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 void printk_address(unsigned long address
, int reliable
)
105 #ifdef CONFIG_KALLSYMS
106 unsigned long offset
= 0, symsize
;
110 char namebuf
[KSYM_NAME_LEN
];
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
);
127 printk(" [<%016lx>]\n", address
);
131 static unsigned long *in_exception_stack(unsigned cpu
, unsigned long stack
,
132 unsigned *usedp
, char **idp
)
134 static char ids
[][8] = {
135 [DEBUG_STACK
- 1] = "#DB",
136 [NMI_STACK
- 1] = "NMI",
137 [DOUBLEFAULT_STACK
- 1] = "#DF",
138 [STACKFAULT_STACK
- 1] = "#SS",
139 [MCE_STACK
- 1] = "#MC",
140 #if DEBUG_STKSZ > EXCEPTION_STKSZ
141 [N_EXCEPTION_STACKS
... N_EXCEPTION_STACKS
+ DEBUG_STKSZ
/ EXCEPTION_STKSZ
- 2] = "#DB[?]"
147 * Iterate over all exception stacks, and figure out whether
148 * 'stack' is in one of them:
150 for (k
= 0; k
< N_EXCEPTION_STACKS
; k
++) {
151 unsigned long end
= per_cpu(orig_ist
, cpu
).ist
[k
];
153 * Is 'stack' above this exception frame's end?
154 * If yes then skip to the next frame.
159 * Is 'stack' above this exception frame's start address?
160 * If yes then we found the right frame.
162 if (stack
>= end
- EXCEPTION_STKSZ
) {
164 * Make sure we only iterate through an exception
165 * stack once. If it comes up for the second time
166 * then there's something wrong going on - just
167 * break out and return NULL:
169 if (*usedp
& (1U << k
))
173 return (unsigned long *)end
;
176 * If this is a debug stack, and if it has a larger size than
177 * the usual exception stacks, then 'stack' might still
178 * be within the lower portion of the debug stack:
180 #if DEBUG_STKSZ > EXCEPTION_STKSZ
181 if (k
== DEBUG_STACK
- 1 && stack
>= end
- DEBUG_STKSZ
) {
182 unsigned j
= N_EXCEPTION_STACKS
- 1;
185 * Black magic. A large debug stack is composed of
186 * multiple exception stack entries, which we
187 * iterate through now. Dont look:
191 end
-= EXCEPTION_STKSZ
;
192 ids
[j
][4] = '1' + (j
- N_EXCEPTION_STACKS
);
193 } while (stack
< end
- EXCEPTION_STKSZ
);
194 if (*usedp
& (1U << j
))
198 return (unsigned long *)end
;
205 #define MSG(txt) ops->warning(data, txt)
208 * x86-64 can have up to three kernel stacks:
211 * severe exception (double fault, nmi, stack fault, debug, mce) hardware stack
214 static inline int valid_stack_ptr(struct thread_info
*tinfo
,
215 void *p
, unsigned int size
, void *end
)
219 if (p
< end
&& p
>= (end
-THREAD_SIZE
))
224 return p
> t
&& p
< t
+ THREAD_SIZE
- size
;
227 /* The form of the top of the frame on the stack */
229 struct stack_frame
*next_frame
;
230 unsigned long return_address
;
234 static inline unsigned long print_context_stack(struct thread_info
*tinfo
,
235 unsigned long *stack
, unsigned long bp
,
236 const struct stacktrace_ops
*ops
, void *data
,
239 struct stack_frame
*frame
= (struct stack_frame
*)bp
;
241 while (valid_stack_ptr(tinfo
, stack
, sizeof(*stack
), end
)) {
245 if (__kernel_text_address(addr
)) {
246 if ((unsigned long) stack
== bp
+ 8) {
247 ops
->address(data
, addr
, 1);
248 frame
= frame
->next_frame
;
249 bp
= (unsigned long) frame
;
251 ops
->address(data
, addr
, bp
== 0);
259 void dump_trace(struct task_struct
*tsk
, struct pt_regs
*regs
,
260 unsigned long *stack
, unsigned long bp
,
261 const struct stacktrace_ops
*ops
, void *data
)
263 const unsigned cpu
= get_cpu();
264 unsigned long *irqstack_end
= (unsigned long*)cpu_pda(cpu
)->irqstackptr
;
266 struct thread_info
*tinfo
;
270 tinfo
= task_thread_info(tsk
);
275 if (tsk
&& tsk
!= current
)
276 stack
= (unsigned long *)tsk
->thread
.sp
;
279 #ifdef CONFIG_FRAME_POINTER
281 if (tsk
== current
) {
282 /* Grab bp right from our regs */
283 asm("movq %%rbp, %0" : "=r" (bp
):);
285 /* bp is the last reg pushed by switch_to */
286 bp
= *(unsigned long *) tsk
->thread
.sp
;
294 * Print function call entries in all stacks, starting at the
295 * current stack address. If the stacks consist of nested
300 unsigned long *estack_end
;
301 estack_end
= in_exception_stack(cpu
, (unsigned long)stack
,
305 if (ops
->stack(data
, id
) < 0)
308 bp
= print_context_stack(tinfo
, stack
, bp
, ops
,
310 ops
->stack(data
, "<EOE>");
312 * We link to the next stack via the
313 * second-to-last pointer (index -2 to end) in the
316 stack
= (unsigned long *) estack_end
[-2];
320 unsigned long *irqstack
;
321 irqstack
= irqstack_end
-
322 (IRQSTACKSIZE
- 64) / sizeof(*irqstack
);
324 if (stack
>= irqstack
&& stack
< irqstack_end
) {
325 if (ops
->stack(data
, "IRQ") < 0)
327 bp
= print_context_stack(tinfo
, stack
, bp
,
328 ops
, data
, irqstack_end
);
330 * We link to the next stack (which would be
331 * the process stack normally) the last
332 * pointer (index -1 to end) in the IRQ stack:
334 stack
= (unsigned long *) (irqstack_end
[-1]);
336 ops
->stack(data
, "EOI");
344 * This handles the process stack:
346 bp
= print_context_stack(tinfo
, stack
, bp
, ops
, data
, NULL
);
349 EXPORT_SYMBOL(dump_trace
);
352 print_trace_warning_symbol(void *data
, char *msg
, unsigned long symbol
)
354 print_symbol(msg
, symbol
);
358 static void print_trace_warning(void *data
, char *msg
)
363 static int print_trace_stack(void *data
, char *name
)
365 printk(" <%s> ", name
);
369 static void print_trace_address(void *data
, unsigned long addr
, int reliable
)
371 touch_nmi_watchdog();
372 printk_address(addr
, reliable
);
375 static const struct stacktrace_ops print_trace_ops
= {
376 .warning
= print_trace_warning
,
377 .warning_symbol
= print_trace_warning_symbol
,
378 .stack
= print_trace_stack
,
379 .address
= print_trace_address
,
383 show_trace(struct task_struct
*tsk
, struct pt_regs
*regs
, unsigned long *stack
,
386 printk("\nCall Trace:\n");
387 dump_trace(tsk
, regs
, stack
, bp
, &print_trace_ops
, NULL
);
392 _show_stack(struct task_struct
*tsk
, struct pt_regs
*regs
, unsigned long *sp
,
395 unsigned long *stack
;
397 const int cpu
= smp_processor_id();
398 unsigned long *irqstack_end
= (unsigned long *) (cpu_pda(cpu
)->irqstackptr
);
399 unsigned long *irqstack
= (unsigned long *) (cpu_pda(cpu
)->irqstackptr
- IRQSTACKSIZE
);
401 // debugging aid: "show_stack(NULL, NULL);" prints the
402 // back trace for this cpu.
406 sp
= (unsigned long *)tsk
->thread
.sp
;
408 sp
= (unsigned long *)&sp
;
412 for(i
=0; i
< kstack_depth_to_print
; i
++) {
413 if (stack
>= irqstack
&& stack
<= irqstack_end
) {
414 if (stack
== irqstack_end
) {
415 stack
= (unsigned long *) (irqstack_end
[-1]);
419 if (((long) stack
& (THREAD_SIZE
-1)) == 0)
422 if (i
&& ((i
% 4) == 0))
424 printk(" %016lx", *stack
++);
425 touch_nmi_watchdog();
427 show_trace(tsk
, regs
, sp
, bp
);
430 void show_stack(struct task_struct
*tsk
, unsigned long * sp
)
432 _show_stack(tsk
, NULL
, sp
, 0);
436 * The architecture-independent dump_stack generator
438 void dump_stack(void)
441 unsigned long bp
= 0;
443 #ifdef CONFIG_FRAME_POINTER
445 asm("movq %%rbp, %0" : "=r" (bp
):);
448 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
449 current
->pid
, current
->comm
, print_tainted(),
450 init_utsname()->release
,
451 (int)strcspn(init_utsname()->version
, " "),
452 init_utsname()->version
);
453 show_trace(NULL
, NULL
, &dummy
, bp
);
456 EXPORT_SYMBOL(dump_stack
);
458 void show_registers(struct pt_regs
*regs
)
462 const int cpu
= smp_processor_id();
463 struct task_struct
*cur
= cpu_pda(cpu
)->pcurrent
;
465 unsigned int code_prologue
= code_bytes
* 43 / 64;
466 unsigned int code_len
= code_bytes
;
469 ip
= (u8
*) regs
->ip
- code_prologue
;
470 printk("CPU %d ", cpu
);
472 printk("Process %s (pid: %d, threadinfo %p, task %p)\n",
473 cur
->comm
, cur
->pid
, task_thread_info(cur
), cur
);
476 * When in-kernel, we also print out the stack and code at the
477 * time of the fault..
479 if (!user_mode(regs
)) {
482 _show_stack(NULL
, regs
, (unsigned long *)sp
, regs
->bp
);
485 printk(KERN_EMERG
"Code: ");
486 if (ip
< (u8
*)PAGE_OFFSET
|| probe_kernel_address(ip
, c
)) {
487 /* try starting at RIP */
488 ip
= (u8
*) regs
->ip
;
489 code_len
= code_len
- code_prologue
+ 1;
491 for (i
= 0; i
< code_len
; i
++, ip
++) {
492 if (ip
< (u8
*)PAGE_OFFSET
||
493 probe_kernel_address(ip
, c
)) {
494 printk(" Bad RIP value.");
497 if (ip
== (u8
*)regs
->ip
)
498 printk("<%02x> ", c
);
506 int is_valid_bugaddr(unsigned long ip
)
510 if (__copy_from_user(&ud2
, (const void __user
*) ip
, sizeof(ud2
)))
513 return ud2
== 0x0b0f;
516 static raw_spinlock_t die_lock
= __RAW_SPIN_LOCK_UNLOCKED
;
517 static int die_owner
= -1;
518 static unsigned int die_nest_count
;
520 unsigned __kprobes
long oops_begin(void)
527 /* racy, but better than risking deadlock. */
528 raw_local_irq_save(flags
);
529 cpu
= smp_processor_id();
530 if (!__raw_spin_trylock(&die_lock
)) {
531 if (cpu
== die_owner
)
532 /* nested oops. should stop eventually */;
534 __raw_spin_lock(&die_lock
);
543 void __kprobes
oops_end(unsigned long flags
, struct pt_regs
*regs
, int signr
)
549 /* Nest count reaches zero, release the lock. */
550 __raw_spin_unlock(&die_lock
);
551 raw_local_irq_restore(flags
);
557 panic("Fatal exception");
562 int __kprobes
__die(const char * str
, struct pt_regs
* regs
, long err
)
564 static int die_counter
;
565 printk(KERN_EMERG
"%s: %04lx [%u] ", str
, err
& 0xffff,++die_counter
);
566 #ifdef CONFIG_PREEMPT
572 #ifdef CONFIG_DEBUG_PAGEALLOC
573 printk("DEBUG_PAGEALLOC");
576 if (notify_die(DIE_OOPS
, str
, regs
, err
, current
->thread
.trap_no
, SIGSEGV
) == NOTIFY_STOP
)
578 show_registers(regs
);
579 add_taint(TAINT_DIE
);
580 /* Executive summary in case the oops scrolled away */
581 printk(KERN_ALERT
"RIP ");
582 printk_address(regs
->ip
, 1);
583 printk(" RSP <%016lx>\n", regs
->sp
);
584 if (kexec_should_crash(current
))
589 void die(const char * str
, struct pt_regs
* regs
, long err
)
591 unsigned long flags
= oops_begin();
593 if (!user_mode(regs
))
594 report_bug(regs
->ip
, regs
);
596 if (__die(str
, regs
, err
))
598 oops_end(flags
, regs
, SIGSEGV
);
601 void __kprobes
die_nmi(char *str
, struct pt_regs
*regs
, int do_panic
)
603 unsigned long flags
= oops_begin();
606 * We are in trouble anyway, lets at least try
607 * to get a message out.
609 printk(str
, smp_processor_id());
610 show_registers(regs
);
611 if (kexec_should_crash(current
))
613 if (do_panic
|| panic_on_oops
)
614 panic("Non maskable interrupt");
615 oops_end(flags
, NULL
, SIGBUS
);
621 static void __kprobes
do_trap(int trapnr
, int signr
, char *str
,
622 struct pt_regs
* regs
, long error_code
,
625 struct task_struct
*tsk
= current
;
627 if (user_mode(regs
)) {
629 * We want error_code and trap_no set for userspace
630 * faults and kernelspace faults which result in
631 * die(), but not kernelspace faults which are fixed
632 * up. die() gives the process no chance to handle
633 * the signal and notice the kernel fault information,
634 * so that won't result in polluting the information
635 * about previously queued, but not yet delivered,
636 * faults. See also do_general_protection below.
638 tsk
->thread
.error_code
= error_code
;
639 tsk
->thread
.trap_no
= trapnr
;
641 if (show_unhandled_signals
&& unhandled_signal(tsk
, signr
) &&
642 printk_ratelimit()) {
644 "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
645 tsk
->comm
, tsk
->pid
, str
,
646 regs
->ip
, regs
->sp
, error_code
);
647 print_vma_addr(" in ", regs
->ip
);
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
) &&
744 printk_ratelimit()) {
746 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
748 regs
->ip
, regs
->sp
, error_code
);
749 print_vma_addr(" in ", regs
->ip
);
753 force_sig(SIGSEGV
, tsk
);
757 if (fixup_exception(regs
))
760 tsk
->thread
.error_code
= error_code
;
761 tsk
->thread
.trap_no
= 13;
762 if (notify_die(DIE_GPF
, "general protection fault", regs
,
763 error_code
, 13, SIGSEGV
) == NOTIFY_STOP
)
765 die("general protection fault", regs
, error_code
);
768 static __kprobes
void
769 mem_parity_error(unsigned char reason
, struct pt_regs
* regs
)
771 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x.\n",
773 printk(KERN_EMERG
"You have some hardware problem, likely on the PCI bus.\n");
775 #if defined(CONFIG_EDAC)
776 if(edac_handler_set()) {
777 edac_atomic_assert_error();
782 if (panic_on_unrecovered_nmi
)
783 panic("NMI: Not continuing");
785 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
787 /* Clear and disable the memory parity error line. */
788 reason
= (reason
& 0xf) | 4;
792 static __kprobes
void
793 io_check_error(unsigned char reason
, struct pt_regs
* regs
)
795 printk("NMI: IOCK error (debug interrupt?)\n");
796 show_registers(regs
);
798 /* Re-enable the IOCK line, wait for a few seconds */
799 reason
= (reason
& 0xf) | 8;
806 static __kprobes
void
807 unknown_nmi_error(unsigned char reason
, struct pt_regs
* regs
)
809 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x.\n",
811 printk(KERN_EMERG
"Do you have a strange power saving mode enabled?\n");
813 if (panic_on_unrecovered_nmi
)
814 panic("NMI: Not continuing");
816 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
819 /* Runs on IST stack. This code must keep interrupts off all the time.
820 Nested NMIs are prevented by the CPU. */
821 asmlinkage __kprobes
void default_do_nmi(struct pt_regs
*regs
)
823 unsigned char reason
= 0;
826 cpu
= smp_processor_id();
828 /* Only the BSP gets external NMIs from the system. */
830 reason
= get_nmi_reason();
832 if (!(reason
& 0xc0)) {
833 if (notify_die(DIE_NMI_IPI
, "nmi_ipi", regs
, reason
, 2, SIGINT
)
837 * Ok, so this is none of the documented NMI sources,
838 * so it must be the NMI watchdog.
840 if (nmi_watchdog_tick(regs
,reason
))
842 if (!do_nmi_callback(regs
,cpu
))
843 unknown_nmi_error(reason
, regs
);
847 if (notify_die(DIE_NMI
, "nmi", regs
, reason
, 2, SIGINT
) == NOTIFY_STOP
)
850 /* AK: following checks seem to be broken on modern chipsets. FIXME */
853 mem_parity_error(reason
, regs
);
855 io_check_error(reason
, regs
);
858 /* runs on IST stack. */
859 asmlinkage
void __kprobes
do_int3(struct pt_regs
* regs
, long error_code
)
861 trace_hardirqs_fixup();
863 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, 3, SIGTRAP
) == NOTIFY_STOP
) {
866 preempt_conditional_sti(regs
);
867 do_trap(3, SIGTRAP
, "int3", regs
, error_code
, NULL
);
868 preempt_conditional_cli(regs
);
871 /* Help handler running on IST stack to switch back to user stack
872 for scheduling or signal handling. The actual stack switch is done in
874 asmlinkage __kprobes
struct pt_regs
*sync_regs(struct pt_regs
*eregs
)
876 struct pt_regs
*regs
= eregs
;
877 /* Did already sync */
878 if (eregs
== (struct pt_regs
*)eregs
->sp
)
880 /* Exception from user space */
881 else if (user_mode(eregs
))
882 regs
= task_pt_regs(current
);
883 /* Exception from kernel and interrupts are enabled. Move to
884 kernel process stack. */
885 else if (eregs
->flags
& X86_EFLAGS_IF
)
886 regs
= (struct pt_regs
*)(eregs
->sp
-= sizeof(struct pt_regs
));
892 /* runs on IST stack. */
893 asmlinkage
void __kprobes
do_debug(struct pt_regs
* regs
,
894 unsigned long error_code
)
896 unsigned long condition
;
897 struct task_struct
*tsk
= current
;
900 trace_hardirqs_fixup();
902 get_debugreg(condition
, 6);
905 * The processor cleared BTF, so don't mark that we need it set.
907 clear_tsk_thread_flag(tsk
, TIF_DEBUGCTLMSR
);
908 tsk
->thread
.debugctlmsr
= 0;
910 if (notify_die(DIE_DEBUG
, "debug", regs
, condition
, error_code
,
911 SIGTRAP
) == NOTIFY_STOP
)
914 preempt_conditional_sti(regs
);
916 /* Mask out spurious debug traps due to lazy DR7 setting */
917 if (condition
& (DR_TRAP0
|DR_TRAP1
|DR_TRAP2
|DR_TRAP3
)) {
918 if (!tsk
->thread
.debugreg7
) {
923 tsk
->thread
.debugreg6
= condition
;
927 * Single-stepping through TF: make sure we ignore any events in
928 * kernel space (but re-enable TF when returning to user mode).
930 if (condition
& DR_STEP
) {
931 if (!user_mode(regs
))
932 goto clear_TF_reenable
;
935 /* Ok, finally something we can handle */
936 tsk
->thread
.trap_no
= 1;
937 tsk
->thread
.error_code
= error_code
;
938 info
.si_signo
= SIGTRAP
;
940 info
.si_code
= TRAP_BRKPT
;
941 info
.si_addr
= user_mode(regs
) ? (void __user
*)regs
->ip
: NULL
;
942 force_sig_info(SIGTRAP
, &info
, tsk
);
945 set_debugreg(0UL, 7);
946 preempt_conditional_cli(regs
);
950 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
951 regs
->flags
&= ~X86_EFLAGS_TF
;
952 preempt_conditional_cli(regs
);
955 static int kernel_math_error(struct pt_regs
*regs
, const char *str
, int trapnr
)
957 if (fixup_exception(regs
))
960 notify_die(DIE_GPF
, str
, regs
, 0, trapnr
, SIGFPE
);
961 /* Illegal floating point operation in the kernel */
962 current
->thread
.trap_no
= trapnr
;
968 * Note that we play around with the 'TS' bit in an attempt to get
969 * the correct behaviour even in the presence of the asynchronous
972 asmlinkage
void do_coprocessor_error(struct pt_regs
*regs
)
974 void __user
*ip
= (void __user
*)(regs
->ip
);
975 struct task_struct
* task
;
977 unsigned short cwd
, swd
;
979 conditional_sti(regs
);
980 if (!user_mode(regs
) &&
981 kernel_math_error(regs
, "kernel x87 math error", 16))
985 * Save the info for the exception handler and clear the error.
989 task
->thread
.trap_no
= 16;
990 task
->thread
.error_code
= 0;
991 info
.si_signo
= SIGFPE
;
993 info
.si_code
= __SI_FAULT
;
996 * (~cwd & swd) will mask out exceptions that are not set to unmasked
997 * status. 0x3f is the exception bits in these regs, 0x200 is the
998 * C1 reg you need in case of a stack fault, 0x040 is the stack
999 * fault bit. We should only be taking one exception at a time,
1000 * so if this combination doesn't produce any single exception,
1001 * then we have a bad program that isn't synchronizing its FPU usage
1002 * and it will suffer the consequences since we won't be able to
1003 * fully reproduce the context of the exception
1005 cwd
= get_fpu_cwd(task
);
1006 swd
= get_fpu_swd(task
);
1007 switch (swd
& ~cwd
& 0x3f) {
1011 case 0x001: /* Invalid Op */
1013 * swd & 0x240 == 0x040: Stack Underflow
1014 * swd & 0x240 == 0x240: Stack Overflow
1015 * User must clear the SF bit (0x40) if set
1017 info
.si_code
= FPE_FLTINV
;
1019 case 0x002: /* Denormalize */
1020 case 0x010: /* Underflow */
1021 info
.si_code
= FPE_FLTUND
;
1023 case 0x004: /* Zero Divide */
1024 info
.si_code
= FPE_FLTDIV
;
1026 case 0x008: /* Overflow */
1027 info
.si_code
= FPE_FLTOVF
;
1029 case 0x020: /* Precision */
1030 info
.si_code
= FPE_FLTRES
;
1033 force_sig_info(SIGFPE
, &info
, task
);
1036 asmlinkage
void bad_intr(void)
1038 printk("bad interrupt");
1041 asmlinkage
void do_simd_coprocessor_error(struct pt_regs
*regs
)
1043 void __user
*ip
= (void __user
*)(regs
->ip
);
1044 struct task_struct
* task
;
1046 unsigned short mxcsr
;
1048 conditional_sti(regs
);
1049 if (!user_mode(regs
) &&
1050 kernel_math_error(regs
, "kernel simd math error", 19))
1054 * Save the info for the exception handler and clear the error.
1057 save_init_fpu(task
);
1058 task
->thread
.trap_no
= 19;
1059 task
->thread
.error_code
= 0;
1060 info
.si_signo
= SIGFPE
;
1062 info
.si_code
= __SI_FAULT
;
1065 * The SIMD FPU exceptions are handled a little differently, as there
1066 * is only a single status/control register. Thus, to determine which
1067 * unmasked exception was caught we must mask the exception mask bits
1068 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1070 mxcsr
= get_fpu_mxcsr(task
);
1071 switch (~((mxcsr
& 0x1f80) >> 7) & (mxcsr
& 0x3f)) {
1075 case 0x001: /* Invalid Op */
1076 info
.si_code
= FPE_FLTINV
;
1078 case 0x002: /* Denormalize */
1079 case 0x010: /* Underflow */
1080 info
.si_code
= FPE_FLTUND
;
1082 case 0x004: /* Zero Divide */
1083 info
.si_code
= FPE_FLTDIV
;
1085 case 0x008: /* Overflow */
1086 info
.si_code
= FPE_FLTOVF
;
1088 case 0x020: /* Precision */
1089 info
.si_code
= FPE_FLTRES
;
1092 force_sig_info(SIGFPE
, &info
, task
);
1095 asmlinkage
void do_spurious_interrupt_bug(struct pt_regs
* regs
)
1099 asmlinkage
void __attribute__((weak
)) smp_thermal_interrupt(void)
1103 asmlinkage
void __attribute__((weak
)) mce_threshold_interrupt(void)
1108 * 'math_state_restore()' saves the current math information in the
1109 * old math state array, and gets the new ones from the current task
1111 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1112 * Don't touch unless you *really* know how it works.
1114 asmlinkage
void math_state_restore(void)
1116 struct task_struct
*me
= current
;
1117 clts(); /* Allow maths ops (or we recurse) */
1121 restore_fpu_checking(&me
->thread
.i387
.fxsave
);
1122 task_thread_info(me
)->status
|= TS_USEDFPU
;
1125 EXPORT_SYMBOL_GPL(math_state_restore
);
1127 void __init
trap_init(void)
1129 set_intr_gate(0,÷_error
);
1130 set_intr_gate_ist(1,&debug
,DEBUG_STACK
);
1131 set_intr_gate_ist(2,&nmi
,NMI_STACK
);
1132 set_system_gate_ist(3,&int3
,DEBUG_STACK
); /* int3 can be called from all */
1133 set_system_gate(4,&overflow
); /* int4 can be called from all */
1134 set_intr_gate(5,&bounds
);
1135 set_intr_gate(6,&invalid_op
);
1136 set_intr_gate(7,&device_not_available
);
1137 set_intr_gate_ist(8,&double_fault
, DOUBLEFAULT_STACK
);
1138 set_intr_gate(9,&coprocessor_segment_overrun
);
1139 set_intr_gate(10,&invalid_TSS
);
1140 set_intr_gate(11,&segment_not_present
);
1141 set_intr_gate_ist(12,&stack_segment
,STACKFAULT_STACK
);
1142 set_intr_gate(13,&general_protection
);
1143 set_intr_gate(14,&page_fault
);
1144 set_intr_gate(15,&spurious_interrupt_bug
);
1145 set_intr_gate(16,&coprocessor_error
);
1146 set_intr_gate(17,&alignment_check
);
1147 #ifdef CONFIG_X86_MCE
1148 set_intr_gate_ist(18,&machine_check
, MCE_STACK
);
1150 set_intr_gate(19,&simd_coprocessor_error
);
1152 #ifdef CONFIG_IA32_EMULATION
1153 set_system_gate(IA32_SYSCALL_VECTOR
, ia32_syscall
);
1157 * Should be a barrier for any external CPU state.
1163 static int __init
oops_setup(char *s
)
1167 if (!strcmp(s
, "panic"))
1171 early_param("oops", oops_setup
);
1173 static int __init
kstack_setup(char *s
)
1177 kstack_depth_to_print
= simple_strtoul(s
,NULL
,0);
1180 early_param("kstack", kstack_setup
);
1183 static int __init
code_bytes_setup(char *s
)
1185 code_bytes
= simple_strtoul(s
, NULL
, 0);
1186 if (code_bytes
> 8192)
1191 __setup("code_bytes=", code_bytes_setup
);