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>
33 #include <linux/uaccess.h>
35 #include <asm/system.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 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;
228 unsigned long sp = UNW_SP(info);
230 if (arch_unw_user_mode(info))
232 while (unwind(info) == 0 && UNW_PC(info)) {
234 oad->ops->address(oad->data, UNW_PC(info));
235 if (arch_unw_user_mode(info))
237 if ((sp & ~(PAGE_SIZE - 1)) == (UNW_SP(info) & ~(PAGE_SIZE - 1))
238 && sp > UNW_SP(info))
245 #define MSG(txt) ops->warning(data, txt)
248 * x86-64 can have upto three kernel stacks:
251 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
254 static inline int valid_stack_ptr(struct thread_info *tinfo, void *p)
256 void *t = (void *)tinfo;
257 return p > t && p < t + THREAD_SIZE - 3;
260 void dump_trace(struct task_struct *tsk, struct pt_regs *regs,
261 unsigned long *stack,
262 struct stacktrace_ops *ops, void *data)
264 const unsigned cpu = get_cpu();
265 unsigned long *irqstack_end = (unsigned long*)cpu_pda(cpu)->irqstackptr;
267 struct thread_info *tinfo;
272 if (call_trace >= 0) {
274 struct unwind_frame_info info;
275 struct ops_and_data oad = { .ops = ops, .data = data };
278 if (unwind_init_frame_info(&info, tsk, regs) == 0)
279 unw_ret = dump_trace_unwind(&info, &oad);
280 } else if (tsk == current)
281 unw_ret = unwind_init_running(&info, dump_trace_unwind,
284 if (unwind_init_blocked(&info, tsk) == 0)
285 unw_ret = dump_trace_unwind(&info, &oad);
288 if (call_trace == 1 && !arch_unw_user_mode(&info)) {
289 ops->warning_symbol(data,
290 "DWARF2 unwinder stuck at %s",
292 if ((long)UNW_SP(&info) < 0) {
293 MSG("Leftover inexact backtrace:");
294 stack = (unsigned long *)UNW_SP(&info);
298 MSG("Full inexact backtrace again:");
299 } else if (call_trace >= 1)
302 MSG("Full inexact backtrace again:");
304 MSG("Inexact backtrace:");
309 if (tsk && tsk != current)
310 stack = (unsigned long *)tsk->thread.rsp;
314 * Print function call entries within a stack. 'cond' is the
315 * "end of stackframe" condition, that the 'stack++'
316 * iteration will eventually trigger.
318 #define HANDLE_STACK(cond) \
320 unsigned long addr = *stack++; \
321 /* Use unlocked access here because except for NMIs \
322 we should be already protected against module unloads */ \
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 tinfo = current_thread_info();
387 HANDLE_STACK (valid_stack_ptr(tinfo, stack));
392 EXPORT_SYMBOL(dump_trace);
395 print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
397 print_symbol(msg, symbol);
401 static void print_trace_warning(void *data, char *msg)
406 static int print_trace_stack(void *data, char *name)
408 printk(" <%s> ", name);
412 static void print_trace_address(void *data, unsigned long addr)
414 printk_address(addr);
417 static struct stacktrace_ops print_trace_ops = {
418 .warning = print_trace_warning,
419 .warning_symbol = print_trace_warning_symbol,
420 .stack = print_trace_stack,
421 .address = print_trace_address,
425 show_trace(struct task_struct *tsk, struct pt_regs *regs, unsigned long *stack)
427 printk("\nCall Trace:\n");
428 dump_trace(tsk, regs, stack, &print_trace_ops, NULL);
433 _show_stack(struct task_struct *tsk, struct pt_regs *regs, unsigned long *rsp)
435 unsigned long *stack;
437 const int cpu = smp_processor_id();
438 unsigned long *irqstack_end = (unsigned long *) (cpu_pda(cpu)->irqstackptr);
439 unsigned long *irqstack = (unsigned long *) (cpu_pda(cpu)->irqstackptr - IRQSTACKSIZE);
441 // debugging aid: "show_stack(NULL, NULL);" prints the
442 // back trace for this cpu.
446 rsp = (unsigned long *)tsk->thread.rsp;
448 rsp = (unsigned long *)&rsp;
452 for(i=0; i < kstack_depth_to_print; i++) {
453 if (stack >= irqstack && stack <= irqstack_end) {
454 if (stack == irqstack_end) {
455 stack = (unsigned long *) (irqstack_end[-1]);
459 if (((long) stack & (THREAD_SIZE-1)) == 0)
462 if (i && ((i % 4) == 0))
464 printk(" %016lx", *stack++);
465 touch_nmi_watchdog();
467 show_trace(tsk, regs, rsp);
470 void show_stack(struct task_struct *tsk, unsigned long * rsp)
472 _show_stack(tsk, NULL, rsp);
476 * The architecture-independent dump_stack generator
478 void dump_stack(void)
481 show_trace(NULL, NULL, &dummy);
484 EXPORT_SYMBOL(dump_stack);
486 void show_registers(struct pt_regs *regs)
489 int in_kernel = !user_mode(regs);
491 const int cpu = smp_processor_id();
492 struct task_struct *cur = cpu_pda(cpu)->pcurrent;
496 printk("CPU %d ", cpu);
498 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
499 cur->comm, cur->pid, task_thread_info(cur), cur);
502 * When in-kernel, we also print out the stack and code at the
503 * time of the fault..
508 _show_stack(NULL, regs, (unsigned long*)rsp);
511 if (regs->rip < PAGE_OFFSET)
514 for (i=0; i<20; i++) {
516 if (__get_user(c, &((unsigned char*)regs->rip)[i])) {
518 printk(" Bad RIP value.");
527 void handle_BUG(struct pt_regs *regs)
531 const char *prefix = "";
535 if (__copy_from_user(&f, (const void __user *) regs->rip,
536 sizeof(struct bug_frame)))
538 if (f.filename >= 0 ||
539 f.ud2[0] != 0x0f || f.ud2[1] != 0x0b)
541 len = __strnlen_user((char *)(long)f.filename, PATH_MAX) - 1;
542 if (len < 0 || len >= PATH_MAX)
543 f.filename = (int)(long)"unmapped filename";
545 f.filename += len - 50;
548 printk("----------- [cut here ] --------- [please bite here ] ---------\n");
549 printk(KERN_ALERT "Kernel BUG at %s%.50s:%d\n", prefix, (char *)(long)f.filename, f.line);
553 void out_of_line_bug(void)
557 EXPORT_SYMBOL(out_of_line_bug);
560 static DEFINE_SPINLOCK(die_lock);
561 static int die_owner = -1;
562 static unsigned int die_nest_count;
564 unsigned __kprobes long oops_begin(void)
566 int cpu = smp_processor_id();
571 /* racy, but better than risking deadlock. */
572 local_irq_save(flags);
573 if (!spin_trylock(&die_lock)) {
574 if (cpu == die_owner)
575 /* nested oops. should stop eventually */;
577 spin_lock(&die_lock);
586 void __kprobes oops_end(unsigned long flags)
592 /* We still own the lock */
593 local_irq_restore(flags);
595 /* Nest count reaches zero, release the lock. */
596 spin_unlock_irqrestore(&die_lock, flags);
598 panic("Fatal exception");
602 void __kprobes __die(const char * str, struct pt_regs * regs, long err)
604 static int die_counter;
605 printk(KERN_EMERG "%s: %04lx [%u] ", str, err & 0xffff,++die_counter);
606 #ifdef CONFIG_PREEMPT
612 #ifdef CONFIG_DEBUG_PAGEALLOC
613 printk("DEBUG_PAGEALLOC");
616 notify_die(DIE_OOPS, str, regs, err, current->thread.trap_no, SIGSEGV);
617 show_registers(regs);
618 /* Executive summary in case the oops scrolled away */
619 printk(KERN_ALERT "RIP ");
620 printk_address(regs->rip);
621 printk(" RSP <%016lx>\n", regs->rsp);
622 if (kexec_should_crash(current))
626 void die(const char * str, struct pt_regs * regs, long err)
628 unsigned long flags = oops_begin();
631 __die(str, regs, err);
636 void __kprobes die_nmi(char *str, struct pt_regs *regs, int do_panic)
638 unsigned long flags = oops_begin();
641 * We are in trouble anyway, lets at least try
642 * to get a message out.
644 printk(str, smp_processor_id());
645 show_registers(regs);
646 if (kexec_should_crash(current))
648 if (do_panic || panic_on_oops)
649 panic("Non maskable interrupt");
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(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
797 printk(KERN_EMERG "You have some hardware problem, likely on the PCI bus.\n");
799 if (panic_on_unrecovered_nmi)
800 panic("NMI: Not continuing");
802 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
804 /* Clear and disable the memory parity error line. */
805 reason = (reason & 0xf) | 4;
809 static __kprobes void
810 io_check_error(unsigned char reason, struct pt_regs * regs)
812 printk("NMI: IOCK error (debug interrupt?)\n");
813 show_registers(regs);
815 /* Re-enable the IOCK line, wait for a few seconds */
816 reason = (reason & 0xf) | 8;
823 static __kprobes void
824 unknown_nmi_error(unsigned char reason, struct pt_regs * regs)
826 printk(KERN_EMERG "Uhhuh. NMI received for unknown reason %02x.\n",
828 printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
830 if (panic_on_unrecovered_nmi)
831 panic("NMI: Not continuing");
833 printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
836 /* Runs on IST stack. This code must keep interrupts off all the time.
837 Nested NMIs are prevented by the CPU. */
838 asmlinkage __kprobes void default_do_nmi(struct pt_regs *regs)
840 unsigned char reason = 0;
843 cpu = smp_processor_id();
845 /* Only the BSP gets external NMIs from the system. */
847 reason = get_nmi_reason();
849 if (!(reason & 0xc0)) {
850 if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
854 * Ok, so this is none of the documented NMI sources,
855 * so it must be the NMI watchdog.
857 if (nmi_watchdog_tick(regs,reason))
859 if (!do_nmi_callback(regs,cpu))
860 unknown_nmi_error(reason, regs);
864 if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
867 /* AK: following checks seem to be broken on modern chipsets. FIXME */
870 mem_parity_error(reason, regs);
872 io_check_error(reason, regs);
875 /* runs on IST stack. */
876 asmlinkage void __kprobes do_int3(struct pt_regs * regs, long error_code)
878 if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP) == NOTIFY_STOP) {
881 preempt_conditional_sti(regs);
882 do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
883 preempt_conditional_cli(regs);
886 /* Help handler running on IST stack to switch back to user stack
887 for scheduling or signal handling. The actual stack switch is done in
889 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
891 struct pt_regs *regs = eregs;
892 /* Did already sync */
893 if (eregs == (struct pt_regs *)eregs->rsp)
895 /* Exception from user space */
896 else if (user_mode(eregs))
897 regs = task_pt_regs(current);
898 /* Exception from kernel and interrupts are enabled. Move to
899 kernel process stack. */
900 else if (eregs->eflags & X86_EFLAGS_IF)
901 regs = (struct pt_regs *)(eregs->rsp -= sizeof(struct pt_regs));
907 /* runs on IST stack. */
908 asmlinkage void __kprobes do_debug(struct pt_regs * regs,
909 unsigned long error_code)
911 unsigned long condition;
912 struct task_struct *tsk = current;
915 get_debugreg(condition, 6);
917 if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
918 SIGTRAP) == NOTIFY_STOP)
921 preempt_conditional_sti(regs);
923 /* Mask out spurious debug traps due to lazy DR7 setting */
924 if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
925 if (!tsk->thread.debugreg7) {
930 tsk->thread.debugreg6 = condition;
932 /* Mask out spurious TF errors due to lazy TF clearing */
933 if (condition & DR_STEP) {
935 * The TF error should be masked out only if the current
936 * process is not traced and if the TRAP flag has been set
937 * previously by a tracing process (condition detected by
938 * the PT_DTRACE flag); remember that the i386 TRAP flag
939 * can be modified by the process itself in user mode,
940 * allowing programs to debug themselves without the ptrace()
943 if (!user_mode(regs))
944 goto clear_TF_reenable;
946 * Was the TF flag set by a debugger? If so, clear it now,
947 * so that register information is correct.
949 if (tsk->ptrace & PT_DTRACE) {
950 regs->eflags &= ~TF_MASK;
951 tsk->ptrace &= ~PT_DTRACE;
955 /* Ok, finally something we can handle */
956 tsk->thread.trap_no = 1;
957 tsk->thread.error_code = error_code;
958 info.si_signo = SIGTRAP;
960 info.si_code = TRAP_BRKPT;
961 info.si_addr = user_mode(regs) ? (void __user *)regs->rip : NULL;
962 force_sig_info(SIGTRAP, &info, tsk);
965 set_debugreg(0UL, 7);
966 preempt_conditional_cli(regs);
970 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
971 regs->eflags &= ~TF_MASK;
972 preempt_conditional_cli(regs);
975 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
977 const struct exception_table_entry *fixup;
978 fixup = search_exception_tables(regs->rip);
980 regs->rip = fixup->fixup;
983 notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
984 /* Illegal floating point operation in the kernel */
985 current->thread.trap_no = trapnr;
991 * Note that we play around with the 'TS' bit in an attempt to get
992 * the correct behaviour even in the presence of the asynchronous
995 asmlinkage void do_coprocessor_error(struct pt_regs *regs)
997 void __user *rip = (void __user *)(regs->rip);
998 struct task_struct * task;
1000 unsigned short cwd, swd;
1002 conditional_sti(regs);
1003 if (!user_mode(regs) &&
1004 kernel_math_error(regs, "kernel x87 math error", 16))
1008 * Save the info for the exception handler and clear the error.
1011 save_init_fpu(task);
1012 task->thread.trap_no = 16;
1013 task->thread.error_code = 0;
1014 info.si_signo = SIGFPE;
1016 info.si_code = __SI_FAULT;
1019 * (~cwd & swd) will mask out exceptions that are not set to unmasked
1020 * status. 0x3f is the exception bits in these regs, 0x200 is the
1021 * C1 reg you need in case of a stack fault, 0x040 is the stack
1022 * fault bit. We should only be taking one exception at a time,
1023 * so if this combination doesn't produce any single exception,
1024 * then we have a bad program that isn't synchronizing its FPU usage
1025 * and it will suffer the consequences since we won't be able to
1026 * fully reproduce the context of the exception
1028 cwd = get_fpu_cwd(task);
1029 swd = get_fpu_swd(task);
1030 switch (swd & ~cwd & 0x3f) {
1034 case 0x001: /* Invalid Op */
1036 * swd & 0x240 == 0x040: Stack Underflow
1037 * swd & 0x240 == 0x240: Stack Overflow
1038 * User must clear the SF bit (0x40) if set
1040 info.si_code = FPE_FLTINV;
1042 case 0x002: /* Denormalize */
1043 case 0x010: /* Underflow */
1044 info.si_code = FPE_FLTUND;
1046 case 0x004: /* Zero Divide */
1047 info.si_code = FPE_FLTDIV;
1049 case 0x008: /* Overflow */
1050 info.si_code = FPE_FLTOVF;
1052 case 0x020: /* Precision */
1053 info.si_code = FPE_FLTRES;
1056 force_sig_info(SIGFPE, &info, task);
1059 asmlinkage void bad_intr(void)
1061 printk("bad interrupt");
1064 asmlinkage void do_simd_coprocessor_error(struct pt_regs *regs)
1066 void __user *rip = (void __user *)(regs->rip);
1067 struct task_struct * task;
1069 unsigned short mxcsr;
1071 conditional_sti(regs);
1072 if (!user_mode(regs) &&
1073 kernel_math_error(regs, "kernel simd math error", 19))
1077 * Save the info for the exception handler and clear the error.
1080 save_init_fpu(task);
1081 task->thread.trap_no = 19;
1082 task->thread.error_code = 0;
1083 info.si_signo = SIGFPE;
1085 info.si_code = __SI_FAULT;
1088 * The SIMD FPU exceptions are handled a little differently, as there
1089 * is only a single status/control register. Thus, to determine which
1090 * unmasked exception was caught we must mask the exception mask bits
1091 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1093 mxcsr = get_fpu_mxcsr(task);
1094 switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
1098 case 0x001: /* Invalid Op */
1099 info.si_code = FPE_FLTINV;
1101 case 0x002: /* Denormalize */
1102 case 0x010: /* Underflow */
1103 info.si_code = FPE_FLTUND;
1105 case 0x004: /* Zero Divide */
1106 info.si_code = FPE_FLTDIV;
1108 case 0x008: /* Overflow */
1109 info.si_code = FPE_FLTOVF;
1111 case 0x020: /* Precision */
1112 info.si_code = FPE_FLTRES;
1115 force_sig_info(SIGFPE, &info, task);
1118 asmlinkage void do_spurious_interrupt_bug(struct pt_regs * regs)
1122 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
1126 asmlinkage void __attribute__((weak)) mce_threshold_interrupt(void)
1131 * 'math_state_restore()' saves the current math information in the
1132 * old math state array, and gets the new ones from the current task
1134 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1135 * Don't touch unless you *really* know how it works.
1137 asmlinkage void math_state_restore(void)
1139 struct task_struct *me = current;
1140 clts(); /* Allow maths ops (or we recurse) */
1144 restore_fpu_checking(&me->thread.i387.fxsave);
1145 task_thread_info(me)->status |= TS_USEDFPU;
1149 void __init trap_init(void)
1151 set_intr_gate(0,÷_error);
1152 set_intr_gate_ist(1,&debug,DEBUG_STACK);
1153 set_intr_gate_ist(2,&nmi,NMI_STACK);
1154 set_system_gate_ist(3,&int3,DEBUG_STACK); /* int3 can be called from all */
1155 set_system_gate(4,&overflow); /* int4 can be called from all */
1156 set_intr_gate(5,&bounds);
1157 set_intr_gate(6,&invalid_op);
1158 set_intr_gate(7,&device_not_available);
1159 set_intr_gate_ist(8,&double_fault, DOUBLEFAULT_STACK);
1160 set_intr_gate(9,&coprocessor_segment_overrun);
1161 set_intr_gate(10,&invalid_TSS);
1162 set_intr_gate(11,&segment_not_present);
1163 set_intr_gate_ist(12,&stack_segment,STACKFAULT_STACK);
1164 set_intr_gate(13,&general_protection);
1165 set_intr_gate(14,&page_fault);
1166 set_intr_gate(15,&spurious_interrupt_bug);
1167 set_intr_gate(16,&coprocessor_error);
1168 set_intr_gate(17,&alignment_check);
1169 #ifdef CONFIG_X86_MCE
1170 set_intr_gate_ist(18,&machine_check, MCE_STACK);
1172 set_intr_gate(19,&simd_coprocessor_error);
1174 #ifdef CONFIG_IA32_EMULATION
1175 set_system_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
1179 * Should be a barrier for any external CPU state.
1185 static int __init oops_setup(char *s)
1189 if (!strcmp(s, "panic"))
1193 early_param("oops", oops_setup);
1195 static int __init kstack_setup(char *s)
1199 kstack_depth_to_print = simple_strtoul(s,NULL,0);
1202 early_param("kstack", kstack_setup);
1204 #ifdef CONFIG_STACK_UNWIND
1205 static int __init call_trace_setup(char *s)
1209 if (strcmp(s, "old") == 0)
1211 else if (strcmp(s, "both") == 0)
1213 else if (strcmp(s, "newfallback") == 0)
1215 else if (strcmp(s, "new") == 0)
1219 early_param("call_trace", call_trace_setup);