2 * linux/arch/i386/traps.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Pentium III FXSR, SSE support
7 * Gareth Hughes <gareth@valinux.com>, May 2000
11 * 'Traps.c' handles hardware traps and faults after we have saved some
14 #include <linux/sched.h>
15 #include <linux/kernel.h>
16 #include <linux/string.h>
17 #include <linux/errno.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/highmem.h>
25 #include <linux/kallsyms.h>
26 #include <linux/ptrace.h>
27 #include <linux/utsname.h>
28 #include <linux/kprobes.h>
29 #include <linux/kexec.h>
30 #include <linux/unwind.h>
31 #include <linux/uaccess.h>
34 #include <linux/ioport.h>
35 #include <linux/eisa.h>
39 #include <linux/mca.h>
42 #include <asm/processor.h>
43 #include <asm/system.h>
45 #include <asm/atomic.h>
46 #include <asm/debugreg.h>
50 #include <asm/unwind.h>
52 #include <asm/arch_hooks.h>
53 #include <asm/kdebug.h>
54 #include <asm/stacktrace.h>
56 #include <linux/module.h>
58 #include "mach_traps.h"
60 asmlinkage
int system_call(void);
62 struct desc_struct default_ldt
[] = { { 0, 0 }, { 0, 0 }, { 0, 0 },
65 /* Do we ignore FPU interrupts ? */
66 char ignore_fpu_irq
= 0;
69 * The IDT has to be page-aligned to simplify the Pentium
70 * F0 0F bug workaround.. We have a special link segment
73 struct desc_struct idt_table
[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
75 asmlinkage
void divide_error(void);
76 asmlinkage
void debug(void);
77 asmlinkage
void nmi(void);
78 asmlinkage
void int3(void);
79 asmlinkage
void overflow(void);
80 asmlinkage
void bounds(void);
81 asmlinkage
void invalid_op(void);
82 asmlinkage
void device_not_available(void);
83 asmlinkage
void coprocessor_segment_overrun(void);
84 asmlinkage
void invalid_TSS(void);
85 asmlinkage
void segment_not_present(void);
86 asmlinkage
void stack_segment(void);
87 asmlinkage
void general_protection(void);
88 asmlinkage
void page_fault(void);
89 asmlinkage
void coprocessor_error(void);
90 asmlinkage
void simd_coprocessor_error(void);
91 asmlinkage
void alignment_check(void);
92 asmlinkage
void spurious_interrupt_bug(void);
93 asmlinkage
void machine_check(void);
95 static int kstack_depth_to_print
= 24;
96 #ifdef CONFIG_STACK_UNWIND
97 static int call_trace
= 1;
99 #define call_trace (-1)
101 ATOMIC_NOTIFIER_HEAD(i386die_chain
);
103 int register_die_notifier(struct notifier_block
*nb
)
106 return atomic_notifier_chain_register(&i386die_chain
, nb
);
108 EXPORT_SYMBOL(register_die_notifier
); /* used modular by kdb */
110 int unregister_die_notifier(struct notifier_block
*nb
)
112 return atomic_notifier_chain_unregister(&i386die_chain
, nb
);
114 EXPORT_SYMBOL(unregister_die_notifier
); /* used modular by kdb */
116 static inline int valid_stack_ptr(struct thread_info
*tinfo
, void *p
)
118 return p
> (void *)tinfo
&&
119 p
< (void *)tinfo
+ THREAD_SIZE
- 3;
122 static inline unsigned long print_context_stack(struct thread_info
*tinfo
,
123 unsigned long *stack
, unsigned long ebp
,
124 struct stacktrace_ops
*ops
, void *data
)
128 #ifdef CONFIG_FRAME_POINTER
129 while (valid_stack_ptr(tinfo
, (void *)ebp
)) {
130 addr
= *(unsigned long *)(ebp
+ 4);
131 ops
->address(data
, addr
);
133 * break out of recursive entries (such as
134 * end_of_stack_stop_unwind_function):
136 if (ebp
== *(unsigned long *)ebp
)
138 ebp
= *(unsigned long *)ebp
;
141 while (valid_stack_ptr(tinfo
, stack
)) {
143 if (__kernel_text_address(addr
))
144 ops
->address(data
, addr
);
150 struct ops_and_data
{
151 struct stacktrace_ops
*ops
;
155 static asmlinkage
int
156 dump_trace_unwind(struct unwind_frame_info
*info
, void *data
)
158 struct ops_and_data
*oad
= (struct ops_and_data
*)data
;
161 while (unwind(info
) == 0 && UNW_PC(info
)) {
163 oad
->ops
->address(oad
->data
, UNW_PC(info
));
164 if (arch_unw_user_mode(info
))
170 void dump_trace(struct task_struct
*task
, struct pt_regs
*regs
,
171 unsigned long *stack
,
172 struct stacktrace_ops
*ops
, void *data
)
174 unsigned long ebp
= 0;
179 if (call_trace
>= 0) {
181 struct unwind_frame_info info
;
182 struct ops_and_data oad
= { .ops
= ops
, .data
= data
};
185 if (unwind_init_frame_info(&info
, task
, regs
) == 0)
186 unw_ret
= dump_trace_unwind(&info
, &oad
);
187 } else if (task
== current
)
188 unw_ret
= unwind_init_running(&info
, dump_trace_unwind
, &oad
);
190 if (unwind_init_blocked(&info
, task
) == 0)
191 unw_ret
= dump_trace_unwind(&info
, &oad
);
194 if (call_trace
== 1 && !arch_unw_user_mode(&info
)) {
195 ops
->warning_symbol(data
, "DWARF2 unwinder stuck at %s\n",
197 if (UNW_SP(&info
) >= PAGE_OFFSET
) {
198 ops
->warning(data
, "Leftover inexact backtrace:\n");
199 stack
= (void *)UNW_SP(&info
);
204 ops
->warning(data
, "Full inexact backtrace again:\n");
205 } else if (call_trace
>= 1)
208 ops
->warning(data
, "Full inexact backtrace again:\n");
210 ops
->warning(data
, "Inexact backtrace:\n");
215 if (task
&& task
!= current
)
216 stack
= (unsigned long *)task
->thread
.esp
;
219 #ifdef CONFIG_FRAME_POINTER
221 if (task
== current
) {
222 /* Grab ebp right from our regs */
223 asm ("movl %%ebp, %0" : "=r" (ebp
) : );
225 /* ebp is the last reg pushed by switch_to */
226 ebp
= *(unsigned long *) task
->thread
.esp
;
232 struct thread_info
*context
;
233 context
= (struct thread_info
*)
234 ((unsigned long)stack
& (~(THREAD_SIZE
- 1)));
235 ebp
= print_context_stack(context
, stack
, ebp
, ops
, data
);
236 /* Should be after the line below, but somewhere
237 in early boot context comes out corrupted and we
238 can't reference it -AK */
239 if (ops
->stack(data
, "IRQ") < 0)
241 stack
= (unsigned long*)context
->previous_esp
;
246 EXPORT_SYMBOL(dump_trace
);
249 print_trace_warning_symbol(void *data
, char *msg
, unsigned long symbol
)
252 print_symbol(msg
, symbol
);
256 static void print_trace_warning(void *data
, char *msg
)
258 printk("%s%s\n", (char *)data
, msg
);
261 static int print_trace_stack(void *data
, char *name
)
267 * Print one address/symbol entries per line.
269 static void print_trace_address(void *data
, unsigned long addr
)
271 printk("%s [<%08lx>] ", (char *)data
, addr
);
272 print_symbol("%s\n", addr
);
275 static struct stacktrace_ops print_trace_ops
= {
276 .warning
= print_trace_warning
,
277 .warning_symbol
= print_trace_warning_symbol
,
278 .stack
= print_trace_stack
,
279 .address
= print_trace_address
,
283 show_trace_log_lvl(struct task_struct
*task
, struct pt_regs
*regs
,
284 unsigned long * stack
, char *log_lvl
)
286 dump_trace(task
, regs
, stack
, &print_trace_ops
, log_lvl
);
287 printk("%s =======================\n", log_lvl
);
290 void show_trace(struct task_struct
*task
, struct pt_regs
*regs
,
291 unsigned long * stack
)
293 show_trace_log_lvl(task
, regs
, stack
, "");
296 static void show_stack_log_lvl(struct task_struct
*task
, struct pt_regs
*regs
,
297 unsigned long *esp
, char *log_lvl
)
299 unsigned long *stack
;
304 esp
= (unsigned long*)task
->thread
.esp
;
306 esp
= (unsigned long *)&esp
;
310 for(i
= 0; i
< kstack_depth_to_print
; i
++) {
311 if (kstack_end(stack
))
313 if (i
&& ((i
% 8) == 0))
314 printk("\n%s ", log_lvl
);
315 printk("%08lx ", *stack
++);
317 printk("\n%sCall Trace:\n", log_lvl
);
318 show_trace_log_lvl(task
, regs
, esp
, log_lvl
);
321 void show_stack(struct task_struct
*task
, unsigned long *esp
)
324 show_stack_log_lvl(task
, NULL
, esp
, "");
328 * The architecture-independent dump_stack generator
330 void dump_stack(void)
334 show_trace(current
, NULL
, &stack
);
337 EXPORT_SYMBOL(dump_stack
);
339 void show_registers(struct pt_regs
*regs
)
346 esp
= (unsigned long) (®s
->esp
);
348 if (user_mode_vm(regs
)) {
351 ss
= regs
->xss
& 0xffff;
354 printk(KERN_EMERG
"CPU: %d\n"
355 KERN_EMERG
"EIP: %04x:[<%08lx>] %s VLI\n"
356 KERN_EMERG
"EFLAGS: %08lx (%s %.*s)\n",
357 smp_processor_id(), 0xffff & regs
->xcs
, regs
->eip
,
358 print_tainted(), regs
->eflags
, system_utsname
.release
,
359 (int)strcspn(system_utsname
.version
, " "),
360 system_utsname
.version
);
361 print_symbol(KERN_EMERG
"EIP is at %s\n", regs
->eip
);
362 printk(KERN_EMERG
"eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
363 regs
->eax
, regs
->ebx
, regs
->ecx
, regs
->edx
);
364 printk(KERN_EMERG
"esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
365 regs
->esi
, regs
->edi
, regs
->ebp
, esp
);
366 printk(KERN_EMERG
"ds: %04x es: %04x ss: %04x\n",
367 regs
->xds
& 0xffff, regs
->xes
& 0xffff, ss
);
368 printk(KERN_EMERG
"Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
369 TASK_COMM_LEN
, current
->comm
, current
->pid
,
370 current_thread_info(), current
, current
->thread_info
);
372 * When in-kernel, we also print out the stack and code at the
373 * time of the fault..
380 printk("\n" KERN_EMERG
"Stack: ");
381 show_stack_log_lvl(NULL
, regs
, (unsigned long *)esp
, KERN_EMERG
);
383 printk(KERN_EMERG
"Code: ");
385 eip
= (u8 __user
*)regs
->eip
- 43;
386 if (eip
< (u8 __user
*)PAGE_OFFSET
|| __get_user(c
, eip
)) {
387 /* try starting at EIP */
388 eip
= (u8 __user
*)regs
->eip
;
391 for (i
= 0; i
< code_bytes
; i
++, eip
++) {
392 if (eip
< (u8 __user
*)PAGE_OFFSET
|| __get_user(c
, eip
)) {
393 printk(" Bad EIP value.");
396 if (eip
== (u8 __user
*)regs
->eip
)
397 printk("<%02x> ", c
);
405 static void handle_BUG(struct pt_regs
*regs
)
407 unsigned long eip
= regs
->eip
;
410 if (eip
< PAGE_OFFSET
)
412 if (probe_kernel_address((unsigned short __user
*)eip
, ud2
))
417 printk(KERN_EMERG
"------------[ cut here ]------------\n");
419 #ifdef CONFIG_DEBUG_BUGVERBOSE
425 if (probe_kernel_address((unsigned short __user
*)(eip
+ 2),
428 if (__get_user(file
, (char * __user
*)(eip
+ 4)) ||
429 (unsigned long)file
< PAGE_OFFSET
|| __get_user(c
, file
))
430 file
= "<bad filename>";
432 printk(KERN_EMERG
"kernel BUG at %s:%d!\n", file
, line
);
436 printk(KERN_EMERG
"Kernel BUG at [verbose debug info unavailable]\n");
439 /* This is gone through when something in the kernel
440 * has done something bad and is about to be terminated.
442 void die(const char * str
, struct pt_regs
* regs
, long err
)
447 int lock_owner_depth
;
449 .lock
= SPIN_LOCK_UNLOCKED
,
451 .lock_owner_depth
= 0
453 static int die_counter
;
458 if (die
.lock_owner
!= raw_smp_processor_id()) {
460 spin_lock_irqsave(&die
.lock
, flags
);
461 die
.lock_owner
= smp_processor_id();
462 die
.lock_owner_depth
= 0;
466 local_save_flags(flags
);
468 if (++die
.lock_owner_depth
< 3) {
474 printk(KERN_EMERG
"%s: %04lx [#%d]\n", str
, err
& 0xffff, ++die_counter
);
475 #ifdef CONFIG_PREEMPT
476 printk(KERN_EMERG
"PREEMPT ");
485 #ifdef CONFIG_DEBUG_PAGEALLOC
488 printk("DEBUG_PAGEALLOC");
493 if (notify_die(DIE_OOPS
, str
, regs
, err
,
494 current
->thread
.trap_no
, SIGSEGV
) !=
496 show_registers(regs
);
497 /* Executive summary in case the oops scrolled away */
498 esp
= (unsigned long) (®s
->esp
);
500 if (user_mode(regs
)) {
502 ss
= regs
->xss
& 0xffff;
504 printk(KERN_EMERG
"EIP: [<%08lx>] ", regs
->eip
);
505 print_symbol("%s", regs
->eip
);
506 printk(" SS:ESP %04x:%08lx\n", ss
, esp
);
511 printk(KERN_EMERG
"Recursive die() failure, output suppressed\n");
515 spin_unlock_irqrestore(&die
.lock
, flags
);
520 if (kexec_should_crash(current
))
524 panic("Fatal exception in interrupt");
527 panic("Fatal exception");
533 static inline void die_if_kernel(const char * str
, struct pt_regs
* regs
, long err
)
535 if (!user_mode_vm(regs
))
539 static void __kprobes
do_trap(int trapnr
, int signr
, char *str
, int vm86
,
540 struct pt_regs
* regs
, long error_code
,
543 struct task_struct
*tsk
= current
;
544 tsk
->thread
.error_code
= error_code
;
545 tsk
->thread
.trap_no
= trapnr
;
547 if (regs
->eflags
& VM_MASK
) {
553 if (!user_mode(regs
))
558 force_sig_info(signr
, info
, tsk
);
560 force_sig(signr
, tsk
);
565 if (!fixup_exception(regs
))
566 die(str
, regs
, error_code
);
571 int ret
= handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
, trapnr
);
572 if (ret
) goto trap_signal
;
577 #define DO_ERROR(trapnr, signr, str, name) \
578 fastcall void do_##name(struct pt_regs * regs, long error_code) \
580 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
583 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
586 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
587 fastcall void do_##name(struct pt_regs * regs, long error_code) \
590 info.si_signo = signr; \
592 info.si_code = sicode; \
593 info.si_addr = (void __user *)siaddr; \
594 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
597 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
600 #define DO_VM86_ERROR(trapnr, signr, str, name) \
601 fastcall void do_##name(struct pt_regs * regs, long error_code) \
603 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
606 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
609 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
610 fastcall void do_##name(struct pt_regs * regs, long error_code) \
613 info.si_signo = signr; \
615 info.si_code = sicode; \
616 info.si_addr = (void __user *)siaddr; \
617 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
620 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
623 DO_VM86_ERROR_INFO( 0, SIGFPE
, "divide error", divide_error
, FPE_INTDIV
, regs
->eip
)
624 #ifndef CONFIG_KPROBES
625 DO_VM86_ERROR( 3, SIGTRAP
, "int3", int3
)
627 DO_VM86_ERROR( 4, SIGSEGV
, "overflow", overflow
)
628 DO_VM86_ERROR( 5, SIGSEGV
, "bounds", bounds
)
629 DO_ERROR_INFO( 6, SIGILL
, "invalid opcode", invalid_op
, ILL_ILLOPN
, regs
->eip
)
630 DO_ERROR( 9, SIGFPE
, "coprocessor segment overrun", coprocessor_segment_overrun
)
631 DO_ERROR(10, SIGSEGV
, "invalid TSS", invalid_TSS
)
632 DO_ERROR(11, SIGBUS
, "segment not present", segment_not_present
)
633 DO_ERROR(12, SIGBUS
, "stack segment", stack_segment
)
634 DO_ERROR_INFO(17, SIGBUS
, "alignment check", alignment_check
, BUS_ADRALN
, 0)
635 DO_ERROR_INFO(32, SIGSEGV
, "iret exception", iret_error
, ILL_BADSTK
, 0)
637 fastcall
void __kprobes
do_general_protection(struct pt_regs
* regs
,
641 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
642 struct thread_struct
*thread
= ¤t
->thread
;
645 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
646 * invalid offset set (the LAZY one) and the faulting thread has
647 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
648 * and we set the offset field correctly. Then we let the CPU to
649 * restart the faulting instruction.
651 if (tss
->io_bitmap_base
== INVALID_IO_BITMAP_OFFSET_LAZY
&&
652 thread
->io_bitmap_ptr
) {
653 memcpy(tss
->io_bitmap
, thread
->io_bitmap_ptr
,
654 thread
->io_bitmap_max
);
656 * If the previously set map was extending to higher ports
657 * than the current one, pad extra space with 0xff (no access).
659 if (thread
->io_bitmap_max
< tss
->io_bitmap_max
)
660 memset((char *) tss
->io_bitmap
+
661 thread
->io_bitmap_max
, 0xff,
662 tss
->io_bitmap_max
- thread
->io_bitmap_max
);
663 tss
->io_bitmap_max
= thread
->io_bitmap_max
;
664 tss
->io_bitmap_base
= IO_BITMAP_OFFSET
;
665 tss
->io_bitmap_owner
= thread
;
671 current
->thread
.error_code
= error_code
;
672 current
->thread
.trap_no
= 13;
674 if (regs
->eflags
& VM_MASK
)
677 if (!user_mode(regs
))
680 current
->thread
.error_code
= error_code
;
681 current
->thread
.trap_no
= 13;
682 force_sig(SIGSEGV
, current
);
687 handle_vm86_fault((struct kernel_vm86_regs
*) regs
, error_code
);
691 if (!fixup_exception(regs
)) {
692 if (notify_die(DIE_GPF
, "general protection fault", regs
,
693 error_code
, 13, SIGSEGV
) == NOTIFY_STOP
)
695 die("general protection fault", regs
, error_code
);
699 static __kprobes
void
700 mem_parity_error(unsigned char reason
, struct pt_regs
* regs
)
702 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x on "
703 "CPU %d.\n", reason
, smp_processor_id());
704 printk(KERN_EMERG
"You probably have a hardware problem with your RAM "
706 if (panic_on_unrecovered_nmi
)
707 panic("NMI: Not continuing");
709 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
711 /* Clear and disable the memory parity error line. */
712 clear_mem_error(reason
);
715 static __kprobes
void
716 io_check_error(unsigned char reason
, struct pt_regs
* regs
)
720 printk(KERN_EMERG
"NMI: IOCK error (debug interrupt?)\n");
721 show_registers(regs
);
723 /* Re-enable the IOCK line, wait for a few seconds */
724 reason
= (reason
& 0xf) | 8;
727 while (--i
) udelay(1000);
732 static __kprobes
void
733 unknown_nmi_error(unsigned char reason
, struct pt_regs
* regs
)
736 /* Might actually be able to figure out what the guilty party
743 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x on "
744 "CPU %d.\n", reason
, smp_processor_id());
745 printk(KERN_EMERG
"Do you have a strange power saving mode enabled?\n");
746 if (panic_on_unrecovered_nmi
)
747 panic("NMI: Not continuing");
749 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
752 static DEFINE_SPINLOCK(nmi_print_lock
);
754 void __kprobes
die_nmi(struct pt_regs
*regs
, const char *msg
)
756 if (notify_die(DIE_NMIWATCHDOG
, msg
, regs
, 0, 2, SIGINT
) ==
760 spin_lock(&nmi_print_lock
);
762 * We are in trouble anyway, lets at least try
763 * to get a message out.
766 printk(KERN_EMERG
"%s", msg
);
767 printk(" on CPU%d, eip %08lx, registers:\n",
768 smp_processor_id(), regs
->eip
);
769 show_registers(regs
);
770 printk(KERN_EMERG
"console shuts up ...\n");
772 spin_unlock(&nmi_print_lock
);
775 /* If we are in kernel we are probably nested up pretty bad
776 * and might aswell get out now while we still can.
778 if (!user_mode_vm(regs
)) {
779 current
->thread
.trap_no
= 2;
786 static __kprobes
void default_do_nmi(struct pt_regs
* regs
)
788 unsigned char reason
= 0;
790 /* Only the BSP gets external NMIs from the system. */
791 if (!smp_processor_id())
792 reason
= get_nmi_reason();
794 if (!(reason
& 0xc0)) {
795 if (notify_die(DIE_NMI_IPI
, "nmi_ipi", regs
, reason
, 2, SIGINT
)
798 #ifdef CONFIG_X86_LOCAL_APIC
800 * Ok, so this is none of the documented NMI sources,
801 * so it must be the NMI watchdog.
803 if (nmi_watchdog_tick(regs
, reason
))
805 if (!do_nmi_callback(regs
, smp_processor_id()))
807 unknown_nmi_error(reason
, regs
);
811 if (notify_die(DIE_NMI
, "nmi", regs
, reason
, 2, SIGINT
) == NOTIFY_STOP
)
814 mem_parity_error(reason
, regs
);
816 io_check_error(reason
, regs
);
818 * Reassert NMI in case it became active meanwhile
819 * as it's edge-triggered.
824 fastcall __kprobes
void do_nmi(struct pt_regs
* regs
, long error_code
)
830 cpu
= smp_processor_id();
834 default_do_nmi(regs
);
839 #ifdef CONFIG_KPROBES
840 fastcall
void __kprobes
do_int3(struct pt_regs
*regs
, long error_code
)
842 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, 3, SIGTRAP
)
845 /* This is an interrupt gate, because kprobes wants interrupts
846 disabled. Normal trap handlers don't. */
847 restore_interrupts(regs
);
848 do_trap(3, SIGTRAP
, "int3", 1, regs
, error_code
, NULL
);
853 * Our handling of the processor debug registers is non-trivial.
854 * We do not clear them on entry and exit from the kernel. Therefore
855 * it is possible to get a watchpoint trap here from inside the kernel.
856 * However, the code in ./ptrace.c has ensured that the user can
857 * only set watchpoints on userspace addresses. Therefore the in-kernel
858 * watchpoint trap can only occur in code which is reading/writing
859 * from user space. Such code must not hold kernel locks (since it
860 * can equally take a page fault), therefore it is safe to call
861 * force_sig_info even though that claims and releases locks.
863 * Code in ./signal.c ensures that the debug control register
864 * is restored before we deliver any signal, and therefore that
865 * user code runs with the correct debug control register even though
868 * Being careful here means that we don't have to be as careful in a
869 * lot of more complicated places (task switching can be a bit lazy
870 * about restoring all the debug state, and ptrace doesn't have to
871 * find every occurrence of the TF bit that could be saved away even
874 fastcall
void __kprobes
do_debug(struct pt_regs
* regs
, long error_code
)
876 unsigned int condition
;
877 struct task_struct
*tsk
= current
;
879 get_debugreg(condition
, 6);
881 if (notify_die(DIE_DEBUG
, "debug", regs
, condition
, error_code
,
882 SIGTRAP
) == NOTIFY_STOP
)
884 /* It's safe to allow irq's after DR6 has been saved */
885 if (regs
->eflags
& X86_EFLAGS_IF
)
888 /* Mask out spurious debug traps due to lazy DR7 setting */
889 if (condition
& (DR_TRAP0
|DR_TRAP1
|DR_TRAP2
|DR_TRAP3
)) {
890 if (!tsk
->thread
.debugreg
[7])
894 if (regs
->eflags
& VM_MASK
)
897 /* Save debug status register where ptrace can see it */
898 tsk
->thread
.debugreg
[6] = condition
;
901 * Single-stepping through TF: make sure we ignore any events in
902 * kernel space (but re-enable TF when returning to user mode).
904 if (condition
& DR_STEP
) {
906 * We already checked v86 mode above, so we can
907 * check for kernel mode by just checking the CPL
910 if (!user_mode(regs
))
911 goto clear_TF_reenable
;
914 /* Ok, finally something we can handle */
915 send_sigtrap(tsk
, regs
, error_code
);
917 /* Disable additional traps. They'll be re-enabled when
918 * the signal is delivered.
925 handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
, 1);
929 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
930 regs
->eflags
&= ~TF_MASK
;
935 * Note that we play around with the 'TS' bit in an attempt to get
936 * the correct behaviour even in the presence of the asynchronous
939 void math_error(void __user
*eip
)
941 struct task_struct
* task
;
943 unsigned short cwd
, swd
;
946 * Save the info for the exception handler and clear the error.
950 task
->thread
.trap_no
= 16;
951 task
->thread
.error_code
= 0;
952 info
.si_signo
= SIGFPE
;
954 info
.si_code
= __SI_FAULT
;
957 * (~cwd & swd) will mask out exceptions that are not set to unmasked
958 * status. 0x3f is the exception bits in these regs, 0x200 is the
959 * C1 reg you need in case of a stack fault, 0x040 is the stack
960 * fault bit. We should only be taking one exception at a time,
961 * so if this combination doesn't produce any single exception,
962 * then we have a bad program that isn't syncronizing its FPU usage
963 * and it will suffer the consequences since we won't be able to
964 * fully reproduce the context of the exception
966 cwd
= get_fpu_cwd(task
);
967 swd
= get_fpu_swd(task
);
968 switch (swd
& ~cwd
& 0x3f) {
969 case 0x000: /* No unmasked exception */
971 default: /* Multiple exceptions */
973 case 0x001: /* Invalid Op */
975 * swd & 0x240 == 0x040: Stack Underflow
976 * swd & 0x240 == 0x240: Stack Overflow
977 * User must clear the SF bit (0x40) if set
979 info
.si_code
= FPE_FLTINV
;
981 case 0x002: /* Denormalize */
982 case 0x010: /* Underflow */
983 info
.si_code
= FPE_FLTUND
;
985 case 0x004: /* Zero Divide */
986 info
.si_code
= FPE_FLTDIV
;
988 case 0x008: /* Overflow */
989 info
.si_code
= FPE_FLTOVF
;
991 case 0x020: /* Precision */
992 info
.si_code
= FPE_FLTRES
;
995 force_sig_info(SIGFPE
, &info
, task
);
998 fastcall
void do_coprocessor_error(struct pt_regs
* regs
, long error_code
)
1001 math_error((void __user
*)regs
->eip
);
1004 static void simd_math_error(void __user
*eip
)
1006 struct task_struct
* task
;
1008 unsigned short mxcsr
;
1011 * Save the info for the exception handler and clear the error.
1014 save_init_fpu(task
);
1015 task
->thread
.trap_no
= 19;
1016 task
->thread
.error_code
= 0;
1017 info
.si_signo
= SIGFPE
;
1019 info
.si_code
= __SI_FAULT
;
1022 * The SIMD FPU exceptions are handled a little differently, as there
1023 * is only a single status/control register. Thus, to determine which
1024 * unmasked exception was caught we must mask the exception mask bits
1025 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1027 mxcsr
= get_fpu_mxcsr(task
);
1028 switch (~((mxcsr
& 0x1f80) >> 7) & (mxcsr
& 0x3f)) {
1032 case 0x001: /* Invalid Op */
1033 info
.si_code
= FPE_FLTINV
;
1035 case 0x002: /* Denormalize */
1036 case 0x010: /* Underflow */
1037 info
.si_code
= FPE_FLTUND
;
1039 case 0x004: /* Zero Divide */
1040 info
.si_code
= FPE_FLTDIV
;
1042 case 0x008: /* Overflow */
1043 info
.si_code
= FPE_FLTOVF
;
1045 case 0x020: /* Precision */
1046 info
.si_code
= FPE_FLTRES
;
1049 force_sig_info(SIGFPE
, &info
, task
);
1052 fastcall
void do_simd_coprocessor_error(struct pt_regs
* regs
,
1056 /* Handle SIMD FPU exceptions on PIII+ processors. */
1058 simd_math_error((void __user
*)regs
->eip
);
1061 * Handle strange cache flush from user space exception
1062 * in all other cases. This is undocumented behaviour.
1064 if (regs
->eflags
& VM_MASK
) {
1065 handle_vm86_fault((struct kernel_vm86_regs
*)regs
,
1069 current
->thread
.trap_no
= 19;
1070 current
->thread
.error_code
= error_code
;
1071 die_if_kernel("cache flush denied", regs
, error_code
);
1072 force_sig(SIGSEGV
, current
);
1076 fastcall
void do_spurious_interrupt_bug(struct pt_regs
* regs
,
1080 /* No need to warn about this any longer. */
1081 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1085 fastcall
void setup_x86_bogus_stack(unsigned char * stk
)
1087 unsigned long *switch16_ptr
, *switch32_ptr
;
1088 struct pt_regs
*regs
;
1089 unsigned long stack_top
, stack_bot
;
1090 unsigned short iret_frame16_off
;
1091 int cpu
= smp_processor_id();
1092 /* reserve the space on 32bit stack for the magic switch16 pointer */
1093 memmove(stk
, stk
+ 8, sizeof(struct pt_regs
));
1094 switch16_ptr
= (unsigned long *)(stk
+ sizeof(struct pt_regs
));
1095 regs
= (struct pt_regs
*)stk
;
1096 /* now the switch32 on 16bit stack */
1097 stack_bot
= (unsigned long)&per_cpu(cpu_16bit_stack
, cpu
);
1098 stack_top
= stack_bot
+ CPU_16BIT_STACK_SIZE
;
1099 switch32_ptr
= (unsigned long *)(stack_top
- 8);
1100 iret_frame16_off
= CPU_16BIT_STACK_SIZE
- 8 - 20;
1101 /* copy iret frame on 16bit stack */
1102 memcpy((void *)(stack_bot
+ iret_frame16_off
), ®s
->eip
, 20);
1103 /* fill in the switch pointers */
1104 switch16_ptr
[0] = (regs
->esp
& 0xffff0000) | iret_frame16_off
;
1105 switch16_ptr
[1] = __ESPFIX_SS
;
1106 switch32_ptr
[0] = (unsigned long)stk
+ sizeof(struct pt_regs
) +
1107 8 - CPU_16BIT_STACK_SIZE
;
1108 switch32_ptr
[1] = __KERNEL_DS
;
1111 fastcall
unsigned char * fixup_x86_bogus_stack(unsigned short sp
)
1113 unsigned long *switch32_ptr
;
1114 unsigned char *stack16
, *stack32
;
1115 unsigned long stack_top
, stack_bot
;
1117 int cpu
= smp_processor_id();
1118 stack_bot
= (unsigned long)&per_cpu(cpu_16bit_stack
, cpu
);
1119 stack_top
= stack_bot
+ CPU_16BIT_STACK_SIZE
;
1120 switch32_ptr
= (unsigned long *)(stack_top
- 8);
1121 /* copy the data from 16bit stack to 32bit stack */
1122 len
= CPU_16BIT_STACK_SIZE
- 8 - sp
;
1123 stack16
= (unsigned char *)(stack_bot
+ sp
);
1124 stack32
= (unsigned char *)
1125 (switch32_ptr
[0] + CPU_16BIT_STACK_SIZE
- 8 - len
);
1126 memcpy(stack32
, stack16
, len
);
1131 * 'math_state_restore()' saves the current math information in the
1132 * old math state array, and gets the new ones from the current task
1134 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1135 * Don't touch unless you *really* know how it works.
1137 * Must be called with kernel preemption disabled (in this case,
1138 * local interrupts are disabled at the call-site in entry.S).
1140 asmlinkage
void math_state_restore(struct pt_regs regs
)
1142 struct thread_info
*thread
= current_thread_info();
1143 struct task_struct
*tsk
= thread
->task
;
1145 clts(); /* Allow maths ops (or we recurse) */
1146 if (!tsk_used_math(tsk
))
1149 thread
->status
|= TS_USEDFPU
; /* So we fnsave on switch_to() */
1152 #ifndef CONFIG_MATH_EMULATION
1154 asmlinkage
void math_emulate(long arg
)
1156 printk(KERN_EMERG
"math-emulation not enabled and no coprocessor found.\n");
1157 printk(KERN_EMERG
"killing %s.\n",current
->comm
);
1158 force_sig(SIGFPE
,current
);
1162 #endif /* CONFIG_MATH_EMULATION */
1164 #ifdef CONFIG_X86_F00F_BUG
1165 void __init
trap_init_f00f_bug(void)
1167 __set_fixmap(FIX_F00F_IDT
, __pa(&idt_table
), PAGE_KERNEL_RO
);
1170 * Update the IDT descriptor and reload the IDT so that
1171 * it uses the read-only mapped virtual address.
1173 idt_descr
.address
= fix_to_virt(FIX_F00F_IDT
);
1174 load_idt(&idt_descr
);
1179 * This needs to use 'idt_table' rather than 'idt', and
1180 * thus use the _nonmapped_ version of the IDT, as the
1181 * Pentium F0 0F bugfix can have resulted in the mapped
1182 * IDT being write-protected.
1184 void set_intr_gate(unsigned int n
, void *addr
)
1186 _set_gate(n
, DESCTYPE_INT
, addr
, __KERNEL_CS
);
1190 * This routine sets up an interrupt gate at directory privilege level 3.
1192 static inline void set_system_intr_gate(unsigned int n
, void *addr
)
1194 _set_gate(n
, DESCTYPE_INT
| DESCTYPE_DPL3
, addr
, __KERNEL_CS
);
1197 static void __init
set_trap_gate(unsigned int n
, void *addr
)
1199 _set_gate(n
, DESCTYPE_TRAP
, addr
, __KERNEL_CS
);
1202 static void __init
set_system_gate(unsigned int n
, void *addr
)
1204 _set_gate(n
, DESCTYPE_TRAP
| DESCTYPE_DPL3
, addr
, __KERNEL_CS
);
1207 static void __init
set_task_gate(unsigned int n
, unsigned int gdt_entry
)
1209 _set_gate(n
, DESCTYPE_TASK
, (void *)0, (gdt_entry
<<3));
1213 void __init
trap_init(void)
1216 void __iomem
*p
= ioremap(0x0FFFD9, 4);
1217 if (readl(p
) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1223 #ifdef CONFIG_X86_LOCAL_APIC
1224 init_apic_mappings();
1227 set_trap_gate(0,÷_error
);
1228 set_intr_gate(1,&debug
);
1229 set_intr_gate(2,&nmi
);
1230 set_system_intr_gate(3, &int3
); /* int3/4 can be called from all */
1231 set_system_gate(4,&overflow
);
1232 set_trap_gate(5,&bounds
);
1233 set_trap_gate(6,&invalid_op
);
1234 set_trap_gate(7,&device_not_available
);
1235 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS
);
1236 set_trap_gate(9,&coprocessor_segment_overrun
);
1237 set_trap_gate(10,&invalid_TSS
);
1238 set_trap_gate(11,&segment_not_present
);
1239 set_trap_gate(12,&stack_segment
);
1240 set_trap_gate(13,&general_protection
);
1241 set_intr_gate(14,&page_fault
);
1242 set_trap_gate(15,&spurious_interrupt_bug
);
1243 set_trap_gate(16,&coprocessor_error
);
1244 set_trap_gate(17,&alignment_check
);
1245 #ifdef CONFIG_X86_MCE
1246 set_trap_gate(18,&machine_check
);
1248 set_trap_gate(19,&simd_coprocessor_error
);
1252 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1253 * Generates a compile-time "error: zero width for bit-field" if
1254 * the alignment is wrong.
1256 struct fxsrAlignAssert
{
1257 int _
:!(offsetof(struct task_struct
,
1258 thread
.i387
.fxsave
) & 15);
1261 printk(KERN_INFO
"Enabling fast FPU save and restore... ");
1262 set_in_cr4(X86_CR4_OSFXSR
);
1266 printk(KERN_INFO
"Enabling unmasked SIMD FPU exception "
1268 set_in_cr4(X86_CR4_OSXMMEXCPT
);
1272 set_system_gate(SYSCALL_VECTOR
,&system_call
);
1275 * Should be a barrier for any external CPU state.
1282 static int __init
kstack_setup(char *s
)
1284 kstack_depth_to_print
= simple_strtoul(s
, NULL
, 0);
1287 __setup("kstack=", kstack_setup
);
1289 #ifdef CONFIG_STACK_UNWIND
1290 static int __init
call_trace_setup(char *s
)
1292 if (strcmp(s
, "old") == 0)
1294 else if (strcmp(s
, "both") == 0)
1296 else if (strcmp(s
, "newfallback") == 0)
1298 else if (strcmp(s
, "new") == 2)
1302 __setup("call_trace=", call_trace_setup
);