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 gate_desc idt_table[256]
80 __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
82 asmlinkage void divide_error(void);
83 asmlinkage void debug(void);
84 asmlinkage void nmi(void);
85 asmlinkage void int3(void);
86 asmlinkage void overflow(void);
87 asmlinkage void bounds(void);
88 asmlinkage void invalid_op(void);
89 asmlinkage void device_not_available(void);
90 asmlinkage void coprocessor_segment_overrun(void);
91 asmlinkage void invalid_TSS(void);
92 asmlinkage void segment_not_present(void);
93 asmlinkage void stack_segment(void);
94 asmlinkage void general_protection(void);
95 asmlinkage void page_fault(void);
96 asmlinkage void coprocessor_error(void);
97 asmlinkage void simd_coprocessor_error(void);
98 asmlinkage void alignment_check(void);
99 asmlinkage void spurious_interrupt_bug(void);
100 asmlinkage void machine_check(void);
102 int kstack_depth_to_print = 24;
103 static unsigned int code_bytes = 64;
105 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p, unsigned size)
107 return p > (void *)tinfo &&
108 p <= (void *)tinfo + THREAD_SIZE - size;
111 /* The form of the top of the frame on the stack */
113 struct stack_frame *next_frame;
114 unsigned long return_address;
117 static inline unsigned long print_context_stack(struct thread_info *tinfo,
118 unsigned long *stack, unsigned long bp,
119 const struct stacktrace_ops *ops, void *data)
121 struct stack_frame *frame = (struct stack_frame *)bp;
123 while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) {
127 if (__kernel_text_address(addr)) {
128 if ((unsigned long) stack == bp + 4) {
129 ops->address(data, addr, 1);
130 frame = frame->next_frame;
131 bp = (unsigned long) frame;
133 ops->address(data, addr, bp == 0);
141 #define MSG(msg) ops->warning(data, msg)
143 void dump_trace(struct task_struct *task, struct pt_regs *regs,
144 unsigned long *stack, unsigned long bp,
145 const struct stacktrace_ops *ops, void *data)
154 stack = (unsigned long *)task->thread.sp;
157 #ifdef CONFIG_FRAME_POINTER
159 if (task == current) {
160 /* Grab bp right from our regs */
161 asm ("movl %%ebp, %0" : "=r" (bp) : );
163 /* bp is the last reg pushed by switch_to */
164 bp = *(unsigned long *) task->thread.sp;
170 struct thread_info *context;
171 context = (struct thread_info *)
172 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
173 bp = print_context_stack(context, stack, bp, ops, data);
174 /* Should be after the line below, but somewhere
175 in early boot context comes out corrupted and we
176 can't reference it -AK */
177 if (ops->stack(data, "IRQ") < 0)
179 stack = (unsigned long*)context->previous_esp;
182 touch_nmi_watchdog();
185 EXPORT_SYMBOL(dump_trace);
188 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
191 print_symbol(msg, symbol);
195 static void print_trace_warning(void *data, char *msg)
197 printk("%s%s\n", (char *)data, msg);
200 static int print_trace_stack(void *data, char *name)
206 * Print one address/symbol entries per line.
208 static void print_trace_address(void *data, unsigned long addr, int reliable)
210 printk("%s [<%08lx>] ", (char *)data, addr);
213 print_symbol("%s\n", addr);
214 touch_nmi_watchdog();
217 static const struct stacktrace_ops print_trace_ops = {
218 .warning = print_trace_warning,
219 .warning_symbol = print_trace_warning_symbol,
220 .stack = print_trace_stack,
221 .address = print_trace_address,
225 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
226 unsigned long *stack, unsigned long bp, char *log_lvl)
228 dump_trace(task, regs, stack, bp, &print_trace_ops, log_lvl);
229 printk("%s =======================\n", log_lvl);
232 void show_trace(struct task_struct *task, struct pt_regs *regs,
233 unsigned long *stack, unsigned long bp)
235 show_trace_log_lvl(task, regs, stack, bp, "");
238 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
239 unsigned long *sp, unsigned long bp, char *log_lvl)
241 unsigned long *stack;
246 sp = (unsigned long*)task->thread.sp;
248 sp = (unsigned long *)&sp;
252 for(i = 0; i < kstack_depth_to_print; i++) {
253 if (kstack_end(stack))
255 if (i && ((i % 8) == 0))
256 printk("\n%s ", log_lvl);
257 printk("%08lx ", *stack++);
259 printk("\n%sCall Trace:\n", log_lvl);
260 show_trace_log_lvl(task, regs, sp, bp, log_lvl);
263 void show_stack(struct task_struct *task, unsigned long *sp)
266 show_stack_log_lvl(task, NULL, sp, 0, "");
270 * The architecture-independent dump_stack generator
272 void dump_stack(void)
275 unsigned long bp = 0;
277 #ifdef CONFIG_FRAME_POINTER
279 asm("movl %%ebp, %0" : "=r" (bp):);
282 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
283 current->pid, current->comm, print_tainted(),
284 init_utsname()->release,
285 (int)strcspn(init_utsname()->version, " "),
286 init_utsname()->version);
287 show_trace(current, NULL, &stack, bp);
290 EXPORT_SYMBOL(dump_stack);
292 void show_registers(struct pt_regs *regs)
297 __show_registers(regs, 0);
298 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
299 TASK_COMM_LEN, current->comm, task_pid_nr(current),
300 current_thread_info(), current, task_thread_info(current));
302 * When in-kernel, we also print out the stack and code at the
303 * time of the fault..
305 if (!user_mode_vm(regs)) {
307 unsigned int code_prologue = code_bytes * 43 / 64;
308 unsigned int code_len = code_bytes;
311 printk("\n" KERN_EMERG "Stack: ");
312 show_stack_log_lvl(NULL, regs, ®s->sp, 0, KERN_EMERG);
314 printk(KERN_EMERG "Code: ");
316 ip = (u8 *)regs->ip - code_prologue;
317 if (ip < (u8 *)PAGE_OFFSET ||
318 probe_kernel_address(ip, c)) {
319 /* try starting at EIP */
321 code_len = code_len - code_prologue + 1;
323 for (i = 0; i < code_len; i++, ip++) {
324 if (ip < (u8 *)PAGE_OFFSET ||
325 probe_kernel_address(ip, c)) {
326 printk(" Bad EIP value.");
329 if (ip == (u8 *)regs->ip)
330 printk("<%02x> ", c);
338 int is_valid_bugaddr(unsigned long ip)
342 if (ip < PAGE_OFFSET)
344 if (probe_kernel_address((unsigned short *)ip, ud2))
347 return ud2 == 0x0b0f;
350 static int die_counter;
352 int __kprobes __die(const char * str, struct pt_regs * regs, long err)
357 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
358 #ifdef CONFIG_PREEMPT
364 #ifdef CONFIG_DEBUG_PAGEALLOC
365 printk("DEBUG_PAGEALLOC");
369 if (notify_die(DIE_OOPS, str, regs, err,
370 current->thread.trap_no, SIGSEGV) !=
372 show_registers(regs);
373 /* Executive summary in case the oops scrolled away */
374 sp = (unsigned long) (®s->sp);
376 if (user_mode(regs)) {
378 ss = regs->ss & 0xffff;
380 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
381 print_symbol("%s", regs->ip);
382 printk(" SS:ESP %04x:%08lx\n", ss, sp);
390 * This is gone through when something in the kernel has done something bad and
391 * is about to be terminated.
393 void die(const char * str, struct pt_regs * regs, long err)
398 int lock_owner_depth;
400 .lock = __RAW_SPIN_LOCK_UNLOCKED,
402 .lock_owner_depth = 0
408 if (die.lock_owner != raw_smp_processor_id()) {
410 raw_local_irq_save(flags);
411 __raw_spin_lock(&die.lock);
412 die.lock_owner = smp_processor_id();
413 die.lock_owner_depth = 0;
416 raw_local_irq_save(flags);
418 if (++die.lock_owner_depth < 3) {
419 report_bug(regs->ip, regs);
421 if (__die(str, regs, err))
424 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
429 add_taint(TAINT_DIE);
430 __raw_spin_unlock(&die.lock);
431 raw_local_irq_restore(flags);
436 if (kexec_should_crash(current))
440 panic("Fatal exception in interrupt");
443 panic("Fatal exception");
449 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
451 if (!user_mode_vm(regs))
455 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
456 struct pt_regs * regs, long error_code,
459 struct task_struct *tsk = current;
461 if (regs->flags & VM_MASK) {
467 if (!user_mode(regs))
472 * We want error_code and trap_no set for userspace faults and
473 * kernelspace faults which result in die(), but not
474 * kernelspace faults which are fixed up. die() gives the
475 * process no chance to handle the signal and notice the
476 * kernel fault information, so that won't result in polluting
477 * the information about previously queued, but not yet
478 * delivered, faults. See also do_general_protection below.
480 tsk->thread.error_code = error_code;
481 tsk->thread.trap_no = trapnr;
484 force_sig_info(signr, info, tsk);
486 force_sig(signr, tsk);
491 if (!fixup_exception(regs)) {
492 tsk->thread.error_code = error_code;
493 tsk->thread.trap_no = trapnr;
494 die(str, regs, error_code);
500 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
501 if (ret) goto trap_signal;
506 #define DO_ERROR(trapnr, signr, str, name) \
507 void do_##name(struct pt_regs * regs, long error_code) \
509 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
512 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
515 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
516 void do_##name(struct pt_regs * regs, long error_code) \
520 local_irq_enable(); \
521 info.si_signo = signr; \
523 info.si_code = sicode; \
524 info.si_addr = (void __user *)siaddr; \
525 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
528 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
531 #define DO_VM86_ERROR(trapnr, signr, str, name) \
532 void do_##name(struct pt_regs * regs, long error_code) \
534 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
537 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
540 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
541 void do_##name(struct pt_regs * regs, long error_code) \
544 info.si_signo = signr; \
546 info.si_code = sicode; \
547 info.si_addr = (void __user *)siaddr; \
548 trace_hardirqs_fixup(); \
549 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
552 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
555 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
556 #ifndef CONFIG_KPROBES
557 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
559 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
560 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
561 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip, 0)
562 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
563 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
564 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
565 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
566 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0)
567 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0, 1)
569 void __kprobes do_general_protection(struct pt_regs * regs,
573 struct tss_struct *tss = &per_cpu(init_tss, cpu);
574 struct thread_struct *thread = ¤t->thread;
577 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
578 * invalid offset set (the LAZY one) and the faulting thread has
579 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
580 * and we set the offset field correctly. Then we let the CPU to
581 * restart the faulting instruction.
583 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
584 thread->io_bitmap_ptr) {
585 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
586 thread->io_bitmap_max);
588 * If the previously set map was extending to higher ports
589 * than the current one, pad extra space with 0xff (no access).
591 if (thread->io_bitmap_max < tss->io_bitmap_max)
592 memset((char *) tss->io_bitmap +
593 thread->io_bitmap_max, 0xff,
594 tss->io_bitmap_max - thread->io_bitmap_max);
595 tss->io_bitmap_max = thread->io_bitmap_max;
596 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
597 tss->io_bitmap_owner = thread;
603 if (regs->flags & VM_MASK)
606 if (!user_mode(regs))
609 current->thread.error_code = error_code;
610 current->thread.trap_no = 13;
611 if (show_unhandled_signals && unhandled_signal(current, SIGSEGV) &&
614 "%s[%d] general protection ip:%lx sp:%lx error:%lx\n",
615 current->comm, task_pid_nr(current),
616 regs->ip, regs->sp, error_code);
618 force_sig(SIGSEGV, current);
623 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
627 if (!fixup_exception(regs)) {
628 current->thread.error_code = error_code;
629 current->thread.trap_no = 13;
630 if (notify_die(DIE_GPF, "general protection fault", regs,
631 error_code, 13, SIGSEGV) == NOTIFY_STOP)
633 die("general protection fault", regs, error_code);
637 static __kprobes void
638 mem_parity_error(unsigned char reason, struct pt_regs * regs)
640 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
641 "CPU %d.\n", reason, smp_processor_id());
642 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
644 #if defined(CONFIG_EDAC)
645 if(edac_handler_set()) {
646 edac_atomic_assert_error();
651 if (panic_on_unrecovered_nmi)
652 panic("NMI: Not continuing");
654 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
656 /* Clear and disable the memory parity error line. */
657 clear_mem_error(reason);
660 static __kprobes void
661 io_check_error(unsigned char reason, struct pt_regs * regs)
665 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
666 show_registers(regs);
668 /* Re-enable the IOCK line, wait for a few seconds */
669 reason = (reason & 0xf) | 8;
672 while (--i) udelay(1000);
677 static __kprobes void
678 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
681 /* Might actually be able to figure out what the guilty party
688 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
689 "CPU %d.\n", reason, smp_processor_id());
690 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
691 if (panic_on_unrecovered_nmi)
692 panic("NMI: Not continuing");
694 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
697 static DEFINE_SPINLOCK(nmi_print_lock);
699 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
701 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
705 spin_lock(&nmi_print_lock);
707 * We are in trouble anyway, lets at least try
708 * to get a message out.
711 printk(KERN_EMERG "%s", msg);
712 printk(" on CPU%d, ip %08lx, registers:\n",
713 smp_processor_id(), regs->ip);
714 show_registers(regs);
716 spin_unlock(&nmi_print_lock);
719 /* If we are in kernel we are probably nested up pretty bad
720 * and might aswell get out now while we still can.
722 if (!user_mode_vm(regs)) {
723 current->thread.trap_no = 2;
730 static __kprobes void default_do_nmi(struct pt_regs * regs)
732 unsigned char reason = 0;
734 /* Only the BSP gets external NMIs from the system. */
735 if (!smp_processor_id())
736 reason = get_nmi_reason();
738 if (!(reason & 0xc0)) {
739 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
742 #ifdef CONFIG_X86_LOCAL_APIC
744 * Ok, so this is none of the documented NMI sources,
745 * so it must be the NMI watchdog.
747 if (nmi_watchdog_tick(regs, reason))
749 if (!do_nmi_callback(regs, smp_processor_id()))
751 unknown_nmi_error(reason, regs);
755 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
758 mem_parity_error(reason, regs);
760 io_check_error(reason, regs);
762 * Reassert NMI in case it became active meanwhile
763 * as it's edge-triggered.
768 static int ignore_nmis;
770 __kprobes void do_nmi(struct pt_regs * regs, long error_code)
776 cpu = smp_processor_id();
781 default_do_nmi(regs);
792 void restart_nmi(void)
798 #ifdef CONFIG_KPROBES
799 void __kprobes do_int3(struct pt_regs *regs, long error_code)
801 trace_hardirqs_fixup();
803 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
806 /* This is an interrupt gate, because kprobes wants interrupts
807 disabled. Normal trap handlers don't. */
808 restore_interrupts(regs);
809 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
814 * Our handling of the processor debug registers is non-trivial.
815 * We do not clear them on entry and exit from the kernel. Therefore
816 * it is possible to get a watchpoint trap here from inside the kernel.
817 * However, the code in ./ptrace.c has ensured that the user can
818 * only set watchpoints on userspace addresses. Therefore the in-kernel
819 * watchpoint trap can only occur in code which is reading/writing
820 * from user space. Such code must not hold kernel locks (since it
821 * can equally take a page fault), therefore it is safe to call
822 * force_sig_info even though that claims and releases locks.
824 * Code in ./signal.c ensures that the debug control register
825 * is restored before we deliver any signal, and therefore that
826 * user code runs with the correct debug control register even though
829 * Being careful here means that we don't have to be as careful in a
830 * lot of more complicated places (task switching can be a bit lazy
831 * about restoring all the debug state, and ptrace doesn't have to
832 * find every occurrence of the TF bit that could be saved away even
835 void __kprobes do_debug(struct pt_regs * regs, long error_code)
837 unsigned int condition;
838 struct task_struct *tsk = current;
840 trace_hardirqs_fixup();
842 get_debugreg(condition, 6);
845 * The processor cleared BTF, so don't mark that we need it set.
847 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
848 tsk->thread.debugctlmsr = 0;
850 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
851 SIGTRAP) == NOTIFY_STOP)
853 /* It's safe to allow irq's after DR6 has been saved */
854 if (regs->flags & X86_EFLAGS_IF)
857 /* Mask out spurious debug traps due to lazy DR7 setting */
858 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
859 if (!tsk->thread.debugreg7)
863 if (regs->flags & VM_MASK)
866 /* Save debug status register where ptrace can see it */
867 tsk->thread.debugreg6 = condition;
870 * Single-stepping through TF: make sure we ignore any events in
871 * kernel space (but re-enable TF when returning to user mode).
873 if (condition & DR_STEP) {
875 * We already checked v86 mode above, so we can
876 * check for kernel mode by just checking the CPL
879 if (!user_mode(regs))
880 goto clear_TF_reenable;
883 /* Ok, finally something we can handle */
884 send_sigtrap(tsk, regs, error_code);
886 /* Disable additional traps. They'll be re-enabled when
887 * the signal is delivered.
894 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
898 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
899 regs->flags &= ~TF_MASK;
904 * Note that we play around with the 'TS' bit in an attempt to get
905 * the correct behaviour even in the presence of the asynchronous
908 void math_error(void __user *ip)
910 struct task_struct * task;
912 unsigned short cwd, swd;
915 * Save the info for the exception handler and clear the error.
919 task->thread.trap_no = 16;
920 task->thread.error_code = 0;
921 info.si_signo = SIGFPE;
923 info.si_code = __SI_FAULT;
926 * (~cwd & swd) will mask out exceptions that are not set to unmasked
927 * status. 0x3f is the exception bits in these regs, 0x200 is the
928 * C1 reg you need in case of a stack fault, 0x040 is the stack
929 * fault bit. We should only be taking one exception at a time,
930 * so if this combination doesn't produce any single exception,
931 * then we have a bad program that isn't syncronizing its FPU usage
932 * and it will suffer the consequences since we won't be able to
933 * fully reproduce the context of the exception
935 cwd = get_fpu_cwd(task);
936 swd = get_fpu_swd(task);
937 switch (swd & ~cwd & 0x3f) {
938 case 0x000: /* No unmasked exception */
940 default: /* Multiple exceptions */
942 case 0x001: /* Invalid Op */
944 * swd & 0x240 == 0x040: Stack Underflow
945 * swd & 0x240 == 0x240: Stack Overflow
946 * User must clear the SF bit (0x40) if set
948 info.si_code = FPE_FLTINV;
950 case 0x002: /* Denormalize */
951 case 0x010: /* Underflow */
952 info.si_code = FPE_FLTUND;
954 case 0x004: /* Zero Divide */
955 info.si_code = FPE_FLTDIV;
957 case 0x008: /* Overflow */
958 info.si_code = FPE_FLTOVF;
960 case 0x020: /* Precision */
961 info.si_code = FPE_FLTRES;
964 force_sig_info(SIGFPE, &info, task);
967 void do_coprocessor_error(struct pt_regs * regs, long error_code)
970 math_error((void __user *)regs->ip);
973 static void simd_math_error(void __user *ip)
975 struct task_struct * task;
977 unsigned short mxcsr;
980 * Save the info for the exception handler and clear the error.
984 task->thread.trap_no = 19;
985 task->thread.error_code = 0;
986 info.si_signo = SIGFPE;
988 info.si_code = __SI_FAULT;
991 * The SIMD FPU exceptions are handled a little differently, as there
992 * is only a single status/control register. Thus, to determine which
993 * unmasked exception was caught we must mask the exception mask bits
994 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
996 mxcsr = get_fpu_mxcsr(task);
997 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1001 case 0x001: /* Invalid Op */
1002 info.si_code = FPE_FLTINV;
1004 case 0x002: /* Denormalize */
1005 case 0x010: /* Underflow */
1006 info.si_code = FPE_FLTUND;
1008 case 0x004: /* Zero Divide */
1009 info.si_code = FPE_FLTDIV;
1011 case 0x008: /* Overflow */
1012 info.si_code = FPE_FLTOVF;
1014 case 0x020: /* Precision */
1015 info.si_code = FPE_FLTRES;
1018 force_sig_info(SIGFPE, &info, task);
1021 void do_simd_coprocessor_error(struct pt_regs * regs,
1025 /* Handle SIMD FPU exceptions on PIII+ processors. */
1027 simd_math_error((void __user *)regs->ip);
1030 * Handle strange cache flush from user space exception
1031 * in all other cases. This is undocumented behaviour.
1033 if (regs->flags & VM_MASK) {
1034 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1038 current->thread.trap_no = 19;
1039 current->thread.error_code = error_code;
1040 die_if_kernel("cache flush denied", regs, error_code);
1041 force_sig(SIGSEGV, current);
1045 void do_spurious_interrupt_bug(struct pt_regs * regs,
1049 /* No need to warn about this any longer. */
1050 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1054 unsigned long patch_espfix_desc(unsigned long uesp,
1057 struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt;
1058 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1059 unsigned long new_kesp = kesp - base;
1060 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1061 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1062 /* Set up base for espfix segment */
1063 desc &= 0x00f0ff0000000000ULL;
1064 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1065 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1066 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1067 (lim_pages & 0xffff);
1068 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1073 * 'math_state_restore()' saves the current math information in the
1074 * old math state array, and gets the new ones from the current task
1076 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1077 * Don't touch unless you *really* know how it works.
1079 * Must be called with kernel preemption disabled (in this case,
1080 * local interrupts are disabled at the call-site in entry.S).
1082 asmlinkage void math_state_restore(void)
1084 struct thread_info *thread = current_thread_info();
1085 struct task_struct *tsk = thread->task;
1087 clts(); /* Allow maths ops (or we recurse) */
1088 if (!tsk_used_math(tsk))
1091 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1094 EXPORT_SYMBOL_GPL(math_state_restore);
1096 #ifndef CONFIG_MATH_EMULATION
1098 asmlinkage void math_emulate(long arg)
1100 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1101 printk(KERN_EMERG "killing %s.\n",current->comm);
1102 force_sig(SIGFPE,current);
1106 #endif /* CONFIG_MATH_EMULATION */
1109 void __init trap_init(void)
1114 void __iomem *p = ioremap(0x0FFFD9, 4);
1115 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1121 #ifdef CONFIG_X86_LOCAL_APIC
1122 init_apic_mappings();
1125 set_trap_gate(0,÷_error);
1126 set_intr_gate(1,&debug);
1127 set_intr_gate(2,&nmi);
1128 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1129 set_system_gate(4,&overflow);
1130 set_trap_gate(5,&bounds);
1131 set_trap_gate(6,&invalid_op);
1132 set_trap_gate(7,&device_not_available);
1133 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1134 set_trap_gate(9,&coprocessor_segment_overrun);
1135 set_trap_gate(10,&invalid_TSS);
1136 set_trap_gate(11,&segment_not_present);
1137 set_trap_gate(12,&stack_segment);
1138 set_trap_gate(13,&general_protection);
1139 set_intr_gate(14,&page_fault);
1140 set_trap_gate(15,&spurious_interrupt_bug);
1141 set_trap_gate(16,&coprocessor_error);
1142 set_trap_gate(17,&alignment_check);
1143 #ifdef CONFIG_X86_MCE
1144 set_trap_gate(18,&machine_check);
1146 set_trap_gate(19,&simd_coprocessor_error);
1150 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1151 * Generates a compile-time "error: zero width for bit-field" if
1152 * the alignment is wrong.
1154 struct fxsrAlignAssert {
1155 int _:!(offsetof(struct task_struct,
1156 thread.i387.fxsave) & 15);
1159 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1160 set_in_cr4(X86_CR4_OSFXSR);
1164 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1166 set_in_cr4(X86_CR4_OSXMMEXCPT);
1170 set_system_gate(SYSCALL_VECTOR,&system_call);
1172 /* Reserve all the builtin and the syscall vector. */
1173 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1174 set_bit(i, used_vectors);
1175 set_bit(SYSCALL_VECTOR, used_vectors);
1178 * Should be a barrier for any external CPU state.
1185 static int __init kstack_setup(char *s)
1187 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1190 __setup("kstack=", kstack_setup);
1192 static int __init code_bytes_setup(char *s)
1194 code_bytes = simple_strtoul(s, NULL, 0);
1195 if (code_bytes > 8192)
1200 __setup("code_bytes=", code_bytes_setup);