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 #ifdef CONFIG_FRAME_POINTER
122 struct stack_frame *frame = (struct stack_frame *)bp;
123 while (valid_stack_ptr(tinfo, frame, sizeof(*frame))) {
124 struct stack_frame *next;
127 addr = frame->return_address;
128 if (__kernel_text_address(addr))
129 ops->address(data, addr);
131 * break out of recursive entries (such as
132 * end_of_stack_stop_unwind_function). Also,
133 * we can never allow a frame pointer to
136 next = frame->next_frame;
137 bp = (unsigned long) next;
143 while (valid_stack_ptr(tinfo, stack, sizeof(*stack))) {
147 if (__kernel_text_address(addr))
148 ops->address(data, addr);
154 #define MSG(msg) ops->warning(data, msg)
156 void dump_trace(struct task_struct *task, struct pt_regs *regs,
157 unsigned long *stack,
158 const struct stacktrace_ops *ops, void *data)
160 unsigned long bp = 0;
169 stack = (unsigned long *)task->thread.sp;
172 #ifdef CONFIG_FRAME_POINTER
174 if (task == current) {
175 /* Grab bp right from our regs */
176 asm ("movl %%ebp, %0" : "=r" (bp) : );
178 /* bp is the last reg pushed by switch_to */
179 bp = *(unsigned long *) task->thread.sp;
185 struct thread_info *context;
186 context = (struct thread_info *)
187 ((unsigned long)stack & (~(THREAD_SIZE - 1)));
188 bp = print_context_stack(context, stack, bp, ops, data);
189 /* Should be after the line below, but somewhere
190 in early boot context comes out corrupted and we
191 can't reference it -AK */
192 if (ops->stack(data, "IRQ") < 0)
194 stack = (unsigned long*)context->previous_esp;
197 touch_nmi_watchdog();
200 EXPORT_SYMBOL(dump_trace);
203 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
206 print_symbol(msg, symbol);
210 static void print_trace_warning(void *data, char *msg)
212 printk("%s%s\n", (char *)data, msg);
215 static int print_trace_stack(void *data, char *name)
221 * Print one address/symbol entries per line.
223 static void print_trace_address(void *data, unsigned long addr)
225 printk("%s [<%08lx>] ", (char *)data, addr);
226 print_symbol("%s\n", addr);
227 touch_nmi_watchdog();
230 static const struct stacktrace_ops print_trace_ops = {
231 .warning = print_trace_warning,
232 .warning_symbol = print_trace_warning_symbol,
233 .stack = print_trace_stack,
234 .address = print_trace_address,
238 show_trace_log_lvl(struct task_struct *task, struct pt_regs *regs,
239 unsigned long * stack, char *log_lvl)
241 dump_trace(task, regs, stack, &print_trace_ops, log_lvl);
242 printk("%s =======================\n", log_lvl);
245 void show_trace(struct task_struct *task, struct pt_regs *regs,
246 unsigned long * stack)
248 show_trace_log_lvl(task, regs, stack, "");
251 static void show_stack_log_lvl(struct task_struct *task, struct pt_regs *regs,
252 unsigned long *sp, char *log_lvl)
254 unsigned long *stack;
259 sp = (unsigned long*)task->thread.sp;
261 sp = (unsigned long *)&sp;
265 for(i = 0; i < kstack_depth_to_print; i++) {
266 if (kstack_end(stack))
268 if (i && ((i % 8) == 0))
269 printk("\n%s ", log_lvl);
270 printk("%08lx ", *stack++);
272 printk("\n%sCall Trace:\n", log_lvl);
273 show_trace_log_lvl(task, regs, sp, log_lvl);
276 void show_stack(struct task_struct *task, unsigned long *sp)
279 show_stack_log_lvl(task, NULL, sp, "");
283 * The architecture-independent dump_stack generator
285 void dump_stack(void)
289 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
290 current->pid, current->comm, print_tainted(),
291 init_utsname()->release,
292 (int)strcspn(init_utsname()->version, " "),
293 init_utsname()->version);
294 show_trace(current, NULL, &stack);
297 EXPORT_SYMBOL(dump_stack);
299 void show_registers(struct pt_regs *regs)
304 __show_registers(regs, 0);
305 printk(KERN_EMERG "Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
306 TASK_COMM_LEN, current->comm, task_pid_nr(current),
307 current_thread_info(), current, task_thread_info(current));
309 * When in-kernel, we also print out the stack and code at the
310 * time of the fault..
312 if (!user_mode_vm(regs)) {
314 unsigned int code_prologue = code_bytes * 43 / 64;
315 unsigned int code_len = code_bytes;
318 printk("\n" KERN_EMERG "Stack: ");
319 show_stack_log_lvl(NULL, regs, ®s->sp, KERN_EMERG);
321 printk(KERN_EMERG "Code: ");
323 ip = (u8 *)regs->ip - code_prologue;
324 if (ip < (u8 *)PAGE_OFFSET ||
325 probe_kernel_address(ip, c)) {
326 /* try starting at EIP */
328 code_len = code_len - code_prologue + 1;
330 for (i = 0; i < code_len; i++, ip++) {
331 if (ip < (u8 *)PAGE_OFFSET ||
332 probe_kernel_address(ip, c)) {
333 printk(" Bad EIP value.");
336 if (ip == (u8 *)regs->ip)
337 printk("<%02x> ", c);
345 int is_valid_bugaddr(unsigned long ip)
349 if (ip < PAGE_OFFSET)
351 if (probe_kernel_address((unsigned short *)ip, ud2))
354 return ud2 == 0x0b0f;
357 static int die_counter;
359 int __kprobes __die(const char * str, struct pt_regs * regs, long err)
364 printk(KERN_EMERG "%s: %04lx [#%d] ", str, err & 0xffff, ++die_counter);
365 #ifdef CONFIG_PREEMPT
371 #ifdef CONFIG_DEBUG_PAGEALLOC
372 printk("DEBUG_PAGEALLOC");
376 if (notify_die(DIE_OOPS, str, regs, err,
377 current->thread.trap_no, SIGSEGV) !=
379 show_registers(regs);
380 /* Executive summary in case the oops scrolled away */
381 sp = (unsigned long) (®s->sp);
383 if (user_mode(regs)) {
385 ss = regs->ss & 0xffff;
387 printk(KERN_EMERG "EIP: [<%08lx>] ", regs->ip);
388 print_symbol("%s", regs->ip);
389 printk(" SS:ESP %04x:%08lx\n", ss, sp);
397 * This is gone through when something in the kernel has done something bad and
398 * is about to be terminated.
400 void die(const char * str, struct pt_regs * regs, long err)
405 int lock_owner_depth;
407 .lock = __RAW_SPIN_LOCK_UNLOCKED,
409 .lock_owner_depth = 0
415 if (die.lock_owner != raw_smp_processor_id()) {
417 raw_local_irq_save(flags);
418 __raw_spin_lock(&die.lock);
419 die.lock_owner = smp_processor_id();
420 die.lock_owner_depth = 0;
423 raw_local_irq_save(flags);
425 if (++die.lock_owner_depth < 3) {
426 report_bug(regs->ip, regs);
428 if (__die(str, regs, err))
431 printk(KERN_EMERG "Recursive die() failure, output suppressed\n");
436 add_taint(TAINT_DIE);
437 __raw_spin_unlock(&die.lock);
438 raw_local_irq_restore(flags);
443 if (kexec_should_crash(current))
447 panic("Fatal exception in interrupt");
450 panic("Fatal exception");
456 static inline void die_if_kernel(const char * str, struct pt_regs * regs, long err)
458 if (!user_mode_vm(regs))
462 static void __kprobes do_trap(int trapnr, int signr, char *str, int vm86,
463 struct pt_regs * regs, long error_code,
466 struct task_struct *tsk = current;
468 if (regs->flags & VM_MASK) {
474 if (!user_mode(regs))
479 * We want error_code and trap_no set for userspace faults and
480 * kernelspace faults which result in die(), but not
481 * kernelspace faults which are fixed up. die() gives the
482 * process no chance to handle the signal and notice the
483 * kernel fault information, so that won't result in polluting
484 * the information about previously queued, but not yet
485 * delivered, faults. See also do_general_protection below.
487 tsk->thread.error_code = error_code;
488 tsk->thread.trap_no = trapnr;
491 force_sig_info(signr, info, tsk);
493 force_sig(signr, tsk);
498 if (!fixup_exception(regs)) {
499 tsk->thread.error_code = error_code;
500 tsk->thread.trap_no = trapnr;
501 die(str, regs, error_code);
507 int ret = handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, trapnr);
508 if (ret) goto trap_signal;
513 #define DO_ERROR(trapnr, signr, str, name) \
514 void do_##name(struct pt_regs * regs, long error_code) \
516 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
519 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
522 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
523 void do_##name(struct pt_regs * regs, long error_code) \
527 local_irq_enable(); \
528 info.si_signo = signr; \
530 info.si_code = sicode; \
531 info.si_addr = (void __user *)siaddr; \
532 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
535 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
538 #define DO_VM86_ERROR(trapnr, signr, str, name) \
539 void do_##name(struct pt_regs * regs, long error_code) \
541 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
544 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
547 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
548 void do_##name(struct pt_regs * regs, long error_code) \
551 info.si_signo = signr; \
553 info.si_code = sicode; \
554 info.si_addr = (void __user *)siaddr; \
555 trace_hardirqs_fixup(); \
556 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
559 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
562 DO_VM86_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
563 #ifndef CONFIG_KPROBES
564 DO_VM86_ERROR( 3, SIGTRAP, "int3", int3)
566 DO_VM86_ERROR( 4, SIGSEGV, "overflow", overflow)
567 DO_VM86_ERROR( 5, SIGSEGV, "bounds", bounds)
568 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip, 0)
569 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
570 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
571 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
572 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
573 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0, 0)
574 DO_ERROR_INFO(32, SIGSEGV, "iret exception", iret_error, ILL_BADSTK, 0, 1)
576 void __kprobes do_general_protection(struct pt_regs * regs,
580 struct tss_struct *tss = &per_cpu(init_tss, cpu);
581 struct thread_struct *thread = ¤t->thread;
584 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
585 * invalid offset set (the LAZY one) and the faulting thread has
586 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
587 * and we set the offset field correctly. Then we let the CPU to
588 * restart the faulting instruction.
590 if (tss->x86_tss.io_bitmap_base == INVALID_IO_BITMAP_OFFSET_LAZY &&
591 thread->io_bitmap_ptr) {
592 memcpy(tss->io_bitmap, thread->io_bitmap_ptr,
593 thread->io_bitmap_max);
595 * If the previously set map was extending to higher ports
596 * than the current one, pad extra space with 0xff (no access).
598 if (thread->io_bitmap_max < tss->io_bitmap_max)
599 memset((char *) tss->io_bitmap +
600 thread->io_bitmap_max, 0xff,
601 tss->io_bitmap_max - thread->io_bitmap_max);
602 tss->io_bitmap_max = thread->io_bitmap_max;
603 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
604 tss->io_bitmap_owner = thread;
610 if (regs->flags & VM_MASK)
613 if (!user_mode(regs))
616 current->thread.error_code = error_code;
617 current->thread.trap_no = 13;
618 if (show_unhandled_signals && unhandled_signal(current, SIGSEGV) &&
621 "%s[%d] general protection ip:%lx sp:%lx error:%lx\n",
622 current->comm, task_pid_nr(current),
623 regs->ip, regs->sp, error_code);
625 force_sig(SIGSEGV, current);
630 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
634 if (!fixup_exception(regs)) {
635 current->thread.error_code = error_code;
636 current->thread.trap_no = 13;
637 if (notify_die(DIE_GPF, "general protection fault", regs,
638 error_code, 13, SIGSEGV) == NOTIFY_STOP)
640 die("general protection fault", regs, error_code);
644 static __kprobes void
645 mem_parity_error(unsigned char reason, struct pt_regs * regs)
647 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
648 "CPU %d.\n", reason, smp_processor_id());
649 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
651 #if defined(CONFIG_EDAC)
652 if(edac_handler_set()) {
653 edac_atomic_assert_error();
658 if (panic_on_unrecovered_nmi)
659 panic("NMI: Not continuing");
661 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
663 /* Clear and disable the memory parity error line. */
664 clear_mem_error(reason);
667 static __kprobes void
668 io_check_error(unsigned char reason, struct pt_regs * regs)
672 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
673 show_registers(regs);
675 /* Re-enable the IOCK line, wait for a few seconds */
676 reason = (reason & 0xf) | 8;
679 while (--i) udelay(1000);
684 static __kprobes void
685 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
688 /* Might actually be able to figure out what the guilty party
695 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
696 "CPU %d.\n", reason, smp_processor_id());
697 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
698 if (panic_on_unrecovered_nmi)
699 panic("NMI: Not continuing");
701 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
704 static DEFINE_SPINLOCK(nmi_print_lock);
706 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
708 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
712 spin_lock(&nmi_print_lock);
714 * We are in trouble anyway, lets at least try
715 * to get a message out.
718 printk(KERN_EMERG "%s", msg);
719 printk(" on CPU%d, ip %08lx, registers:\n",
720 smp_processor_id(), regs->ip);
721 show_registers(regs);
723 spin_unlock(&nmi_print_lock);
726 /* If we are in kernel we are probably nested up pretty bad
727 * and might aswell get out now while we still can.
729 if (!user_mode_vm(regs)) {
730 current->thread.trap_no = 2;
737 static __kprobes void default_do_nmi(struct pt_regs * regs)
739 unsigned char reason = 0;
741 /* Only the BSP gets external NMIs from the system. */
742 if (!smp_processor_id())
743 reason = get_nmi_reason();
745 if (!(reason & 0xc0)) {
746 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
749 #ifdef CONFIG_X86_LOCAL_APIC
751 * Ok, so this is none of the documented NMI sources,
752 * so it must be the NMI watchdog.
754 if (nmi_watchdog_tick(regs, reason))
756 if (!do_nmi_callback(regs, smp_processor_id()))
758 unknown_nmi_error(reason, regs);
762 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
765 mem_parity_error(reason, regs);
767 io_check_error(reason, regs);
769 * Reassert NMI in case it became active meanwhile
770 * as it's edge-triggered.
775 static int ignore_nmis;
777 __kprobes void do_nmi(struct pt_regs * regs, long error_code)
783 cpu = smp_processor_id();
788 default_do_nmi(regs);
799 void restart_nmi(void)
805 #ifdef CONFIG_KPROBES
806 void __kprobes do_int3(struct pt_regs *regs, long error_code)
808 trace_hardirqs_fixup();
810 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
813 /* This is an interrupt gate, because kprobes wants interrupts
814 disabled. Normal trap handlers don't. */
815 restore_interrupts(regs);
816 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
821 * Our handling of the processor debug registers is non-trivial.
822 * We do not clear them on entry and exit from the kernel. Therefore
823 * it is possible to get a watchpoint trap here from inside the kernel.
824 * However, the code in ./ptrace.c has ensured that the user can
825 * only set watchpoints on userspace addresses. Therefore the in-kernel
826 * watchpoint trap can only occur in code which is reading/writing
827 * from user space. Such code must not hold kernel locks (since it
828 * can equally take a page fault), therefore it is safe to call
829 * force_sig_info even though that claims and releases locks.
831 * Code in ./signal.c ensures that the debug control register
832 * is restored before we deliver any signal, and therefore that
833 * user code runs with the correct debug control register even though
836 * Being careful here means that we don't have to be as careful in a
837 * lot of more complicated places (task switching can be a bit lazy
838 * about restoring all the debug state, and ptrace doesn't have to
839 * find every occurrence of the TF bit that could be saved away even
842 void __kprobes do_debug(struct pt_regs * regs, long error_code)
844 unsigned int condition;
845 struct task_struct *tsk = current;
847 trace_hardirqs_fixup();
849 get_debugreg(condition, 6);
852 * The processor cleared BTF, so don't mark that we need it set.
854 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
855 tsk->thread.debugctlmsr = 0;
857 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
858 SIGTRAP) == NOTIFY_STOP)
860 /* It's safe to allow irq's after DR6 has been saved */
861 if (regs->flags & X86_EFLAGS_IF)
864 /* Mask out spurious debug traps due to lazy DR7 setting */
865 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
866 if (!tsk->thread.debugreg7)
870 if (regs->flags & VM_MASK)
873 /* Save debug status register where ptrace can see it */
874 tsk->thread.debugreg6 = condition;
877 * Single-stepping through TF: make sure we ignore any events in
878 * kernel space (but re-enable TF when returning to user mode).
880 if (condition & DR_STEP) {
882 * We already checked v86 mode above, so we can
883 * check for kernel mode by just checking the CPL
886 if (!user_mode(regs))
887 goto clear_TF_reenable;
890 /* Ok, finally something we can handle */
891 send_sigtrap(tsk, regs, error_code);
893 /* Disable additional traps. They'll be re-enabled when
894 * the signal is delivered.
901 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
905 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
906 regs->flags &= ~TF_MASK;
911 * Note that we play around with the 'TS' bit in an attempt to get
912 * the correct behaviour even in the presence of the asynchronous
915 void math_error(void __user *ip)
917 struct task_struct * task;
919 unsigned short cwd, swd;
922 * Save the info for the exception handler and clear the error.
926 task->thread.trap_no = 16;
927 task->thread.error_code = 0;
928 info.si_signo = SIGFPE;
930 info.si_code = __SI_FAULT;
933 * (~cwd & swd) will mask out exceptions that are not set to unmasked
934 * status. 0x3f is the exception bits in these regs, 0x200 is the
935 * C1 reg you need in case of a stack fault, 0x040 is the stack
936 * fault bit. We should only be taking one exception at a time,
937 * so if this combination doesn't produce any single exception,
938 * then we have a bad program that isn't syncronizing its FPU usage
939 * and it will suffer the consequences since we won't be able to
940 * fully reproduce the context of the exception
942 cwd = get_fpu_cwd(task);
943 swd = get_fpu_swd(task);
944 switch (swd & ~cwd & 0x3f) {
945 case 0x000: /* No unmasked exception */
947 default: /* Multiple exceptions */
949 case 0x001: /* Invalid Op */
951 * swd & 0x240 == 0x040: Stack Underflow
952 * swd & 0x240 == 0x240: Stack Overflow
953 * User must clear the SF bit (0x40) if set
955 info.si_code = FPE_FLTINV;
957 case 0x002: /* Denormalize */
958 case 0x010: /* Underflow */
959 info.si_code = FPE_FLTUND;
961 case 0x004: /* Zero Divide */
962 info.si_code = FPE_FLTDIV;
964 case 0x008: /* Overflow */
965 info.si_code = FPE_FLTOVF;
967 case 0x020: /* Precision */
968 info.si_code = FPE_FLTRES;
971 force_sig_info(SIGFPE, &info, task);
974 void do_coprocessor_error(struct pt_regs * regs, long error_code)
977 math_error((void __user *)regs->ip);
980 static void simd_math_error(void __user *ip)
982 struct task_struct * task;
984 unsigned short mxcsr;
987 * Save the info for the exception handler and clear the error.
991 task->thread.trap_no = 19;
992 task->thread.error_code = 0;
993 info.si_signo = SIGFPE;
995 info.si_code = __SI_FAULT;
998 * The SIMD FPU exceptions are handled a little differently, as there
999 * is only a single status/control register. Thus, to determine which
1000 * unmasked exception was caught we must mask the exception mask bits
1001 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1003 mxcsr = get_fpu_mxcsr(task);
1004 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1008 case 0x001: /* Invalid Op */
1009 info.si_code = FPE_FLTINV;
1011 case 0x002: /* Denormalize */
1012 case 0x010: /* Underflow */
1013 info.si_code = FPE_FLTUND;
1015 case 0x004: /* Zero Divide */
1016 info.si_code = FPE_FLTDIV;
1018 case 0x008: /* Overflow */
1019 info.si_code = FPE_FLTOVF;
1021 case 0x020: /* Precision */
1022 info.si_code = FPE_FLTRES;
1025 force_sig_info(SIGFPE, &info, task);
1028 void do_simd_coprocessor_error(struct pt_regs * regs,
1032 /* Handle SIMD FPU exceptions on PIII+ processors. */
1034 simd_math_error((void __user *)regs->ip);
1037 * Handle strange cache flush from user space exception
1038 * in all other cases. This is undocumented behaviour.
1040 if (regs->flags & VM_MASK) {
1041 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1045 current->thread.trap_no = 19;
1046 current->thread.error_code = error_code;
1047 die_if_kernel("cache flush denied", regs, error_code);
1048 force_sig(SIGSEGV, current);
1052 void do_spurious_interrupt_bug(struct pt_regs * regs,
1056 /* No need to warn about this any longer. */
1057 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1061 unsigned long patch_espfix_desc(unsigned long uesp,
1064 struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt;
1065 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1066 unsigned long new_kesp = kesp - base;
1067 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1068 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1069 /* Set up base for espfix segment */
1070 desc &= 0x00f0ff0000000000ULL;
1071 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1072 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1073 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1074 (lim_pages & 0xffff);
1075 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1080 * 'math_state_restore()' saves the current math information in the
1081 * old math state array, and gets the new ones from the current task
1083 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1084 * Don't touch unless you *really* know how it works.
1086 * Must be called with kernel preemption disabled (in this case,
1087 * local interrupts are disabled at the call-site in entry.S).
1089 asmlinkage void math_state_restore(void)
1091 struct thread_info *thread = current_thread_info();
1092 struct task_struct *tsk = thread->task;
1094 clts(); /* Allow maths ops (or we recurse) */
1095 if (!tsk_used_math(tsk))
1098 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1101 EXPORT_SYMBOL_GPL(math_state_restore);
1103 #ifndef CONFIG_MATH_EMULATION
1105 asmlinkage void math_emulate(long arg)
1107 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1108 printk(KERN_EMERG "killing %s.\n",current->comm);
1109 force_sig(SIGFPE,current);
1113 #endif /* CONFIG_MATH_EMULATION */
1116 void __init trap_init(void)
1121 void __iomem *p = ioremap(0x0FFFD9, 4);
1122 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1128 #ifdef CONFIG_X86_LOCAL_APIC
1129 init_apic_mappings();
1132 set_trap_gate(0,÷_error);
1133 set_intr_gate(1,&debug);
1134 set_intr_gate(2,&nmi);
1135 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1136 set_system_gate(4,&overflow);
1137 set_trap_gate(5,&bounds);
1138 set_trap_gate(6,&invalid_op);
1139 set_trap_gate(7,&device_not_available);
1140 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1141 set_trap_gate(9,&coprocessor_segment_overrun);
1142 set_trap_gate(10,&invalid_TSS);
1143 set_trap_gate(11,&segment_not_present);
1144 set_trap_gate(12,&stack_segment);
1145 set_trap_gate(13,&general_protection);
1146 set_intr_gate(14,&page_fault);
1147 set_trap_gate(15,&spurious_interrupt_bug);
1148 set_trap_gate(16,&coprocessor_error);
1149 set_trap_gate(17,&alignment_check);
1150 #ifdef CONFIG_X86_MCE
1151 set_trap_gate(18,&machine_check);
1153 set_trap_gate(19,&simd_coprocessor_error);
1157 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1158 * Generates a compile-time "error: zero width for bit-field" if
1159 * the alignment is wrong.
1161 struct fxsrAlignAssert {
1162 int _:!(offsetof(struct task_struct,
1163 thread.i387.fxsave) & 15);
1166 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1167 set_in_cr4(X86_CR4_OSFXSR);
1171 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1173 set_in_cr4(X86_CR4_OSXMMEXCPT);
1177 set_system_gate(SYSCALL_VECTOR,&system_call);
1179 /* Reserve all the builtin and the syscall vector. */
1180 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1181 set_bit(i, used_vectors);
1182 set_bit(SYSCALL_VECTOR, used_vectors);
1185 * Should be a barrier for any external CPU state.
1192 static int __init kstack_setup(char *s)
1194 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1197 __setup("kstack=", kstack_setup);
1199 static int __init code_bytes_setup(char *s)
1201 code_bytes = simple_strtoul(s, NULL, 0);
1202 if (code_bytes > 8192)
1207 __setup("code_bytes=", code_bytes_setup);