2 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
9 * 'Traps.c' handles hardware traps and faults after we have saved some
12 #include <linux/sched.h>
13 #include <linux/kernel.h>
14 #include <linux/string.h>
15 #include <linux/errno.h>
16 #include <linux/timer.h>
18 #include <linux/init.h>
19 #include <linux/delay.h>
20 #include <linux/spinlock.h>
21 #include <linux/interrupt.h>
22 #include <linux/highmem.h>
23 #include <linux/kallsyms.h>
24 #include <linux/ptrace.h>
25 #include <linux/utsname.h>
26 #include <linux/kprobes.h>
27 #include <linux/kexec.h>
28 #include <linux/unwind.h>
29 #include <linux/uaccess.h>
30 #include <linux/nmi.h>
31 #include <linux/bug.h>
34 #include <linux/ioport.h>
35 #include <linux/eisa.h>
39 #include <linux/mca.h>
42 #if defined(CONFIG_EDAC)
43 #include <linux/edac.h>
46 #include <asm/processor.h>
47 #include <asm/system.h>
49 #include <asm/atomic.h>
50 #include <asm/debugreg.h>
54 #include <asm/unwind.h>
56 #include <asm/arch_hooks.h>
57 #include <linux/kdebug.h>
58 #include <asm/stacktrace.h>
60 #include <linux/module.h>
62 #include "mach_traps.h"
64 int panic_on_unrecovered_nmi;
66 DECLARE_BITMAP(used_vectors, NR_VECTORS);
67 EXPORT_SYMBOL_GPL(used_vectors);
69 asmlinkage int system_call(void);
71 /* Do we ignore FPU interrupts ? */
72 char ignore_fpu_irq = 0;
75 * The IDT has to be page-aligned to simplify the Pentium
76 * F0 0F bug workaround.. We have a special link segment
79 struct desc_struct idt_table[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
81 asmlinkage void divide_error(void);
82 asmlinkage void debug(void);
83 asmlinkage void nmi(void);
84 asmlinkage void int3(void);
85 asmlinkage void overflow(void);
86 asmlinkage void bounds(void);
87 asmlinkage void invalid_op(void);
88 asmlinkage void device_not_available(void);
89 asmlinkage void coprocessor_segment_overrun(void);
90 asmlinkage void invalid_TSS(void);
91 asmlinkage void segment_not_present(void);
92 asmlinkage void stack_segment(void);
93 asmlinkage void general_protection(void);
94 asmlinkage void page_fault(void);
95 asmlinkage void coprocessor_error(void);
96 asmlinkage void simd_coprocessor_error(void);
97 asmlinkage void alignment_check(void);
98 asmlinkage void spurious_interrupt_bug(void);
99 asmlinkage void machine_check(void);
101 int kstack_depth_to_print = 24;
102 static unsigned int code_bytes = 64;
104 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p, unsigned size)
106 return p > (void *)tinfo &&
107 p <= (void *)tinfo + THREAD_SIZE - size;
110 /* The form of the top of the frame on the stack */
112 struct stack_frame *next_frame;
113 unsigned long return_address;
116 static inline unsigned long print_context_stack(struct thread_info *tinfo,
117 unsigned long *stack, unsigned long ebp,
118 const struct stacktrace_ops *ops, void *data)
120 #ifdef CONFIG_FRAME_POINTER
121 struct stack_frame *frame = (struct stack_frame *)ebp;
122 while (valid_stack_ptr(tinfo, frame, sizeof(*frame))) {
123 struct stack_frame *next;
126 addr = frame->return_address;
127 ops->address(data, addr);
129 * break out of recursive entries (such as
130 * end_of_stack_stop_unwind_function). Also,
131 * we can never allow a frame pointer to
134 next = frame->next_frame;
140 while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) {
144 if (__kernel_text_address(addr))
145 ops->address(data, addr);
151 #define MSG(msg) ops->warning(data, msg)
153 void dump_trace(struct task_struct *task, struct pt_regs *regs,
154 unsigned long *stack,
155 const struct stacktrace_ops *ops, void *data)
157 unsigned long ebp = 0;
166 stack = (unsigned long *)task->thread.esp;
169 #ifdef CONFIG_FRAME_POINTER
171 if (task == current) {
172 /* Grab ebp right from our regs */
173 asm ("movl %%ebp, %0" : "=r" (ebp) : );
175 /* ebp is the last reg pushed by switch_to */
176 ebp = *(unsigned long *) task->thread.esp;
182 struct thread_info *context;
183 context = (struct thread_info *)
184 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
185 ebp = print_context_stack(context, stack, ebp, ops, data);
186 /* Should be after the line below, but somewhere
187 in early boot context comes out corrupted and we
188 can't reference it -AK */
189 if (ops->stack(data, "IRQ") < 0)
191 stack = (unsigned long*)context->previous_esp;
194 touch_nmi_watchdog();
197 EXPORT_SYMBOL(dump_trace);
200 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
203 print_symbol(msg, symbol);
207 static void print_trace_warning(void *data, char *msg)
209 printk("%s%s\n", (char *)data, msg);
212 static int print_trace_stack(void *data, char *name)
218 * Print one address/symbol entries per line.
220 static void print_trace_address(void *data, unsigned long addr)
222 printk("%s [<%08lx>] ", (char *)data, addr);
223 print_symbol("%s\n", addr);
224 touch_nmi_watchdog();
227 static const struct stacktrace_ops print_trace_ops = {
228 .warning = print_trace_warning,
229 .warning_symbol = print_trace_warning_symbol,
230 .stack = print_trace_stack,
231 .address = print_trace_address,
235 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
236 unsigned long * stack, char *log_lvl)
238 dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
239 printk("%s =======================\n", log_lvl);
242 void show_trace(struct task_struct *task, struct pt_regs *regs,
243 unsigned long * stack)
245 show_trace_log_lvl(task, regs, stack, "");
248 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
249 unsigned long *esp, char *log_lvl)
251 unsigned long *stack;
256 esp = (unsigned long*)task->thread.esp;
258 esp = (unsigned long *)&esp;
262 for(i = 0; i < kstack_depth_to_print; i++) {
263 if (kstack_end(stack))
265 if (i && ((i % 8) == 0))
266 printk("\n%s ", log_lvl);
267 printk("%08lx ", *stack++);
269 printk("\n%sCall Trace:\n", log_lvl);
270 show_trace_log_lvl(task, regs, esp, log_lvl);
273 void show_stack(struct task_struct *task, unsigned long *esp)
276 show_stack_log_lvl(task, NULL, esp, "");
280 * The architecture-independent dump_stack generator
282 void dump_stack(void)
286 show_trace(current, NULL, &stack);
289 EXPORT_SYMBOL(dump_stack);
291 void show_registers(struct pt_regs *regs)
296 __show_registers(regs, 0);
297 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
298 TASK_COMM_LEN, current->comm, task_pid_nr(current),
299 current_thread_info(), current, task_thread_info(current));
301 * When in-kernel, we also print out the stack and code at the
302 * time of the fault..
304 if (!user_mode_vm(regs)) {
306 unsigned int code_prologue = code_bytes * 43 / 64;
307 unsigned int code_len = code_bytes;
310 printk("\n" KERN_EMERG "Stack: ");
311 show_stack_log_lvl(NULL, regs, ®s->esp, KERN_EMERG);
313 printk(KERN_EMERG "Code: ");
315 eip = (u8 *)regs->eip - code_prologue;
316 if (eip < (u8 *)PAGE_OFFSET ||
317 probe_kernel_address(eip, c)) {
318 /* try starting at EIP */
319 eip = (u8 *)regs->eip;
320 code_len = code_len - code_prologue + 1;
322 for (i = 0; i < code_len; i++, eip++) {
323 if (eip < (u8 *)PAGE_OFFSET ||
324 probe_kernel_address(eip, c)) {
325 printk(" Bad EIP value.");
328 if (eip == (u8 *)regs->eip)
329 printk("<%02x> ", c);
337 int is_valid_bugaddr(unsigned long eip)
341 if (eip < PAGE_OFFSET)
343 if (probe_kernel_address((unsigned short *)eip, ud2))
346 return ud2 == 0x0b0f;
350 * This is gone through when something in the kernel has done something bad and
351 * is about to be terminated.
353 void die(const char * str, struct pt_regs * regs, long err)
358 int lock_owner_depth;
360 .lock = __RAW_SPIN_LOCK_UNLOCKED,
362 .lock_owner_depth = 0
364 static int die_counter;
369 if (die.lock_owner != raw_smp_processor_id()) {
371 __raw_spin_lock(&die.lock);
372 raw_local_save_flags(flags);
373 die.lock_owner = smp_processor_id();
374 die.lock_owner_depth = 0;
378 raw_local_save_flags(flags);
380 if (++die.lock_owner_depth < 3) {
384 report_bug(regs->eip, regs);
386 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff,
388 #ifdef CONFIG_PREEMPT
394 #ifdef CONFIG_DEBUG_PAGEALLOC
395 printk("DEBUG_PAGEALLOC");
399 if (notify_die(DIE_OOPS, str, regs, err,
400 current->thread.trap_no, SIGSEGV) !=
402 show_registers(regs);
403 /* Executive summary in case the oops scrolled away */
404 esp = (unsigned long) (®s->esp);
406 if (user_mode(regs)) {
408 ss = regs->xss & 0xffff;
410 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->eip);
411 print_symbol("%s", regs->eip);
412 printk(" SS:ESP %04x:%08lx\n", ss, esp);
417 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
421 add_taint(TAINT_DIE);
422 __raw_spin_unlock(&die.lock);
423 raw_local_irq_restore(flags);
428 if (kexec_should_crash(current))
432 panic("Fatal exception in interrupt");
435 panic("Fatal exception");
441 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
443 if (!user_mode_vm(regs))
447 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
448 struct pt_regs * regs, long error_code,
451 struct task_struct *tsk = current;
453 if (regs->eflags & VM_MASK) {
459 if (!user_mode(regs))
464 * We want error_code and trap_no set for userspace faults and
465 * kernelspace faults which result in die(), but not
466 * kernelspace faults which are fixed up. die() gives the
467 * process no chance to handle the signal and notice the
468 * kernel fault information, so that won't result in polluting
469 * the information about previously queued, but not yet
470 * delivered, faults. See also do_general_protection below.
472 tsk->thread.error_code = error_code;
473 tsk->thread.trap_no = trapnr;
476 force_sig_info(signr, info, tsk);
478 force_sig(signr, tsk);
483 if (!fixup_exception(regs)) {
484 tsk->thread.error_code = error_code;
485 tsk->thread.trap_no = trapnr;
486 die(str, regs, error_code);
492 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
493 if (ret) goto trap_signal;
498 #define DO_ERROR(trapnr, signr, str, name) \
499 fastcall void do_##name(struct pt_regs * regs, long error_code) \
501 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
504 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
507 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
508 fastcall void do_##name(struct pt_regs * regs, long error_code) \
512 local_irq_enable(); \
513 info.si_signo = signr; \
515 info.si_code = sicode; \
516 info.si_addr = (void __user *)siaddr; \
517 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
520 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
523 #define DO_VM86_ERROR(trapnr, signr, str, name) \
524 fastcall void do_##name(struct pt_regs * regs, long error_code) \
526 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
529 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
532 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
533 fastcall void do_##name(struct pt_regs * regs, long error_code) \
536 info.si_signo = signr; \
538 info.si_code = sicode; \
539 info.si_addr = (void __user *)siaddr; \
540 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
543 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
546 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->eip)
547 #ifndef CONFIG_KPROBES
548 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
550 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
551 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
552 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->eip, 0)
553 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
554 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
555 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
556 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
557 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0)
558 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0, 1)
560 fastcall void __kprobes do_general_protection(struct pt_regs * regs,
564 struct tss_struct *tss = &per_cpu(init_tss, cpu);
565 struct thread_struct *thread = ¤t->thread;
568 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
569 * invalid offset set (the LAZY one) and the faulting thread has
570 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
571 * and we set the offset field correctly. Then we let the CPU to
572 * restart the faulting instruction.
574 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
575 thread->io_bitmap_ptr) {
576 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
577 thread->io_bitmap_max);
579 * If the previously set map was extending to higher ports
580 * than the current one, pad extra space with 0xff (no access).
582 if (thread->io_bitmap_max < tss->io_bitmap_max)
583 memset((char *) tss->io_bitmap +
584 thread->io_bitmap_max, 0xff,
585 tss->io_bitmap_max - thread->io_bitmap_max);
586 tss->io_bitmap_max = thread->io_bitmap_max;
587 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
588 tss->io_bitmap_owner = thread;
594 if (regs->eflags & VM_MASK)
597 if (!user_mode(regs))
600 current->thread.error_code = error_code;
601 current->thread.trap_no = 13;
602 if (show_unhandled_signals && unhandled_signal(current, SIGSEGV) &&
605 "%s[%d] general protection eip:%lx esp:%lx error:%lx\n",
606 current->comm, task_pid_nr(current),
607 regs->eip, regs->esp, error_code);
609 force_sig(SIGSEGV, current);
614 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
618 if (!fixup_exception(regs)) {
619 current->thread.error_code = error_code;
620 current->thread.trap_no = 13;
621 if (notify_die(DIE_GPF, "general protection fault", regs,
622 error_code, 13, SIGSEGV) == NOTIFY_STOP)
624 die("general protection fault", regs, error_code);
628 static __kprobes void
629 mem_parity_error(unsigned char reason, struct pt_regs * regs)
631 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
632 "CPU %d.\n", reason, smp_processor_id());
633 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
635 #if defined(CONFIG_EDAC)
636 if(edac_handler_set()) {
637 edac_atomic_assert_error();
642 if (panic_on_unrecovered_nmi)
643 panic("NMI: Not continuing");
645 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
647 /* Clear and disable the memory parity error line. */
648 clear_mem_error(reason);
651 static __kprobes void
652 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 __kprobes void
669 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
672 /* Might actually be able to figure out what the guilty party
679 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
680 "CPU %d.\n", reason, smp_processor_id());
681 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
682 if (panic_on_unrecovered_nmi)
683 panic("NMI: Not continuing");
685 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
688 static DEFINE_SPINLOCK(nmi_print_lock);
690 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
692 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
696 spin_lock(&nmi_print_lock);
698 * We are in trouble anyway, lets at least try
699 * to get a message out.
702 printk(KERN_EMERG "%s", msg);
703 printk(" on CPU%d, eip %08lx, registers:\n",
704 smp_processor_id(), regs->eip);
705 show_registers(regs);
707 spin_unlock(&nmi_print_lock);
710 /* If we are in kernel we are probably nested up pretty bad
711 * and might aswell get out now while we still can.
713 if (!user_mode_vm(regs)) {
714 current->thread.trap_no = 2;
721 static __kprobes void default_do_nmi(struct pt_regs * regs)
723 unsigned char reason = 0;
725 /* Only the BSP gets external NMIs from the system. */
726 if (!smp_processor_id())
727 reason = get_nmi_reason();
729 if (!(reason & 0xc0)) {
730 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
733 #ifdef CONFIG_X86_LOCAL_APIC
735 * Ok, so this is none of the documented NMI sources,
736 * so it must be the NMI watchdog.
738 if (nmi_watchdog_tick(regs, reason))
740 if (!do_nmi_callback(regs, smp_processor_id()))
742 unknown_nmi_error(reason, regs);
746 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
749 mem_parity_error(reason, regs);
751 io_check_error(reason, regs);
753 * Reassert NMI in case it became active meanwhile
754 * as it's edge-triggered.
759 static int ignore_nmis;
761 fastcall __kprobes void do_nmi(struct pt_regs * regs, long error_code)
767 cpu = smp_processor_id();
772 default_do_nmi(regs);
783 void restart_nmi(void)
789 #ifdef CONFIG_KPROBES
790 fastcall void __kprobes do_int3(struct pt_regs *regs, long error_code)
792 trace_hardirqs_fixup();
794 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
797 /* This is an interrupt gate, because kprobes wants interrupts
798 disabled. Normal trap handlers don't. */
799 restore_interrupts(regs);
800 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
805 * Our handling of the processor debug registers is non-trivial.
806 * We do not clear them on entry and exit from the kernel. Therefore
807 * it is possible to get a watchpoint trap here from inside the kernel.
808 * However, the code in ./ptrace.c has ensured that the user can
809 * only set watchpoints on userspace addresses. Therefore the in-kernel
810 * watchpoint trap can only occur in code which is reading/writing
811 * from user space. Such code must not hold kernel locks (since it
812 * can equally take a page fault), therefore it is safe to call
813 * force_sig_info even though that claims and releases locks.
815 * Code in ./signal.c ensures that the debug control register
816 * is restored before we deliver any signal, and therefore that
817 * user code runs with the correct debug control register even though
820 * Being careful here means that we don't have to be as careful in a
821 * lot of more complicated places (task switching can be a bit lazy
822 * about restoring all the debug state, and ptrace doesn't have to
823 * find every occurrence of the TF bit that could be saved away even
826 fastcall void __kprobes do_debug(struct pt_regs * regs, long error_code)
828 unsigned int condition;
829 struct task_struct *tsk = current;
831 get_debugreg(condition, 6);
833 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
834 SIGTRAP) == NOTIFY_STOP)
836 /* It's safe to allow irq's after DR6 has been saved */
837 if (regs->eflags & X86_EFLAGS_IF)
840 /* Mask out spurious debug traps due to lazy DR7 setting */
841 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
842 if (!tsk->thread.debugreg[7])
846 if (regs->eflags & VM_MASK)
849 /* Save debug status register where ptrace can see it */
850 tsk->thread.debugreg[6] = condition;
853 * Single-stepping through TF: make sure we ignore any events in
854 * kernel space (but re-enable TF when returning to user mode).
856 if (condition & DR_STEP) {
858 * We already checked v86 mode above, so we can
859 * check for kernel mode by just checking the CPL
862 if (!user_mode(regs))
863 goto clear_TF_reenable;
866 /* Ok, finally something we can handle */
867 send_sigtrap(tsk, regs, error_code);
869 /* Disable additional traps. They'll be re-enabled when
870 * the signal is delivered.
877 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
881 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
882 regs->eflags &= ~TF_MASK;
887 * Note that we play around with the 'TS' bit in an attempt to get
888 * the correct behaviour even in the presence of the asynchronous
891 void math_error(void __user *eip)
893 struct task_struct * task;
895 unsigned short cwd, swd;
898 * Save the info for the exception handler and clear the error.
902 task->thread.trap_no = 16;
903 task->thread.error_code = 0;
904 info.si_signo = SIGFPE;
906 info.si_code = __SI_FAULT;
909 * (~cwd & swd) will mask out exceptions that are not set to unmasked
910 * status. 0x3f is the exception bits in these regs, 0x200 is the
911 * C1 reg you need in case of a stack fault, 0x040 is the stack
912 * fault bit. We should only be taking one exception at a time,
913 * so if this combination doesn't produce any single exception,
914 * then we have a bad program that isn't syncronizing its FPU usage
915 * and it will suffer the consequences since we won't be able to
916 * fully reproduce the context of the exception
918 cwd = get_fpu_cwd(task);
919 swd = get_fpu_swd(task);
920 switch (swd & ~cwd & 0x3f) {
921 case 0x000: /* No unmasked exception */
923 default: /* Multiple exceptions */
925 case 0x001: /* Invalid Op */
927 * swd & 0x240 == 0x040: Stack Underflow
928 * swd & 0x240 == 0x240: Stack Overflow
929 * User must clear the SF bit (0x40) if set
931 info.si_code = FPE_FLTINV;
933 case 0x002: /* Denormalize */
934 case 0x010: /* Underflow */
935 info.si_code = FPE_FLTUND;
937 case 0x004: /* Zero Divide */
938 info.si_code = FPE_FLTDIV;
940 case 0x008: /* Overflow */
941 info.si_code = FPE_FLTOVF;
943 case 0x020: /* Precision */
944 info.si_code = FPE_FLTRES;
947 force_sig_info(SIGFPE, &info, task);
950 fastcall void do_coprocessor_error(struct pt_regs * regs, long error_code)
953 math_error((void __user *)regs->eip);
956 static void simd_math_error(void __user *eip)
958 struct task_struct * task;
960 unsigned short mxcsr;
963 * Save the info for the exception handler and clear the error.
967 task->thread.trap_no = 19;
968 task->thread.error_code = 0;
969 info.si_signo = SIGFPE;
971 info.si_code = __SI_FAULT;
974 * The SIMD FPU exceptions are handled a little differently, as there
975 * is only a single status/control register. Thus, to determine which
976 * unmasked exception was caught we must mask the exception mask bits
977 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
979 mxcsr = get_fpu_mxcsr(task);
980 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
984 case 0x001: /* Invalid Op */
985 info.si_code = FPE_FLTINV;
987 case 0x002: /* Denormalize */
988 case 0x010: /* Underflow */
989 info.si_code = FPE_FLTUND;
991 case 0x004: /* Zero Divide */
992 info.si_code = FPE_FLTDIV;
994 case 0x008: /* Overflow */
995 info.si_code = FPE_FLTOVF;
997 case 0x020: /* Precision */
998 info.si_code = FPE_FLTRES;
1001 force_sig_info(SIGFPE, &info, task);
1004 fastcall void do_simd_coprocessor_error(struct pt_regs * regs,
1008 /* Handle SIMD FPU exceptions on PIII+ processors. */
1010 simd_math_error((void __user *)regs->eip);
1013 * Handle strange cache flush from user space exception
1014 * in all other cases. This is undocumented behaviour.
1016 if (regs->eflags & VM_MASK) {
1017 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1021 current->thread.trap_no = 19;
1022 current->thread.error_code = error_code;
1023 die_if_kernel("cache flush denied", regs, error_code);
1024 force_sig(SIGSEGV, current);
1028 fastcall void do_spurious_interrupt_bug(struct pt_regs * regs,
1032 /* No need to warn about this any longer. */
1033 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1037 fastcall unsigned long patch_espfix_desc(unsigned long uesp,
1040 struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt;
1041 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1042 unsigned long new_kesp = kesp - base;
1043 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1044 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1045 /* Set up base for espfix segment */
1046 desc &= 0x00f0ff0000000000ULL;
1047 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1048 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1049 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1050 (lim_pages & 0xffff);
1051 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1056 * 'math_state_restore()' saves the current math information in the
1057 * old math state array, and gets the new ones from the current task
1059 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1060 * Don't touch unless you *really* know how it works.
1062 * Must be called with kernel preemption disabled (in this case,
1063 * local interrupts are disabled at the call-site in entry.S).
1065 asmlinkage void math_state_restore(void)
1067 struct thread_info *thread = current_thread_info();
1068 struct task_struct *tsk = thread->task;
1070 clts(); /* Allow maths ops (or we recurse) */
1071 if (!tsk_used_math(tsk))
1074 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1077 EXPORT_SYMBOL_GPL(math_state_restore);
1079 #ifndef CONFIG_MATH_EMULATION
1081 asmlinkage void math_emulate(long arg)
1083 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1084 printk(KERN_EMERG "killing %s.\n",current->comm);
1085 force_sig(SIGFPE,current);
1089 #endif /* CONFIG_MATH_EMULATION */
1092 * This needs to use 'idt_table' rather than 'idt', and
1093 * thus use the _nonmapped_ version of the IDT, as the
1094 * Pentium F0 0F bugfix can have resulted in the mapped
1095 * IDT being write-protected.
1097 void set_intr_gate(unsigned int n, void *addr)
1099 _set_gate(n, DESCTYPE_INT, addr, __KERNEL_CS);
1103 * This routine sets up an interrupt gate at directory privilege level 3.
1105 static inline void set_system_intr_gate(unsigned int n, void *addr)
1107 _set_gate(n, DESCTYPE_INT | DESCTYPE_DPL3, addr, __KERNEL_CS);
1110 static void __init set_trap_gate(unsigned int n, void *addr)
1112 _set_gate(n, DESCTYPE_TRAP, addr, __KERNEL_CS);
1115 static void __init set_system_gate(unsigned int n, void *addr)
1117 _set_gate(n, DESCTYPE_TRAP | DESCTYPE_DPL3, addr, __KERNEL_CS);
1120 static void __init set_task_gate(unsigned int n, unsigned int gdt_entry)
1122 _set_gate(n, DESCTYPE_TASK, (void *)0, (gdt_entry<<3));
1126 void __init trap_init(void)
1131 void __iomem *p = ioremap(0x0FFFD9, 4);
1132 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1138 #ifdef CONFIG_X86_LOCAL_APIC
1139 init_apic_mappings();
1142 set_trap_gate(0,÷_error);
1143 set_intr_gate(1,&debug);
1144 set_intr_gate(2,&nmi);
1145 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1146 set_system_gate(4,&overflow);
1147 set_trap_gate(5,&bounds);
1148 set_trap_gate(6,&invalid_op);
1149 set_trap_gate(7,&device_not_available);
1150 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1151 set_trap_gate(9,&coprocessor_segment_overrun);
1152 set_trap_gate(10,&invalid_TSS);
1153 set_trap_gate(11,&segment_not_present);
1154 set_trap_gate(12,&stack_segment);
1155 set_trap_gate(13,&general_protection);
1156 set_intr_gate(14,&page_fault);
1157 set_trap_gate(15,&spurious_interrupt_bug);
1158 set_trap_gate(16,&coprocessor_error);
1159 set_trap_gate(17,&alignment_check);
1160 #ifdef CONFIG_X86_MCE
1161 set_trap_gate(18,&machine_check);
1163 set_trap_gate(19,&simd_coprocessor_error);
1167 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1168 * Generates a compile-time "error: zero width for bit-field" if
1169 * the alignment is wrong.
1171 struct fxsrAlignAssert {
1172 int _:!(offsetof(struct task_struct,
1173 thread.i387.fxsave) & 15);
1176 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1177 set_in_cr4(X86_CR4_OSFXSR);
1181 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1183 set_in_cr4(X86_CR4_OSXMMEXCPT);
1187 set_system_gate(SYSCALL_VECTOR,&system_call);
1189 /* Reserve all the builtin and the syscall vector. */
1190 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1191 set_bit(i, used_vectors);
1192 set_bit(SYSCALL_VECTOR, used_vectors);
1195 * Should be a barrier for any external CPU state.
1202 static int __init kstack_setup(char *s)
1204 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1207 __setup("kstack=", kstack_setup);
1209 static int __init code_bytes_setup(char *s)
1211 code_bytes = simple_strtoul(s, NULL, 0);
1212 if (code_bytes > 8192)
1217 __setup("code_bytes=", code_bytes_setup);