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/moduleparam.h>
14 #include <linux/interrupt.h>
15 #include <linux/kallsyms.h>
16 #include <linux/spinlock.h>
17 #include <linux/kprobes.h>
18 #include <linux/uaccess.h>
19 #include <linux/utsname.h>
20 #include <linux/kdebug.h>
21 #include <linux/kernel.h>
22 #include <linux/module.h>
23 #include <linux/ptrace.h>
24 #include <linux/string.h>
25 #include <linux/unwind.h>
26 #include <linux/delay.h>
27 #include <linux/errno.h>
28 #include <linux/kexec.h>
29 #include <linux/sched.h>
30 #include <linux/timer.h>
31 #include <linux/init.h>
32 #include <linux/bug.h>
33 #include <linux/nmi.h>
35 #include <linux/smp.h>
38 #if defined(CONFIG_EDAC)
39 #include <linux/edac.h>
42 #include <asm/stacktrace.h>
43 #include <asm/processor.h>
44 #include <asm/debugreg.h>
45 #include <asm/atomic.h>
46 #include <asm/system.h>
47 #include <asm/unwind.h>
50 #include <asm/pgalloc.h>
51 #include <asm/proto.h>
53 #include <asm/traps.h>
55 #include <mach_traps.h>
57 int panic_on_unrecovered_nmi
;
58 int kstack_depth_to_print
= 12;
59 static unsigned int code_bytes
= 64;
60 static int ignore_nmis
;
61 static int die_counter
;
63 static inline void conditional_sti(struct pt_regs
*regs
)
65 if (regs
->flags
& X86_EFLAGS_IF
)
69 static inline void preempt_conditional_sti(struct pt_regs
*regs
)
72 if (regs
->flags
& X86_EFLAGS_IF
)
76 static inline void preempt_conditional_cli(struct pt_regs
*regs
)
78 if (regs
->flags
& X86_EFLAGS_IF
)
80 /* Make sure to not schedule here because we could be running
81 on an exception stack. */
85 void printk_address(unsigned long address
, int reliable
)
87 printk(" [<%016lx>] %s%pS\n",
88 address
, reliable
? "" : "? ", (void *) address
);
91 static unsigned long *in_exception_stack(unsigned cpu
, unsigned long stack
,
92 unsigned *usedp
, char **idp
)
94 static char ids
[][8] = {
95 [DEBUG_STACK
- 1] = "#DB",
96 [NMI_STACK
- 1] = "NMI",
97 [DOUBLEFAULT_STACK
- 1] = "#DF",
98 [STACKFAULT_STACK
- 1] = "#SS",
99 [MCE_STACK
- 1] = "#MC",
100 #if DEBUG_STKSZ > EXCEPTION_STKSZ
101 [N_EXCEPTION_STACKS
...
102 N_EXCEPTION_STACKS
+ DEBUG_STKSZ
/ EXCEPTION_STKSZ
- 2] = "#DB[?]"
108 * Iterate over all exception stacks, and figure out whether
109 * 'stack' is in one of them:
111 for (k
= 0; k
< N_EXCEPTION_STACKS
; k
++) {
112 unsigned long end
= per_cpu(orig_ist
, cpu
).ist
[k
];
114 * Is 'stack' above this exception frame's end?
115 * If yes then skip to the next frame.
120 * Is 'stack' above this exception frame's start address?
121 * If yes then we found the right frame.
123 if (stack
>= end
- EXCEPTION_STKSZ
) {
125 * Make sure we only iterate through an exception
126 * stack once. If it comes up for the second time
127 * then there's something wrong going on - just
128 * break out and return NULL:
130 if (*usedp
& (1U << k
))
134 return (unsigned long *)end
;
137 * If this is a debug stack, and if it has a larger size than
138 * the usual exception stacks, then 'stack' might still
139 * be within the lower portion of the debug stack:
141 #if DEBUG_STKSZ > EXCEPTION_STKSZ
142 if (k
== DEBUG_STACK
- 1 && stack
>= end
- DEBUG_STKSZ
) {
143 unsigned j
= N_EXCEPTION_STACKS
- 1;
146 * Black magic. A large debug stack is composed of
147 * multiple exception stack entries, which we
148 * iterate through now. Dont look:
152 end
-= EXCEPTION_STKSZ
;
153 ids
[j
][4] = '1' + (j
- N_EXCEPTION_STACKS
);
154 } while (stack
< end
- EXCEPTION_STKSZ
);
155 if (*usedp
& (1U << j
))
159 return (unsigned long *)end
;
167 * x86-64 can have up to three kernel stacks:
170 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
173 static inline int valid_stack_ptr(struct thread_info
*tinfo
,
174 void *p
, unsigned int size
, void *end
)
178 if (p
< end
&& p
>= (end
-THREAD_SIZE
))
183 return p
> t
&& p
< t
+ THREAD_SIZE
- size
;
186 /* The form of the top of the frame on the stack */
188 struct stack_frame
*next_frame
;
189 unsigned long return_address
;
192 static inline unsigned long
193 print_context_stack(struct thread_info
*tinfo
,
194 unsigned long *stack
, unsigned long bp
,
195 const struct stacktrace_ops
*ops
, void *data
,
198 struct stack_frame
*frame
= (struct stack_frame
*)bp
;
200 while (valid_stack_ptr(tinfo
, stack
, sizeof(*stack
), end
)) {
204 if (__kernel_text_address(addr
)) {
205 if ((unsigned long) stack
== bp
+ 8) {
206 ops
->address(data
, addr
, 1);
207 frame
= frame
->next_frame
;
208 bp
= (unsigned long) frame
;
210 ops
->address(data
, addr
, bp
== 0);
218 void dump_trace(struct task_struct
*task
, struct pt_regs
*regs
,
219 unsigned long *stack
, unsigned long bp
,
220 const struct stacktrace_ops
*ops
, void *data
)
222 const unsigned cpu
= get_cpu();
223 unsigned long *irqstack_end
= (unsigned long *)cpu_pda(cpu
)->irqstackptr
;
225 struct thread_info
*tinfo
;
233 if (task
&& task
!= current
)
234 stack
= (unsigned long *)task
->thread
.sp
;
237 #ifdef CONFIG_FRAME_POINTER
239 if (task
== current
) {
240 /* Grab bp right from our regs */
241 asm("movq %%rbp, %0" : "=r" (bp
) : );
243 /* bp is the last reg pushed by switch_to */
244 bp
= *(unsigned long *) task
->thread
.sp
;
250 * Print function call entries in all stacks, starting at the
251 * current stack address. If the stacks consist of nested
254 tinfo
= task_thread_info(task
);
257 unsigned long *estack_end
;
258 estack_end
= in_exception_stack(cpu
, (unsigned long)stack
,
262 if (ops
->stack(data
, id
) < 0)
265 bp
= print_context_stack(tinfo
, stack
, bp
, ops
,
267 ops
->stack(data
, "<EOE>");
269 * We link to the next stack via the
270 * second-to-last pointer (index -2 to end) in the
273 stack
= (unsigned long *) estack_end
[-2];
277 unsigned long *irqstack
;
278 irqstack
= irqstack_end
-
279 (IRQSTACKSIZE
- 64) / sizeof(*irqstack
);
281 if (stack
>= irqstack
&& stack
< irqstack_end
) {
282 if (ops
->stack(data
, "IRQ") < 0)
284 bp
= print_context_stack(tinfo
, stack
, bp
,
285 ops
, data
, irqstack_end
);
287 * We link to the next stack (which would be
288 * the process stack normally) the last
289 * pointer (index -1 to end) in the IRQ stack:
291 stack
= (unsigned long *) (irqstack_end
[-1]);
293 ops
->stack(data
, "EOI");
301 * This handles the process stack:
303 bp
= print_context_stack(tinfo
, stack
, bp
, ops
, data
, NULL
);
306 EXPORT_SYMBOL(dump_trace
);
309 print_trace_warning_symbol(void *data
, char *msg
, unsigned long symbol
)
311 print_symbol(msg
, symbol
);
315 static void print_trace_warning(void *data
, char *msg
)
320 static int print_trace_stack(void *data
, char *name
)
322 printk(" <%s> ", name
);
326 static void print_trace_address(void *data
, unsigned long addr
, int reliable
)
328 touch_nmi_watchdog();
329 printk_address(addr
, reliable
);
332 static const struct stacktrace_ops print_trace_ops
= {
333 .warning
= print_trace_warning
,
334 .warning_symbol
= print_trace_warning_symbol
,
335 .stack
= print_trace_stack
,
336 .address
= print_trace_address
,
340 show_trace_log_lvl(struct task_struct
*task
, struct pt_regs
*regs
,
341 unsigned long *stack
, unsigned long bp
, char *log_lvl
)
343 printk("Call Trace:\n");
344 dump_trace(task
, regs
, stack
, bp
, &print_trace_ops
, log_lvl
);
347 void show_trace(struct task_struct
*task
, struct pt_regs
*regs
,
348 unsigned long *stack
, unsigned long bp
)
350 show_trace_log_lvl(task
, regs
, stack
, bp
, "");
354 show_stack_log_lvl(struct task_struct
*task
, struct pt_regs
*regs
,
355 unsigned long *sp
, unsigned long bp
, char *log_lvl
)
357 unsigned long *stack
;
359 const int cpu
= smp_processor_id();
360 unsigned long *irqstack_end
=
361 (unsigned long *) (cpu_pda(cpu
)->irqstackptr
);
362 unsigned long *irqstack
=
363 (unsigned long *) (cpu_pda(cpu
)->irqstackptr
- IRQSTACKSIZE
);
366 * debugging aid: "show_stack(NULL, NULL);" prints the
367 * back trace for this cpu.
372 sp
= (unsigned long *)task
->thread
.sp
;
374 sp
= (unsigned long *)&sp
;
378 for (i
= 0; i
< kstack_depth_to_print
; i
++) {
379 if (stack
>= irqstack
&& stack
<= irqstack_end
) {
380 if (stack
== irqstack_end
) {
381 stack
= (unsigned long *) (irqstack_end
[-1]);
385 if (((long) stack
& (THREAD_SIZE
-1)) == 0)
388 if (i
&& ((i
% 4) == 0))
390 printk(" %016lx", *stack
++);
391 touch_nmi_watchdog();
394 show_trace_log_lvl(task
, regs
, sp
, bp
, log_lvl
);
397 void show_stack(struct task_struct
*task
, unsigned long *sp
)
399 show_stack_log_lvl(task
, NULL
, sp
, 0, "");
403 * The architecture-independent dump_stack generator
405 void dump_stack(void)
407 unsigned long bp
= 0;
410 #ifdef CONFIG_FRAME_POINTER
412 asm("movq %%rbp, %0" : "=r" (bp
) : );
415 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
416 current
->pid
, current
->comm
, print_tainted(),
417 init_utsname()->release
,
418 (int)strcspn(init_utsname()->version
, " "),
419 init_utsname()->version
);
420 show_trace(NULL
, NULL
, &stack
, bp
);
422 EXPORT_SYMBOL(dump_stack
);
424 void show_registers(struct pt_regs
*regs
)
428 const int cpu
= smp_processor_id();
429 struct task_struct
*cur
= cpu_pda(cpu
)->pcurrent
;
432 printk("CPU %d ", cpu
);
434 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
435 cur
->comm
, cur
->pid
, task_thread_info(cur
), cur
);
438 * When in-kernel, we also print out the stack and code at the
439 * time of the fault..
441 if (!user_mode(regs
)) {
442 unsigned int code_prologue
= code_bytes
* 43 / 64;
443 unsigned int code_len
= code_bytes
;
448 show_stack_log_lvl(NULL
, regs
, (unsigned long *)sp
,
451 printk(KERN_EMERG
"Code: ");
453 ip
= (u8
*)regs
->ip
- code_prologue
;
454 if (ip
< (u8
*)PAGE_OFFSET
|| probe_kernel_address(ip
, c
)) {
455 /* try starting at RIP */
457 code_len
= code_len
- code_prologue
+ 1;
459 for (i
= 0; i
< code_len
; i
++, ip
++) {
460 if (ip
< (u8
*)PAGE_OFFSET
||
461 probe_kernel_address(ip
, c
)) {
462 printk(" Bad RIP value.");
465 if (ip
== (u8
*)regs
->ip
)
466 printk("<%02x> ", c
);
474 int is_valid_bugaddr(unsigned long ip
)
478 if (__copy_from_user(&ud2
, (const void __user
*) ip
, sizeof(ud2
)))
481 return ud2
== 0x0b0f;
484 static raw_spinlock_t die_lock
= __RAW_SPIN_LOCK_UNLOCKED
;
485 static int die_owner
= -1;
486 static unsigned int die_nest_count
;
488 unsigned __kprobes
long oops_begin(void)
495 /* racy, but better than risking deadlock. */
496 raw_local_irq_save(flags
);
497 cpu
= smp_processor_id();
498 if (!__raw_spin_trylock(&die_lock
)) {
499 if (cpu
== die_owner
)
500 /* nested oops. should stop eventually */;
502 __raw_spin_lock(&die_lock
);
511 void __kprobes
oops_end(unsigned long flags
, struct pt_regs
*regs
, int signr
)
517 /* Nest count reaches zero, release the lock. */
518 __raw_spin_unlock(&die_lock
);
519 raw_local_irq_restore(flags
);
525 panic("Fatal exception");
530 int __kprobes
__die(const char *str
, struct pt_regs
*regs
, long err
)
532 printk(KERN_EMERG
"%s: %04lx [%u] ", str
, err
& 0xffff, ++die_counter
);
533 #ifdef CONFIG_PREEMPT
539 #ifdef CONFIG_DEBUG_PAGEALLOC
540 printk("DEBUG_PAGEALLOC");
543 if (notify_die(DIE_OOPS
, str
, regs
, err
,
544 current
->thread
.trap_no
, SIGSEGV
) == NOTIFY_STOP
)
547 show_registers(regs
);
548 add_taint(TAINT_DIE
);
549 /* Executive summary in case the oops scrolled away */
550 printk(KERN_ALERT
"RIP ");
551 printk_address(regs
->ip
, 1);
552 printk(" RSP <%016lx>\n", regs
->sp
);
553 if (kexec_should_crash(current
))
558 void die(const char *str
, struct pt_regs
*regs
, long err
)
560 unsigned long flags
= oops_begin();
562 if (!user_mode(regs
))
563 report_bug(regs
->ip
, regs
);
565 if (__die(str
, regs
, err
))
567 oops_end(flags
, regs
, SIGSEGV
);
570 notrace __kprobes
void
571 die_nmi(char *str
, struct pt_regs
*regs
, int do_panic
)
575 if (notify_die(DIE_NMIWATCHDOG
, str
, regs
, 0, 2, SIGINT
) == NOTIFY_STOP
)
578 flags
= oops_begin();
580 * We are in trouble anyway, lets at least try
581 * to get a message out.
583 printk(KERN_EMERG
"%s", str
);
584 printk(" on CPU%d, ip %08lx, registers:\n",
585 smp_processor_id(), regs
->ip
);
586 show_registers(regs
);
587 if (kexec_should_crash(current
))
589 if (do_panic
|| panic_on_oops
)
590 panic("Non maskable interrupt");
591 oops_end(flags
, NULL
, SIGBUS
);
597 static void __kprobes
598 do_trap(int trapnr
, int signr
, char *str
, struct pt_regs
*regs
,
599 long error_code
, siginfo_t
*info
)
601 struct task_struct
*tsk
= current
;
603 if (!user_mode(regs
))
607 * We want error_code and trap_no set for userspace faults and
608 * kernelspace faults which result in die(), but not
609 * kernelspace faults which are fixed up. die() gives the
610 * process no chance to handle the signal and notice the
611 * kernel fault information, so that won't result in polluting
612 * the information about previously queued, but not yet
613 * delivered, faults. See also do_general_protection below.
615 tsk
->thread
.error_code
= error_code
;
616 tsk
->thread
.trap_no
= trapnr
;
618 if (show_unhandled_signals
&& unhandled_signal(tsk
, signr
) &&
619 printk_ratelimit()) {
621 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
622 tsk
->comm
, tsk
->pid
, str
,
623 regs
->ip
, regs
->sp
, error_code
);
624 print_vma_addr(" in ", regs
->ip
);
629 force_sig_info(signr
, info
, tsk
);
631 force_sig(signr
, tsk
);
635 if (!fixup_exception(regs
)) {
636 tsk
->thread
.error_code
= error_code
;
637 tsk
->thread
.trap_no
= trapnr
;
638 die(str
, regs
, error_code
);
643 #define DO_ERROR(trapnr, signr, str, name) \
644 asmlinkage void do_##name(struct pt_regs *regs, long error_code) \
646 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
649 conditional_sti(regs); \
650 do_trap(trapnr, signr, str, regs, error_code, NULL); \
653 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
654 asmlinkage void do_##name(struct pt_regs *regs, long error_code) \
657 info.si_signo = signr; \
659 info.si_code = sicode; \
660 info.si_addr = (void __user *)siaddr; \
661 trace_hardirqs_fixup(); \
662 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
665 conditional_sti(regs); \
666 do_trap(trapnr, signr, str, regs, error_code, &info); \
669 DO_ERROR_INFO(0, SIGFPE
, "divide error", divide_error
, FPE_INTDIV
, regs
->ip
)
670 DO_ERROR(4, SIGSEGV
, "overflow", overflow
)
671 DO_ERROR(5, SIGSEGV
, "bounds", bounds
)
672 DO_ERROR_INFO(6, SIGILL
, "invalid opcode", invalid_op
, ILL_ILLOPN
, regs
->ip
)
673 DO_ERROR(9, SIGFPE
, "coprocessor segment overrun", coprocessor_segment_overrun
)
674 DO_ERROR(10, SIGSEGV
, "invalid TSS", invalid_TSS
)
675 DO_ERROR(11, SIGBUS
, "segment not present", segment_not_present
)
676 DO_ERROR_INFO(17, SIGBUS
, "alignment check", alignment_check
, BUS_ADRALN
, 0)
678 /* Runs on IST stack */
679 asmlinkage
void do_stack_segment(struct pt_regs
*regs
, long error_code
)
681 if (notify_die(DIE_TRAP
, "stack segment", regs
, error_code
,
682 12, SIGBUS
) == NOTIFY_STOP
)
684 preempt_conditional_sti(regs
);
685 do_trap(12, SIGBUS
, "stack segment", regs
, error_code
, NULL
);
686 preempt_conditional_cli(regs
);
689 asmlinkage
void do_double_fault(struct pt_regs
*regs
, long error_code
)
691 static const char str
[] = "double fault";
692 struct task_struct
*tsk
= current
;
694 /* Return not checked because double check cannot be ignored */
695 notify_die(DIE_TRAP
, str
, regs
, error_code
, 8, SIGSEGV
);
697 tsk
->thread
.error_code
= error_code
;
698 tsk
->thread
.trap_no
= 8;
700 /* This is always a kernel trap and never fixable (and thus must
703 die(str
, regs
, error_code
);
706 asmlinkage
void __kprobes
707 do_general_protection(struct pt_regs
*regs
, long error_code
)
709 struct task_struct
*tsk
;
711 conditional_sti(regs
);
714 if (!user_mode(regs
))
717 tsk
->thread
.error_code
= error_code
;
718 tsk
->thread
.trap_no
= 13;
720 if (show_unhandled_signals
&& unhandled_signal(tsk
, SIGSEGV
) &&
721 printk_ratelimit()) {
723 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
725 regs
->ip
, regs
->sp
, error_code
);
726 print_vma_addr(" in ", regs
->ip
);
730 force_sig(SIGSEGV
, tsk
);
734 if (fixup_exception(regs
))
737 tsk
->thread
.error_code
= error_code
;
738 tsk
->thread
.trap_no
= 13;
739 if (notify_die(DIE_GPF
, "general protection fault", regs
,
740 error_code
, 13, SIGSEGV
) == NOTIFY_STOP
)
742 die("general protection fault", regs
, error_code
);
745 static notrace __kprobes
void
746 mem_parity_error(unsigned char reason
, struct pt_regs
*regs
)
748 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x.\n",
750 printk(KERN_EMERG
"You have some hardware problem, likely on the PCI bus.\n");
752 #if defined(CONFIG_EDAC)
753 if (edac_handler_set()) {
754 edac_atomic_assert_error();
759 if (panic_on_unrecovered_nmi
)
760 panic("NMI: Not continuing");
762 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
764 /* Clear and disable the memory parity error line. */
765 reason
= (reason
& 0xf) | 4;
769 static notrace __kprobes
void
770 io_check_error(unsigned char reason
, struct pt_regs
*regs
)
772 printk("NMI: IOCK error (debug interrupt?)\n");
773 show_registers(regs
);
775 /* Re-enable the IOCK line, wait for a few seconds */
776 reason
= (reason
& 0xf) | 8;
783 static notrace __kprobes
void
784 unknown_nmi_error(unsigned char reason
, struct pt_regs
*regs
)
786 if (notify_die(DIE_NMIUNKNOWN
, "nmi", regs
, reason
, 2, SIGINT
) ==
789 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x.\n",
791 printk(KERN_EMERG
"Do you have a strange power saving mode enabled?\n");
793 if (panic_on_unrecovered_nmi
)
794 panic("NMI: Not continuing");
796 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
799 /* Runs on IST stack. This code must keep interrupts off all the time.
800 Nested NMIs are prevented by the CPU. */
801 asmlinkage notrace __kprobes
void default_do_nmi(struct pt_regs
*regs
)
803 unsigned char reason
= 0;
806 cpu
= smp_processor_id();
808 /* Only the BSP gets external NMIs from the system. */
810 reason
= get_nmi_reason();
812 if (!(reason
& 0xc0)) {
813 if (notify_die(DIE_NMI_IPI
, "nmi_ipi", regs
, reason
, 2, SIGINT
)
817 * Ok, so this is none of the documented NMI sources,
818 * so it must be the NMI watchdog.
820 if (nmi_watchdog_tick(regs
, reason
))
822 if (!do_nmi_callback(regs
, cpu
))
823 unknown_nmi_error(reason
, regs
);
827 if (notify_die(DIE_NMI
, "nmi", regs
, reason
, 2, SIGINT
) == NOTIFY_STOP
)
830 /* AK: following checks seem to be broken on modern chipsets. FIXME */
832 mem_parity_error(reason
, regs
);
834 io_check_error(reason
, regs
);
837 asmlinkage notrace __kprobes
void
838 do_nmi(struct pt_regs
*regs
, long error_code
)
842 add_pda(__nmi_count
, 1);
845 default_do_nmi(regs
);
856 void restart_nmi(void)
862 /* runs on IST stack. */
863 asmlinkage
void __kprobes
do_int3(struct pt_regs
*regs
, long error_code
)
865 trace_hardirqs_fixup();
867 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, 3, SIGTRAP
)
871 preempt_conditional_sti(regs
);
872 do_trap(3, SIGTRAP
, "int3", regs
, error_code
, NULL
);
873 preempt_conditional_cli(regs
);
876 /* Help handler running on IST stack to switch back to user stack
877 for scheduling or signal handling. The actual stack switch is done in
879 asmlinkage __kprobes
struct pt_regs
*sync_regs(struct pt_regs
*eregs
)
881 struct pt_regs
*regs
= eregs
;
882 /* Did already sync */
883 if (eregs
== (struct pt_regs
*)eregs
->sp
)
885 /* Exception from user space */
886 else if (user_mode(eregs
))
887 regs
= task_pt_regs(current
);
888 /* Exception from kernel and interrupts are enabled. Move to
889 kernel process stack. */
890 else if (eregs
->flags
& X86_EFLAGS_IF
)
891 regs
= (struct pt_regs
*)(eregs
->sp
-= sizeof(struct pt_regs
));
897 /* runs on IST stack. */
898 asmlinkage
void __kprobes
do_debug(struct pt_regs
*regs
,
899 unsigned long error_code
)
901 struct task_struct
*tsk
= current
;
902 unsigned long condition
;
905 trace_hardirqs_fixup();
907 get_debugreg(condition
, 6);
910 * The processor cleared BTF, so don't mark that we need it set.
912 clear_tsk_thread_flag(tsk
, TIF_DEBUGCTLMSR
);
913 tsk
->thread
.debugctlmsr
= 0;
915 if (notify_die(DIE_DEBUG
, "debug", regs
, condition
, error_code
,
916 SIGTRAP
) == NOTIFY_STOP
)
919 preempt_conditional_sti(regs
);
921 /* Mask out spurious debug traps due to lazy DR7 setting */
922 if (condition
& (DR_TRAP0
|DR_TRAP1
|DR_TRAP2
|DR_TRAP3
)) {
923 if (!tsk
->thread
.debugreg7
)
927 tsk
->thread
.debugreg6
= condition
;
930 * Single-stepping through TF: make sure we ignore any events in
931 * kernel space (but re-enable TF when returning to user mode).
933 if (condition
& DR_STEP
) {
934 if (!user_mode(regs
))
935 goto clear_TF_reenable
;
938 /* Ok, finally something we can handle */
939 tsk
->thread
.trap_no
= 1;
940 tsk
->thread
.error_code
= error_code
;
941 info
.si_signo
= SIGTRAP
;
943 info
.si_code
= TRAP_BRKPT
;
944 info
.si_addr
= user_mode(regs
) ? (void __user
*)regs
->ip
: NULL
;
945 force_sig_info(SIGTRAP
, &info
, tsk
);
949 preempt_conditional_cli(regs
);
953 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
954 regs
->flags
&= ~X86_EFLAGS_TF
;
955 preempt_conditional_cli(regs
);
959 static int kernel_math_error(struct pt_regs
*regs
, const char *str
, int trapnr
)
961 if (fixup_exception(regs
))
964 notify_die(DIE_GPF
, str
, regs
, 0, trapnr
, SIGFPE
);
965 /* Illegal floating point operation in the kernel */
966 current
->thread
.trap_no
= trapnr
;
972 * Note that we play around with the 'TS' bit in an attempt to get
973 * the correct behaviour even in the presence of the asynchronous
976 asmlinkage
void do_coprocessor_error(struct pt_regs
*regs
)
978 void __user
*ip
= (void __user
*)(regs
->ip
);
979 struct task_struct
*task
;
981 unsigned short cwd
, swd
;
983 conditional_sti(regs
);
984 if (!user_mode(regs
) &&
985 kernel_math_error(regs
, "kernel x87 math error", 16))
989 * Save the info for the exception handler and clear the error.
993 task
->thread
.trap_no
= 16;
994 task
->thread
.error_code
= 0;
995 info
.si_signo
= SIGFPE
;
997 info
.si_code
= __SI_FAULT
;
1000 * (~cwd & swd) will mask out exceptions that are not set to unmasked
1001 * status. 0x3f is the exception bits in these regs, 0x200 is the
1002 * C1 reg you need in case of a stack fault, 0x040 is the stack
1003 * fault bit. We should only be taking one exception at a time,
1004 * so if this combination doesn't produce any single exception,
1005 * then we have a bad program that isn't synchronizing its FPU usage
1006 * and it will suffer the consequences since we won't be able to
1007 * fully reproduce the context of the exception
1009 cwd
= get_fpu_cwd(task
);
1010 swd
= get_fpu_swd(task
);
1011 switch (swd
& ~cwd
& 0x3f) {
1012 case 0x000: /* No unmasked exception */
1013 default: /* Multiple exceptions */
1015 case 0x001: /* Invalid Op */
1017 * swd & 0x240 == 0x040: Stack Underflow
1018 * swd & 0x240 == 0x240: Stack Overflow
1019 * User must clear the SF bit (0x40) if set
1021 info
.si_code
= FPE_FLTINV
;
1023 case 0x002: /* Denormalize */
1024 case 0x010: /* Underflow */
1025 info
.si_code
= FPE_FLTUND
;
1027 case 0x004: /* Zero Divide */
1028 info
.si_code
= FPE_FLTDIV
;
1030 case 0x008: /* Overflow */
1031 info
.si_code
= FPE_FLTOVF
;
1033 case 0x020: /* Precision */
1034 info
.si_code
= FPE_FLTRES
;
1037 force_sig_info(SIGFPE
, &info
, task
);
1040 asmlinkage
void bad_intr(void)
1042 printk("bad interrupt");
1045 asmlinkage
void do_simd_coprocessor_error(struct pt_regs
*regs
)
1047 void __user
*ip
= (void __user
*)(regs
->ip
);
1048 struct task_struct
*task
;
1050 unsigned short mxcsr
;
1052 conditional_sti(regs
);
1053 if (!user_mode(regs
) &&
1054 kernel_math_error(regs
, "kernel simd math error", 19))
1058 * Save the info for the exception handler and clear the error.
1061 save_init_fpu(task
);
1062 task
->thread
.trap_no
= 19;
1063 task
->thread
.error_code
= 0;
1064 info
.si_signo
= SIGFPE
;
1066 info
.si_code
= __SI_FAULT
;
1069 * The SIMD FPU exceptions are handled a little differently, as there
1070 * is only a single status/control register. Thus, to determine which
1071 * unmasked exception was caught we must mask the exception mask bits
1072 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1074 mxcsr
= get_fpu_mxcsr(task
);
1075 switch (~((mxcsr
& 0x1f80) >> 7) & (mxcsr
& 0x3f)) {
1079 case 0x001: /* Invalid Op */
1080 info
.si_code
= FPE_FLTINV
;
1082 case 0x002: /* Denormalize */
1083 case 0x010: /* Underflow */
1084 info
.si_code
= FPE_FLTUND
;
1086 case 0x004: /* Zero Divide */
1087 info
.si_code
= FPE_FLTDIV
;
1089 case 0x008: /* Overflow */
1090 info
.si_code
= FPE_FLTOVF
;
1092 case 0x020: /* Precision */
1093 info
.si_code
= FPE_FLTRES
;
1096 force_sig_info(SIGFPE
, &info
, task
);
1099 asmlinkage
void do_spurious_interrupt_bug(struct pt_regs
*regs
)
1103 asmlinkage
void __attribute__((weak
)) smp_thermal_interrupt(void)
1107 asmlinkage
void __attribute__((weak
)) mce_threshold_interrupt(void)
1112 * 'math_state_restore()' saves the current math information in the
1113 * old math state array, and gets the new ones from the current task
1115 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1116 * Don't touch unless you *really* know how it works.
1118 asmlinkage
void math_state_restore(void)
1120 struct task_struct
*me
= current
;
1125 * does a slab alloc which can sleep
1129 * ran out of memory!
1131 do_group_exit(SIGKILL
);
1134 local_irq_disable();
1137 clts(); /* Allow maths ops (or we recurse) */
1139 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
1141 if (unlikely(restore_fpu_checking(me
))) {
1143 force_sig(SIGSEGV
, me
);
1146 task_thread_info(me
)->status
|= TS_USEDFPU
;
1149 EXPORT_SYMBOL_GPL(math_state_restore
);
1151 void __init
trap_init(void)
1153 set_intr_gate(0, ÷_error
);
1154 set_intr_gate_ist(1, &debug
, DEBUG_STACK
);
1155 set_intr_gate_ist(2, &nmi
, NMI_STACK
);
1156 /* int3 can be called from all */
1157 set_system_gate_ist(3, &int3
, DEBUG_STACK
);
1158 /* int4 can be called from all */
1159 set_system_gate(4, &overflow
);
1160 set_intr_gate(5, &bounds
);
1161 set_intr_gate(6, &invalid_op
);
1162 set_intr_gate(7, &device_not_available
);
1163 set_intr_gate_ist(8, &double_fault
, DOUBLEFAULT_STACK
);
1164 set_intr_gate(9, &coprocessor_segment_overrun
);
1165 set_intr_gate(10, &invalid_TSS
);
1166 set_intr_gate(11, &segment_not_present
);
1167 set_intr_gate_ist(12, &stack_segment
, STACKFAULT_STACK
);
1168 set_intr_gate(13, &general_protection
);
1169 set_intr_gate(14, &page_fault
);
1170 set_intr_gate(15, &spurious_interrupt_bug
);
1171 set_intr_gate(16, &coprocessor_error
);
1172 set_intr_gate(17, &alignment_check
);
1173 #ifdef CONFIG_X86_MCE
1174 set_intr_gate_ist(18, &machine_check
, MCE_STACK
);
1176 set_intr_gate(19, &simd_coprocessor_error
);
1178 #ifdef CONFIG_IA32_EMULATION
1179 set_system_gate(IA32_SYSCALL_VECTOR
, ia32_syscall
);
1182 * Should be a barrier for any external CPU state:
1187 static int __init
oops_setup(char *s
)
1191 if (!strcmp(s
, "panic"))
1195 early_param("oops", oops_setup
);
1197 static int __init
kstack_setup(char *s
)
1201 kstack_depth_to_print
= simple_strtoul(s
, NULL
, 0);
1204 early_param("kstack", kstack_setup
);
1206 static int __init
code_bytes_setup(char *s
)
1208 code_bytes
= simple_strtoul(s
, NULL
, 0);
1209 if (code_bytes
> 8192)
1214 __setup("code_bytes=", code_bytes_setup
);