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
);
73 EXPORT_SYMBOL(die_chain
);
75 int register_die_notifier(struct notifier_block
*nb
)
78 return atomic_notifier_chain_register(&die_chain
, nb
);
80 EXPORT_SYMBOL(register_die_notifier
); /* used modular by kdb */
82 int unregister_die_notifier(struct notifier_block
*nb
)
84 return atomic_notifier_chain_unregister(&die_chain
, nb
);
86 EXPORT_SYMBOL(unregister_die_notifier
); /* used modular by kdb */
88 static inline void conditional_sti(struct pt_regs
*regs
)
90 if (regs
->eflags
& X86_EFLAGS_IF
)
94 static inline void preempt_conditional_sti(struct pt_regs
*regs
)
97 if (regs
->eflags
& X86_EFLAGS_IF
)
101 static inline void preempt_conditional_cli(struct pt_regs
*regs
)
103 if (regs
->eflags
& X86_EFLAGS_IF
)
105 /* Make sure to not schedule here because we could be running
106 on an exception stack. */
107 preempt_enable_no_resched();
110 static int kstack_depth_to_print
= 12;
111 #ifdef CONFIG_STACK_UNWIND
112 static int call_trace
= 1;
114 #define call_trace (-1)
117 #ifdef CONFIG_KALLSYMS
118 # include <linux/kallsyms.h>
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
++) {
168 * set 'end' to the end of the exception stack.
172 * TODO: this block is not needed i think, because
173 * setup64.c:cpu_init() sets up t->ist[DEBUG_STACK]
176 #if DEBUG_STKSZ > EXCEPTION_STKSZ
178 end
= cpu_pda(cpu
)->debugstack
+ DEBUG_STKSZ
;
182 end
= per_cpu(orig_ist
, cpu
).ist
[k
];
186 * Is 'stack' above this exception frame's end?
187 * If yes then skip to the next frame.
192 * Is 'stack' above this exception frame's start address?
193 * If yes then we found the right frame.
195 if (stack
>= end
- EXCEPTION_STKSZ
) {
197 * Make sure we only iterate through an exception
198 * stack once. If it comes up for the second time
199 * then there's something wrong going on - just
200 * break out and return NULL:
202 if (*usedp
& (1U << k
))
206 return (unsigned long *)end
;
209 * If this is a debug stack, and if it has a larger size than
210 * the usual exception stacks, then 'stack' might still
211 * be within the lower portion of the debug stack:
213 #if DEBUG_STKSZ > EXCEPTION_STKSZ
214 if (k
== DEBUG_STACK
- 1 && stack
>= end
- DEBUG_STKSZ
) {
215 unsigned j
= N_EXCEPTION_STACKS
- 1;
218 * Black magic. A large debug stack is composed of
219 * multiple exception stack entries, which we
220 * iterate through now. Dont look:
224 end
-= EXCEPTION_STKSZ
;
225 ids
[j
][4] = '1' + (j
- N_EXCEPTION_STACKS
);
226 } while (stack
< end
- EXCEPTION_STKSZ
);
227 if (*usedp
& (1U << j
))
231 return (unsigned long *)end
;
238 struct ops_and_data
{
239 struct stacktrace_ops
*ops
;
243 static int dump_trace_unwind(struct unwind_frame_info
*info
, void *context
)
245 struct ops_and_data
*oad
= (struct ops_and_data
*)context
;
248 while (unwind(info
) == 0 && UNW_PC(info
)) {
250 oad
->ops
->address(oad
->data
, UNW_PC(info
));
251 if (arch_unw_user_mode(info
))
258 * x86-64 can have upto three kernel stacks:
261 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
264 void dump_trace(struct task_struct
*tsk
, struct pt_regs
*regs
, unsigned long * stack
,
265 struct stacktrace_ops
*ops
, void *data
)
267 const unsigned cpu
= safe_smp_processor_id();
268 unsigned long *irqstack_end
= (unsigned long *)cpu_pda(cpu
)->irqstackptr
;
274 if (call_trace
>= 0) {
276 struct unwind_frame_info info
;
277 struct ops_and_data oad
= { .ops
= ops
, .data
= data
};
280 if (unwind_init_frame_info(&info
, tsk
, regs
) == 0)
281 unw_ret
= dump_trace_unwind(&info
, &oad
);
282 } else if (tsk
== current
)
283 unw_ret
= unwind_init_running(&info
, dump_trace_unwind
, &oad
);
285 if (unwind_init_blocked(&info
, tsk
) == 0)
286 unw_ret
= dump_trace_unwind(&info
, &oad
);
289 if (call_trace
== 1 && !arch_unw_user_mode(&info
)) {
290 ops
->warning_symbol(data
, "DWARF2 unwinder stuck at %s\n",
292 if ((long)UNW_SP(&info
) < 0) {
293 ops
->warning(data
, "Leftover inexact backtrace:\n");
294 stack
= (unsigned long *)UNW_SP(&info
);
296 ops
->warning(data
, "Full inexact backtrace again:\n");
297 } else if (call_trace
>= 1)
300 ops
->warning(data
, "Full inexact backtrace again:\n");
302 ops
->warning(data
, "Inexact backtrace:\n");
307 if (tsk
&& tsk
!= current
)
308 stack
= (unsigned long *)tsk
->thread
.rsp
;
312 * Print function call entries within a stack. 'cond' is the
313 * "end of stackframe" condition, that the 'stack++'
314 * iteration will eventually trigger.
316 #define HANDLE_STACK(cond) \
318 unsigned long addr = *stack++; \
319 if (kernel_text_address(addr)) { \
321 * If the address is either in the text segment of the \
322 * kernel, or in the region which contains vmalloc'ed \
323 * memory, it *may* be the address of a calling \
324 * routine; if so, print it so that someone tracing \
325 * down the cause of the crash will be able to figure \
326 * out the call path that was taken. \
328 ops->address(data, addr); \
333 * Print function call entries in all stacks, starting at the
334 * current stack address. If the stacks consist of nested
339 unsigned long *estack_end
;
340 estack_end
= in_exception_stack(cpu
, (unsigned long)stack
,
344 if (ops
->stack(data
, id
) < 0)
346 HANDLE_STACK (stack
< estack_end
);
347 ops
->stack(data
, "<EOE>");
349 * We link to the next stack via the
350 * second-to-last pointer (index -2 to end) in the
353 stack
= (unsigned long *) estack_end
[-2];
357 unsigned long *irqstack
;
358 irqstack
= irqstack_end
-
359 (IRQSTACKSIZE
- 64) / sizeof(*irqstack
);
361 if (stack
>= irqstack
&& stack
< irqstack_end
) {
362 if (ops
->stack(data
, "IRQ") < 0)
364 HANDLE_STACK (stack
< irqstack_end
);
366 * We link to the next stack (which would be
367 * the process stack normally) the last
368 * pointer (index -1 to end) in the IRQ stack:
370 stack
= (unsigned long *) (irqstack_end
[-1]);
372 ops
->stack(data
, "EOI");
380 * This handles the process stack:
382 HANDLE_STACK (((long) stack
& (THREAD_SIZE
-1)) != 0);
385 EXPORT_SYMBOL(dump_trace
);
388 print_trace_warning_symbol(void *data
, char *msg
, unsigned long symbol
)
390 print_symbol(msg
, symbol
);
394 static void print_trace_warning(void *data
, char *msg
)
399 static int print_trace_stack(void *data
, char *name
)
401 printk(" <%s> ", name
);
405 static void print_trace_address(void *data
, unsigned long addr
)
407 printk_address(addr
);
410 static struct stacktrace_ops print_trace_ops
= {
411 .warning
= print_trace_warning
,
412 .warning_symbol
= print_trace_warning_symbol
,
413 .stack
= print_trace_stack
,
414 .address
= print_trace_address
,
418 show_trace(struct task_struct
*tsk
, struct pt_regs
*regs
, unsigned long *stack
)
420 printk("\nCall Trace:\n");
421 dump_trace(tsk
, regs
, stack
, &print_trace_ops
, NULL
);
426 _show_stack(struct task_struct
*tsk
, struct pt_regs
*regs
, unsigned long *rsp
)
428 unsigned long *stack
;
430 const int cpu
= safe_smp_processor_id();
431 unsigned long *irqstack_end
= (unsigned long *) (cpu_pda(cpu
)->irqstackptr
);
432 unsigned long *irqstack
= (unsigned long *) (cpu_pda(cpu
)->irqstackptr
- IRQSTACKSIZE
);
434 // debugging aid: "show_stack(NULL, NULL);" prints the
435 // back trace for this cpu.
439 rsp
= (unsigned long *)tsk
->thread
.rsp
;
441 rsp
= (unsigned long *)&rsp
;
445 for(i
=0; i
< kstack_depth_to_print
; i
++) {
446 if (stack
>= irqstack
&& stack
<= irqstack_end
) {
447 if (stack
== irqstack_end
) {
448 stack
= (unsigned long *) (irqstack_end
[-1]);
452 if (((long) stack
& (THREAD_SIZE
-1)) == 0)
455 if (i
&& ((i
% 4) == 0))
457 printk(" %016lx", *stack
++);
458 touch_nmi_watchdog();
460 show_trace(tsk
, regs
, rsp
);
463 void show_stack(struct task_struct
*tsk
, unsigned long * rsp
)
465 _show_stack(tsk
, NULL
, rsp
);
469 * The architecture-independent dump_stack generator
471 void dump_stack(void)
474 show_trace(NULL
, NULL
, &dummy
);
477 EXPORT_SYMBOL(dump_stack
);
479 void show_registers(struct pt_regs
*regs
)
482 int in_kernel
= !user_mode(regs
);
484 const int cpu
= safe_smp_processor_id();
485 struct task_struct
*cur
= cpu_pda(cpu
)->pcurrent
;
489 printk("CPU %d ", cpu
);
491 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
492 cur
->comm
, cur
->pid
, task_thread_info(cur
), cur
);
495 * When in-kernel, we also print out the stack and code at the
496 * time of the fault..
501 _show_stack(NULL
, regs
, (unsigned long*)rsp
);
504 if (regs
->rip
< PAGE_OFFSET
)
507 for (i
=0; i
<20; i
++) {
509 if (__get_user(c
, &((unsigned char*)regs
->rip
)[i
])) {
511 printk(" Bad RIP value.");
520 void handle_BUG(struct pt_regs
*regs
)
524 const char *prefix
= "";
528 if (__copy_from_user(&f
, (const void __user
*) regs
->rip
,
529 sizeof(struct bug_frame
)))
531 if (f
.filename
>= 0 ||
532 f
.ud2
[0] != 0x0f || f
.ud2
[1] != 0x0b)
534 len
= __strnlen_user((char *)(long)f
.filename
, PATH_MAX
) - 1;
535 if (len
< 0 || len
>= PATH_MAX
)
536 f
.filename
= (int)(long)"unmapped filename";
538 f
.filename
+= len
- 50;
541 printk("----------- [cut here ] --------- [please bite here ] ---------\n");
542 printk(KERN_ALERT
"Kernel BUG at %s%.50s:%d\n", prefix
, (char *)(long)f
.filename
, f
.line
);
546 void out_of_line_bug(void)
550 EXPORT_SYMBOL(out_of_line_bug
);
553 static DEFINE_SPINLOCK(die_lock
);
554 static int die_owner
= -1;
555 static unsigned int die_nest_count
;
557 unsigned __kprobes
long oops_begin(void)
559 int cpu
= safe_smp_processor_id();
562 /* racy, but better than risking deadlock. */
563 local_irq_save(flags
);
564 if (!spin_trylock(&die_lock
)) {
565 if (cpu
== die_owner
)
566 /* nested oops. should stop eventually */;
568 spin_lock(&die_lock
);
577 void __kprobes
oops_end(unsigned long flags
)
583 /* We still own the lock */
584 local_irq_restore(flags
);
586 /* Nest count reaches zero, release the lock. */
587 spin_unlock_irqrestore(&die_lock
, flags
);
589 panic("Fatal exception");
592 void __kprobes
__die(const char * str
, struct pt_regs
* regs
, long err
)
594 static int die_counter
;
595 printk(KERN_EMERG
"%s: %04lx [%u] ", str
, err
& 0xffff,++die_counter
);
596 #ifdef CONFIG_PREEMPT
602 #ifdef CONFIG_DEBUG_PAGEALLOC
603 printk("DEBUG_PAGEALLOC");
606 notify_die(DIE_OOPS
, str
, regs
, err
, current
->thread
.trap_no
, SIGSEGV
);
607 show_registers(regs
);
608 /* Executive summary in case the oops scrolled away */
609 printk(KERN_ALERT
"RIP ");
610 printk_address(regs
->rip
);
611 printk(" RSP <%016lx>\n", regs
->rsp
);
612 if (kexec_should_crash(current
))
616 void die(const char * str
, struct pt_regs
* regs
, long err
)
618 unsigned long flags
= oops_begin();
621 __die(str
, regs
, err
);
626 void __kprobes
die_nmi(char *str
, struct pt_regs
*regs
, int do_panic
)
628 unsigned long flags
= oops_begin();
631 * We are in trouble anyway, lets at least try
632 * to get a message out.
634 printk(str
, safe_smp_processor_id());
635 show_registers(regs
);
636 if (kexec_should_crash(current
))
638 if (do_panic
|| panic_on_oops
)
639 panic("Non maskable interrupt");
646 static void __kprobes
do_trap(int trapnr
, int signr
, char *str
,
647 struct pt_regs
* regs
, long error_code
,
650 struct task_struct
*tsk
= current
;
652 tsk
->thread
.error_code
= error_code
;
653 tsk
->thread
.trap_no
= trapnr
;
655 if (user_mode(regs
)) {
656 if (exception_trace
&& unhandled_signal(tsk
, signr
))
658 "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
659 tsk
->comm
, tsk
->pid
, str
,
660 regs
->rip
, regs
->rsp
, error_code
);
663 force_sig_info(signr
, info
, tsk
);
665 force_sig(signr
, tsk
);
672 const struct exception_table_entry
*fixup
;
673 fixup
= search_exception_tables(regs
->rip
);
675 regs
->rip
= fixup
->fixup
;
677 die(str
, regs
, error_code
);
682 #define DO_ERROR(trapnr, signr, str, name) \
683 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
685 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
688 conditional_sti(regs); \
689 do_trap(trapnr, signr, str, regs, error_code, NULL); \
692 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
693 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
696 info.si_signo = signr; \
698 info.si_code = sicode; \
699 info.si_addr = (void __user *)siaddr; \
700 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
703 conditional_sti(regs); \
704 do_trap(trapnr, signr, str, regs, error_code, &info); \
707 DO_ERROR_INFO( 0, SIGFPE
, "divide error", divide_error
, FPE_INTDIV
, regs
->rip
)
708 DO_ERROR( 4, SIGSEGV
, "overflow", overflow
)
709 DO_ERROR( 5, SIGSEGV
, "bounds", bounds
)
710 DO_ERROR_INFO( 6, SIGILL
, "invalid opcode", invalid_op
, ILL_ILLOPN
, regs
->rip
)
711 DO_ERROR( 7, SIGSEGV
, "device not available", device_not_available
)
712 DO_ERROR( 9, SIGFPE
, "coprocessor segment overrun", coprocessor_segment_overrun
)
713 DO_ERROR(10, SIGSEGV
, "invalid TSS", invalid_TSS
)
714 DO_ERROR(11, SIGBUS
, "segment not present", segment_not_present
)
715 DO_ERROR_INFO(17, SIGBUS
, "alignment check", alignment_check
, BUS_ADRALN
, 0)
716 DO_ERROR(18, SIGSEGV
, "reserved", reserved
)
718 /* Runs on IST stack */
719 asmlinkage
void do_stack_segment(struct pt_regs
*regs
, long error_code
)
721 if (notify_die(DIE_TRAP
, "stack segment", regs
, error_code
,
722 12, SIGBUS
) == NOTIFY_STOP
)
724 preempt_conditional_sti(regs
);
725 do_trap(12, SIGBUS
, "stack segment", regs
, error_code
, NULL
);
726 preempt_conditional_cli(regs
);
729 asmlinkage
void do_double_fault(struct pt_regs
* regs
, long error_code
)
731 static const char str
[] = "double fault";
732 struct task_struct
*tsk
= current
;
734 /* Return not checked because double check cannot be ignored */
735 notify_die(DIE_TRAP
, str
, regs
, error_code
, 8, SIGSEGV
);
737 tsk
->thread
.error_code
= error_code
;
738 tsk
->thread
.trap_no
= 8;
740 /* This is always a kernel trap and never fixable (and thus must
743 die(str
, regs
, error_code
);
746 asmlinkage
void __kprobes
do_general_protection(struct pt_regs
* regs
,
749 struct task_struct
*tsk
= current
;
751 conditional_sti(regs
);
753 tsk
->thread
.error_code
= error_code
;
754 tsk
->thread
.trap_no
= 13;
756 if (user_mode(regs
)) {
757 if (exception_trace
&& unhandled_signal(tsk
, SIGSEGV
))
759 "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
761 regs
->rip
, regs
->rsp
, error_code
);
763 force_sig(SIGSEGV
, tsk
);
769 const struct exception_table_entry
*fixup
;
770 fixup
= search_exception_tables(regs
->rip
);
772 regs
->rip
= fixup
->fixup
;
775 if (notify_die(DIE_GPF
, "general protection fault", regs
,
776 error_code
, 13, SIGSEGV
) == NOTIFY_STOP
)
778 die("general protection fault", regs
, error_code
);
782 static __kprobes
void
783 mem_parity_error(unsigned char reason
, struct pt_regs
* regs
)
785 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x.\n",
787 printk(KERN_EMERG
"You probably have a hardware problem with your "
790 if (panic_on_unrecovered_nmi
)
791 panic("NMI: Not continuing");
793 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
795 /* Clear and disable the memory parity error line. */
796 reason
= (reason
& 0xf) | 4;
800 static __kprobes
void
801 io_check_error(unsigned char reason
, struct pt_regs
* regs
)
803 printk("NMI: IOCK error (debug interrupt?)\n");
804 show_registers(regs
);
806 /* Re-enable the IOCK line, wait for a few seconds */
807 reason
= (reason
& 0xf) | 8;
814 static __kprobes
void
815 unknown_nmi_error(unsigned char reason
, struct pt_regs
* regs
)
817 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x.\n",
819 printk(KERN_EMERG
"Do you have a strange power saving mode enabled?\n");
821 if (panic_on_unrecovered_nmi
)
822 panic("NMI: Not continuing");
824 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
827 /* Runs on IST stack. This code must keep interrupts off all the time.
828 Nested NMIs are prevented by the CPU. */
829 asmlinkage __kprobes
void default_do_nmi(struct pt_regs
*regs
)
831 unsigned char reason
= 0;
834 cpu
= smp_processor_id();
836 /* Only the BSP gets external NMIs from the system. */
838 reason
= get_nmi_reason();
840 if (!(reason
& 0xc0)) {
841 if (notify_die(DIE_NMI_IPI
, "nmi_ipi", regs
, reason
, 2, SIGINT
)
845 * Ok, so this is none of the documented NMI sources,
846 * so it must be the NMI watchdog.
848 if (nmi_watchdog_tick(regs
,reason
))
850 if (!do_nmi_callback(regs
,cpu
))
851 unknown_nmi_error(reason
, regs
);
855 if (notify_die(DIE_NMI
, "nmi", regs
, reason
, 2, SIGINT
) == NOTIFY_STOP
)
858 /* AK: following checks seem to be broken on modern chipsets. FIXME */
861 mem_parity_error(reason
, regs
);
863 io_check_error(reason
, regs
);
866 /* runs on IST stack. */
867 asmlinkage
void __kprobes
do_int3(struct pt_regs
* regs
, long error_code
)
869 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, 3, SIGTRAP
) == NOTIFY_STOP
) {
872 preempt_conditional_sti(regs
);
873 do_trap(3, SIGTRAP
, "int3", regs
, error_code
, NULL
);
874 preempt_conditional_cli(regs
);
877 /* Help handler running on IST stack to switch back to user stack
878 for scheduling or signal handling. The actual stack switch is done in
880 asmlinkage __kprobes
struct pt_regs
*sync_regs(struct pt_regs
*eregs
)
882 struct pt_regs
*regs
= eregs
;
883 /* Did already sync */
884 if (eregs
== (struct pt_regs
*)eregs
->rsp
)
886 /* Exception from user space */
887 else if (user_mode(eregs
))
888 regs
= task_pt_regs(current
);
889 /* Exception from kernel and interrupts are enabled. Move to
890 kernel process stack. */
891 else if (eregs
->eflags
& X86_EFLAGS_IF
)
892 regs
= (struct pt_regs
*)(eregs
->rsp
-= sizeof(struct pt_regs
));
898 /* runs on IST stack. */
899 asmlinkage
void __kprobes
do_debug(struct pt_regs
* regs
,
900 unsigned long error_code
)
902 unsigned long condition
;
903 struct task_struct
*tsk
= current
;
906 get_debugreg(condition
, 6);
908 if (notify_die(DIE_DEBUG
, "debug", regs
, condition
, error_code
,
909 SIGTRAP
) == NOTIFY_STOP
)
912 preempt_conditional_sti(regs
);
914 /* Mask out spurious debug traps due to lazy DR7 setting */
915 if (condition
& (DR_TRAP0
|DR_TRAP1
|DR_TRAP2
|DR_TRAP3
)) {
916 if (!tsk
->thread
.debugreg7
) {
921 tsk
->thread
.debugreg6
= condition
;
923 /* Mask out spurious TF errors due to lazy TF clearing */
924 if (condition
& DR_STEP
) {
926 * The TF error should be masked out only if the current
927 * process is not traced and if the TRAP flag has been set
928 * previously by a tracing process (condition detected by
929 * the PT_DTRACE flag); remember that the i386 TRAP flag
930 * can be modified by the process itself in user mode,
931 * allowing programs to debug themselves without the ptrace()
934 if (!user_mode(regs
))
935 goto clear_TF_reenable
;
937 * Was the TF flag set by a debugger? If so, clear it now,
938 * so that register information is correct.
940 if (tsk
->ptrace
& PT_DTRACE
) {
941 regs
->eflags
&= ~TF_MASK
;
942 tsk
->ptrace
&= ~PT_DTRACE
;
946 /* Ok, finally something we can handle */
947 tsk
->thread
.trap_no
= 1;
948 tsk
->thread
.error_code
= error_code
;
949 info
.si_signo
= SIGTRAP
;
951 info
.si_code
= TRAP_BRKPT
;
952 info
.si_addr
= user_mode(regs
) ? (void __user
*)regs
->rip
: NULL
;
953 force_sig_info(SIGTRAP
, &info
, tsk
);
956 set_debugreg(0UL, 7);
957 preempt_conditional_cli(regs
);
961 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
962 regs
->eflags
&= ~TF_MASK
;
963 preempt_conditional_cli(regs
);
966 static int kernel_math_error(struct pt_regs
*regs
, const char *str
, int trapnr
)
968 const struct exception_table_entry
*fixup
;
969 fixup
= search_exception_tables(regs
->rip
);
971 regs
->rip
= fixup
->fixup
;
974 notify_die(DIE_GPF
, str
, regs
, 0, trapnr
, SIGFPE
);
975 /* Illegal floating point operation in the kernel */
976 current
->thread
.trap_no
= trapnr
;
982 * Note that we play around with the 'TS' bit in an attempt to get
983 * the correct behaviour even in the presence of the asynchronous
986 asmlinkage
void do_coprocessor_error(struct pt_regs
*regs
)
988 void __user
*rip
= (void __user
*)(regs
->rip
);
989 struct task_struct
* task
;
991 unsigned short cwd
, swd
;
993 conditional_sti(regs
);
994 if (!user_mode(regs
) &&
995 kernel_math_error(regs
, "kernel x87 math error", 16))
999 * Save the info for the exception handler and clear the error.
1002 save_init_fpu(task
);
1003 task
->thread
.trap_no
= 16;
1004 task
->thread
.error_code
= 0;
1005 info
.si_signo
= SIGFPE
;
1007 info
.si_code
= __SI_FAULT
;
1010 * (~cwd & swd) will mask out exceptions that are not set to unmasked
1011 * status. 0x3f is the exception bits in these regs, 0x200 is the
1012 * C1 reg you need in case of a stack fault, 0x040 is the stack
1013 * fault bit. We should only be taking one exception at a time,
1014 * so if this combination doesn't produce any single exception,
1015 * then we have a bad program that isn't synchronizing its FPU usage
1016 * and it will suffer the consequences since we won't be able to
1017 * fully reproduce the context of the exception
1019 cwd
= get_fpu_cwd(task
);
1020 swd
= get_fpu_swd(task
);
1021 switch (swd
& ~cwd
& 0x3f) {
1025 case 0x001: /* Invalid Op */
1027 * swd & 0x240 == 0x040: Stack Underflow
1028 * swd & 0x240 == 0x240: Stack Overflow
1029 * User must clear the SF bit (0x40) if set
1031 info
.si_code
= FPE_FLTINV
;
1033 case 0x002: /* Denormalize */
1034 case 0x010: /* Underflow */
1035 info
.si_code
= FPE_FLTUND
;
1037 case 0x004: /* Zero Divide */
1038 info
.si_code
= FPE_FLTDIV
;
1040 case 0x008: /* Overflow */
1041 info
.si_code
= FPE_FLTOVF
;
1043 case 0x020: /* Precision */
1044 info
.si_code
= FPE_FLTRES
;
1047 force_sig_info(SIGFPE
, &info
, task
);
1050 asmlinkage
void bad_intr(void)
1052 printk("bad interrupt");
1055 asmlinkage
void do_simd_coprocessor_error(struct pt_regs
*regs
)
1057 void __user
*rip
= (void __user
*)(regs
->rip
);
1058 struct task_struct
* task
;
1060 unsigned short mxcsr
;
1062 conditional_sti(regs
);
1063 if (!user_mode(regs
) &&
1064 kernel_math_error(regs
, "kernel simd math error", 19))
1068 * Save the info for the exception handler and clear the error.
1071 save_init_fpu(task
);
1072 task
->thread
.trap_no
= 19;
1073 task
->thread
.error_code
= 0;
1074 info
.si_signo
= SIGFPE
;
1076 info
.si_code
= __SI_FAULT
;
1079 * The SIMD FPU exceptions are handled a little differently, as there
1080 * is only a single status/control register. Thus, to determine which
1081 * unmasked exception was caught we must mask the exception mask bits
1082 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1084 mxcsr
= get_fpu_mxcsr(task
);
1085 switch (~((mxcsr
& 0x1f80) >> 7) & (mxcsr
& 0x3f)) {
1089 case 0x001: /* Invalid Op */
1090 info
.si_code
= FPE_FLTINV
;
1092 case 0x002: /* Denormalize */
1093 case 0x010: /* Underflow */
1094 info
.si_code
= FPE_FLTUND
;
1096 case 0x004: /* Zero Divide */
1097 info
.si_code
= FPE_FLTDIV
;
1099 case 0x008: /* Overflow */
1100 info
.si_code
= FPE_FLTOVF
;
1102 case 0x020: /* Precision */
1103 info
.si_code
= FPE_FLTRES
;
1106 force_sig_info(SIGFPE
, &info
, task
);
1109 asmlinkage
void do_spurious_interrupt_bug(struct pt_regs
* regs
)
1113 asmlinkage
void __attribute__((weak
)) smp_thermal_interrupt(void)
1117 asmlinkage
void __attribute__((weak
)) mce_threshold_interrupt(void)
1122 * 'math_state_restore()' saves the current math information in the
1123 * old math state array, and gets the new ones from the current task
1125 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1126 * Don't touch unless you *really* know how it works.
1128 asmlinkage
void math_state_restore(void)
1130 struct task_struct
*me
= current
;
1131 clts(); /* Allow maths ops (or we recurse) */
1135 restore_fpu_checking(&me
->thread
.i387
.fxsave
);
1136 task_thread_info(me
)->status
|= TS_USEDFPU
;
1139 void __init
trap_init(void)
1141 set_intr_gate(0,÷_error
);
1142 set_intr_gate_ist(1,&debug
,DEBUG_STACK
);
1143 set_intr_gate_ist(2,&nmi
,NMI_STACK
);
1144 set_system_gate_ist(3,&int3
,DEBUG_STACK
); /* int3 can be called from all */
1145 set_system_gate(4,&overflow
); /* int4 can be called from all */
1146 set_intr_gate(5,&bounds
);
1147 set_intr_gate(6,&invalid_op
);
1148 set_intr_gate(7,&device_not_available
);
1149 set_intr_gate_ist(8,&double_fault
, DOUBLEFAULT_STACK
);
1150 set_intr_gate(9,&coprocessor_segment_overrun
);
1151 set_intr_gate(10,&invalid_TSS
);
1152 set_intr_gate(11,&segment_not_present
);
1153 set_intr_gate_ist(12,&stack_segment
,STACKFAULT_STACK
);
1154 set_intr_gate(13,&general_protection
);
1155 set_intr_gate(14,&page_fault
);
1156 set_intr_gate(15,&spurious_interrupt_bug
);
1157 set_intr_gate(16,&coprocessor_error
);
1158 set_intr_gate(17,&alignment_check
);
1159 #ifdef CONFIG_X86_MCE
1160 set_intr_gate_ist(18,&machine_check
, MCE_STACK
);
1162 set_intr_gate(19,&simd_coprocessor_error
);
1164 #ifdef CONFIG_IA32_EMULATION
1165 set_system_gate(IA32_SYSCALL_VECTOR
, ia32_syscall
);
1169 * Should be a barrier for any external CPU state.
1175 static int __init
oops_setup(char *s
)
1179 if (!strcmp(s
, "panic"))
1183 early_param("oops", oops_setup
);
1185 static int __init
kstack_setup(char *s
)
1189 kstack_depth_to_print
= simple_strtoul(s
,NULL
,0);
1192 early_param("kstack", kstack_setup
);
1194 #ifdef CONFIG_STACK_UNWIND
1195 static int __init
call_trace_setup(char *s
)
1199 if (strcmp(s
, "old") == 0)
1201 else if (strcmp(s
, "both") == 0)
1203 else if (strcmp(s
, "newfallback") == 0)
1205 else if (strcmp(s
, "new") == 0)
1209 early_param("call_trace", call_trace_setup
);