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) &&
612 printk_ratelimit()) {
614 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
615 current->comm, task_pid_nr(current),
616 regs->ip, regs->sp, error_code);
617 print_vma_addr(" in ", regs->ip);
621 force_sig(SIGSEGV, current);
626 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
630 if (!fixup_exception(regs)) {
631 current->thread.error_code = error_code;
632 current->thread.trap_no = 13;
633 if (notify_die(DIE_GPF, "general protection fault", regs,
634 error_code, 13, SIGSEGV) == NOTIFY_STOP)
636 die("general protection fault", regs, error_code);
640 static __kprobes void
641 mem_parity_error(unsigned char reason, struct pt_regs * regs)
643 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
644 "CPU %d.\n", reason, smp_processor_id());
645 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
647 #if defined(CONFIG_EDAC)
648 if(edac_handler_set()) {
649 edac_atomic_assert_error();
654 if (panic_on_unrecovered_nmi)
655 panic("NMI: Not continuing");
657 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
659 /* Clear and disable the memory parity error line. */
660 clear_mem_error(reason);
663 static __kprobes void
664 io_check_error(unsigned char reason, struct pt_regs * regs)
668 printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
669 show_registers(regs);
671 /* Re-enable the IOCK line, wait for a few seconds */
672 reason = (reason & 0xf) | 8;
675 while (--i) udelay(1000);
680 static __kprobes void
681 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
684 /* Might actually be able to figure out what the guilty party
691 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x on "
692 "CPU %d.\n", reason, smp_processor_id());
693 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
694 if (panic_on_unrecovered_nmi)
695 panic("NMI: Not continuing");
697 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
700 static DEFINE_SPINLOCK(nmi_print_lock);
702 void __kprobes die_nmi(struct pt_regs *regs, const char *msg)
704 if (notify_die(DIE_NMIWATCHDOG, msg, regs, 0, 2, SIGINT) ==
708 spin_lock(&nmi_print_lock);
710 * We are in trouble anyway, lets at least try
711 * to get a message out.
714 printk(KERN_EMERG "%s", msg);
715 printk(" on CPU%d, ip %08lx, registers:\n",
716 smp_processor_id(), regs->ip);
717 show_registers(regs);
719 spin_unlock(&nmi_print_lock);
722 /* If we are in kernel we are probably nested up pretty bad
723 * and might aswell get out now while we still can.
725 if (!user_mode_vm(regs)) {
726 current->thread.trap_no = 2;
733 static __kprobes void default_do_nmi(struct pt_regs * regs)
735 unsigned char reason = 0;
737 /* Only the BSP gets external NMIs from the system. */
738 if (!smp_processor_id())
739 reason = get_nmi_reason();
741 if (!(reason & 0xc0)) {
742 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
745 #ifdef CONFIG_X86_LOCAL_APIC
747 * Ok, so this is none of the documented NMI sources,
748 * so it must be the NMI watchdog.
750 if (nmi_watchdog_tick(regs, reason))
752 if (!do_nmi_callback(regs, smp_processor_id()))
754 unknown_nmi_error(reason, regs);
758 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
761 mem_parity_error(reason, regs);
763 io_check_error(reason, regs);
765 * Reassert NMI in case it became active meanwhile
766 * as it's edge-triggered.
771 static int ignore_nmis;
773 __kprobes void do_nmi(struct pt_regs * regs, long error_code)
779 cpu = smp_processor_id();
784 default_do_nmi(regs);
795 void restart_nmi(void)
801 #ifdef CONFIG_KPROBES
802 void __kprobes do_int3(struct pt_regs *regs, long error_code)
804 trace_hardirqs_fixup();
806 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
809 /* This is an interrupt gate, because kprobes wants interrupts
810 disabled. Normal trap handlers don't. */
811 restore_interrupts(regs);
812 do_trap(3, SIGTRAP, "int3", 1, regs, error_code, NULL);
817 * Our handling of the processor debug registers is non-trivial.
818 * We do not clear them on entry and exit from the kernel. Therefore
819 * it is possible to get a watchpoint trap here from inside the kernel.
820 * However, the code in ./ptrace.c has ensured that the user can
821 * only set watchpoints on userspace addresses. Therefore the in-kernel
822 * watchpoint trap can only occur in code which is reading/writing
823 * from user space. Such code must not hold kernel locks (since it
824 * can equally take a page fault), therefore it is safe to call
825 * force_sig_info even though that claims and releases locks.
827 * Code in ./signal.c ensures that the debug control register
828 * is restored before we deliver any signal, and therefore that
829 * user code runs with the correct debug control register even though
832 * Being careful here means that we don't have to be as careful in a
833 * lot of more complicated places (task switching can be a bit lazy
834 * about restoring all the debug state, and ptrace doesn't have to
835 * find every occurrence of the TF bit that could be saved away even
838 void __kprobes do_debug(struct pt_regs * regs, long error_code)
840 unsigned int condition;
841 struct task_struct *tsk = current;
843 trace_hardirqs_fixup();
845 get_debugreg(condition, 6);
848 * The processor cleared BTF, so don't mark that we need it set.
850 clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
851 tsk->thread.debugctlmsr = 0;
853 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
854 SIGTRAP) == NOTIFY_STOP)
856 /* It's safe to allow irq's after DR6 has been saved */
857 if (regs->flags & X86_EFLAGS_IF)
860 /* Mask out spurious debug traps due to lazy DR7 setting */
861 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
862 if (!tsk->thread.debugreg7)
866 if (regs->flags & VM_MASK)
869 /* Save debug status register where ptrace can see it */
870 tsk->thread.debugreg6 = condition;
873 * Single-stepping through TF: make sure we ignore any events in
874 * kernel space (but re-enable TF when returning to user mode).
876 if (condition & DR_STEP) {
878 * We already checked v86 mode above, so we can
879 * check for kernel mode by just checking the CPL
882 if (!user_mode(regs))
883 goto clear_TF_reenable;
886 /* Ok, finally something we can handle */
887 send_sigtrap(tsk, regs, error_code);
889 /* Disable additional traps. They'll be re-enabled when
890 * the signal is delivered.
897 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
901 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
902 regs->flags &= ~TF_MASK;
907 * Note that we play around with the 'TS' bit in an attempt to get
908 * the correct behaviour even in the presence of the asynchronous
911 void math_error(void __user *ip)
913 struct task_struct * task;
915 unsigned short cwd, swd;
918 * Save the info for the exception handler and clear the error.
922 task->thread.trap_no = 16;
923 task->thread.error_code = 0;
924 info.si_signo = SIGFPE;
926 info.si_code = __SI_FAULT;
929 * (~cwd & swd) will mask out exceptions that are not set to unmasked
930 * status. 0x3f is the exception bits in these regs, 0x200 is the
931 * C1 reg you need in case of a stack fault, 0x040 is the stack
932 * fault bit. We should only be taking one exception at a time,
933 * so if this combination doesn't produce any single exception,
934 * then we have a bad program that isn't syncronizing its FPU usage
935 * and it will suffer the consequences since we won't be able to
936 * fully reproduce the context of the exception
938 cwd = get_fpu_cwd(task);
939 swd = get_fpu_swd(task);
940 switch (swd & ~cwd & 0x3f) {
941 case 0x000: /* No unmasked exception */
943 default: /* Multiple exceptions */
945 case 0x001: /* Invalid Op */
947 * swd & 0x240 == 0x040: Stack Underflow
948 * swd & 0x240 == 0x240: Stack Overflow
949 * User must clear the SF bit (0x40) if set
951 info.si_code = FPE_FLTINV;
953 case 0x002: /* Denormalize */
954 case 0x010: /* Underflow */
955 info.si_code = FPE_FLTUND;
957 case 0x004: /* Zero Divide */
958 info.si_code = FPE_FLTDIV;
960 case 0x008: /* Overflow */
961 info.si_code = FPE_FLTOVF;
963 case 0x020: /* Precision */
964 info.si_code = FPE_FLTRES;
967 force_sig_info(SIGFPE, &info, task);
970 void do_coprocessor_error(struct pt_regs * regs, long error_code)
973 math_error((void __user *)regs->ip);
976 static void simd_math_error(void __user *ip)
978 struct task_struct * task;
980 unsigned short mxcsr;
983 * Save the info for the exception handler and clear the error.
987 task->thread.trap_no = 19;
988 task->thread.error_code = 0;
989 info.si_signo = SIGFPE;
991 info.si_code = __SI_FAULT;
994 * The SIMD FPU exceptions are handled a little differently, as there
995 * is only a single status/control register. Thus, to determine which
996 * unmasked exception was caught we must mask the exception mask bits
997 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
999 mxcsr = get_fpu_mxcsr(task);
1000 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1004 case 0x001: /* Invalid Op */
1005 info.si_code = FPE_FLTINV;
1007 case 0x002: /* Denormalize */
1008 case 0x010: /* Underflow */
1009 info.si_code = FPE_FLTUND;
1011 case 0x004: /* Zero Divide */
1012 info.si_code = FPE_FLTDIV;
1014 case 0x008: /* Overflow */
1015 info.si_code = FPE_FLTOVF;
1017 case 0x020: /* Precision */
1018 info.si_code = FPE_FLTRES;
1021 force_sig_info(SIGFPE, &info, task);
1024 void do_simd_coprocessor_error(struct pt_regs * regs,
1028 /* Handle SIMD FPU exceptions on PIII+ processors. */
1030 simd_math_error((void __user *)regs->ip);
1033 * Handle strange cache flush from user space exception
1034 * in all other cases. This is undocumented behaviour.
1036 if (regs->flags & VM_MASK) {
1037 handle_vm86_fault((struct kernel_vm86_regs *)regs,
1041 current->thread.trap_no = 19;
1042 current->thread.error_code = error_code;
1043 die_if_kernel("cache flush denied", regs, error_code);
1044 force_sig(SIGSEGV, current);
1048 void do_spurious_interrupt_bug(struct pt_regs * regs,
1052 /* No need to warn about this any longer. */
1053 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1057 unsigned long patch_espfix_desc(unsigned long uesp,
1060 struct desc_struct *gdt = __get_cpu_var(gdt_page).gdt;
1061 unsigned long base = (kesp - uesp) & -THREAD_SIZE;
1062 unsigned long new_kesp = kesp - base;
1063 unsigned long lim_pages = (new_kesp | (THREAD_SIZE - 1)) >> PAGE_SHIFT;
1064 __u64 desc = *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS];
1065 /* Set up base for espfix segment */
1066 desc &= 0x00f0ff0000000000ULL;
1067 desc |= ((((__u64)base) << 16) & 0x000000ffffff0000ULL) |
1068 ((((__u64)base) << 32) & 0xff00000000000000ULL) |
1069 ((((__u64)lim_pages) << 32) & 0x000f000000000000ULL) |
1070 (lim_pages & 0xffff);
1071 *(__u64 *)&gdt[GDT_ENTRY_ESPFIX_SS] = desc;
1076 * 'math_state_restore()' saves the current math information in the
1077 * old math state array, and gets the new ones from the current task
1079 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1080 * Don't touch unless you *really* know how it works.
1082 * Must be called with kernel preemption disabled (in this case,
1083 * local interrupts are disabled at the call-site in entry.S).
1085 asmlinkage void math_state_restore(void)
1087 struct thread_info *thread = current_thread_info();
1088 struct task_struct *tsk = thread->task;
1090 clts(); /* Allow maths ops (or we recurse) */
1091 if (!tsk_used_math(tsk))
1094 thread->status |= TS_USEDFPU; /* So we fnsave on switch_to() */
1097 EXPORT_SYMBOL_GPL(math_state_restore);
1099 #ifndef CONFIG_MATH_EMULATION
1101 asmlinkage void math_emulate(long arg)
1103 printk(KERN_EMERG "math-emulation not enabled and no coprocessor found.\n");
1104 printk(KERN_EMERG "killing %s.\n",current->comm);
1105 force_sig(SIGFPE,current);
1109 #endif /* CONFIG_MATH_EMULATION */
1112 void __init trap_init(void)
1117 void __iomem *p = ioremap(0x0FFFD9, 4);
1118 if (readl(p) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1124 #ifdef CONFIG_X86_LOCAL_APIC
1125 init_apic_mappings();
1128 set_trap_gate(0,÷_error);
1129 set_intr_gate(1,&debug);
1130 set_intr_gate(2,&nmi);
1131 set_system_intr_gate(3, &int3); /* int3/4 can be called from all */
1132 set_system_gate(4,&overflow);
1133 set_trap_gate(5,&bounds);
1134 set_trap_gate(6,&invalid_op);
1135 set_trap_gate(7,&device_not_available);
1136 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS);
1137 set_trap_gate(9,&coprocessor_segment_overrun);
1138 set_trap_gate(10,&invalid_TSS);
1139 set_trap_gate(11,&segment_not_present);
1140 set_trap_gate(12,&stack_segment);
1141 set_trap_gate(13,&general_protection);
1142 set_intr_gate(14,&page_fault);
1143 set_trap_gate(15,&spurious_interrupt_bug);
1144 set_trap_gate(16,&coprocessor_error);
1145 set_trap_gate(17,&alignment_check);
1146 #ifdef CONFIG_X86_MCE
1147 set_trap_gate(18,&machine_check);
1149 set_trap_gate(19,&simd_coprocessor_error);
1153 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1154 * Generates a compile-time "error: zero width for bit-field" if
1155 * the alignment is wrong.
1157 struct fxsrAlignAssert {
1158 int _:!(offsetof(struct task_struct,
1159 thread.i387.fxsave) & 15);
1162 printk(KERN_INFO "Enabling fast FPU save and restore... ");
1163 set_in_cr4(X86_CR4_OSFXSR);
1167 printk(KERN_INFO "Enabling unmasked SIMD FPU exception "
1169 set_in_cr4(X86_CR4_OSXMMEXCPT);
1173 set_system_gate(SYSCALL_VECTOR,&system_call);
1175 /* Reserve all the builtin and the syscall vector. */
1176 for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
1177 set_bit(i, used_vectors);
1178 set_bit(SYSCALL_VECTOR, used_vectors);
1181 * Should be a barrier for any external CPU state.
1188 static int __init kstack_setup(char *s)
1190 kstack_depth_to_print = simple_strtoul(s, NULL, 0);
1193 __setup("kstack=", kstack_setup);
1195 static int __init code_bytes_setup(char *s)
1197 code_bytes = simple_strtoul(s, NULL, 0);
1198 if (code_bytes > 8192)
1203 __setup("code_bytes=", code_bytes_setup);