2 * linux/arch/x86-64/traps.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
7 * Pentium III FXSR, SSE support
8 * Gareth Hughes <gareth@valinux.com>, May 2000
12 * 'Traps.c' handles hardware traps and faults after we have saved some
15 #include <linux/sched.h>
16 #include <linux/kernel.h>
17 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/ptrace.h>
20 #include <linux/timer.h>
22 #include <linux/init.h>
23 #include <linux/delay.h>
24 #include <linux/spinlock.h>
25 #include <linux/interrupt.h>
26 #include <linux/module.h>
27 #include <linux/moduleparam.h>
28 #include <linux/nmi.h>
29 #include <linux/kprobes.h>
30 #include <linux/kexec.h>
31 #include <linux/unwind.h>
33 #include <asm/system.h>
34 #include <asm/uaccess.h>
36 #include <asm/atomic.h>
37 #include <asm/debugreg.h>
40 #include <asm/kdebug.h>
41 #include <asm/processor.h>
42 #include <asm/unwind.h>
44 #include <asm/pgalloc.h>
46 #include <asm/proto.h>
48 #include <asm/stacktrace.h>
50 asmlinkage void divide_error(void);
51 asmlinkage void debug(void);
52 asmlinkage void nmi(void);
53 asmlinkage void int3(void);
54 asmlinkage void overflow(void);
55 asmlinkage void bounds(void);
56 asmlinkage void invalid_op(void);
57 asmlinkage void device_not_available(void);
58 asmlinkage void double_fault(void);
59 asmlinkage void coprocessor_segment_overrun(void);
60 asmlinkage void invalid_TSS(void);
61 asmlinkage void segment_not_present(void);
62 asmlinkage void stack_segment(void);
63 asmlinkage void general_protection(void);
64 asmlinkage void page_fault(void);
65 asmlinkage void coprocessor_error(void);
66 asmlinkage void simd_coprocessor_error(void);
67 asmlinkage void reserved(void);
68 asmlinkage void alignment_check(void);
69 asmlinkage void machine_check(void);
70 asmlinkage void spurious_interrupt_bug(void);
72 ATOMIC_NOTIFIER_HEAD(die_chain);
74 extern char last_sysfs_file[];
76 int register_die_notifier(struct notifier_block *nb)
79 return atomic_notifier_chain_register(&die_chain, nb);
81 EXPORT_SYMBOL(register_die_notifier); /* used modular by kdb */
83 int unregister_die_notifier(struct notifier_block *nb)
85 return atomic_notifier_chain_unregister(&die_chain, nb);
87 EXPORT_SYMBOL(unregister_die_notifier); /* used modular by kdb */
89 static inline void conditional_sti(struct pt_regs *regs)
91 if (regs->eflags & X86_EFLAGS_IF)
95 static inline void preempt_conditional_sti(struct pt_regs *regs)
98 if (regs->eflags & X86_EFLAGS_IF)
102 static inline void preempt_conditional_cli(struct pt_regs *regs)
104 if (regs->eflags & X86_EFLAGS_IF)
106 /* Make sure to not schedule here because we could be running
107 on an exception stack. */
108 preempt_enable_no_resched();
111 static int kstack_depth_to_print = 10;
112 #ifdef CONFIG_STACK_UNWIND
113 static int call_trace = 1;
115 #define call_trace (-1)
119 #ifdef CONFIG_KALLSYMS
120 # include <linux/kallsyms.h>
121 void printk_address(unsigned long address)
123 unsigned long offset = 0, symsize;
129 symname = kallsyms_lookup(address, &symsize, &offset,
132 printk(" [<%016lx>]\n", address);
136 modname = delim = "";
137 printk(" [<%016lx>] %s%s%s%s+0x%lx/0x%lx\n",
138 address, delim, modname, delim, symname, offset, symsize);
141 void printk_address(unsigned long address)
143 printk(" [<%016lx>]\n", address);
147 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
148 unsigned *usedp, char **idp)
150 #ifndef CONFIG_X86_NO_TSS
151 static char ids[][8] = {
152 [DEBUG_STACK - 1] = "#DB",
153 [NMI_STACK - 1] = "NMI",
154 [DOUBLEFAULT_STACK - 1] = "#DF",
155 [STACKFAULT_STACK - 1] = "#SS",
156 [MCE_STACK - 1] = "#MC",
157 #if DEBUG_STKSZ > EXCEPTION_STKSZ
158 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
164 * Iterate over all exception stacks, and figure out whether
165 * 'stack' is in one of them:
167 for (k = 0; k < N_EXCEPTION_STACKS; k++) {
171 * set 'end' to the end of the exception stack.
175 * TODO: this block is not needed i think, because
176 * setup64.c:cpu_init() sets up t->ist[DEBUG_STACK]
179 #if DEBUG_STKSZ > EXCEPTION_STKSZ
181 end = cpu_pda(cpu)->debugstack + DEBUG_STKSZ;
185 end = per_cpu(orig_tss, cpu).ist[k];
189 * Is 'stack' above this exception frame's end?
190 * If yes then skip to the next frame.
195 * Is 'stack' above this exception frame's start address?
196 * If yes then we found the right frame.
198 if (stack >= end - EXCEPTION_STKSZ) {
200 * Make sure we only iterate through an exception
201 * stack once. If it comes up for the second time
202 * then there's something wrong going on - just
203 * break out and return NULL:
205 if (*usedp & (1U << k))
209 return (unsigned long *)end;
212 * If this is a debug stack, and if it has a larger size than
213 * the usual exception stacks, then 'stack' might still
214 * be within the lower portion of the debug stack:
216 #if DEBUG_STKSZ > EXCEPTION_STKSZ
217 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
218 unsigned j = N_EXCEPTION_STACKS - 1;
221 * Black magic. A large debug stack is composed of
222 * multiple exception stack entries, which we
223 * iterate through now. Dont look:
227 end -= EXCEPTION_STKSZ;
228 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
229 } while (stack < end - EXCEPTION_STKSZ);
230 if (*usedp & (1U << j))
234 return (unsigned long *)end;
242 struct ops_and_data {
243 struct stacktrace_ops *ops;
247 static int dump_trace_unwind(struct unwind_frame_info *info, void *context)
249 struct ops_and_data *oad = (struct ops_and_data *)context;
252 while (unwind(info) == 0 && UNW_PC(info)) {
253 if (arch_unw_user_mode(info))
256 oad->ops->address(oad->data, UNW_PC(info));
262 * x86-64 can have upto three kernel stacks:
265 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
268 void dump_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long * stack,
269 struct stacktrace_ops *ops, void *data)
271 const unsigned cpu = safe_smp_processor_id();
272 unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
278 if (call_trace >= 0) {
280 struct unwind_frame_info info;
281 struct ops_and_data oad = { .ops = ops, .data = data };
284 if (unwind_init_frame_info(&info, tsk, regs) == 0)
285 unw_ret = dump_trace_unwind(&info, &oad);
286 } else if (tsk == current)
287 unw_ret = unwind_init_running(&info, dump_trace_unwind, &oad);
289 if (unwind_init_blocked(&info, tsk) == 0)
290 unw_ret = dump_trace_unwind(&info, &oad);
293 if (call_trace == 1 && !arch_unw_user_mode(&info)) {
294 ops->warning_symbol(data, "DWARF2 unwinder stuck at %s\n",
296 if ((long)UNW_SP(&info) < 0) {
297 ops->warning(data, "Leftover inexact backtrace:\n");
298 stack = (unsigned long *)UNW_SP(&info);
300 ops->warning(data, "Full inexact backtrace again:\n");
301 } else if (call_trace >= 1)
304 ops->warning(data, "Full inexact backtrace again:\n");
306 ops->warning(data, "Inexact backtrace:\n");
311 if (tsk && tsk != current)
312 stack = (unsigned long *)tsk->thread.rsp;
316 * Print function call entries within a stack. 'cond' is the
317 * "end of stackframe" condition, that the 'stack++'
318 * iteration will eventually trigger.
320 #define HANDLE_STACK(cond) \
322 unsigned long addr = *stack++; \
323 if (kernel_text_address(addr)) { \
325 * If the address is either in the text segment of the \
326 * kernel, or in the region which contains vmalloc'ed \
327 * memory, it *may* be the address of a calling \
328 * routine; if so, print it so that someone tracing \
329 * down the cause of the crash will be able to figure \
330 * out the call path that was taken. \
332 ops->address(data, addr); \
337 * Print function call entries in all stacks, starting at the
338 * current stack address. If the stacks consist of nested
343 unsigned long *estack_end;
344 estack_end = in_exception_stack(cpu, (unsigned long)stack,
348 if (ops->stack(data, id) < 0)
350 HANDLE_STACK (stack < estack_end);
351 ops->stack(data, "<EOE>");
353 * We link to the next stack via the
354 * second-to-last pointer (index -2 to end) in the
357 stack = (unsigned long *) estack_end[-2];
361 unsigned long *irqstack;
362 irqstack = irqstack_end -
363 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
365 if (stack >= irqstack && stack < irqstack_end) {
366 if (ops->stack(data, "IRQ") < 0)
368 HANDLE_STACK (stack < irqstack_end);
370 * We link to the next stack (which would be
371 * the process stack normally) the last
372 * pointer (index -1 to end) in the IRQ stack:
374 stack = (unsigned long *) (irqstack_end[-1]);
376 ops->stack(data, "EOI");
384 * This handles the process stack:
386 HANDLE_STACK (((long) stack & (THREAD_SIZE-1)) != 0);
389 EXPORT_SYMBOL(dump_trace);
392 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
394 print_symbol(msg, symbol);
398 static void print_trace_warning(void *data, char *msg)
403 static int print_trace_stack(void *data, char *name)
405 printk(" <%s> ", name);
409 static void print_trace_address(void *data, unsigned long addr)
411 printk_address(addr);
414 static struct stacktrace_ops print_trace_ops = {
415 .warning = print_trace_warning,
416 .warning_symbol = print_trace_warning_symbol,
417 .stack = print_trace_stack,
418 .address = print_trace_address,
422 show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack)
424 printk("\nCall Trace:\n");
425 dump_trace(tsk, regs, stack, &print_trace_ops, NULL);
430 _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
432 unsigned long *stack;
434 const int cpu = safe_smp_processor_id();
435 unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
436 unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
438 // debugging aid: "show_stack(NULL, NULL);" prints the
439 // back trace for this cpu.
443 rsp = (unsigned long *)tsk->thread.rsp;
445 rsp = (unsigned long *)&rsp;
449 for(i=0; i < kstack_depth_to_print; i++) {
450 if (stack >= irqstack && stack <= irqstack_end) {
451 if (stack == irqstack_end) {
452 stack = (unsigned long *) (irqstack_end[-1]);
456 if (((long) stack & (THREAD_SIZE-1)) == 0)
459 if (i && ((i % 4) == 0))
461 printk(" %016lx ", *stack++);
462 touch_nmi_watchdog();
464 show_trace(tsk, regs, rsp);
467 void show_stack(struct task_struct *tsk, unsigned long * rsp)
469 _show_stack(tsk, NULL, rsp);
473 * The architecture-independent dump_stack generator
475 void dump_stack(void)
478 show_trace(NULL, NULL, &dummy);
481 EXPORT_SYMBOL(dump_stack);
483 void show_registers(struct pt_regs *regs)
486 int in_kernel = !user_mode(regs);
488 const int cpu = safe_smp_processor_id();
489 struct task_struct *cur = cpu_pda(cpu)->pcurrent;
493 printk("CPU %d ", cpu);
495 printk("Process %s (pid: %d[#%u], threadinfo %p, task %p)\n",
496 cur->comm, cur->pid, cur->xid,
497 task_thread_info(cur), cur);
500 * When in-kernel, we also print out the stack and code at the
501 * time of the fault..
506 _show_stack(NULL, regs, (unsigned long*)rsp);
509 if (regs->rip < PAGE_OFFSET)
512 for (i=0; i<20; i++) {
514 if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
516 printk(" Bad RIP value.");
525 void handle_BUG(struct pt_regs *regs)
529 const char *prefix = "";
533 if (__copy_from_user(&f, (const void __user *) regs->rip,
534 sizeof(struct bug_frame)))
536 if (f.filename >= 0 ||
537 f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
539 len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1;
540 if (len < 0 || len >= PATH_MAX)
541 f.filename = (int)(long)"unmapped filename";
543 f.filename += len - 50;
546 printk("----------- [cut here ] --------- [please bite here ] ---------\n");
547 printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line);
551 void out_of_line_bug(void)
557 static DEFINE_SPINLOCK(die_lock);
558 static int die_owner = -1;
559 static unsigned int die_nest_count;
561 unsigned __kprobes long oops_begin(void)
563 int cpu = safe_smp_processor_id();
566 /* racy, but better than risking deadlock. */
567 local_irq_save(flags);
568 if (!spin_trylock(&die_lock)) {
569 if (cpu == die_owner)
570 /* nested oops. should stop eventually */;
572 spin_lock(&die_lock);
581 void __kprobes oops_end(unsigned long flags)
587 /* We still own the lock */
588 local_irq_restore(flags);
590 /* Nest count reaches zero, release the lock. */
591 spin_unlock_irqrestore(&die_lock, flags);
593 panic("Fatal exception");
596 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
598 static int die_counter;
599 printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
600 #ifdef CONFIG_PREEMPT
606 #ifdef CONFIG_DEBUG_PAGEALLOC
607 printk("DEBUG_PAGEALLOC");
611 printk(KERN_ALERT "last sysfs file: %s\n", last_sysfs_file);
613 notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
614 show_registers(regs);
615 /* Executive summary in case the oops scrolled away */
616 printk(KERN_ALERT "RIP ");
617 printk_address(regs->rip);
618 printk(" RSP <%016lx>\n", regs->rsp);
619 if (kexec_should_crash(current))
623 void die(const char * str, struct pt_regs * regs, long err)
625 unsigned long flags = oops_begin();
628 __die(str, regs, err);
633 #ifdef CONFIG_X86_LOCAL_APIC
634 void __kprobes die_nmi(char *str, struct pt_regs *regs)
636 unsigned long flags = oops_begin();
639 * We are in trouble anyway, lets at least try
640 * to get a message out.
642 printk(str, safe_smp_processor_id());
643 show_registers(regs);
644 if (kexec_should_crash(current))
646 if (panic_on_timeout || panic_on_oops)
647 panic("nmi watchdog");
648 printk("console shuts up ...\n");
656 static void __kprobes do_trap(int trapnr, int signr, char *str,
657 struct pt_regs * regs, long error_code,
660 struct task_struct *tsk = current;
662 tsk->thread.error_code = error_code;
663 tsk->thread.trap_no = trapnr;
665 if (user_mode(regs)) {
666 if (exception_trace && unhandled_signal(tsk, signr))
668 "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
669 tsk->comm, tsk->pid, str,
670 regs->rip, regs->rsp, error_code);
673 force_sig_info(signr, info, tsk);
675 force_sig(signr, tsk);
682 const struct exception_table_entry *fixup;
683 fixup = search_exception_tables(regs->rip);
685 regs->rip = fixup->fixup;
687 die(str, regs, error_code);
692 #define DO_ERROR(trapnr, signr, str, name) \
693 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
695 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
698 conditional_sti(regs); \
699 do_trap(trapnr, signr, str, regs, error_code, NULL); \
702 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
703 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
706 info.si_signo = signr; \
708 info.si_code = sicode; \
709 info.si_addr = (void __user *)siaddr; \
710 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
713 conditional_sti(regs); \
714 do_trap(trapnr, signr, str, regs, error_code, &info); \
717 DO_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->rip)
718 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
719 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
720 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
721 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
722 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
723 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
724 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
725 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
726 DO_ERROR(18, SIGSEGV, "reserved", reserved)
728 /* Runs on IST stack */
729 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
731 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
732 12, SIGBUS) == NOTIFY_STOP)
734 preempt_conditional_sti(regs);
735 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
736 preempt_conditional_cli(regs);
739 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
741 static const char str[] = "double fault";
742 struct task_struct *tsk = current;
744 /* Return not checked because double check cannot be ignored */
745 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
747 tsk->thread.error_code = error_code;
748 tsk->thread.trap_no = 8;
750 /* This is always a kernel trap and never fixable (and thus must
753 die(str, regs, error_code);
756 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
759 struct task_struct *tsk = current;
761 conditional_sti(regs);
763 tsk->thread.error_code = error_code;
764 tsk->thread.trap_no = 13;
766 if (user_mode(regs)) {
767 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
769 "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
771 regs->rip, regs->rsp, error_code);
773 force_sig(SIGSEGV, tsk);
779 const struct exception_table_entry *fixup;
780 fixup = search_exception_tables(regs->rip);
782 regs->rip = fixup->fixup;
785 if (notify_die(DIE_GPF, "general protection fault", regs,
786 error_code, 13, SIGSEGV) == NOTIFY_STOP)
788 die("general protection fault", regs, error_code);
792 static __kprobes void
793 mem_parity_error(unsigned char reason, struct pt_regs * regs)
795 printk("Uhhuh. NMI received. Dazed and confused, but trying to continue\n");
796 printk("You probably have a hardware problem with your RAM chips\n");
799 /* Clear and disable the memory parity error line. */
800 reason = (reason & 0xf) | 4;
805 static __kprobes void
806 io_check_error(unsigned char reason, struct pt_regs * regs)
808 printk("NMI: IOCK error (debug interrupt?)\n");
809 show_registers(regs);
812 /* Re-enable the IOCK line, wait for a few seconds */
813 reason = (reason & 0xf) | 8;
821 static __kprobes void
822 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
823 { printk("Uhhuh. NMI received for unknown reason %02x.\n", reason);
824 printk("Dazed and confused, but trying to continue\n");
825 printk("Do you have a strange power saving mode enabled?\n");
828 /* Runs on IST stack. This code must keep interrupts off all the time.
829 Nested NMIs are prevented by the CPU. */
830 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
832 unsigned char reason = 0;
835 cpu = smp_processor_id();
837 /* Only the BSP gets external NMIs from the system. */
839 reason = get_nmi_reason();
841 if (!(reason & 0xc0)) {
842 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
845 #ifdef CONFIG_X86_LOCAL_APIC
847 * Ok, so this is none of the documented NMI sources,
848 * so it must be the NMI watchdog.
850 if (nmi_watchdog > 0) {
851 nmi_watchdog_tick(regs,reason);
855 unknown_nmi_error(reason, regs);
858 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
861 /* AK: following checks seem to be broken on modern chipsets. FIXME */
864 mem_parity_error(reason, regs);
866 io_check_error(reason, regs);
869 /* runs on IST stack. */
870 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
872 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
875 preempt_conditional_sti(regs);
876 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
877 preempt_conditional_cli(regs);
880 /* Help handler running on IST stack to switch back to user stack
881 for scheduling or signal handling. The actual stack switch is done in
883 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
885 struct pt_regs *regs = eregs;
886 /* Did already sync */
887 if (eregs == (struct pt_regs *)eregs->rsp)
889 /* Exception from user space */
890 else if (user_mode(eregs))
891 regs = task_pt_regs(current);
892 /* Exception from kernel and interrupts are enabled. Move to
893 kernel process stack. */
894 else if (eregs->eflags & X86_EFLAGS_IF)
895 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
901 /* runs on IST stack. */
902 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
903 unsigned long error_code)
905 unsigned long condition;
906 struct task_struct *tsk = current;
909 get_debugreg(condition, 6);
911 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
912 SIGTRAP) == NOTIFY_STOP)
915 preempt_conditional_sti(regs);
917 /* Mask out spurious debug traps due to lazy DR7 setting */
918 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
919 if (!tsk->thread.debugreg7) {
924 tsk->thread.debugreg6 = condition;
926 /* Mask out spurious TF errors due to lazy TF clearing */
927 if (condition & DR_STEP) {
929 * The TF error should be masked out only if the current
930 * process is not traced and if the TRAP flag has been set
931 * previously by a tracing process (condition detected by
932 * the PT_DTRACE flag); remember that the i386 TRAP flag
933 * can be modified by the process itself in user mode,
934 * allowing programs to debug themselves without the ptrace()
937 if (!user_mode(regs))
938 goto clear_TF_reenable;
941 /* Ok, finally something we can handle */
942 tsk->thread.trap_no = 1;
943 tsk->thread.error_code = error_code;
944 info.si_signo = SIGTRAP;
946 info.si_code = TRAP_BRKPT;
947 info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
948 force_sig_info(SIGTRAP, &info, tsk);
951 set_debugreg(0UL, 7);
952 preempt_conditional_cli(regs);
956 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
957 regs->eflags &= ~TF_MASK;
958 preempt_conditional_cli(regs);
961 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
963 const struct exception_table_entry *fixup;
964 fixup = search_exception_tables(regs->rip);
966 regs->rip = fixup->fixup;
969 notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
970 /* Illegal floating point operation in the kernel */
971 current->thread.trap_no = trapnr;
977 * Note that we play around with the 'TS' bit in an attempt to get
978 * the correct behaviour even in the presence of the asynchronous
981 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
983 void __user *rip = (void __user *)(regs->rip);
984 struct task_struct * task;
986 unsigned short cwd, swd;
988 conditional_sti(regs);
989 if (!user_mode(regs) &&
990 kernel_math_error(regs, "kernel x87 math error", 16))
994 * Save the info for the exception handler and clear the error.
998 task->thread.trap_no = 16;
999 task->thread.error_code = 0;
1000 info.si_signo = SIGFPE;
1002 info.si_code = __SI_FAULT;
1005 * (~cwd & swd) will mask out exceptions that are not set to unmasked
1006 * status. 0x3f is the exception bits in these regs, 0x200 is the
1007 * C1 reg you need in case of a stack fault, 0x040 is the stack
1008 * fault bit. We should only be taking one exception at a time,
1009 * so if this combination doesn't produce any single exception,
1010 * then we have a bad program that isn't synchronizing its FPU usage
1011 * and it will suffer the consequences since we won't be able to
1012 * fully reproduce the context of the exception
1014 cwd = get_fpu_cwd(task);
1015 swd = get_fpu_swd(task);
1016 switch (swd & ~cwd & 0x3f) {
1020 case 0x001: /* Invalid Op */
1022 * swd & 0x240 == 0x040: Stack Underflow
1023 * swd & 0x240 == 0x240: Stack Overflow
1024 * User must clear the SF bit (0x40) if set
1026 info.si_code = FPE_FLTINV;
1028 case 0x002: /* Denormalize */
1029 case 0x010: /* Underflow */
1030 info.si_code = FPE_FLTUND;
1032 case 0x004: /* Zero Divide */
1033 info.si_code = FPE_FLTDIV;
1035 case 0x008: /* Overflow */
1036 info.si_code = FPE_FLTOVF;
1038 case 0x020: /* Precision */
1039 info.si_code = FPE_FLTRES;
1042 force_sig_info(SIGFPE, &info, task);
1045 asmlinkage void bad_intr(void)
1047 printk("bad interrupt");
1050 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1052 void __user *rip = (void __user *)(regs->rip);
1053 struct task_struct * task;
1055 unsigned short mxcsr;
1057 conditional_sti(regs);
1058 if (!user_mode(regs) &&
1059 kernel_math_error(regs, "kernel simd math error", 19))
1063 * Save the info for the exception handler and clear the error.
1066 save_init_fpu(task);
1067 task->thread.trap_no = 19;
1068 task->thread.error_code = 0;
1069 info.si_signo = SIGFPE;
1071 info.si_code = __SI_FAULT;
1074 * The SIMD FPU exceptions are handled a little differently, as there
1075 * is only a single status/control register. Thus, to determine which
1076 * unmasked exception was caught we must mask the exception mask bits
1077 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1079 mxcsr = get_fpu_mxcsr(task);
1080 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1084 case 0x001: /* Invalid Op */
1085 info.si_code = FPE_FLTINV;
1087 case 0x002: /* Denormalize */
1088 case 0x010: /* Underflow */
1089 info.si_code = FPE_FLTUND;
1091 case 0x004: /* Zero Divide */
1092 info.si_code = FPE_FLTDIV;
1094 case 0x008: /* Overflow */
1095 info.si_code = FPE_FLTOVF;
1097 case 0x020: /* Precision */
1098 info.si_code = FPE_FLTRES;
1101 force_sig_info(SIGFPE, &info, task);
1104 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1109 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1114 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1119 * 'math_state_restore()' saves the current math information in the
1120 * old math state array, and gets the new ones from the current task
1122 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1123 * Don't touch unless you *really* know how it works.
1125 asmlinkage void math_state_restore(void)
1127 struct task_struct *me = current;
1128 /* clts(); */ /* 'clts' is done for us by Xen during virtual trap. */
1132 restore_fpu_checking(&me->thread.i387.fxsave);
1133 task_thread_info(me)->status |= TS_USEDFPU;
1138 * NB. All these are "interrupt gates" (i.e. events_mask is set) because we
1139 * specify <dpl>|4 in the second field.
1141 static trap_info_t trap_table[] = {
1142 { 0, 0|4, __KERNEL_CS, (unsigned long)divide_error },
1143 { 1, 0|4, __KERNEL_CS, (unsigned long)debug },
1144 { 3, 3|4, __KERNEL_CS, (unsigned long)int3 },
1145 { 4, 3|4, __KERNEL_CS, (unsigned long)overflow },
1146 { 5, 0|4, __KERNEL_CS, (unsigned long)bounds },
1147 { 6, 0|4, __KERNEL_CS, (unsigned long)invalid_op },
1148 { 7, 0|4, __KERNEL_CS, (unsigned long)device_not_available },
1149 { 9, 0|4, __KERNEL_CS, (unsigned long)coprocessor_segment_overrun},
1150 { 10, 0|4, __KERNEL_CS, (unsigned long)invalid_TSS },
1151 { 11, 0|4, __KERNEL_CS, (unsigned long)segment_not_present },
1152 { 12, 0|4, __KERNEL_CS, (unsigned long)stack_segment },
1153 { 13, 0|4, __KERNEL_CS, (unsigned long)general_protection },
1154 { 14, 0|4, __KERNEL_CS, (unsigned long)page_fault },
1155 { 15, 0|4, __KERNEL_CS, (unsigned long)spurious_interrupt_bug },
1156 { 16, 0|4, __KERNEL_CS, (unsigned long)coprocessor_error },
1157 { 17, 0|4, __KERNEL_CS, (unsigned long)alignment_check },
1158 #ifdef CONFIG_X86_MCE
1159 { 18, 0|4, __KERNEL_CS, (unsigned long)machine_check },
1161 { 19, 0|4, __KERNEL_CS, (unsigned long)simd_coprocessor_error },
1162 #ifdef CONFIG_IA32_EMULATION
1163 { IA32_SYSCALL_VECTOR, 3|4, __KERNEL_CS, (unsigned long)ia32_syscall},
1168 void __init trap_init(void)
1172 ret = HYPERVISOR_set_trap_table(trap_table);
1175 printk("HYPERVISOR_set_trap_table faild: error %d\n",
1179 * Should be a barrier for any external CPU state.
1184 void smp_trap_init(trap_info_t *trap_ctxt)
1186 trap_info_t *t = trap_table;
1188 for (t = trap_table; t->address; t++) {
1189 trap_ctxt[t->vector].flags = t->flags;
1190 trap_ctxt[t->vector].cs = t->cs;
1191 trap_ctxt[t->vector].address = t->address;
1196 /* Actual parsing is done early in setup.c. */
1197 static int __init oops_dummy(char *s)
1202 __setup("oops=", oops_dummy);
1204 static int __init kstack_setup(char *s)
1206 kstack_depth_to_print = simple_strtoul(s,NULL,0);
1209 __setup("kstack=", kstack_setup);
1211 #ifdef CONFIG_STACK_UNWIND
1212 static int __init call_trace_setup(char *s)
1214 if (strcmp(s, "old") == 0)
1216 else if (strcmp(s, "both") == 0)
1218 else if (strcmp(s, "newfallback") == 0)
1220 else if (strcmp(s, "new") == 0)
1224 __setup("call_trace=", call_trace_setup);