "CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\n"
302 "EFLAGS: %08lx (%s %.*s) \n",
303 smp_processor_id(), 0xffff & regs->xcs, regs->eip,
304 print_tainted(), regs->eflags, system_utsname.release,
305 (int)strcspn(system_utsname.version, " "),
306 system_utsname.version);
307 print_symbol(KERN_EMERG "EIP is at %s\n", regs->eip);
308 printk(KERN_EMERG "eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
309 regs->eax, regs->ebx, regs->ecx, regs->edx);
310 printk(KERN_EMERG "esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
311 regs->esi, regs->edi, regs->ebp, esp);
312 printk(KERN_EMERG "ds: %04x es: %04x ss: %04x\n",
313 regs->xds & 0xffff, regs->xes & 0xffff, ss);
314 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
315 TASK_COMM_LEN, current->comm, current->pid,
316 current_thread_info(), current, current->thread_info);
318 * When in-kernel, we also print out the stack and code at the
319 * time of the fault..
324 printk("\n" KERN_EMERG "Stack: ");
325 show_stack_log_lvl(NULL, regs, (unsigned long *)esp, KERN_EMERG);
327 printk(KERN_EMERG "Code: ");
329 eip = (u8 __user *)regs->eip - 43;
330 for (i = 0; i < 64; i++, eip++) {
333 if (eip < (u8 __user *)PAGE_OFFSET || __get_user(c, eip)) {
334 printk(" Bad EIP value.");
337 if (eip == (u8 __user *)regs->eip)
338 printk("<%02x> ", c);
346 static void handle_BUG(struct pt_regs *regs)
356 if (eip < PAGE_OFFSET)
358 if (__get_user(ud2, (unsigned short __user *)eip))
362 if (__get_user(line, (unsigned short __user *)(eip + 2)))
364 if (__get_user(file, (char * __user *)(eip + 4)) ||
365 (unsigned long)file < PAGE_OFFSET || __get_user(c, file))
366 file = "<bad filename>";
368 printk(KERN_EMERG "------------[ cut here ]------------\n");
369 printk(KERN_EMERG "kernel BUG at %s:%d!\n", file, line);
374 /* Here we know it was a BUG but file-n-line is unavailable */
376 printk(KERN_EMERG "Kernel BUG\n");
379 /* This is gone through when something in the kernel
380 * has done something bad and is about to be terminated.
382 void die(const char * str, struct pt_regs * regs, long err)
387 int lock_owner_depth;
389 .lock = SPIN_LOCK_UNLOCKED,
391 .lock_owner_depth = 0
393 static int die_counter;
398 if (die.lock_owner != raw_smp_processor_id()) {
400 spin_lock_irqsave(&die.lock, flags);
401 die.lock_owner = smp_processor_id();
402 die.lock_owner_depth = 0;
406 local_save_flags(flags);
408 if (++die.lock_owner_depth < 3) {
414 printk(KERN_EMERG "%s: %04lx [#%d]\n", str, err & 0xffff, ++die_counter);
415 #ifdef CONFIG_PREEMPT
416 printk(KERN_EMERG "PREEMPT ");
425 #ifdef CONFIG_DEBUG_PAGEALLOC
428 printk("DEBUG_PAGEALLOC");
433 if (notify_die(DIE_OOPS, str, regs, err,
434 current->thread.trap_no, SIGSEGV) !=
436 show_registers(regs);
437 /* Executive summary in case the oops scrolled away */
438 esp = (unsigned long) (®s->esp);
440 if (user_mode(regs)) {
442 ss = regs->xss & 0xffff;
444 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
445 print_symbol("%s", regs->eip);
446 printk(" SS:ESP %04x:%08lx\n", ss, esp);
451 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
455 spin_unlock_irqrestore(&die.lock, flags);
460 if (kexec_should_crash(current))
464 panic("Fatal exception in interrupt");
467 printk(KERN_EMERG "Fatal exception: panic in 5 seconds\n");
469 panic("Fatal exception");
475 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
477 if (!user_mode_vm(regs))
481 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
482 struct pt_regs * regs, long error_code,
485 struct task_struct *tsk = current;
486 tsk->thread.error_code = error_code;
487 tsk->thread.trap_no = trapnr;
489 if (regs->eflags & VM_MASK) {
495 if (!user_mode(regs))
500 force_sig_info(signr, info, tsk);
502 force_sig(signr, tsk);
507 if (!fixup_exception(regs))
508 die(str, regs, error_code);
513 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
514 if (ret) goto trap_signal;
519 #define DO_ERROR(trapnr, signr, str, name) \
520 fastcall void do_##name(struct pt_regs * regs, long error_code) \
522 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
525 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
528 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
529 fastcall void do_##name(struct pt_regs * regs, long error_code) \
532 info.si_signo = signr; \
534 info.si_code = sicode; \
535 info.si_addr = (void __user *)siaddr; \
536 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
539 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
542 #define DO_VM86_ERROR(trapnr, signr, str, name) \
543 fastcall void do_##name(struct pt_regs * regs, long error_code) \
545 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
548 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
551 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
552 fastcall void do_##name(struct pt_regs * regs, long error_code) \
555 info.si_signo = signr; \
557 info.si_code = sicode; \
558 info.si_addr = (void __user *)siaddr; \
559 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
562 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
565 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
566 #ifndef CONFIG_KPROBES
567 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
569 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
570 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
571 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip)
572 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
573 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
574 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
575 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
576 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
577 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0)
579 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
583 struct tss_struct *tss = &per_cpu(init_tss, cpu);
584 struct thread_struct *thread = ¤t->thread;
587 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
588 * invalid offset set (the LAZY one) and the faulting thread has
589 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
590 * and we set the offset field correctly. Then we let the CPU to
591 * restart the faulting instruction.
593 if (tss->io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
594 thread->io_bitmap_ptr) {
595 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
596 thread->io_bitmap_max);
598 * If the previously set map was extending to higher ports
599 * than the current one, pad extra space with 0xff (no access).
601 if (thread->io_bitmap_max < tss->io_bitmap_max)
602 memset((char *) tss->io_bitmap +
603 thread->io_bitmap_max, 0xff,
604 tss->io_bitmap_max - thread->io_bitmap_max);
605 tss->io_bitmap_max = thread->io_bitmap_max;
606 tss->io_bitmap_base = IO_BITMAP_OFFSET;
607 tss->io_bitmap_owner = thread;
613 current->thread.error_code = error_code;
614 current->thread.trap_no = 13;
616 if (regs->eflags & VM_MASK)
619 if (!user_mode(regs))
622 current->thread.error_code = error_code;
623 current->thread.trap_no = 13;
624 force_sig(SIGSEGV, current);
629 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
633 if (!fixup_exception(regs)) {
634 if (notify_die(DIE_GPF, "general protection fault", regs,
635 error_code, 13, SIGSEGV) == NOTIFY_STOP)
637 die("general protection fault", regs, error_code);
641 static void mem_parity_error(unsigned char reason, struct pt_regs * regs)
643 printk(KERN_EMERG "Uhhuh. NMI received. Dazed and confused, but trying "
645 printk(KERN_EMERG "You probably have a hardware problem with your RAM "
648 /* Clear and disable the memory parity error line. */
649 clear_mem_error(reason);
652 static void io_check_error(unsigned char reason, struct pt_regs * regs)
656 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
657 show_registers(regs);
659 /* Re-enable the IOCK line, wait for a few seconds */
660 reason = (reason & 0xf) | 8;
663 while (--i) udelay(1000);
668 static void unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
671 /* Might actually be able to figure out what the guilty party
678 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
679 reason, smp_processor_id());
680 printk("Dazed and confused, but trying to continue\n");
681 printk("Do you have a strange power saving mode enabled?\n");
684 static DEFINE_SPINLOCK(nmi_print_lock);
686 void die_nmi (struct pt_regs *regs, const char *msg)
688 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
692 spin_lock(&nmi_print_lock);
694 * We are in trouble anyway, lets at least try
695 * to get a message out.
698 printk(KERN_EMERG "%s", msg);
699 printk(" on CPU%d, eip %08lx, registers:\n",
700 smp_processor_id(), regs->eip);
701 show_registers(regs);
702 printk(KERN_EMERG "console shuts up ...\n");
704 spin_unlock(&nmi_print_lock);
707 /* If we are in kernel we are probably nested up pretty bad
708 * and might aswell get out now while we still can.
710 if (!user_mode_vm(regs)) {
711 current->thread.trap_no = 2;
718 static void default_do_nmi(struct pt_regs * regs)
720 unsigned char reason = 0;
722 /* Only the BSP gets external NMIs from the system. */
723 if (!smp_processor_id())
724 reason = get_nmi_reason();
726 if (!(reason & 0xc0)) {
727 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
730 #ifdef CONFIG_X86_LOCAL_APIC
732 * Ok, so this is none of the documented NMI sources,
733 * so it must be the NMI watchdog.
736 nmi_watchdog_tick(regs);
740 unknown_nmi_error(reason, regs);
743 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
746 mem_parity_error(reason, regs);
748 io_check_error(reason, regs);
750 * Reassert NMI in case it became active meanwhile
751 * as it's edge-triggered.
756 static int dummy_nmi_callback(struct pt_regs * regs, int cpu)
761 static nmi_callback_t nmi_callback = dummy_nmi_callback;
763 fastcall void do_nmi(struct pt_regs * regs, long error_code)
769 cpu = smp_processor_id();
773 if (!rcu_dereference(nmi_callback)(regs, cpu))
774 default_do_nmi(regs);
779 void set_nmi_callback(nmi_callback_t callback)
782 rcu_assign_pointer(nmi_callback, callback);
784 EXPORT_SYMBOL_GPL(set_nmi_callback);
786 void unset_nmi_callback(void)
788 nmi_callback = dummy_nmi_callback;
790 EXPORT_SYMBOL_GPL(unset_nmi_callback);
792 #ifdef CONFIG_KPROBES
793 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
795 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
798 /* This is an interrupt gate, because kprobes wants interrupts
799 disabled. Normal trap handlers don't. */
800 restore_interrupts(regs);
801 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
806 * Our handling of the processor debug registers is non-trivial.
807 * We do not clear them on entry and exit from the kernel. Therefore
808 * it is possible to get a watchpoint trap here from inside the kernel.
809 * However, the code in ./ptrace.c has ensured that the user can
810 * only set watchpoints on userspace addresses. Therefore the in-kernel
811 * watchpoint trap can only occur in code which is reading/writing
812 * from user space. Such code must not hold kernel locks (since it
813 * can equally take a page fault), therefore it is safe to call
814 * force_sig_info even though that claims and releases locks.
816 * Code in ./signal.c ensures that the debug control register
817 * is restored before we deliver any signal, and therefore that
818 * user code runs with the correct debug control register even though
821 * Being careful here means that we don't have to be as careful in a
822 * lot of more complicated places (task switching can be a bit lazy
823 * about restoring all the debug state, and ptrace doesn't have to
824 * find every occurrence of the TF bit that could be saved away even
827 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
829 unsigned int condition;
830 struct task_struct *tsk = current;
832 get_debugreg(condition, 6);
834 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
835 SIGTRAP) == NOTIFY_STOP)
837 /* It's safe to allow irq's after DR6 has been saved */
838 if (regs->eflags & X86_EFLAGS_IF)
841 /* Mask out spurious debug traps due to lazy DR7 setting */
842 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
843 if (!tsk->thread.debugreg[7])
847 if (regs->eflags & VM_MASK)
850 /* Save debug status register where ptrace can see it */
851 tsk->thread.debugreg[6] = condition;
854 * Single-stepping through TF: make sure we ignore any events in
855 * kernel space (but re-enable TF when returning to user mode).
857 if (condition & DR_STEP) {
859 * We already checked v86 mode above, so we can
860 * check for kernel mode by just checking the CPL
863 if (!user_mode(regs))
864 goto clear_TF_reenable;
867 /* Ok, finally something we can handle */
868 send_sigtrap(tsk, regs, error_code);
870 /* Disable additional traps. They'll be re-enabled when
871 * the signal is delivered.
878 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
882 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
883 regs->eflags &= ~TF_MASK;
888 * Note that we play around with the 'TS' bit in an attempt to get
889 * the correct behaviour even in the presence of the asynchronous
892 void math_error(void __user *eip)
894 struct task_struct * task;
896 unsigned short cwd, swd;
899 * Save the info for the exception handler and clear the error.
903 task->thread.trap_no = 16;
904 task->thread.error_code = 0;
905 info.si_signo = SIGFPE;
907 info.si_code = __SI_FAULT;
910 * (~cwd & swd) will mask out exceptions that are not set to unmasked
911 * status. 0x3f is the exception bits in these regs, 0x200 is the
912 * C1 reg you need in case of a stack fault, 0x040 is the stack
913 * fault bit. We should only be taking one exception at a time,
914 * so if this combination doesn't produce any single exception,
915 * then we have a bad program that isn't syncronizing its FPU usage
916 * and it will suffer the consequences since we won't be able to
917 * fully reproduce the context of the exception
919 cwd = get_fpu_cwd(task);
920 swd = get_fpu_swd(task);
921 switch (swd & ~cwd & 0x3f) {
922 case 0x000: /* No unmasked exception */
924 default: /* Multiple exceptions */
926 case 0x001: /* Invalid Op */
928 * swd & 0x240 == 0x040: Stack Underflow
929 * swd & 0x240 == 0x240: Stack Overflow
930 * User must clear the SF bit (0x40) if set
932 info.si_code = FPE_FLTINV;
934 case 0x002: /* Denormalize */
935 case 0x010: /* Underflow */
936 info.si_code = FPE_FLTUND;
938 case 0x004: /* Zero Divide */
939 info.si_code = FPE_FLTDIV;
941 case 0x008: /* Overflow */
942 info.si_code = FPE_FLTOVF;
944 case 0x020: /* Precision */
945 info.si_code = FPE_FLTRES;
948 force_sig_info(SIGFPE, &info, task);
951 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
954 math_error((void __user *)regs->eip);
957 static void simd_math_error(void __user *eip)
959 struct task_struct * task;
961 unsigned short mxcsr;
964 * Save the info for the exception handler and clear the error.
968 task->thread.trap_no = 19;
969 task->thread.error_code = 0;
970 info.si_signo = SIGFPE;
972 info.si_code = __SI_FAULT;
975 * The SIMD FPU exceptions are handled a little differently, as there
976 * is only a single status/control register. Thus, to determine which
977 * unmasked exception was caught we must mask the exception mask bits
978 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
980 mxcsr = get_fpu_mxcsr(task);
981 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
985 case 0x001: /* Invalid Op */
986 info.si_code = FPE_FLTINV;
988 case 0x002: /* Denormalize */
989 case 0x010: /* Underflow */
990 info.si_code = FPE_FLTUND;
992 case 0x004: /* Zero Divide */
993 info.si_code = FPE_FLTDIV;
995 case 0x008: /* Overflow */
996 info.si_code = FPE_FLTOVF;
998 case 0x020: /* Precision */
999 info.si_code = FPE_FLTRES;
1002 force_sig_info(SIGFPE, &info, task);
1005 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1009 /* Handle SIMD FPU exceptions on PIII+ processors. */
1011 simd_math_error((void __user *)regs->eip);
1014 * Handle strange cache flush from user space exception
1015 * in all other cases. This is undocumented behaviour.
1017 if (regs->eflags & VM_MASK) {
1018 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1022 current->thread.trap_no = 19;
1023 current->thread.error_code = error_code;
1024 die_if_kernel("cache flush denied", regs, error_code);
1025 force_sig(SIGSEGV, current);
1029 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1033 /* No need to warn about this any longer. */
1034 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1038 fastcall void setup_x86_bogus_stack(unsigned char * stk)
1040 unsigned long *switch16_ptr, *switch32_ptr;
1041 struct pt_regs *regs;
1042 unsigned long stack_top, stack_bot;
1043 unsigned short iret_frame16_off;
1044 int cpu = smp_processor_id();
1045 /* reserve the space on 32bit stack for the magic switch16 pointer */
1046 memmove(stk, stk + 8, sizeof(struct pt_regs));
1047 switch16_ptr = (unsigned long *)(stk + sizeof(struct pt_regs));
1048 regs = (struct pt_regs *)stk;
1049 /* now the switch32 on 16bit stack */
1050 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1051 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1052 switch32_ptr = (unsigned long *)(stack_top - 8);
1053 iret_frame16_off = CPU_16BIT_STACK_SIZE - 8 - 20;
1054 /* copy iret frame on 16bit stack */
1055 memcpy((void *)(stack_bot + iret_frame16_off), ®s->eip, 20);
1056 /* fill in the switch pointers */
1057 switch16_ptr[0] = (regs->esp & 0xffff0000) | iret_frame16_off;
1058 switch16_ptr[1] = __ESPFIX_SS;
1059 switch32_ptr[0] = (unsigned long)stk + sizeof(struct pt_regs) +
1060 8 - CPU_16BIT_STACK_SIZE;
1061 switch32_ptr[1] = __KERNEL_DS;
1064 fastcall unsigned char * fixup_x86_bogus_stack(unsigned short sp)
1066 unsigned long *switch32_ptr;
1067 unsigned char *stack16, *stack32;
1068 unsigned long stack_top, stack_bot;
1070 int cpu = smp_processor_id();
1071 stack_bot = (unsigned long)&per_cpu(cpu_16bit_stack, cpu);
1072 stack_top = stack_bot + CPU_16BIT_STACK_SIZE;
1073 switch32_ptr = (unsigned long *)(stack_top - 8);
1074 /* copy the data from 16bit stack to 32bit stack */
1075 len = CPU_16BIT_STACK_SIZE - 8 - sp;
1076 stack16 = (unsigned char *)(stack_bot + sp);
1077 stack32 = (unsigned char *)
1078 (switch32_ptr[0] + CPU_16BIT_STACK_SIZE - 8 - len);
1079 memcpy(stack32, stack16, len);
1084 * 'math_state_restore()' saves the current math information in the
1085 * old math state array, and gets the new ones from the current task
1087 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1088 * Don't touch unless you *really* know how it works.
1090 * Must be called with kernel preemption disabled (in this case,
1091 * local interrupts are disabled at the call-site in entry.S).
1093 asmlinkage void math_state_restore(struct pt_regs regs)
1095 struct thread_info *thread = current_thread_info();
1096 struct task_struct *tsk = thread->task;
1098 clts(); /* Allow maths ops (or we recurse) */
1099 if (!tsk_used_math(tsk))
1102 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1105 #ifndef CONFIG_MATH_EMULATION
1107 asmlinkage void math_emulate(long arg)
1109 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1110 printk(KERN_EMERG "killing %s.\n",current->comm);
1111 force_sig(SIGFPE,current);
1115 #endif /* CONFIG_MATH_EMULATION */
1117 #ifdef CONFIG_X86_F00F_BUG
1118 void __init trap_init_f00f_bug(void)
1120 __set_fixmap(FIX_F00F_IDT, __pa(&idt_table), PAGE_KERNEL_RO);
1123 * Update the IDT descriptor and reload the IDT so that
1124 * it uses the read-only mapped virtual address.
1126 idt_descr.address = fix_to_virt(FIX_F00F_IDT);
1127 load_idt(&idt_descr);
1131 #define _set_gate(gate_addr,type,dpl,addr,seg) \
1134 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1135 "movw %4,%%dx\n\t" \
1136 "movl %%eax,%0\n\t" \
1138 :"=m" (*((long *) (gate_addr))), \
1139 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1140 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1141 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1146 * This needs to use 'idt_table' rather than 'idt', and
1147 * thus use the _nonmapped_ version of the IDT, as the
1148 * Pentium F0 0F bugfix can have resulted in the mapped
1149 * IDT being write-protected.
1151 void set_intr_gate(unsigned int n, void *addr)
1153 _set_gate(idt_table+n,14,0,addr,__KERNEL_CS);
1157 * This routine sets up an interrupt gate at directory privilege level 3.
1159 static inline void set_system_intr_gate(unsigned int n, void *addr)
1161 _set_gate(idt_table+n, 14, 3, addr, __KERNEL_CS);
1164 static void __init set_trap_gate(unsigned int n, void *addr)
1166 _set_gate(idt_table+n,15,0,addr,__KERNEL_CS);
1169 static void __init set_system_gate(unsigned int n, void *addr)
1171 _set_gate(idt_table+n,15,3,addr,__KERNEL_CS);
1174 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1176 _set_gate(idt_table+n,5,0,0,(gdt_entry<<3));
1180 void __init trap_init(void)
1183 void __iomem *p = ioremap(0x0FFFD9, 4);
1184 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1190 #ifdef CONFIG_X86_LOCAL_APIC
1191 init_apic_mappings();
1194 set_trap_gate(0,÷_error);
1195 set_intr_gate(1,&debug);
1196 set_intr_gate(2,&nmi);
1197 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1198 set_system_gate(4,&overflow);
1199 set_trap_gate(5,&bounds);
1200 set_trap_gate(6,&invalid_op);
1201 set_trap_gate(7,&device_not_available);
1202 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1203 set_trap_gate(9,&coprocessor_segment_overrun);
1204 set_trap_gate(10,&invalid_TSS);
1205 set_trap_gate(11,&segment_not_present);
1206 set_trap_gate(12,&stack_segment);
1207 set_trap_gate(13,&general_protection);
1208 set_intr_gate(14,&page_fault);
1209 set_trap_gate(15,&spurious_interrupt_bug);
1210 set_trap_gate(16,&coprocessor_error);
1211 set_trap_gate(17,&alignment_check);
1212 #ifdef CONFIG_X86_MCE
1213 set_trap_gate(18,&machine_check);
1215 set_trap_gate(19,&simd_coprocessor_error);
1219 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1220 * Generates a compile-time "error: zero width for bit-field" if
1221 * the alignment is wrong.
1223 struct fxsrAlignAssert {
1224 int _:!(offsetof(struct task_struct,
1225 thread.i387.fxsave) & 15);
1228 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1229 set_in_cr4(X86_CR4_OSFXSR);
1233 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1235 set_in_cr4(X86_CR4_OSXMMEXCPT);
1239 set_system_gate(SYSCALL_VECTOR,&system_call);
1242 * Should be a barrier for any external CPU state.
1249 static int __init kstack_setup(char *s)
1251 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1254 __setup("kstack=", kstack_setup);
1256 static int __init call_trace_setup(char *s)
1258 if (strcmp(s, "old") == 0)
1260 else if (strcmp(s, "both") == 0)
1262 else if (strcmp(s, "new") == 0)
1266 __setup("call_trace=", call_trace_setup);