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/kallsyms.h>
27 #include <linux/module.h>
28 #include <linux/moduleparam.h>
29 #include <linux/nmi.h>
30 #include <linux/kprobes.h>
31 #include <linux/kexec.h>
32 #include <linux/unwind.h>
34 #include <asm/system.h>
35 #include <asm/uaccess.h>
37 #include <asm/atomic.h>
38 #include <asm/debugreg.h>
41 #include <asm/kdebug.h>
42 #include <asm/processor.h>
43 #include <asm/unwind.h>
45 #include <asm/pgalloc.h>
47 #include <asm/proto.h>
49 #include <asm/stacktrace.h>
51 asmlinkage void divide_error(void);
52 asmlinkage void debug(void);
53 asmlinkage void nmi(void);
54 asmlinkage void int3(void);
55 asmlinkage void overflow(void);
56 asmlinkage void bounds(void);
57 asmlinkage void invalid_op(void);
58 asmlinkage void device_not_available(void);
59 asmlinkage void double_fault(void);
60 asmlinkage void coprocessor_segment_overrun(void);
61 asmlinkage void invalid_TSS(void);
62 asmlinkage void segment_not_present(void);
63 asmlinkage void stack_segment(void);
64 asmlinkage void general_protection(void);
65 asmlinkage void page_fault(void);
66 asmlinkage void coprocessor_error(void);
67 asmlinkage void simd_coprocessor_error(void);
68 asmlinkage void reserved(void);
69 asmlinkage void alignment_check(void);
70 asmlinkage void machine_check(void);
71 asmlinkage void spurious_interrupt_bug(void);
73 ATOMIC_NOTIFIER_HEAD(die_chain);
74 EXPORT_SYMBOL(die_chain);
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 = 12;
112 #ifdef CONFIG_STACK_UNWIND
113 static int call_trace = 1;
115 #define call_trace (-1)
118 #ifdef CONFIG_KALLSYMS
119 void printk_address(unsigned long address)
121 unsigned long offset = 0, symsize;
127 symname = kallsyms_lookup(address, &symsize, &offset,
130 printk(" [<%016lx>]\n", address);
134 modname = delim = "";
135 printk(" [<%016lx>] %s%s%s%s+0x%lx/0x%lx\n",
136 address, delim, modname, delim, symname, offset, symsize);
139 void printk_address(unsigned long address)
141 printk(" [<%016lx>]\n", address);
145 static unsigned long *in_exception_stack(unsigned cpu, unsigned long stack,
146 unsigned *usedp, char **idp)
148 static char ids[][8] = {
149 [DEBUG_STACK - 1] = "#DB",
150 [NMI_STACK - 1] = "NMI",
151 [DOUBLEFAULT_STACK - 1] = "#DF",
152 [STACKFAULT_STACK - 1] = "#SS",
153 [MCE_STACK - 1] = "#MC",
154 #if DEBUG_STKSZ > EXCEPTION_STKSZ
155 [N_EXCEPTION_STACKS ... N_EXCEPTION_STACKS + DEBUG_STKSZ / EXCEPTION_STKSZ - 2] = "#DB[?]"
161 * Iterate over all exception stacks, and figure out whether
162 * 'stack' is in one of them:
164 for (k = 0; k < N_EXCEPTION_STACKS; k++) {
165 unsigned long end = per_cpu(orig_ist, cpu).ist[k];
167 * Is 'stack' above this exception frame's end?
168 * If yes then skip to the next frame.
173 * Is 'stack' above this exception frame's start address?
174 * If yes then we found the right frame.
176 if (stack >= end - EXCEPTION_STKSZ) {
178 * Make sure we only iterate through an exception
179 * stack once. If it comes up for the second time
180 * then there's something wrong going on - just
181 * break out and return NULL:
183 if (*usedp & (1U << k))
187 return (unsigned long *)end;
190 * If this is a debug stack, and if it has a larger size than
191 * the usual exception stacks, then 'stack' might still
192 * be within the lower portion of the debug stack:
194 #if DEBUG_STKSZ > EXCEPTION_STKSZ
195 if (k == DEBUG_STACK - 1 && stack >= end - DEBUG_STKSZ) {
196 unsigned j = N_EXCEPTION_STACKS - 1;
199 * Black magic. A large debug stack is composed of
200 * multiple exception stack entries, which we
201 * iterate through now. Dont look:
205 end -= EXCEPTION_STKSZ;
206 ids[j][4] = '1' + (j - N_EXCEPTION_STACKS);
207 } while (stack < end - EXCEPTION_STKSZ);
208 if (*usedp & (1U << j))
212 return (unsigned long *)end;
219 struct ops_and_data {
220 struct stacktrace_ops *ops;
224 static int dump_trace_unwind(struct unwind_frame_info *info, void *context)
226 struct ops_and_data *oad = (struct ops_and_data *)context;
229 while (unwind(info) == 0 && UNW_PC(info)) {
231 oad->ops->address(oad->data, UNW_PC(info));
232 if (arch_unw_user_mode(info))
239 * x86-64 can have upto three kernel stacks:
242 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
245 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
247 void *t = (void *)tinfo;
248 return p > t && p < t + THREAD_SIZE - 3;
251 void dump_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long * stack,
252 struct stacktrace_ops *ops, void *data)
254 const unsigned cpu = smp_processor_id();
255 unsigned long *irqstack_end = (unsigned long *)cpu_pda(cpu)->irqstackptr;
257 struct thread_info *tinfo;
262 if (call_trace >= 0) {
264 struct unwind_frame_info info;
265 struct ops_and_data oad = { .ops = ops, .data = data };
268 if (unwind_init_frame_info(&info, tsk, regs) == 0)
269 unw_ret = dump_trace_unwind(&info, &oad);
270 } else if (tsk == current)
271 unw_ret = unwind_init_running(&info, dump_trace_unwind, &oad);
273 if (unwind_init_blocked(&info, tsk) == 0)
274 unw_ret = dump_trace_unwind(&info, &oad);
277 if (call_trace == 1 && !arch_unw_user_mode(&info)) {
278 ops->warning_symbol(data, "DWARF2 unwinder stuck at %s\n",
280 if ((long)UNW_SP(&info) < 0) {
281 ops->warning(data, "Leftover inexact backtrace:\n");
282 stack = (unsigned long *)UNW_SP(&info);
286 ops->warning(data, "Full inexact backtrace again:\n");
287 } else if (call_trace >= 1)
290 ops->warning(data, "Full inexact backtrace again:\n");
292 ops->warning(data, "Inexact backtrace:\n");
297 if (tsk && tsk != current)
298 stack = (unsigned long *)tsk->thread.rsp;
301 * Align the stack pointer on word boundary, later loops
302 * rely on that (and corruption / debug info bugs can cause
303 * unaligned values here):
305 stack = (unsigned long *)((unsigned long)stack & ~(sizeof(long)-1));
308 * Print function call entries within a stack. 'cond' is the
309 * "end of stackframe" condition, that the 'stack++'
310 * iteration will eventually trigger.
312 #define HANDLE_STACK(cond) \
314 unsigned long addr = *stack++; \
315 if (oops_in_progress ? \
316 __kernel_text_address(addr) : \
317 kernel_text_address(addr)) { \
319 * If the address is either in the text segment of the \
320 * kernel, or in the region which contains vmalloc'ed \
321 * memory, it *may* be the address of a calling \
322 * routine; if so, print it so that someone tracing \
323 * down the cause of the crash will be able to figure \
324 * out the call path that was taken. \
326 ops->address(data, addr); \
331 * Print function call entries in all stacks, starting at the
332 * current stack address. If the stacks consist of nested
337 unsigned long *estack_end;
338 estack_end = in_exception_stack(cpu, (unsigned long)stack,
342 if (ops->stack(data, id) < 0)
344 HANDLE_STACK (stack < estack_end);
345 ops->stack(data, "<EOE>");
347 * We link to the next stack via the
348 * second-to-last pointer (index -2 to end) in the
351 stack = (unsigned long *) estack_end[-2];
355 unsigned long *irqstack;
356 irqstack = irqstack_end -
357 (IRQSTACKSIZE - 64) / sizeof(*irqstack);
359 if (stack >= irqstack && stack < irqstack_end) {
360 if (ops->stack(data, "IRQ") < 0)
362 HANDLE_STACK (stack < irqstack_end);
364 * We link to the next stack (which would be
365 * the process stack normally) the last
366 * pointer (index -1 to end) in the IRQ stack:
368 stack = (unsigned long *) (irqstack_end[-1]);
370 ops->stack(data, "EOI");
378 * This handles the process stack:
380 tinfo = current_thread_info();
381 HANDLE_STACK (valid_stack_ptr(tinfo, stack));
384 EXPORT_SYMBOL(dump_trace);
387 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
389 print_symbol(msg, symbol);
393 static void print_trace_warning(void *data, char *msg)
398 static int print_trace_stack(void *data, char *name)
400 printk(" <%s> ", name);
404 static void print_trace_address(void *data, unsigned long addr)
406 printk_address(addr);
409 static struct stacktrace_ops print_trace_ops = {
410 .warning = print_trace_warning,
411 .warning_symbol = print_trace_warning_symbol,
412 .stack = print_trace_stack,
413 .address = print_trace_address,
417 show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack)
419 printk("\nCall Trace:\n");
420 dump_trace(tsk, regs, stack, &print_trace_ops, NULL);
425 _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
427 unsigned long *stack;
429 const int cpu = smp_processor_id();
430 unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
431 unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
433 // debugging aid: "show_stack(NULL, NULL);" prints the
434 // back trace for this cpu.
438 rsp = (unsigned long *)tsk->thread.rsp;
440 rsp = (unsigned long *)&rsp;
444 for(i=0; i < kstack_depth_to_print; i++) {
445 if (stack >= irqstack && stack <= irqstack_end) {
446 if (stack == irqstack_end) {
447 stack = (unsigned long *) (irqstack_end[-1]);
451 if (((long) stack & (THREAD_SIZE-1)) == 0)
454 if (i && ((i % 4) == 0))
456 printk(" %016lx", *stack++);
457 touch_nmi_watchdog();
459 show_trace(tsk, regs, rsp);
462 void show_stack(struct task_struct *tsk, unsigned long * rsp)
464 _show_stack(tsk, NULL, rsp);
468 * The architecture-independent dump_stack generator
470 void dump_stack(void)
473 show_trace(NULL, NULL, &dummy);
476 EXPORT_SYMBOL(dump_stack);
478 void show_registers(struct pt_regs *regs)
481 int in_kernel = !user_mode(regs);
483 const int cpu = smp_processor_id();
484 struct task_struct *cur = cpu_pda(cpu)->pcurrent;
488 printk("CPU %d ", cpu);
490 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
491 cur->comm, cur->pid, task_thread_info(cur), cur);
494 * When in-kernel, we also print out the stack and code at the
495 * time of the fault..
500 _show_stack(NULL, regs, (unsigned long*)rsp);
503 if (regs->rip < PAGE_OFFSET)
506 for (i=0; i<20; i++) {
508 if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
510 printk(" Bad RIP value.");
519 void handle_BUG(struct pt_regs *regs)
523 const char *prefix = "";
527 if (__copy_from_user(&f, (const void __user *) regs->rip,
528 sizeof(struct bug_frame)))
530 if (f.filename >= 0 ||
531 f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
533 len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1;
534 if (len < 0 || len >= PATH_MAX)
535 f.filename = (int)(long)"unmapped filename";
537 f.filename += len - 50;
540 printk("----------- [cut here ] --------- [please bite here ] ---------\n");
541 printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line);
545 void out_of_line_bug(void)
549 EXPORT_SYMBOL(out_of_line_bug);
552 static DEFINE_SPINLOCK(die_lock);
553 static int die_owner = -1;
554 static unsigned int die_nest_count;
556 unsigned __kprobes long oops_begin(void)
558 int cpu = smp_processor_id();
563 /* racy, but better than risking deadlock. */
564 local_irq_save(flags);
565 if (!spin_trylock(&die_lock)) {
566 if (cpu == die_owner)
567 /* nested oops. should stop eventually */;
569 spin_lock(&die_lock);
578 void __kprobes oops_end(unsigned long flags)
584 /* We still own the lock */
585 local_irq_restore(flags);
587 /* Nest count reaches zero, release the lock. */
588 spin_unlock_irqrestore(&die_lock, flags);
590 panic("Fatal exception");
594 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
596 static int die_counter;
597 printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
598 #ifdef CONFIG_PREEMPT
604 #ifdef CONFIG_DEBUG_PAGEALLOC
605 printk("DEBUG_PAGEALLOC");
608 notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
609 show_registers(regs);
610 /* Executive summary in case the oops scrolled away */
611 printk(KERN_ALERT "RIP ");
612 printk_address(regs->rip);
613 printk(" RSP <%016lx>\n", regs->rsp);
614 if (kexec_should_crash(current))
618 void die(const char * str, struct pt_regs * regs, long err)
620 unsigned long flags = oops_begin();
623 __die(str, regs, err);
628 void __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
630 unsigned long flags = oops_begin();
633 * We are in trouble anyway, lets at least try
634 * to get a message out.
636 printk(str, smp_processor_id());
637 show_registers(regs);
638 if (kexec_should_crash(current))
640 if (do_panic || panic_on_oops)
641 panic("Non maskable interrupt");
648 static void __kprobes do_trap(int trapnr, int signr, char *str,
649 struct pt_regs * regs, long error_code,
652 struct task_struct *tsk = current;
654 tsk->thread.error_code = error_code;
655 tsk->thread.trap_no = trapnr;
657 if (user_mode(regs)) {
658 if (exception_trace && unhandled_signal(tsk, signr))
660 "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
661 tsk->comm, tsk->pid, str,
662 regs->rip, regs->rsp, error_code);
665 force_sig_info(signr, info, tsk);
667 force_sig(signr, tsk);
674 const struct exception_table_entry *fixup;
675 fixup = search_exception_tables(regs->rip);
677 regs->rip = fixup->fixup;
679 die(str, regs, error_code);
684 #define DO_ERROR(trapnr, signr, str, name) \
685 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
687 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
690 conditional_sti(regs); \
691 do_trap(trapnr, signr, str, regs, error_code, NULL); \
694 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
695 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
698 info.si_signo = signr; \
700 info.si_code = sicode; \
701 info.si_addr = (void __user *)siaddr; \
702 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
705 conditional_sti(regs); \
706 do_trap(trapnr, signr, str, regs, error_code, &info); \
709 DO_ERROR_INFO( 0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->rip)
710 DO_ERROR( 4, SIGSEGV, "overflow", overflow)
711 DO_ERROR( 5, SIGSEGV, "bounds", bounds)
712 DO_ERROR_INFO( 6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->rip)
713 DO_ERROR( 7, SIGSEGV, "device not available", device_not_available)
714 DO_ERROR( 9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
715 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
716 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
717 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
718 DO_ERROR(18, SIGSEGV, "reserved", reserved)
720 /* Runs on IST stack */
721 asmlinkage void do_stack_segment(struct pt_regs *regs, long error_code)
723 if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
724 12, SIGBUS) == NOTIFY_STOP)
726 preempt_conditional_sti(regs);
727 do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
728 preempt_conditional_cli(regs);
731 asmlinkage void do_double_fault(struct pt_regs * regs, long error_code)
733 static const char str[] = "double fault";
734 struct task_struct *tsk = current;
736 /* Return not checked because double check cannot be ignored */
737 notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
739 tsk->thread.error_code = error_code;
740 tsk->thread.trap_no = 8;
742 /* This is always a kernel trap and never fixable (and thus must
745 die(str, regs, error_code);
748 asmlinkage void __kprobes do_general_protection(struct pt_regs * regs,
751 struct task_struct *tsk = current;
753 conditional_sti(regs);
755 tsk->thread.error_code = error_code;
756 tsk->thread.trap_no = 13;
758 if (user_mode(regs)) {
759 if (exception_trace && unhandled_signal(tsk, SIGSEGV))
761 "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
763 regs->rip, regs->rsp, error_code);
765 force_sig(SIGSEGV, tsk);
771 const struct exception_table_entry *fixup;
772 fixup = search_exception_tables(regs->rip);
774 regs->rip = fixup->fixup;
777 if (notify_die(DIE_GPF, "general protection fault", regs,
778 error_code, 13, SIGSEGV) == NOTIFY_STOP)
780 die("general protection fault", regs, error_code);
784 static __kprobes void
785 mem_parity_error(unsigned char reason, struct pt_regs * regs)
787 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
789 printk(KERN_EMERG "You probably have a hardware problem with your "
792 if (panic_on_unrecovered_nmi)
793 panic("NMI: Not continuing");
795 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
797 /* Clear and disable the memory parity error line. */
798 reason = (reason & 0xf) | 4;
802 static __kprobes void
803 io_check_error(unsigned char reason, struct pt_regs * regs)
805 printk("NMI: IOCK error (debug interrupt?)\n");
806 show_registers(regs);
808 /* Re-enable the IOCK line, wait for a few seconds */
809 reason = (reason & 0xf) | 8;
816 static __kprobes void
817 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
819 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
821 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
823 if (panic_on_unrecovered_nmi)
824 panic("NMI: Not continuing");
826 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
829 /* Runs on IST stack. This code must keep interrupts off all the time.
830 Nested NMIs are prevented by the CPU. */
831 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
833 unsigned char reason = 0;
836 cpu = smp_processor_id();
838 /* Only the BSP gets external NMIs from the system. */
840 reason = get_nmi_reason();
842 if (!(reason & 0xc0)) {
843 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
847 * Ok, so this is none of the documented NMI sources,
848 * so it must be the NMI watchdog.
850 if (nmi_watchdog_tick(regs,reason))
852 if (!do_nmi_callback(regs,cpu))
853 unknown_nmi_error(reason, regs);
857 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
860 /* AK: following checks seem to be broken on modern chipsets. FIXME */
863 mem_parity_error(reason, regs);
865 io_check_error(reason, regs);
868 /* runs on IST stack. */
869 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
871 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
874 preempt_conditional_sti(regs);
875 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
876 preempt_conditional_cli(regs);
879 /* Help handler running on IST stack to switch back to user stack
880 for scheduling or signal handling. The actual stack switch is done in
882 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
884 struct pt_regs *regs = eregs;
885 /* Did already sync */
886 if (eregs == (struct pt_regs *)eregs->rsp)
888 /* Exception from user space */
889 else if (user_mode(eregs))
890 regs = task_pt_regs(current);
891 /* Exception from kernel and interrupts are enabled. Move to
892 kernel process stack. */
893 else if (eregs->eflags & X86_EFLAGS_IF)
894 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
900 /* runs on IST stack. */
901 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
902 unsigned long error_code)
904 unsigned long condition;
905 struct task_struct *tsk = current;
908 get_debugreg(condition, 6);
910 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
911 SIGTRAP) == NOTIFY_STOP)
914 preempt_conditional_sti(regs);
916 /* Mask out spurious debug traps due to lazy DR7 setting */
917 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
918 if (!tsk->thread.debugreg7) {
923 tsk->thread.debugreg6 = condition;
925 /* Mask out spurious TF errors due to lazy TF clearing */
926 if (condition & DR_STEP) {
928 * The TF error should be masked out only if the current
929 * process is not traced and if the TRAP flag has been set
930 * previously by a tracing process (condition detected by
931 * the PT_DTRACE flag); remember that the i386 TRAP flag
932 * can be modified by the process itself in user mode,
933 * allowing programs to debug themselves without the ptrace()
936 if (!user_mode(regs))
937 goto clear_TF_reenable;
939 * Was the TF flag set by a debugger? If so, clear it now,
940 * so that register information is correct.
942 if (tsk->ptrace & PT_DTRACE) {
943 regs->eflags &= ~TF_MASK;
944 tsk->ptrace &= ~PT_DTRACE;
948 /* Ok, finally something we can handle */
949 tsk->thread.trap_no = 1;
950 tsk->thread.error_code = error_code;
951 info.si_signo = SIGTRAP;
953 info.si_code = TRAP_BRKPT;
954 info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
955 force_sig_info(SIGTRAP, &info, tsk);
958 set_debugreg(0UL, 7);
959 preempt_conditional_cli(regs);
963 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
964 regs->eflags &= ~TF_MASK;
965 preempt_conditional_cli(regs);
968 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
970 const struct exception_table_entry *fixup;
971 fixup = search_exception_tables(regs->rip);
973 regs->rip = fixup->fixup;
976 notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
977 /* Illegal floating point operation in the kernel */
978 current->thread.trap_no = trapnr;
984 * Note that we play around with the 'TS' bit in an attempt to get
985 * the correct behaviour even in the presence of the asynchronous
988 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
990 void __user *rip = (void __user *)(regs->rip);
991 struct task_struct * task;
993 unsigned short cwd, swd;
995 conditional_sti(regs);
996 if (!user_mode(regs) &&
997 kernel_math_error(regs, "kernel x87 math error", 16))
1001 * Save the info for the exception handler and clear the error.
1004 save_init_fpu(task);
1005 task->thread.trap_no = 16;
1006 task->thread.error_code = 0;
1007 info.si_signo = SIGFPE;
1009 info.si_code = __SI_FAULT;
1012 * (~cwd & swd) will mask out exceptions that are not set to unmasked
1013 * status. 0x3f is the exception bits in these regs, 0x200 is the
1014 * C1 reg you need in case of a stack fault, 0x040 is the stack
1015 * fault bit. We should only be taking one exception at a time,
1016 * so if this combination doesn't produce any single exception,
1017 * then we have a bad program that isn't synchronizing its FPU usage
1018 * and it will suffer the consequences since we won't be able to
1019 * fully reproduce the context of the exception
1021 cwd = get_fpu_cwd(task);
1022 swd = get_fpu_swd(task);
1023 switch (swd & ~cwd & 0x3f) {
1027 case 0x001: /* Invalid Op */
1029 * swd & 0x240 == 0x040: Stack Underflow
1030 * swd & 0x240 == 0x240: Stack Overflow
1031 * User must clear the SF bit (0x40) if set
1033 info.si_code = FPE_FLTINV;
1035 case 0x002: /* Denormalize */
1036 case 0x010: /* Underflow */
1037 info.si_code = FPE_FLTUND;
1039 case 0x004: /* Zero Divide */
1040 info.si_code = FPE_FLTDIV;
1042 case 0x008: /* Overflow */
1043 info.si_code = FPE_FLTOVF;
1045 case 0x020: /* Precision */
1046 info.si_code = FPE_FLTRES;
1049 force_sig_info(SIGFPE, &info, task);
1052 asmlinkage void bad_intr(void)
1054 printk("bad interrupt");
1057 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1059 void __user *rip = (void __user *)(regs->rip);
1060 struct task_struct * task;
1062 unsigned short mxcsr;
1064 conditional_sti(regs);
1065 if (!user_mode(regs) &&
1066 kernel_math_error(regs, "kernel simd math error", 19))
1070 * Save the info for the exception handler and clear the error.
1073 save_init_fpu(task);
1074 task->thread.trap_no = 19;
1075 task->thread.error_code = 0;
1076 info.si_signo = SIGFPE;
1078 info.si_code = __SI_FAULT;
1081 * The SIMD FPU exceptions are handled a little differently, as there
1082 * is only a single status/control register. Thus, to determine which
1083 * unmasked exception was caught we must mask the exception mask bits
1084 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1086 mxcsr = get_fpu_mxcsr(task);
1087 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1091 case 0x001: /* Invalid Op */
1092 info.si_code = FPE_FLTINV;
1094 case 0x002: /* Denormalize */
1095 case 0x010: /* Underflow */
1096 info.si_code = FPE_FLTUND;
1098 case 0x004: /* Zero Divide */
1099 info.si_code = FPE_FLTDIV;
1101 case 0x008: /* Overflow */
1102 info.si_code = FPE_FLTOVF;
1104 case 0x020: /* Precision */
1105 info.si_code = FPE_FLTRES;
1108 force_sig_info(SIGFPE, &info, task);
1111 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1115 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1119 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1124 * 'math_state_restore()' saves the current math information in the
1125 * old math state array, and gets the new ones from the current task
1127 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1128 * Don't touch unless you *really* know how it works.
1130 asmlinkage void math_state_restore(void)
1132 struct task_struct *me = current;
1133 clts(); /* Allow maths ops (or we recurse) */
1137 restore_fpu_checking(&me->thread.i387.fxsave);
1138 task_thread_info(me)->status |= TS_USEDFPU;
1142 void __init trap_init(void)
1144 set_intr_gate(0,÷_error);
1145 set_intr_gate_ist(1,&debug,DEBUG_STACK);
1146 set_intr_gate_ist(2,&nmi,NMI_STACK);
1147 set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
1148 set_system_gate(4,&overflow); /* int4 can be called from all */
1149 set_intr_gate(5,&bounds);
1150 set_intr_gate(6,&invalid_op);
1151 set_intr_gate(7,&device_not_available);
1152 set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
1153 set_intr_gate(9,&coprocessor_segment_overrun);
1154 set_intr_gate(10,&invalid_TSS);
1155 set_intr_gate(11,&segment_not_present);
1156 set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
1157 set_intr_gate(13,&general_protection);
1158 set_intr_gate(14,&page_fault);
1159 set_intr_gate(15,&spurious_interrupt_bug);
1160 set_intr_gate(16,&coprocessor_error);
1161 set_intr_gate(17,&alignment_check);
1162 #ifdef CONFIG_X86_MCE
1163 set_intr_gate_ist(18,&machine_check, MCE_STACK);
1165 set_intr_gate(19,&simd_coprocessor_error);
1167 #ifdef CONFIG_IA32_EMULATION
1168 set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1172 * Should be a barrier for any external CPU state.
1178 static int __init oops_setup(char *s)
1182 if (!strcmp(s, "panic"))
1186 early_param("oops", oops_setup);
1188 static int __init kstack_setup(char *s)
1192 kstack_depth_to_print = simple_strtoul(s,NULL,0);
1195 early_param("kstack", kstack_setup);
1197 #ifdef CONFIG_STACK_UNWIND
1198 static int __init call_trace_setup(char *s)
1202 if (strcmp(s, "old") == 0)
1204 else if (strcmp(s, "both") == 0)
1206 else if (strcmp(s, "newfallback") == 0)
1208 else if (strcmp(s, "new") == 0)
1212 early_param("call_trace", call_trace_setup);