2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
10 * 'Traps.c' handles hardware traps and faults after we have saved some
13 #include <linux/sched.h>
14 #include <linux/kernel.h>
15 #include <linux/string.h>
16 #include <linux/errno.h>
17 #include <linux/ptrace.h>
18 #include <linux/timer.h>
20 #include <linux/init.h>
21 #include <linux/delay.h>
22 #include <linux/spinlock.h>
23 #include <linux/interrupt.h>
24 #include <linux/kallsyms.h>
25 #include <linux/module.h>
26 #include <linux/moduleparam.h>
27 #include <linux/nmi.h>
28 #include <linux/kprobes.h>
29 #include <linux/kexec.h>
30 #include <linux/unwind.h>
31 #include <linux/uaccess.h>
32 #include <linux/bug.h>
33 #include <linux/kdebug.h>
34 #include <linux/utsname.h>
36 #if defined(CONFIG_EDAC)
37 #include <linux/edac.h>
40 #include <asm/system.h>
42 #include <asm/atomic.h>
43 #include <asm/debugreg.h>
46 #include <asm/processor.h>
47 #include <asm/unwind.h>
49 #include <asm/pgalloc.h>
51 #include <asm/proto.h>
53 #include <asm/stacktrace.h>
55 asmlinkage
void divide_error(void);
56 asmlinkage
void debug(void);
57 asmlinkage
void nmi(void);
58 asmlinkage
void int3(void);
59 asmlinkage
void overflow(void);
60 asmlinkage
void bounds(void);
61 asmlinkage
void invalid_op(void);
62 asmlinkage
void device_not_available(void);
63 asmlinkage
void double_fault(void);
64 asmlinkage
void coprocessor_segment_overrun(void);
65 asmlinkage
void invalid_TSS(void);
66 asmlinkage
void segment_not_present(void);
67 asmlinkage
void stack_segment(void);
68 asmlinkage
void general_protection(void);
69 asmlinkage
void page_fault(void);
70 asmlinkage
void coprocessor_error(void);
71 asmlinkage
void simd_coprocessor_error(void);
72 asmlinkage
void reserved(void);
73 asmlinkage
void alignment_check(void);
74 asmlinkage
void machine_check(void);
75 asmlinkage
void spurious_interrupt_bug(void);
77 static unsigned int code_bytes
= 64;
79 static inline void conditional_sti(struct pt_regs
*regs
)
81 if (regs
->flags
& X86_EFLAGS_IF
)
85 static inline void preempt_conditional_sti(struct pt_regs
*regs
)
88 if (regs
->flags
& X86_EFLAGS_IF
)
92 static inline void preempt_conditional_cli(struct pt_regs
*regs
)
94 if (regs
->flags
& X86_EFLAGS_IF
)
96 /* Make sure to not schedule here because we could be running
97 on an exception stack. */
98 preempt_enable_no_resched();
101 int kstack_depth_to_print
= 12;
103 #ifdef CONFIG_KALLSYMS
104 void printk_address(unsigned long address
, int reliable
)
106 unsigned long offset
= 0, symsize
;
111 char reliab
[4] = "";;
113 symname
= kallsyms_lookup(address
, &symsize
, &offset
,
116 printk(" [<%016lx>]\n", address
);
120 strcpy(reliab
, "? ");
123 modname
= delim
= "";
124 printk(" [<%016lx>] %s%s%s%s%s+0x%lx/0x%lx\n",
125 address
, reliab
, delim
, modname
, delim
, symname
, offset
, symsize
);
128 void printk_address(unsigned long address
, int reliable
)
130 printk(" [<%016lx>]\n", address
);
134 static unsigned long *in_exception_stack(unsigned cpu
, unsigned long stack
,
135 unsigned *usedp
, char **idp
)
137 static char ids
[][8] = {
138 [DEBUG_STACK
- 1] = "#DB",
139 [NMI_STACK
- 1] = "NMI",
140 [DOUBLEFAULT_STACK
- 1] = "#DF",
141 [STACKFAULT_STACK
- 1] = "#SS",
142 [MCE_STACK
- 1] = "#MC",
143 #if DEBUG_STKSZ > EXCEPTION_STKSZ
144 [N_EXCEPTION_STACKS
... N_EXCEPTION_STACKS
+ DEBUG_STKSZ
/ EXCEPTION_STKSZ
- 2] = "#DB[?]"
150 * Iterate over all exception stacks, and figure out whether
151 * 'stack' is in one of them:
153 for (k
= 0; k
< N_EXCEPTION_STACKS
; k
++) {
154 unsigned long end
= per_cpu(orig_ist
, cpu
).ist
[k
];
156 * Is 'stack' above this exception frame's end?
157 * If yes then skip to the next frame.
162 * Is 'stack' above this exception frame's start address?
163 * If yes then we found the right frame.
165 if (stack
>= end
- EXCEPTION_STKSZ
) {
167 * Make sure we only iterate through an exception
168 * stack once. If it comes up for the second time
169 * then there's something wrong going on - just
170 * break out and return NULL:
172 if (*usedp
& (1U << k
))
176 return (unsigned long *)end
;
179 * If this is a debug stack, and if it has a larger size than
180 * the usual exception stacks, then 'stack' might still
181 * be within the lower portion of the debug stack:
183 #if DEBUG_STKSZ > EXCEPTION_STKSZ
184 if (k
== DEBUG_STACK
- 1 && stack
>= end
- DEBUG_STKSZ
) {
185 unsigned j
= N_EXCEPTION_STACKS
- 1;
188 * Black magic. A large debug stack is composed of
189 * multiple exception stack entries, which we
190 * iterate through now. Dont look:
194 end
-= EXCEPTION_STKSZ
;
195 ids
[j
][4] = '1' + (j
- N_EXCEPTION_STACKS
);
196 } while (stack
< end
- EXCEPTION_STKSZ
);
197 if (*usedp
& (1U << j
))
201 return (unsigned long *)end
;
208 #define MSG(txt) ops->warning(data, txt)
211 * x86-64 can have up to three kernel stacks:
214 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
217 static inline int valid_stack_ptr(struct thread_info
*tinfo
,
218 void *p
, unsigned int size
, void *end
)
220 void *t
= (void *)tinfo
;
222 if (p
< end
&& p
>= (end
-THREAD_SIZE
))
227 return p
> t
&& p
< t
+ THREAD_SIZE
- size
;
230 /* The form of the top of the frame on the stack */
232 struct stack_frame
*next_frame
;
233 unsigned long return_address
;
237 static inline unsigned long print_context_stack(struct thread_info
*tinfo
,
238 unsigned long *stack
, unsigned long bp
,
239 const struct stacktrace_ops
*ops
, void *data
,
242 struct stack_frame
*frame
= (struct stack_frame
*)bp
;
244 while (valid_stack_ptr(tinfo
, stack
, sizeof(*stack
), end
)) {
248 if (__kernel_text_address(addr
)) {
249 if ((unsigned long) stack
== bp
+ 8) {
250 ops
->address(data
, addr
, 1);
251 frame
= frame
->next_frame
;
252 bp
= (unsigned long) frame
;
254 ops
->address(data
, addr
, bp
== 0);
262 void dump_trace(struct task_struct
*tsk
, struct pt_regs
*regs
,
263 unsigned long *stack
, unsigned long bp
,
264 const struct stacktrace_ops
*ops
, void *data
)
266 const unsigned cpu
= get_cpu();
267 unsigned long *irqstack_end
= (unsigned long*)cpu_pda(cpu
)->irqstackptr
;
269 struct thread_info
*tinfo
;
273 tinfo
= task_thread_info(tsk
);
278 if (tsk
&& tsk
!= current
)
279 stack
= (unsigned long *)tsk
->thread
.sp
;
282 #ifdef CONFIG_FRAME_POINTER
284 if (tsk
== current
) {
285 /* Grab bp right from our regs */
286 asm("movq %%rbp, %0" : "=r" (bp
):);
288 /* bp is the last reg pushed by switch_to */
289 bp
= *(unsigned long *) tsk
->thread
.sp
;
297 * Print function call entries in all stacks, starting at the
298 * current stack address. If the stacks consist of nested
303 unsigned long *estack_end
;
304 estack_end
= in_exception_stack(cpu
, (unsigned long)stack
,
308 if (ops
->stack(data
, id
) < 0)
311 bp
= print_context_stack(tinfo
, stack
, bp
, ops
,
313 ops
->stack(data
, "<EOE>");
315 * We link to the next stack via the
316 * second-to-last pointer (index -2 to end) in the
319 stack
= (unsigned long *) estack_end
[-2];
323 unsigned long *irqstack
;
324 irqstack
= irqstack_end
-
325 (IRQSTACKSIZE
- 64) / sizeof(*irqstack
);
327 if (stack
>= irqstack
&& stack
< irqstack_end
) {
328 if (ops
->stack(data
, "IRQ") < 0)
330 bp
= print_context_stack(tinfo
, stack
, bp
,
331 ops
, data
, irqstack_end
);
333 * We link to the next stack (which would be
334 * the process stack normally) the last
335 * pointer (index -1 to end) in the IRQ stack:
337 stack
= (unsigned long *) (irqstack_end
[-1]);
339 ops
->stack(data
, "EOI");
347 * This handles the process stack:
349 bp
= print_context_stack(tinfo
, stack
, bp
, ops
, data
, NULL
);
352 EXPORT_SYMBOL(dump_trace
);
355 print_trace_warning_symbol(void *data
, char *msg
, unsigned long symbol
)
357 print_symbol(msg
, symbol
);
361 static void print_trace_warning(void *data
, char *msg
)
366 static int print_trace_stack(void *data
, char *name
)
368 printk(" <%s> ", name
);
372 static void print_trace_address(void *data
, unsigned long addr
, int reliable
)
374 touch_nmi_watchdog();
375 printk_address(addr
, reliable
);
378 static const struct stacktrace_ops print_trace_ops
= {
379 .warning
= print_trace_warning
,
380 .warning_symbol
= print_trace_warning_symbol
,
381 .stack
= print_trace_stack
,
382 .address
= print_trace_address
,
386 show_trace(struct task_struct
*tsk
, struct pt_regs
*regs
, unsigned long *stack
,
389 printk("\nCall Trace:\n");
390 dump_trace(tsk
, regs
, stack
, bp
, &print_trace_ops
, NULL
);
395 _show_stack(struct task_struct
*tsk
, struct pt_regs
*regs
, unsigned long *sp
,
398 unsigned long *stack
;
400 const int cpu
= smp_processor_id();
401 unsigned long *irqstack_end
= (unsigned long *) (cpu_pda(cpu
)->irqstackptr
);
402 unsigned long *irqstack
= (unsigned long *) (cpu_pda(cpu
)->irqstackptr
- IRQSTACKSIZE
);
404 // debugging aid: "show_stack(NULL, NULL);" prints the
405 // back trace for this cpu.
409 sp
= (unsigned long *)tsk
->thread
.sp
;
411 sp
= (unsigned long *)&sp
;
415 for(i
=0; i
< kstack_depth_to_print
; i
++) {
416 if (stack
>= irqstack
&& stack
<= irqstack_end
) {
417 if (stack
== irqstack_end
) {
418 stack
= (unsigned long *) (irqstack_end
[-1]);
422 if (((long) stack
& (THREAD_SIZE
-1)) == 0)
425 if (i
&& ((i
% 4) == 0))
427 printk(" %016lx", *stack
++);
428 touch_nmi_watchdog();
430 show_trace(tsk
, regs
, sp
, bp
);
433 void show_stack(struct task_struct
*tsk
, unsigned long * sp
)
435 _show_stack(tsk
, NULL
, sp
, 0);
439 * The architecture-independent dump_stack generator
441 void dump_stack(void)
444 unsigned long bp
= 0;
446 #ifdef CONFIG_FRAME_POINTER
448 asm("movq %%rbp, %0" : "=r" (bp
):);
451 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
452 current
->pid
, current
->comm
, print_tainted(),
453 init_utsname()->release
,
454 (int)strcspn(init_utsname()->version
, " "),
455 init_utsname()->version
);
456 show_trace(NULL
, NULL
, &dummy
, bp
);
459 EXPORT_SYMBOL(dump_stack
);
461 void show_registers(struct pt_regs
*regs
)
465 const int cpu
= smp_processor_id();
466 struct task_struct
*cur
= cpu_pda(cpu
)->pcurrent
;
468 unsigned int code_prologue
= code_bytes
* 43 / 64;
469 unsigned int code_len
= code_bytes
;
472 ip
= (u8
*) regs
->ip
- code_prologue
;
473 printk("CPU %d ", cpu
);
475 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
476 cur
->comm
, cur
->pid
, task_thread_info(cur
), cur
);
479 * When in-kernel, we also print out the stack and code at the
480 * time of the fault..
482 if (!user_mode(regs
)) {
485 _show_stack(NULL
, regs
, (unsigned long *)sp
, regs
->bp
);
488 printk(KERN_EMERG
"Code: ");
489 if (ip
< (u8
*)PAGE_OFFSET
|| probe_kernel_address(ip
, c
)) {
490 /* try starting at RIP */
491 ip
= (u8
*) regs
->ip
;
492 code_len
= code_len
- code_prologue
+ 1;
494 for (i
= 0; i
< code_len
; i
++, ip
++) {
495 if (ip
< (u8
*)PAGE_OFFSET
||
496 probe_kernel_address(ip
, c
)) {
497 printk(" Bad RIP value.");
500 if (ip
== (u8
*)regs
->ip
)
501 printk("<%02x> ", c
);
509 int is_valid_bugaddr(unsigned long ip
)
513 if (__copy_from_user(&ud2
, (const void __user
*) ip
, sizeof(ud2
)))
516 return ud2
== 0x0b0f;
519 static raw_spinlock_t die_lock
= __RAW_SPIN_LOCK_UNLOCKED
;
520 static int die_owner
= -1;
521 static unsigned int die_nest_count
;
523 unsigned __kprobes
long oops_begin(void)
530 /* racy, but better than risking deadlock. */
531 raw_local_irq_save(flags
);
532 cpu
= smp_processor_id();
533 if (!__raw_spin_trylock(&die_lock
)) {
534 if (cpu
== die_owner
)
535 /* nested oops. should stop eventually */;
537 __raw_spin_lock(&die_lock
);
546 void __kprobes
oops_end(unsigned long flags
, struct pt_regs
*regs
, int signr
)
552 /* Nest count reaches zero, release the lock. */
553 __raw_spin_unlock(&die_lock
);
554 raw_local_irq_restore(flags
);
560 panic("Fatal exception");
565 int __kprobes
__die(const char * str
, struct pt_regs
* regs
, long err
)
567 static int die_counter
;
568 printk(KERN_EMERG
"%s: %04lx [%u] ", str
, err
& 0xffff,++die_counter
);
569 #ifdef CONFIG_PREEMPT
575 #ifdef CONFIG_DEBUG_PAGEALLOC
576 printk("DEBUG_PAGEALLOC");
579 if (notify_die(DIE_OOPS
, str
, regs
, err
, current
->thread
.trap_no
, SIGSEGV
) == NOTIFY_STOP
)
581 show_registers(regs
);
582 add_taint(TAINT_DIE
);
583 /* Executive summary in case the oops scrolled away */
584 printk(KERN_ALERT
"RIP ");
585 printk_address(regs
->ip
, regs
->bp
);
586 printk(" RSP <%016lx>\n", regs
->sp
);
587 if (kexec_should_crash(current
))
592 void die(const char * str
, struct pt_regs
* regs
, long err
)
594 unsigned long flags
= oops_begin();
596 if (!user_mode(regs
))
597 report_bug(regs
->ip
, regs
);
599 if (__die(str
, regs
, err
))
601 oops_end(flags
, regs
, SIGSEGV
);
604 void __kprobes
die_nmi(char *str
, struct pt_regs
*regs
, int do_panic
)
606 unsigned long flags
= oops_begin();
609 * We are in trouble anyway, lets at least try
610 * to get a message out.
612 printk(str
, smp_processor_id());
613 show_registers(regs
);
614 if (kexec_should_crash(current
))
616 if (do_panic
|| panic_on_oops
)
617 panic("Non maskable interrupt");
618 oops_end(flags
, NULL
, SIGBUS
);
624 static void __kprobes
do_trap(int trapnr
, int signr
, char *str
,
625 struct pt_regs
* regs
, long error_code
,
628 struct task_struct
*tsk
= current
;
630 if (user_mode(regs
)) {
632 * We want error_code and trap_no set for userspace
633 * faults and kernelspace faults which result in
634 * die(), but not kernelspace faults which are fixed
635 * up. die() gives the process no chance to handle
636 * the signal and notice the kernel fault information,
637 * so that won't result in polluting the information
638 * about previously queued, but not yet delivered,
639 * faults. See also do_general_protection below.
641 tsk
->thread
.error_code
= error_code
;
642 tsk
->thread
.trap_no
= trapnr
;
644 if (show_unhandled_signals
&& unhandled_signal(tsk
, signr
) &&
647 "%s[%d] trap %s ip:%lx sp:%lx error:%lx\n",
648 tsk
->comm
, tsk
->pid
, str
,
649 regs
->ip
, regs
->sp
, error_code
);
652 force_sig_info(signr
, info
, tsk
);
654 force_sig(signr
, tsk
);
659 if (!fixup_exception(regs
)) {
660 tsk
->thread
.error_code
= error_code
;
661 tsk
->thread
.trap_no
= trapnr
;
662 die(str
, regs
, error_code
);
667 #define DO_ERROR(trapnr, signr, str, name) \
668 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
670 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
673 conditional_sti(regs); \
674 do_trap(trapnr, signr, str, regs, error_code, NULL); \
677 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
678 asmlinkage void do_##name(struct pt_regs * regs, long error_code) \
681 info.si_signo = signr; \
683 info.si_code = sicode; \
684 info.si_addr = (void __user *)siaddr; \
685 trace_hardirqs_fixup(); \
686 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
689 conditional_sti(regs); \
690 do_trap(trapnr, signr, str, regs, error_code, &info); \
693 DO_ERROR_INFO( 0, SIGFPE
, "divide error", divide_error
, FPE_INTDIV
, regs
->ip
)
694 DO_ERROR( 4, SIGSEGV
, "overflow", overflow
)
695 DO_ERROR( 5, SIGSEGV
, "bounds", bounds
)
696 DO_ERROR_INFO( 6, SIGILL
, "invalid opcode", invalid_op
, ILL_ILLOPN
, regs
->ip
)
697 DO_ERROR( 7, SIGSEGV
, "device not available", device_not_available
)
698 DO_ERROR( 9, SIGFPE
, "coprocessor segment overrun", coprocessor_segment_overrun
)
699 DO_ERROR(10, SIGSEGV
, "invalid TSS", invalid_TSS
)
700 DO_ERROR(11, SIGBUS
, "segment not present", segment_not_present
)
701 DO_ERROR_INFO(17, SIGBUS
, "alignment check", alignment_check
, BUS_ADRALN
, 0)
702 DO_ERROR(18, SIGSEGV
, "reserved", reserved
)
704 /* Runs on IST stack */
705 asmlinkage
void do_stack_segment(struct pt_regs
*regs
, long error_code
)
707 if (notify_die(DIE_TRAP
, "stack segment", regs
, error_code
,
708 12, SIGBUS
) == NOTIFY_STOP
)
710 preempt_conditional_sti(regs
);
711 do_trap(12, SIGBUS
, "stack segment", regs
, error_code
, NULL
);
712 preempt_conditional_cli(regs
);
715 asmlinkage
void do_double_fault(struct pt_regs
* regs
, long error_code
)
717 static const char str
[] = "double fault";
718 struct task_struct
*tsk
= current
;
720 /* Return not checked because double check cannot be ignored */
721 notify_die(DIE_TRAP
, str
, regs
, error_code
, 8, SIGSEGV
);
723 tsk
->thread
.error_code
= error_code
;
724 tsk
->thread
.trap_no
= 8;
726 /* This is always a kernel trap and never fixable (and thus must
729 die(str
, regs
, error_code
);
732 asmlinkage
void __kprobes
do_general_protection(struct pt_regs
* regs
,
735 struct task_struct
*tsk
= current
;
737 conditional_sti(regs
);
739 if (user_mode(regs
)) {
740 tsk
->thread
.error_code
= error_code
;
741 tsk
->thread
.trap_no
= 13;
743 if (show_unhandled_signals
&& unhandled_signal(tsk
, SIGSEGV
) &&
746 "%s[%d] general protection ip:%lx sp:%lx error:%lx\n",
748 regs
->ip
, regs
->sp
, error_code
);
750 force_sig(SIGSEGV
, tsk
);
754 if (fixup_exception(regs
))
757 tsk
->thread
.error_code
= error_code
;
758 tsk
->thread
.trap_no
= 13;
759 if (notify_die(DIE_GPF
, "general protection fault", regs
,
760 error_code
, 13, SIGSEGV
) == NOTIFY_STOP
)
762 die("general protection fault", regs
, error_code
);
765 static __kprobes
void
766 mem_parity_error(unsigned char reason
, struct pt_regs
* regs
)
768 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x.\n",
770 printk(KERN_EMERG
"You have some hardware problem, likely on the PCI bus.\n");
772 #if defined(CONFIG_EDAC)
773 if(edac_handler_set()) {
774 edac_atomic_assert_error();
779 if (panic_on_unrecovered_nmi
)
780 panic("NMI: Not continuing");
782 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
784 /* Clear and disable the memory parity error line. */
785 reason
= (reason
& 0xf) | 4;
789 static __kprobes
void
790 io_check_error(unsigned char reason
, struct pt_regs
* regs
)
792 printk("NMI: IOCK error (debug interrupt?)\n");
793 show_registers(regs
);
795 /* Re-enable the IOCK line, wait for a few seconds */
796 reason
= (reason
& 0xf) | 8;
803 static __kprobes
void
804 unknown_nmi_error(unsigned char reason
, struct pt_regs
* regs
)
806 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x.\n",
808 printk(KERN_EMERG
"Do you have a strange power saving mode enabled?\n");
810 if (panic_on_unrecovered_nmi
)
811 panic("NMI: Not continuing");
813 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
816 /* Runs on IST stack. This code must keep interrupts off all the time.
817 Nested NMIs are prevented by the CPU. */
818 asmlinkage __kprobes
void default_do_nmi(struct pt_regs
*regs
)
820 unsigned char reason
= 0;
823 cpu
= smp_processor_id();
825 /* Only the BSP gets external NMIs from the system. */
827 reason
= get_nmi_reason();
829 if (!(reason
& 0xc0)) {
830 if (notify_die(DIE_NMI_IPI
, "nmi_ipi", regs
, reason
, 2, SIGINT
)
834 * Ok, so this is none of the documented NMI sources,
835 * so it must be the NMI watchdog.
837 if (nmi_watchdog_tick(regs
,reason
))
839 if (!do_nmi_callback(regs
,cpu
))
840 unknown_nmi_error(reason
, regs
);
844 if (notify_die(DIE_NMI
, "nmi", regs
, reason
, 2, SIGINT
) == NOTIFY_STOP
)
847 /* AK: following checks seem to be broken on modern chipsets. FIXME */
850 mem_parity_error(reason
, regs
);
852 io_check_error(reason
, regs
);
855 /* runs on IST stack. */
856 asmlinkage
void __kprobes
do_int3(struct pt_regs
* regs
, long error_code
)
858 trace_hardirqs_fixup();
860 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, 3, SIGTRAP
) == NOTIFY_STOP
) {
863 preempt_conditional_sti(regs
);
864 do_trap(3, SIGTRAP
, "int3", regs
, error_code
, NULL
);
865 preempt_conditional_cli(regs
);
868 /* Help handler running on IST stack to switch back to user stack
869 for scheduling or signal handling. The actual stack switch is done in
871 asmlinkage __kprobes
struct pt_regs
*sync_regs(struct pt_regs
*eregs
)
873 struct pt_regs
*regs
= eregs
;
874 /* Did already sync */
875 if (eregs
== (struct pt_regs
*)eregs
->sp
)
877 /* Exception from user space */
878 else if (user_mode(eregs
))
879 regs
= task_pt_regs(current
);
880 /* Exception from kernel and interrupts are enabled. Move to
881 kernel process stack. */
882 else if (eregs
->flags
& X86_EFLAGS_IF
)
883 regs
= (struct pt_regs
*)(eregs
->sp
-= sizeof(struct pt_regs
));
889 /* runs on IST stack. */
890 asmlinkage
void __kprobes
do_debug(struct pt_regs
* regs
,
891 unsigned long error_code
)
893 unsigned long condition
;
894 struct task_struct
*tsk
= current
;
897 trace_hardirqs_fixup();
899 get_debugreg(condition
, 6);
902 * The processor cleared BTF, so don't mark that we need it set.
904 clear_tsk_thread_flag(tsk
, TIF_DEBUGCTLMSR
);
905 tsk
->thread
.debugctlmsr
= 0;
907 if (notify_die(DIE_DEBUG
, "debug", regs
, condition
, error_code
,
908 SIGTRAP
) == NOTIFY_STOP
)
911 preempt_conditional_sti(regs
);
913 /* Mask out spurious debug traps due to lazy DR7 setting */
914 if (condition
& (DR_TRAP0
|DR_TRAP1
|DR_TRAP2
|DR_TRAP3
)) {
915 if (!tsk
->thread
.debugreg7
) {
920 tsk
->thread
.debugreg6
= condition
;
924 * Single-stepping through TF: make sure we ignore any events in
925 * kernel space (but re-enable TF when returning to user mode).
927 if (condition
& DR_STEP
) {
928 if (!user_mode(regs
))
929 goto clear_TF_reenable
;
932 /* Ok, finally something we can handle */
933 tsk
->thread
.trap_no
= 1;
934 tsk
->thread
.error_code
= error_code
;
935 info
.si_signo
= SIGTRAP
;
937 info
.si_code
= TRAP_BRKPT
;
938 info
.si_addr
= user_mode(regs
) ? (void __user
*)regs
->ip
: NULL
;
939 force_sig_info(SIGTRAP
, &info
, tsk
);
942 set_debugreg(0UL, 7);
943 preempt_conditional_cli(regs
);
947 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
948 regs
->flags
&= ~X86_EFLAGS_TF
;
949 preempt_conditional_cli(regs
);
952 static int kernel_math_error(struct pt_regs
*regs
, const char *str
, int trapnr
)
954 if (fixup_exception(regs
))
957 notify_die(DIE_GPF
, str
, regs
, 0, trapnr
, SIGFPE
);
958 /* Illegal floating point operation in the kernel */
959 current
->thread
.trap_no
= trapnr
;
965 * Note that we play around with the 'TS' bit in an attempt to get
966 * the correct behaviour even in the presence of the asynchronous
969 asmlinkage
void do_coprocessor_error(struct pt_regs
*regs
)
971 void __user
*ip
= (void __user
*)(regs
->ip
);
972 struct task_struct
* task
;
974 unsigned short cwd
, swd
;
976 conditional_sti(regs
);
977 if (!user_mode(regs
) &&
978 kernel_math_error(regs
, "kernel x87 math error", 16))
982 * Save the info for the exception handler and clear the error.
986 task
->thread
.trap_no
= 16;
987 task
->thread
.error_code
= 0;
988 info
.si_signo
= SIGFPE
;
990 info
.si_code
= __SI_FAULT
;
993 * (~cwd & swd) will mask out exceptions that are not set to unmasked
994 * status. 0x3f is the exception bits in these regs, 0x200 is the
995 * C1 reg you need in case of a stack fault, 0x040 is the stack
996 * fault bit. We should only be taking one exception at a time,
997 * so if this combination doesn't produce any single exception,
998 * then we have a bad program that isn't synchronizing its FPU usage
999 * and it will suffer the consequences since we won't be able to
1000 * fully reproduce the context of the exception
1002 cwd
= get_fpu_cwd(task
);
1003 swd
= get_fpu_swd(task
);
1004 switch (swd
& ~cwd
& 0x3f) {
1008 case 0x001: /* Invalid Op */
1010 * swd & 0x240 == 0x040: Stack Underflow
1011 * swd & 0x240 == 0x240: Stack Overflow
1012 * User must clear the SF bit (0x40) if set
1014 info
.si_code
= FPE_FLTINV
;
1016 case 0x002: /* Denormalize */
1017 case 0x010: /* Underflow */
1018 info
.si_code
= FPE_FLTUND
;
1020 case 0x004: /* Zero Divide */
1021 info
.si_code
= FPE_FLTDIV
;
1023 case 0x008: /* Overflow */
1024 info
.si_code
= FPE_FLTOVF
;
1026 case 0x020: /* Precision */
1027 info
.si_code
= FPE_FLTRES
;
1030 force_sig_info(SIGFPE
, &info
, task
);
1033 asmlinkage
void bad_intr(void)
1035 printk("bad interrupt");
1038 asmlinkage
void do_simd_coprocessor_error(struct pt_regs
*regs
)
1040 void __user
*ip
= (void __user
*)(regs
->ip
);
1041 struct task_struct
* task
;
1043 unsigned short mxcsr
;
1045 conditional_sti(regs
);
1046 if (!user_mode(regs
) &&
1047 kernel_math_error(regs
, "kernel simd math error", 19))
1051 * Save the info for the exception handler and clear the error.
1054 save_init_fpu(task
);
1055 task
->thread
.trap_no
= 19;
1056 task
->thread
.error_code
= 0;
1057 info
.si_signo
= SIGFPE
;
1059 info
.si_code
= __SI_FAULT
;
1062 * The SIMD FPU exceptions are handled a little differently, as there
1063 * is only a single status/control register. Thus, to determine which
1064 * unmasked exception was caught we must mask the exception mask bits
1065 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1067 mxcsr
= get_fpu_mxcsr(task
);
1068 switch (~((mxcsr
& 0x1f80) >> 7) & (mxcsr
& 0x3f)) {
1072 case 0x001: /* Invalid Op */
1073 info
.si_code
= FPE_FLTINV
;
1075 case 0x002: /* Denormalize */
1076 case 0x010: /* Underflow */
1077 info
.si_code
= FPE_FLTUND
;
1079 case 0x004: /* Zero Divide */
1080 info
.si_code
= FPE_FLTDIV
;
1082 case 0x008: /* Overflow */
1083 info
.si_code
= FPE_FLTOVF
;
1085 case 0x020: /* Precision */
1086 info
.si_code
= FPE_FLTRES
;
1089 force_sig_info(SIGFPE
, &info
, task
);
1092 asmlinkage
void do_spurious_interrupt_bug(struct pt_regs
* regs
)
1096 asmlinkage
void __attribute__((weak
)) smp_thermal_interrupt(void)
1100 asmlinkage
void __attribute__((weak
)) mce_threshold_interrupt(void)
1105 * 'math_state_restore()' saves the current math information in the
1106 * old math state array, and gets the new ones from the current task
1108 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1109 * Don't touch unless you *really* know how it works.
1111 asmlinkage
void math_state_restore(void)
1113 struct task_struct
*me
= current
;
1114 clts(); /* Allow maths ops (or we recurse) */
1118 restore_fpu_checking(&me
->thread
.i387
.fxsave
);
1119 task_thread_info(me
)->status
|= TS_USEDFPU
;
1122 EXPORT_SYMBOL_GPL(math_state_restore
);
1124 void __init
trap_init(void)
1126 set_intr_gate(0,÷_error
);
1127 set_intr_gate_ist(1,&debug
,DEBUG_STACK
);
1128 set_intr_gate_ist(2,&nmi
,NMI_STACK
);
1129 set_system_gate_ist(3,&int3
,DEBUG_STACK
); /* int3 can be called from all */
1130 set_system_gate(4,&overflow
); /* int4 can be called from all */
1131 set_intr_gate(5,&bounds
);
1132 set_intr_gate(6,&invalid_op
);
1133 set_intr_gate(7,&device_not_available
);
1134 set_intr_gate_ist(8,&double_fault
, DOUBLEFAULT_STACK
);
1135 set_intr_gate(9,&coprocessor_segment_overrun
);
1136 set_intr_gate(10,&invalid_TSS
);
1137 set_intr_gate(11,&segment_not_present
);
1138 set_intr_gate_ist(12,&stack_segment
,STACKFAULT_STACK
);
1139 set_intr_gate(13,&general_protection
);
1140 set_intr_gate(14,&page_fault
);
1141 set_intr_gate(15,&spurious_interrupt_bug
);
1142 set_intr_gate(16,&coprocessor_error
);
1143 set_intr_gate(17,&alignment_check
);
1144 #ifdef CONFIG_X86_MCE
1145 set_intr_gate_ist(18,&machine_check
, MCE_STACK
);
1147 set_intr_gate(19,&simd_coprocessor_error
);
1149 #ifdef CONFIG_IA32_EMULATION
1150 set_system_gate(IA32_SYSCALL_VECTOR
, ia32_syscall
);
1154 * Should be a barrier for any external CPU state.
1160 static int __init
oops_setup(char *s
)
1164 if (!strcmp(s
, "panic"))
1168 early_param("oops", oops_setup
);
1170 static int __init
kstack_setup(char *s
)
1174 kstack_depth_to_print
= simple_strtoul(s
,NULL
,0);
1177 early_param("kstack", kstack_setup
);
1180 static int __init
code_bytes_setup(char *s
)
1182 code_bytes
= simple_strtoul(s
, NULL
, 0);
1183 if (code_bytes
> 8192)
1188 __setup("code_bytes=", code_bytes_setup
);