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
;
229 while (unwind(info
) == 0 && UNW_PC(info
)) {
231 oad
->ops
->address(oad
->data
, UNW_PC(info
));
232 if (arch_unw_user_mode(info
))
238 #define MSG(txt) ops->warning(data, txt)
241 * x86-64 can have upto three kernel stacks:
244 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
247 static inline int valid_stack_ptr(struct thread_info
*tinfo
, void *p
)
249 void *t
= (void *)tinfo
;
250 return p
> t
&& p
< t
+ THREAD_SIZE
- 3;
253 void dump_trace(struct task_struct
*tsk
, struct pt_regs
*regs
,
254 unsigned long *stack
,
255 struct stacktrace_ops
*ops
, void *data
)
257 const unsigned cpu
= get_cpu();
258 unsigned long *irqstack_end
= (unsigned long*)cpu_pda(cpu
)->irqstackptr
;
260 struct thread_info
*tinfo
;
265 if (call_trace
>= 0) {
267 struct unwind_frame_info info
;
268 struct ops_and_data oad
= { .ops
= ops
, .data
= data
};
271 if (unwind_init_frame_info(&info
, tsk
, regs
) == 0)
272 unw_ret
= dump_trace_unwind(&info
, &oad
);
273 } else if (tsk
== current
)
274 unw_ret
= unwind_init_running(&info
, dump_trace_unwind
,
277 if (unwind_init_blocked(&info
, tsk
) == 0)
278 unw_ret
= dump_trace_unwind(&info
, &oad
);
281 if (call_trace
== 1 && !arch_unw_user_mode(&info
)) {
282 ops
->warning_symbol(data
,
283 "DWARF2 unwinder stuck at %s",
285 if ((long)UNW_SP(&info
) < 0) {
286 MSG("Leftover inexact backtrace:");
287 stack
= (unsigned long *)UNW_SP(&info
);
291 MSG("Full inexact backtrace again:");
292 } else if (call_trace
>= 1)
295 MSG("Full inexact backtrace again:");
297 MSG("Inexact backtrace:");
302 if (tsk
&& tsk
!= current
)
303 stack
= (unsigned long *)tsk
->thread
.rsp
;
306 * Align the stack pointer on word boundary, later loops
307 * rely on that (and corruption / debug info bugs can cause
308 * unaligned values here):
310 stack
= (unsigned long *)((unsigned long)stack
& ~(sizeof(long)-1));
313 * Print function call entries within a stack. 'cond' is the
314 * "end of stackframe" condition, that the 'stack++'
315 * iteration will eventually trigger.
317 #define HANDLE_STACK(cond) \
319 unsigned long addr = *stack++; \
320 if (oops_in_progress ? \
321 __kernel_text_address(addr) : \
322 kernel_text_address(addr)) { \
324 * If the address is either in the text segment of the \
325 * kernel, or in the region which contains vmalloc'ed \
326 * memory, it *may* be the address of a calling \
327 * routine; if so, print it so that someone tracing \
328 * down the cause of the crash will be able to figure \
329 * out the call path that was taken. \
331 ops->address(data, addr); \
336 * Print function call entries in all stacks, starting at the
337 * current stack address. If the stacks consist of nested
342 unsigned long *estack_end
;
343 estack_end
= in_exception_stack(cpu
, (unsigned long)stack
,
347 if (ops
->stack(data
, id
) < 0)
349 HANDLE_STACK (stack
< estack_end
);
350 ops
->stack(data
, "<EOE>");
352 * We link to the next stack via the
353 * second-to-last pointer (index -2 to end) in the
356 stack
= (unsigned long *) estack_end
[-2];
360 unsigned long *irqstack
;
361 irqstack
= irqstack_end
-
362 (IRQSTACKSIZE
- 64) / sizeof(*irqstack
);
364 if (stack
>= irqstack
&& stack
< irqstack_end
) {
365 if (ops
->stack(data
, "IRQ") < 0)
367 HANDLE_STACK (stack
< irqstack_end
);
369 * We link to the next stack (which would be
370 * the process stack normally) the last
371 * pointer (index -1 to end) in the IRQ stack:
373 stack
= (unsigned long *) (irqstack_end
[-1]);
375 ops
->stack(data
, "EOI");
383 * This handles the process stack:
385 tinfo
= current_thread_info();
386 HANDLE_STACK (valid_stack_ptr(tinfo
, stack
));
391 EXPORT_SYMBOL(dump_trace
);
394 print_trace_warning_symbol(void *data
, char *msg
, unsigned long symbol
)
396 print_symbol(msg
, symbol
);
400 static void print_trace_warning(void *data
, char *msg
)
405 static int print_trace_stack(void *data
, char *name
)
407 printk(" <%s> ", name
);
411 static void print_trace_address(void *data
, unsigned long addr
)
413 printk_address(addr
);
416 static struct stacktrace_ops print_trace_ops
= {
417 .warning
= print_trace_warning
,
418 .warning_symbol
= print_trace_warning_symbol
,
419 .stack
= print_trace_stack
,
420 .address
= print_trace_address
,
424 show_trace(struct task_struct
*tsk
, struct pt_regs
*regs
, unsigned long *stack
)
426 printk("\nCall Trace:\n");
427 dump_trace(tsk
, regs
, stack
, &print_trace_ops
, NULL
);
432 _show_stack(struct task_struct
*tsk
, struct pt_regs
*regs
, unsigned long *rsp
)
434 unsigned long *stack
;
436 const int cpu
= smp_processor_id();
437 unsigned long *irqstack_end
= (unsigned long *) (cpu_pda(cpu
)->irqstackptr
);
438 unsigned long *irqstack
= (unsigned long *) (cpu_pda(cpu
)->irqstackptr
- IRQSTACKSIZE
);
440 // debugging aid: "show_stack(NULL, NULL);" prints the
441 // back trace for this cpu.
445 rsp
= (unsigned long *)tsk
->thread
.rsp
;
447 rsp
= (unsigned long *)&rsp
;
451 for(i
=0; i
< kstack_depth_to_print
; i
++) {
452 if (stack
>= irqstack
&& stack
<= irqstack_end
) {
453 if (stack
== irqstack_end
) {
454 stack
= (unsigned long *) (irqstack_end
[-1]);
458 if (((long) stack
& (THREAD_SIZE
-1)) == 0)
461 if (i
&& ((i
% 4) == 0))
463 printk(" %016lx", *stack
++);
464 touch_nmi_watchdog();
466 show_trace(tsk
, regs
, rsp
);
469 void show_stack(struct task_struct
*tsk
, unsigned long * rsp
)
471 _show_stack(tsk
, NULL
, rsp
);
475 * The architecture-independent dump_stack generator
477 void dump_stack(void)
480 show_trace(NULL
, NULL
, &dummy
);
483 EXPORT_SYMBOL(dump_stack
);
485 void show_registers(struct pt_regs
*regs
)
488 int in_kernel
= !user_mode(regs
);
490 const int cpu
= smp_processor_id();
491 struct task_struct
*cur
= cpu_pda(cpu
)->pcurrent
;
495 printk("CPU %d ", cpu
);
497 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
498 cur
->comm
, cur
->pid
, task_thread_info(cur
), cur
);
501 * When in-kernel, we also print out the stack and code at the
502 * time of the fault..
507 _show_stack(NULL
, regs
, (unsigned long*)rsp
);
510 if (regs
->rip
< PAGE_OFFSET
)
513 for (i
=0; i
<20; i
++) {
515 if (__get_user(c
, &((unsigned char*)regs
->rip
)[i
])) {
517 printk(" Bad RIP value.");
526 void handle_BUG(struct pt_regs
*regs
)
530 const char *prefix
= "";
534 if (__copy_from_user(&f
, (const void __user
*) regs
->rip
,
535 sizeof(struct bug_frame
)))
537 if (f
.filename
>= 0 ||
538 f
.ud2
[0] != 0x0f || f
.ud2
[1] != 0x0b)
540 len
= __strnlen_user((char *)(long)f
.filename
, PATH_MAX
) - 1;
541 if (len
< 0 || len
>= PATH_MAX
)
542 f
.filename
= (int)(long)"unmapped filename";
544 f
.filename
+= len
- 50;
547 printk("----------- [cut here ] --------- [please bite here ] ---------\n");
548 printk(KERN_ALERT
"Kernel BUG at %s%.50s:%d\n", prefix
, (char *)(long)f
.filename
, f
.line
);
552 void out_of_line_bug(void)
556 EXPORT_SYMBOL(out_of_line_bug
);
559 static DEFINE_SPINLOCK(die_lock
);
560 static int die_owner
= -1;
561 static unsigned int die_nest_count
;
563 unsigned __kprobes
long oops_begin(void)
565 int cpu
= smp_processor_id();
570 /* racy, but better than risking deadlock. */
571 local_irq_save(flags
);
572 if (!spin_trylock(&die_lock
)) {
573 if (cpu
== die_owner
)
574 /* nested oops. should stop eventually */;
576 spin_lock(&die_lock
);
585 void __kprobes
oops_end(unsigned long flags
)
591 /* We still own the lock */
592 local_irq_restore(flags
);
594 /* Nest count reaches zero, release the lock. */
595 spin_unlock_irqrestore(&die_lock
, flags
);
597 panic("Fatal exception");
601 void __kprobes
__die(const char * str
, struct pt_regs
* regs
, long err
)
603 static int die_counter
;
604 printk(KERN_EMERG
"%s: %04lx [%u] ", str
, err
& 0xffff,++die_counter
);
605 #ifdef CONFIG_PREEMPT
611 #ifdef CONFIG_DEBUG_PAGEALLOC
612 printk("DEBUG_PAGEALLOC");
615 notify_die(DIE_OOPS
, str
, regs
, err
, current
->thread
.trap_no
, SIGSEGV
);
616 show_registers(regs
);
617 /* Executive summary in case the oops scrolled away */
618 printk(KERN_ALERT
"RIP ");
619 printk_address(regs
->rip
);
620 printk(" RSP <%016lx>\n", regs
->rsp
);
621 if (kexec_should_crash(current
))
625 void die(const char * str
, struct pt_regs
* regs
, long err
)
627 unsigned long flags
= oops_begin();
630 __die(str
, regs
, err
);
635 void __kprobes
die_nmi(char *str
, struct pt_regs
*regs
, int do_panic
)
637 unsigned long flags
= oops_begin();
640 * We are in trouble anyway, lets at least try
641 * to get a message out.
643 printk(str
, smp_processor_id());
644 show_registers(regs
);
645 if (kexec_should_crash(current
))
647 if (do_panic
|| panic_on_oops
)
648 panic("Non maskable interrupt");
655 static void __kprobes
do_trap(int trapnr
, int signr
, char *str
,
656 struct pt_regs
* regs
, long error_code
,
659 struct task_struct
*tsk
= current
;
661 tsk
->thread
.error_code
= error_code
;
662 tsk
->thread
.trap_no
= trapnr
;
664 if (user_mode(regs
)) {
665 if (exception_trace
&& unhandled_signal(tsk
, signr
))
667 "%s[%d] trap %s rip:%lx rsp:%lx error:%lx\n",
668 tsk
->comm
, tsk
->pid
, str
,
669 regs
->rip
, regs
->rsp
, error_code
);
672 force_sig_info(signr
, info
, tsk
);
674 force_sig(signr
, tsk
);
681 const struct exception_table_entry
*fixup
;
682 fixup
= search_exception_tables(regs
->rip
);
684 regs
->rip
= fixup
->fixup
;
686 die(str
, regs
, error_code
);
691 #define DO_ERROR(trapnr, signr, str, name) \
692 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
694 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
697 conditional_sti(regs); \
698 do_trap(trapnr, signr, str, regs, error_code, NULL); \
701 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
702 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
705 info.si_signo = signr; \
707 info.si_code = sicode; \
708 info.si_addr = (void __user *)siaddr; \
709 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
712 conditional_sti(regs); \
713 do_trap(trapnr, signr, str, regs, error_code, &info); \
716 DO_ERROR_INFO( 0, SIGFPE
, "divide error", divide_error
, FPE_INTDIV
, regs
->rip
)
717 DO_ERROR( 4, SIGSEGV
, "overflow", overflow
)
718 DO_ERROR( 5, SIGSEGV
, "bounds", bounds
)
719 DO_ERROR_INFO( 6, SIGILL
, "invalid opcode", invalid_op
, ILL_ILLOPN
, regs
->rip
)
720 DO_ERROR( 7, SIGSEGV
, "device not available", device_not_available
)
721 DO_ERROR( 9, SIGFPE
, "coprocessor segment overrun", coprocessor_segment_overrun
)
722 DO_ERROR(10, SIGSEGV
, "invalid TSS", invalid_TSS
)
723 DO_ERROR(11, SIGBUS
, "segment not present", segment_not_present
)
724 DO_ERROR_INFO(17, SIGBUS
, "alignment check", alignment_check
, BUS_ADRALN
, 0)
725 DO_ERROR(18, SIGSEGV
, "reserved", reserved
)
727 /* Runs on IST stack */
728 asmlinkage
void do_stack_segment(struct pt_regs
*regs
, long error_code
)
730 if (notify_die(DIE_TRAP
, "stack segment", regs
, error_code
,
731 12, SIGBUS
) == NOTIFY_STOP
)
733 preempt_conditional_sti(regs
);
734 do_trap(12, SIGBUS
, "stack segment", regs
, error_code
, NULL
);
735 preempt_conditional_cli(regs
);
738 asmlinkage
void do_double_fault(struct pt_regs
* regs
, long error_code
)
740 static const char str
[] = "double fault";
741 struct task_struct
*tsk
= current
;
743 /* Return not checked because double check cannot be ignored */
744 notify_die(DIE_TRAP
, str
, regs
, error_code
, 8, SIGSEGV
);
746 tsk
->thread
.error_code
= error_code
;
747 tsk
->thread
.trap_no
= 8;
749 /* This is always a kernel trap and never fixable (and thus must
752 die(str
, regs
, error_code
);
755 asmlinkage
void __kprobes
do_general_protection(struct pt_regs
* regs
,
758 struct task_struct
*tsk
= current
;
760 conditional_sti(regs
);
762 tsk
->thread
.error_code
= error_code
;
763 tsk
->thread
.trap_no
= 13;
765 if (user_mode(regs
)) {
766 if (exception_trace
&& unhandled_signal(tsk
, SIGSEGV
))
768 "%s[%d] general protection rip:%lx rsp:%lx error:%lx\n",
770 regs
->rip
, regs
->rsp
, error_code
);
772 force_sig(SIGSEGV
, tsk
);
778 const struct exception_table_entry
*fixup
;
779 fixup
= search_exception_tables(regs
->rip
);
781 regs
->rip
= fixup
->fixup
;
784 if (notify_die(DIE_GPF
, "general protection fault", regs
,
785 error_code
, 13, SIGSEGV
) == NOTIFY_STOP
)
787 die("general protection fault", regs
, error_code
);
791 static __kprobes
void
792 mem_parity_error(unsigned char reason
, struct pt_regs
* regs
)
794 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x.\n",
796 printk(KERN_EMERG
"You have some hardware problem, likely on the PCI bus.\n");
798 if (panic_on_unrecovered_nmi
)
799 panic("NMI: Not continuing");
801 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
803 /* Clear and disable the memory parity error line. */
804 reason
= (reason
& 0xf) | 4;
808 static __kprobes
void
809 io_check_error(unsigned char reason
, struct pt_regs
* regs
)
811 printk("NMI: IOCK error (debug interrupt?)\n");
812 show_registers(regs
);
814 /* Re-enable the IOCK line, wait for a few seconds */
815 reason
= (reason
& 0xf) | 8;
822 static __kprobes
void
823 unknown_nmi_error(unsigned char reason
, struct pt_regs
* regs
)
825 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x.\n",
827 printk(KERN_EMERG
"Do you have a strange power saving mode enabled?\n");
829 if (panic_on_unrecovered_nmi
)
830 panic("NMI: Not continuing");
832 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
835 /* Runs on IST stack. This code must keep interrupts off all the time.
836 Nested NMIs are prevented by the CPU. */
837 asmlinkage __kprobes
void default_do_nmi(struct pt_regs
*regs
)
839 unsigned char reason
= 0;
842 cpu
= smp_processor_id();
844 /* Only the BSP gets external NMIs from the system. */
846 reason
= get_nmi_reason();
848 if (!(reason
& 0xc0)) {
849 if (notify_die(DIE_NMI_IPI
, "nmi_ipi", regs
, reason
, 2, SIGINT
)
853 * Ok, so this is none of the documented NMI sources,
854 * so it must be the NMI watchdog.
856 if (nmi_watchdog_tick(regs
,reason
))
858 if (!do_nmi_callback(regs
,cpu
))
859 unknown_nmi_error(reason
, regs
);
863 if (notify_die(DIE_NMI
, "nmi", regs
, reason
, 2, SIGINT
) == NOTIFY_STOP
)
866 /* AK: following checks seem to be broken on modern chipsets. FIXME */
869 mem_parity_error(reason
, regs
);
871 io_check_error(reason
, regs
);
874 /* runs on IST stack. */
875 asmlinkage
void __kprobes
do_int3(struct pt_regs
* regs
, long error_code
)
877 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, 3, SIGTRAP
) == NOTIFY_STOP
) {
880 preempt_conditional_sti(regs
);
881 do_trap(3, SIGTRAP
, "int3", regs
, error_code
, NULL
);
882 preempt_conditional_cli(regs
);
885 /* Help handler running on IST stack to switch back to user stack
886 for scheduling or signal handling. The actual stack switch is done in
888 asmlinkage __kprobes
struct pt_regs
*sync_regs(struct pt_regs
*eregs
)
890 struct pt_regs
*regs
= eregs
;
891 /* Did already sync */
892 if (eregs
== (struct pt_regs
*)eregs
->rsp
)
894 /* Exception from user space */
895 else if (user_mode(eregs
))
896 regs
= task_pt_regs(current
);
897 /* Exception from kernel and interrupts are enabled. Move to
898 kernel process stack. */
899 else if (eregs
->eflags
& X86_EFLAGS_IF
)
900 regs
= (struct pt_regs
*)(eregs
->rsp
-= sizeof(struct pt_regs
));
906 /* runs on IST stack. */
907 asmlinkage
void __kprobes
do_debug(struct pt_regs
* regs
,
908 unsigned long error_code
)
910 unsigned long condition
;
911 struct task_struct
*tsk
= current
;
914 get_debugreg(condition
, 6);
916 if (notify_die(DIE_DEBUG
, "debug", regs
, condition
, error_code
,
917 SIGTRAP
) == NOTIFY_STOP
)
920 preempt_conditional_sti(regs
);
922 /* Mask out spurious debug traps due to lazy DR7 setting */
923 if (condition
& (DR_TRAP0
|DR_TRAP1
|DR_TRAP2
|DR_TRAP3
)) {
924 if (!tsk
->thread
.debugreg7
) {
929 tsk
->thread
.debugreg6
= condition
;
931 /* Mask out spurious TF errors due to lazy TF clearing */
932 if (condition
& DR_STEP
) {
934 * The TF error should be masked out only if the current
935 * process is not traced and if the TRAP flag has been set
936 * previously by a tracing process (condition detected by
937 * the PT_DTRACE flag); remember that the i386 TRAP flag
938 * can be modified by the process itself in user mode,
939 * allowing programs to debug themselves without the ptrace()
942 if (!user_mode(regs
))
943 goto clear_TF_reenable
;
945 * Was the TF flag set by a debugger? If so, clear it now,
946 * so that register information is correct.
948 if (tsk
->ptrace
& PT_DTRACE
) {
949 regs
->eflags
&= ~TF_MASK
;
950 tsk
->ptrace
&= ~PT_DTRACE
;
954 /* Ok, finally something we can handle */
955 tsk
->thread
.trap_no
= 1;
956 tsk
->thread
.error_code
= error_code
;
957 info
.si_signo
= SIGTRAP
;
959 info
.si_code
= TRAP_BRKPT
;
960 info
.si_addr
= user_mode(regs
) ? (void __user
*)regs
->rip
: NULL
;
961 force_sig_info(SIGTRAP
, &info
, tsk
);
964 set_debugreg(0UL, 7);
965 preempt_conditional_cli(regs
);
969 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
970 regs
->eflags
&= ~TF_MASK
;
971 preempt_conditional_cli(regs
);
974 static int kernel_math_error(struct pt_regs
*regs
, const char *str
, int trapnr
)
976 const struct exception_table_entry
*fixup
;
977 fixup
= search_exception_tables(regs
->rip
);
979 regs
->rip
= fixup
->fixup
;
982 notify_die(DIE_GPF
, str
, regs
, 0, trapnr
, SIGFPE
);
983 /* Illegal floating point operation in the kernel */
984 current
->thread
.trap_no
= trapnr
;
990 * Note that we play around with the 'TS' bit in an attempt to get
991 * the correct behaviour even in the presence of the asynchronous
994 asmlinkage
void do_coprocessor_error(struct pt_regs
*regs
)
996 void __user
*rip
= (void __user
*)(regs
->rip
);
997 struct task_struct
* task
;
999 unsigned short cwd
, swd
;
1001 conditional_sti(regs
);
1002 if (!user_mode(regs
) &&
1003 kernel_math_error(regs
, "kernel x87 math error", 16))
1007 * Save the info for the exception handler and clear the error.
1010 save_init_fpu(task
);
1011 task
->thread
.trap_no
= 16;
1012 task
->thread
.error_code
= 0;
1013 info
.si_signo
= SIGFPE
;
1015 info
.si_code
= __SI_FAULT
;
1018 * (~cwd & swd) will mask out exceptions that are not set to unmasked
1019 * status. 0x3f is the exception bits in these regs, 0x200 is the
1020 * C1 reg you need in case of a stack fault, 0x040 is the stack
1021 * fault bit. We should only be taking one exception at a time,
1022 * so if this combination doesn't produce any single exception,
1023 * then we have a bad program that isn't synchronizing its FPU usage
1024 * and it will suffer the consequences since we won't be able to
1025 * fully reproduce the context of the exception
1027 cwd
= get_fpu_cwd(task
);
1028 swd
= get_fpu_swd(task
);
1029 switch (swd
& ~cwd
& 0x3f) {
1033 case 0x001: /* Invalid Op */
1035 * swd & 0x240 == 0x040: Stack Underflow
1036 * swd & 0x240 == 0x240: Stack Overflow
1037 * User must clear the SF bit (0x40) if set
1039 info
.si_code
= FPE_FLTINV
;
1041 case 0x002: /* Denormalize */
1042 case 0x010: /* Underflow */
1043 info
.si_code
= FPE_FLTUND
;
1045 case 0x004: /* Zero Divide */
1046 info
.si_code
= FPE_FLTDIV
;
1048 case 0x008: /* Overflow */
1049 info
.si_code
= FPE_FLTOVF
;
1051 case 0x020: /* Precision */
1052 info
.si_code
= FPE_FLTRES
;
1055 force_sig_info(SIGFPE
, &info
, task
);
1058 asmlinkage
void bad_intr(void)
1060 printk("bad interrupt");
1063 asmlinkage
void do_simd_coprocessor_error(struct pt_regs
*regs
)
1065 void __user
*rip
= (void __user
*)(regs
->rip
);
1066 struct task_struct
* task
;
1068 unsigned short mxcsr
;
1070 conditional_sti(regs
);
1071 if (!user_mode(regs
) &&
1072 kernel_math_error(regs
, "kernel simd math error", 19))
1076 * Save the info for the exception handler and clear the error.
1079 save_init_fpu(task
);
1080 task
->thread
.trap_no
= 19;
1081 task
->thread
.error_code
= 0;
1082 info
.si_signo
= SIGFPE
;
1084 info
.si_code
= __SI_FAULT
;
1087 * The SIMD FPU exceptions are handled a little differently, as there
1088 * is only a single status/control register. Thus, to determine which
1089 * unmasked exception was caught we must mask the exception mask bits
1090 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1092 mxcsr
= get_fpu_mxcsr(task
);
1093 switch (~((mxcsr
& 0x1f80) >> 7) & (mxcsr
& 0x3f)) {
1097 case 0x001: /* Invalid Op */
1098 info
.si_code
= FPE_FLTINV
;
1100 case 0x002: /* Denormalize */
1101 case 0x010: /* Underflow */
1102 info
.si_code
= FPE_FLTUND
;
1104 case 0x004: /* Zero Divide */
1105 info
.si_code
= FPE_FLTDIV
;
1107 case 0x008: /* Overflow */
1108 info
.si_code
= FPE_FLTOVF
;
1110 case 0x020: /* Precision */
1111 info
.si_code
= FPE_FLTRES
;
1114 force_sig_info(SIGFPE
, &info
, task
);
1117 asmlinkage
void do_spurious_interrupt_bug(struct pt_regs
* regs
)
1121 asmlinkage
void __attribute__((weak
)) smp_thermal_interrupt(void)
1125 asmlinkage
void __attribute__((weak
)) mce_threshold_interrupt(void)
1130 * 'math_state_restore()' saves the current math information in the
1131 * old math state array, and gets the new ones from the current task
1133 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1134 * Don't touch unless you *really* know how it works.
1136 asmlinkage
void math_state_restore(void)
1138 struct task_struct
*me
= current
;
1139 clts(); /* Allow maths ops (or we recurse) */
1143 restore_fpu_checking(&me
->thread
.i387
.fxsave
);
1144 task_thread_info(me
)->status
|= TS_USEDFPU
;
1148 void __init
trap_init(void)
1150 set_intr_gate(0,÷_error
);
1151 set_intr_gate_ist(1,&debug
,DEBUG_STACK
);
1152 set_intr_gate_ist(2,&nmi
,NMI_STACK
);
1153 set_system_gate_ist(3,&int3
,DEBUG_STACK
); /* int3 can be called from all */
1154 set_system_gate(4,&overflow
); /* int4 can be called from all */
1155 set_intr_gate(5,&bounds
);
1156 set_intr_gate(6,&invalid_op
);
1157 set_intr_gate(7,&device_not_available
);
1158 set_intr_gate_ist(8,&double_fault
, DOUBLEFAULT_STACK
);
1159 set_intr_gate(9,&coprocessor_segment_overrun
);
1160 set_intr_gate(10,&invalid_TSS
);
1161 set_intr_gate(11,&segment_not_present
);
1162 set_intr_gate_ist(12,&stack_segment
,STACKFAULT_STACK
);
1163 set_intr_gate(13,&general_protection
);
1164 set_intr_gate(14,&page_fault
);
1165 set_intr_gate(15,&spurious_interrupt_bug
);
1166 set_intr_gate(16,&coprocessor_error
);
1167 set_intr_gate(17,&alignment_check
);
1168 #ifdef CONFIG_X86_MCE
1169 set_intr_gate_ist(18,&machine_check
, MCE_STACK
);
1171 set_intr_gate(19,&simd_coprocessor_error
);
1173 #ifdef CONFIG_IA32_EMULATION
1174 set_system_gate(IA32_SYSCALL_VECTOR
, ia32_syscall
);
1178 * Should be a barrier for any external CPU state.
1184 static int __init
oops_setup(char *s
)
1188 if (!strcmp(s
, "panic"))
1192 early_param("oops", oops_setup
);
1194 static int __init
kstack_setup(char *s
)
1198 kstack_depth_to_print
= simple_strtoul(s
,NULL
,0);
1201 early_param("kstack", kstack_setup
);
1203 #ifdef CONFIG_STACK_UNWIND
1204 static int __init
call_trace_setup(char *s
)
1208 if (strcmp(s
, "old") == 0)
1210 else if (strcmp(s
, "both") == 0)
1212 else if (strcmp(s
, "newfallback") == 0)
1214 else if (strcmp(s
, "new") == 0)
1218 early_param("call_trace", call_trace_setup
);