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/interrupt.h>
14 #include <linux/kallsyms.h>
15 #include <linux/spinlock.h>
16 #include <linux/highmem.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>
37 #include <linux/ioport.h>
38 #include <linux/eisa.h>
42 #include <linux/mca.h>
45 #if defined(CONFIG_EDAC)
46 #include <linux/edac.h>
49 #include <asm/arch_hooks.h>
50 #include <asm/stacktrace.h>
51 #include <asm/processor.h>
52 #include <asm/debugreg.h>
53 #include <asm/atomic.h>
54 #include <asm/system.h>
55 #include <asm/unwind.h>
61 #include <asm/traps.h>
63 #include "mach_traps.h"
65 DECLARE_BITMAP(used_vectors
, NR_VECTORS
);
66 EXPORT_SYMBOL_GPL(used_vectors
);
68 asmlinkage
int system_call(void);
70 /* Do we ignore FPU interrupts ? */
74 * The IDT has to be page-aligned to simplify the Pentium
75 * F0 0F bug workaround.. We have a special link segment
78 gate_desc idt_table
[256]
79 __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
81 int panic_on_unrecovered_nmi
;
82 int kstack_depth_to_print
= 24;
83 static unsigned int code_bytes
= 64;
84 static int ignore_nmis
;
85 static int die_counter
;
87 static inline void conditional_sti(struct pt_regs
*regs
)
89 if (regs
->flags
& X86_EFLAGS_IF
)
93 void printk_address(unsigned long address
, int reliable
)
95 #ifdef CONFIG_KALLSYMS
96 unsigned long offset
= 0;
97 unsigned long symsize
;
101 char namebuf
[KSYM_NAME_LEN
];
104 symname
= kallsyms_lookup(address
, &symsize
, &offset
,
107 printk(" [<%08lx>]\n", address
);
111 strcpy(reliab
, "? ");
114 modname
= delim
= "";
115 printk(" [<%08lx>] %s%s%s%s%s+0x%lx/0x%lx\n",
116 address
, reliab
, delim
, modname
, delim
, symname
, offset
, symsize
);
118 printk(" [<%08lx>]\n", address
);
122 static inline int valid_stack_ptr(struct thread_info
*tinfo
,
123 void *p
, unsigned int size
)
126 return p
> t
&& p
<= t
+ THREAD_SIZE
- size
;
129 /* The form of the top of the frame on the stack */
131 struct stack_frame
*next_frame
;
132 unsigned long return_address
;
135 static inline unsigned long
136 print_context_stack(struct thread_info
*tinfo
,
137 unsigned long *stack
, unsigned long bp
,
138 const struct stacktrace_ops
*ops
, void *data
)
140 struct stack_frame
*frame
= (struct stack_frame
*)bp
;
142 while (valid_stack_ptr(tinfo
, stack
, sizeof(*stack
))) {
146 if (__kernel_text_address(addr
)) {
147 if ((unsigned long) stack
== bp
+ 4) {
148 ops
->address(data
, addr
, 1);
149 frame
= frame
->next_frame
;
150 bp
= (unsigned long) frame
;
152 ops
->address(data
, addr
, bp
== 0);
160 void dump_trace(struct task_struct
*task
, struct pt_regs
*regs
,
161 unsigned long *stack
, unsigned long bp
,
162 const struct stacktrace_ops
*ops
, void *data
)
171 stack
= (unsigned long *)task
->thread
.sp
;
174 #ifdef CONFIG_FRAME_POINTER
176 if (task
== current
) {
177 /* Grab bp right from our regs */
178 asm("movl %%ebp, %0" : "=r" (bp
) :);
180 /* bp is the last reg pushed by switch_to */
181 bp
= *(unsigned long *) task
->thread
.sp
;
187 struct thread_info
*context
;
189 context
= (struct thread_info
*)
190 ((unsigned long)stack
& (~(THREAD_SIZE
- 1)));
191 bp
= print_context_stack(context
, stack
, bp
, ops
, data
);
193 * Should be after the line below, but somewhere
194 * in early boot context comes out corrupted and we
195 * can't reference it:
197 if (ops
->stack(data
, "IRQ") < 0)
199 stack
= (unsigned long *)context
->previous_esp
;
202 touch_nmi_watchdog();
205 EXPORT_SYMBOL(dump_trace
);
208 print_trace_warning_symbol(void *data
, char *msg
, unsigned long symbol
)
211 print_symbol(msg
, symbol
);
215 static void print_trace_warning(void *data
, char *msg
)
217 printk("%s%s\n", (char *)data
, msg
);
220 static int print_trace_stack(void *data
, char *name
)
226 * Print one address/symbol entries per line.
228 static void print_trace_address(void *data
, unsigned long addr
, int reliable
)
230 printk("%s [<%08lx>] ", (char *)data
, addr
);
233 print_symbol("%s\n", addr
);
234 touch_nmi_watchdog();
237 static const struct stacktrace_ops print_trace_ops
= {
238 .warning
= print_trace_warning
,
239 .warning_symbol
= print_trace_warning_symbol
,
240 .stack
= print_trace_stack
,
241 .address
= print_trace_address
,
245 show_trace_log_lvl(struct task_struct
*task
, struct pt_regs
*regs
,
246 unsigned long *stack
, unsigned long bp
, char *log_lvl
)
248 dump_trace(task
, regs
, stack
, bp
, &print_trace_ops
, log_lvl
);
249 printk("%s =======================\n", log_lvl
);
252 void show_trace(struct task_struct
*task
, struct pt_regs
*regs
,
253 unsigned long *stack
, unsigned long bp
)
255 show_trace_log_lvl(task
, regs
, stack
, bp
, "");
259 show_stack_log_lvl(struct task_struct
*task
, struct pt_regs
*regs
,
260 unsigned long *sp
, unsigned long bp
, char *log_lvl
)
262 unsigned long *stack
;
267 sp
= (unsigned long *)task
->thread
.sp
;
269 sp
= (unsigned long *)&sp
;
273 for (i
= 0; i
< kstack_depth_to_print
; i
++) {
274 if (kstack_end(stack
))
276 if (i
&& ((i
% 8) == 0))
277 printk("\n%s ", log_lvl
);
278 printk("%08lx ", *stack
++);
280 printk("\n%sCall Trace:\n", log_lvl
);
282 show_trace_log_lvl(task
, regs
, sp
, bp
, log_lvl
);
285 void show_stack(struct task_struct
*task
, unsigned long *sp
)
288 show_stack_log_lvl(task
, NULL
, sp
, 0, "");
292 * The architecture-independent dump_stack generator
294 void dump_stack(void)
296 unsigned long bp
= 0;
299 #ifdef CONFIG_FRAME_POINTER
301 asm("movl %%ebp, %0" : "=r" (bp
):);
304 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
305 current
->pid
, current
->comm
, print_tainted(),
306 init_utsname()->release
,
307 (int)strcspn(init_utsname()->version
, " "),
308 init_utsname()->version
);
310 show_trace(current
, NULL
, &stack
, bp
);
313 EXPORT_SYMBOL(dump_stack
);
315 void show_registers(struct pt_regs
*regs
)
320 __show_registers(regs
, 0);
322 printk(KERN_EMERG
"Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
323 TASK_COMM_LEN
, current
->comm
, task_pid_nr(current
),
324 current_thread_info(), current
, task_thread_info(current
));
326 * When in-kernel, we also print out the stack and code at the
327 * time of the fault..
329 if (!user_mode_vm(regs
)) {
330 unsigned int code_prologue
= code_bytes
* 43 / 64;
331 unsigned int code_len
= code_bytes
;
335 printk("\n" KERN_EMERG
"Stack: ");
336 show_stack_log_lvl(NULL
, regs
, ®s
->sp
, 0, KERN_EMERG
);
338 printk(KERN_EMERG
"Code: ");
340 ip
= (u8
*)regs
->ip
- code_prologue
;
341 if (ip
< (u8
*)PAGE_OFFSET
|| probe_kernel_address(ip
, c
)) {
342 /* try starting at EIP */
344 code_len
= code_len
- code_prologue
+ 1;
346 for (i
= 0; i
< code_len
; i
++, ip
++) {
347 if (ip
< (u8
*)PAGE_OFFSET
||
348 probe_kernel_address(ip
, c
)) {
349 printk(" Bad EIP value.");
352 if (ip
== (u8
*)regs
->ip
)
353 printk("<%02x> ", c
);
361 int is_valid_bugaddr(unsigned long ip
)
365 if (ip
< PAGE_OFFSET
)
367 if (probe_kernel_address((unsigned short *)ip
, ud2
))
370 return ud2
== 0x0b0f;
373 static raw_spinlock_t die_lock
= __RAW_SPIN_LOCK_UNLOCKED
;
374 static int die_owner
= -1;
375 static unsigned int die_nest_count
;
377 unsigned __kprobes
long oops_begin(void)
383 if (die_owner
!= raw_smp_processor_id()) {
385 raw_local_irq_save(flags
);
386 __raw_spin_lock(&die_lock
);
387 die_owner
= smp_processor_id();
391 raw_local_irq_save(flags
);
397 void __kprobes
oops_end(unsigned long flags
, struct pt_regs
*regs
, int signr
)
401 add_taint(TAINT_DIE
);
402 __raw_spin_unlock(&die_lock
);
403 raw_local_irq_restore(flags
);
408 if (kexec_should_crash(current
))
412 panic("Fatal exception in interrupt");
415 panic("Fatal exception");
421 int __kprobes
__die(const char *str
, struct pt_regs
*regs
, long err
)
426 printk(KERN_EMERG
"%s: %04lx [#%d] ", str
, err
& 0xffff, ++die_counter
);
427 #ifdef CONFIG_PREEMPT
433 #ifdef CONFIG_DEBUG_PAGEALLOC
434 printk("DEBUG_PAGEALLOC");
437 if (notify_die(DIE_OOPS
, str
, regs
, err
,
438 current
->thread
.trap_no
, SIGSEGV
) == NOTIFY_STOP
)
441 show_registers(regs
);
442 /* Executive summary in case the oops scrolled away */
443 sp
= (unsigned long) (®s
->sp
);
445 if (user_mode(regs
)) {
447 ss
= regs
->ss
& 0xffff;
449 printk(KERN_EMERG
"EIP: [<%08lx>] ", regs
->ip
);
450 print_symbol("%s", regs
->ip
);
451 printk(" SS:ESP %04x:%08lx\n", ss
, sp
);
456 * This is gone through when something in the kernel has done something bad
457 * and is about to be terminated:
459 void die(const char *str
, struct pt_regs
*regs
, long err
)
461 unsigned long flags
= oops_begin();
463 if (die_nest_count
< 3) {
464 report_bug(regs
->ip
, regs
);
466 if (__die(str
, regs
, err
))
469 printk(KERN_EMERG
"Recursive die() failure, output suppressed\n");
472 oops_end(flags
, regs
, SIGSEGV
);
476 die_if_kernel(const char *str
, struct pt_regs
*regs
, long err
)
478 if (!user_mode_vm(regs
))
482 static void __kprobes
483 do_trap(int trapnr
, int signr
, char *str
, int vm86
, struct pt_regs
*regs
,
484 long error_code
, siginfo_t
*info
)
486 struct task_struct
*tsk
= current
;
488 if (regs
->flags
& X86_VM_MASK
) {
494 if (!user_mode(regs
))
499 * We want error_code and trap_no set for userspace faults and
500 * kernelspace faults which result in die(), but not
501 * kernelspace faults which are fixed up. die() gives the
502 * process no chance to handle the signal and notice the
503 * kernel fault information, so that won't result in polluting
504 * the information about previously queued, but not yet
505 * delivered, faults. See also do_general_protection below.
507 tsk
->thread
.error_code
= error_code
;
508 tsk
->thread
.trap_no
= trapnr
;
511 force_sig_info(signr
, info
, tsk
);
513 force_sig(signr
, tsk
);
517 if (!fixup_exception(regs
)) {
518 tsk
->thread
.error_code
= error_code
;
519 tsk
->thread
.trap_no
= trapnr
;
520 die(str
, regs
, error_code
);
525 if (handle_vm86_trap((struct kernel_vm86_regs
*) regs
,
531 #define DO_TRAP(trapnr, signr, str, name) \
532 void do_##name(struct pt_regs *regs, long error_code) \
534 trace_hardirqs_fixup(); \
535 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
538 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
541 #define DO_TRAP_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
542 void do_##name(struct pt_regs *regs, long error_code) \
546 local_irq_enable(); \
547 info.si_signo = signr; \
549 info.si_code = sicode; \
550 info.si_addr = (void __user *)siaddr; \
551 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
554 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
557 #define DO_VM86_TRAP(trapnr, signr, str, name) \
558 void do_##name(struct pt_regs *regs, long error_code) \
560 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
563 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
566 #define DO_VM86_TRAP_INFO(trapnr, signr, str, name, sicode, siaddr) \
567 void do_##name(struct pt_regs *regs, long error_code) \
570 info.si_signo = signr; \
572 info.si_code = sicode; \
573 info.si_addr = (void __user *)siaddr; \
574 trace_hardirqs_fixup(); \
575 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
578 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
581 #define DO_ERROR(trapnr, signr, str, name) \
582 void do_##name(struct pt_regs *regs, long error_code) \
584 trace_hardirqs_fixup(); \
585 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
588 conditional_sti(regs); \
589 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
592 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
593 void do_##name(struct pt_regs *regs, long error_code) \
597 local_irq_enable(); \
598 info.si_signo = signr; \
600 info.si_code = sicode; \
601 info.si_addr = (void __user *)siaddr; \
602 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
605 conditional_sti(regs); \
606 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
609 #define DO_VM86_ERROR(trapnr, signr, str, name) \
610 void do_##name(struct pt_regs *regs, long error_code) \
612 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
615 conditional_sti(regs); \
616 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
619 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
620 void do_##name(struct pt_regs *regs, long error_code) \
623 info.si_signo = signr; \
625 info.si_code = sicode; \
626 info.si_addr = (void __user *)siaddr; \
627 trace_hardirqs_fixup(); \
628 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
631 conditional_sti(regs); \
632 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
635 DO_VM86_ERROR_INFO(0, SIGFPE
, "divide error", divide_error
, FPE_INTDIV
, regs
->ip
)
636 DO_VM86_ERROR(4, SIGSEGV
, "overflow", overflow
)
637 DO_VM86_ERROR(5, SIGSEGV
, "bounds", bounds
)
638 DO_TRAP_INFO(6, SIGILL
, "invalid opcode", invalid_op
, ILL_ILLOPN
, regs
->ip
, 0)
639 DO_TRAP(9, SIGFPE
, "coprocessor segment overrun", coprocessor_segment_overrun
)
640 DO_TRAP(10, SIGSEGV
, "invalid TSS", invalid_TSS
)
641 DO_TRAP(11, SIGBUS
, "segment not present", segment_not_present
)
642 DO_TRAP(12, SIGBUS
, "stack segment", stack_segment
)
643 DO_TRAP_INFO(17, SIGBUS
, "alignment check", alignment_check
, BUS_ADRALN
, 0, 0)
644 DO_TRAP_INFO(32, SIGILL
, "iret exception", iret_error
, ILL_BADSTK
, 0, 1)
647 do_general_protection(struct pt_regs
*regs
, long error_code
)
649 struct task_struct
*tsk
;
650 struct thread_struct
*thread
;
651 struct tss_struct
*tss
;
655 tss
= &per_cpu(init_tss
, cpu
);
656 thread
= ¤t
->thread
;
659 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
660 * invalid offset set (the LAZY one) and the faulting thread has
661 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
662 * and we set the offset field correctly. Then we let the CPU to
663 * restart the faulting instruction.
665 if (tss
->x86_tss
.io_bitmap_base
== INVALID_IO_BITMAP_OFFSET_LAZY
&&
666 thread
->io_bitmap_ptr
) {
667 memcpy(tss
->io_bitmap
, thread
->io_bitmap_ptr
,
668 thread
->io_bitmap_max
);
670 * If the previously set map was extending to higher ports
671 * than the current one, pad extra space with 0xff (no access).
673 if (thread
->io_bitmap_max
< tss
->io_bitmap_max
) {
674 memset((char *) tss
->io_bitmap
+
675 thread
->io_bitmap_max
, 0xff,
676 tss
->io_bitmap_max
- thread
->io_bitmap_max
);
678 tss
->io_bitmap_max
= thread
->io_bitmap_max
;
679 tss
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
680 tss
->io_bitmap_owner
= thread
;
687 if (regs
->flags
& X86_VM_MASK
)
691 if (!user_mode(regs
))
694 tsk
->thread
.error_code
= error_code
;
695 tsk
->thread
.trap_no
= 13;
697 if (show_unhandled_signals
&& unhandled_signal(tsk
, SIGSEGV
) &&
698 printk_ratelimit()) {
700 "%s[%d] general protection ip:%lx sp:%lx error:%lx",
701 tsk
->comm
, task_pid_nr(tsk
),
702 regs
->ip
, regs
->sp
, error_code
);
703 print_vma_addr(" in ", regs
->ip
);
707 force_sig(SIGSEGV
, tsk
);
712 handle_vm86_fault((struct kernel_vm86_regs
*) regs
, error_code
);
716 if (fixup_exception(regs
))
719 tsk
->thread
.error_code
= error_code
;
720 tsk
->thread
.trap_no
= 13;
721 if (notify_die(DIE_GPF
, "general protection fault", regs
,
722 error_code
, 13, SIGSEGV
) == NOTIFY_STOP
)
724 die("general protection fault", regs
, error_code
);
727 static notrace __kprobes
void
728 mem_parity_error(unsigned char reason
, struct pt_regs
*regs
)
731 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
732 reason
, smp_processor_id());
735 "You have some hardware problem, likely on the PCI bus.\n");
737 #if defined(CONFIG_EDAC)
738 if (edac_handler_set()) {
739 edac_atomic_assert_error();
744 if (panic_on_unrecovered_nmi
)
745 panic("NMI: Not continuing");
747 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
749 /* Clear and disable the memory parity error line. */
750 clear_mem_error(reason
);
753 static notrace __kprobes
void
754 io_check_error(unsigned char reason
, struct pt_regs
*regs
)
758 printk(KERN_EMERG
"NMI: IOCK error (debug interrupt?)\n");
759 show_registers(regs
);
761 /* Re-enable the IOCK line, wait for a few seconds */
762 reason
= (reason
& 0xf) | 8;
773 static notrace __kprobes
void
774 unknown_nmi_error(unsigned char reason
, struct pt_regs
*regs
)
776 if (notify_die(DIE_NMIUNKNOWN
, "nmi", regs
, reason
, 2, SIGINT
) == NOTIFY_STOP
)
780 * Might actually be able to figure out what the guilty party
789 "Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
790 reason
, smp_processor_id());
792 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 static DEFINE_SPINLOCK(nmi_print_lock
);
801 void notrace __kprobes
die_nmi(char *str
, struct pt_regs
*regs
, int do_panic
)
803 if (notify_die(DIE_NMIWATCHDOG
, str
, regs
, 0, 2, SIGINT
) == NOTIFY_STOP
)
806 spin_lock(&nmi_print_lock
);
808 * We are in trouble anyway, lets at least try
809 * to get a message out:
812 printk(KERN_EMERG
"%s", str
);
813 printk(" on CPU%d, ip %08lx, registers:\n",
814 smp_processor_id(), regs
->ip
);
815 show_registers(regs
);
817 panic("Non maskable interrupt");
819 spin_unlock(&nmi_print_lock
);
823 * If we are in kernel we are probably nested up pretty bad
824 * and might aswell get out now while we still can:
826 if (!user_mode_vm(regs
)) {
827 current
->thread
.trap_no
= 2;
834 static notrace __kprobes
void default_do_nmi(struct pt_regs
*regs
)
836 unsigned char reason
= 0;
839 cpu
= smp_processor_id();
841 /* Only the BSP gets external NMIs from the system. */
843 reason
= get_nmi_reason();
845 if (!(reason
& 0xc0)) {
846 if (notify_die(DIE_NMI_IPI
, "nmi_ipi", regs
, reason
, 2, SIGINT
)
849 #ifdef CONFIG_X86_LOCAL_APIC
851 * Ok, so this is none of the documented NMI sources,
852 * so it must be the NMI watchdog.
854 if (nmi_watchdog_tick(regs
, reason
))
856 if (!do_nmi_callback(regs
, cpu
))
857 unknown_nmi_error(reason
, regs
);
859 unknown_nmi_error(reason
, regs
);
864 if (notify_die(DIE_NMI
, "nmi", regs
, reason
, 2, SIGINT
) == NOTIFY_STOP
)
867 /* AK: following checks seem to be broken on modern chipsets. FIXME */
869 mem_parity_error(reason
, regs
);
871 io_check_error(reason
, regs
);
873 * Reassert NMI in case it became active meanwhile
874 * as it's edge-triggered:
879 notrace __kprobes
void do_nmi(struct pt_regs
*regs
, long error_code
)
885 cpu
= smp_processor_id();
890 default_do_nmi(regs
);
901 void restart_nmi(void)
907 void __kprobes
do_int3(struct pt_regs
*regs
, long error_code
)
909 #ifdef CONFIG_KPROBES
910 trace_hardirqs_fixup();
912 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, 3, SIGTRAP
)
916 * This is an interrupt gate, because kprobes wants interrupts
917 * disabled. Normal trap handlers don't.
919 conditional_sti(regs
);
921 if (notify_die(DIE_TRAP
, "int3", regs
, error_code
, 3, SIGTRAP
)
926 do_trap(3, SIGTRAP
, "int3", 1, regs
, error_code
, NULL
);
930 * Our handling of the processor debug registers is non-trivial.
931 * We do not clear them on entry and exit from the kernel. Therefore
932 * it is possible to get a watchpoint trap here from inside the kernel.
933 * However, the code in ./ptrace.c has ensured that the user can
934 * only set watchpoints on userspace addresses. Therefore the in-kernel
935 * watchpoint trap can only occur in code which is reading/writing
936 * from user space. Such code must not hold kernel locks (since it
937 * can equally take a page fault), therefore it is safe to call
938 * force_sig_info even though that claims and releases locks.
940 * Code in ./signal.c ensures that the debug control register
941 * is restored before we deliver any signal, and therefore that
942 * user code runs with the correct debug control register even though
945 * Being careful here means that we don't have to be as careful in a
946 * lot of more complicated places (task switching can be a bit lazy
947 * about restoring all the debug state, and ptrace doesn't have to
948 * find every occurrence of the TF bit that could be saved away even
951 void __kprobes
do_debug(struct pt_regs
*regs
, long error_code
)
953 struct task_struct
*tsk
= current
;
954 unsigned int condition
;
957 trace_hardirqs_fixup();
959 get_debugreg(condition
, 6);
962 * The processor cleared BTF, so don't mark that we need it set.
964 clear_tsk_thread_flag(tsk
, TIF_DEBUGCTLMSR
);
965 tsk
->thread
.debugctlmsr
= 0;
967 if (notify_die(DIE_DEBUG
, "debug", regs
, condition
, error_code
,
968 SIGTRAP
) == NOTIFY_STOP
)
970 /* It's safe to allow irq's after DR6 has been saved */
971 if (regs
->flags
& X86_EFLAGS_IF
)
974 /* Mask out spurious debug traps due to lazy DR7 setting */
975 if (condition
& (DR_TRAP0
|DR_TRAP1
|DR_TRAP2
|DR_TRAP3
)) {
976 if (!tsk
->thread
.debugreg7
)
980 if (regs
->flags
& X86_VM_MASK
)
983 /* Save debug status register where ptrace can see it */
984 tsk
->thread
.debugreg6
= condition
;
987 * Single-stepping through TF: make sure we ignore any events in
988 * kernel space (but re-enable TF when returning to user mode).
990 if (condition
& DR_STEP
) {
992 * We already checked v86 mode above, so we can
993 * check for kernel mode by just checking the CPL
996 if (!user_mode(regs
))
997 goto clear_TF_reenable
;
1000 si_code
= get_si_code((unsigned long)condition
);
1001 /* Ok, finally something we can handle */
1002 send_sigtrap(tsk
, regs
, error_code
, si_code
);
1005 * Disable additional traps. They'll be re-enabled when
1006 * the signal is delivered.
1013 handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
, 1);
1017 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
1018 regs
->flags
&= ~X86_EFLAGS_TF
;
1023 * Note that we play around with the 'TS' bit in an attempt to get
1024 * the correct behaviour even in the presence of the asynchronous
1027 void math_error(void __user
*ip
)
1029 struct task_struct
*task
;
1031 unsigned short cwd
, swd
;
1034 * Save the info for the exception handler and clear the error.
1037 save_init_fpu(task
);
1038 task
->thread
.trap_no
= 16;
1039 task
->thread
.error_code
= 0;
1040 info
.si_signo
= SIGFPE
;
1042 info
.si_code
= __SI_FAULT
;
1045 * (~cwd & swd) will mask out exceptions that are not set to unmasked
1046 * status. 0x3f is the exception bits in these regs, 0x200 is the
1047 * C1 reg you need in case of a stack fault, 0x040 is the stack
1048 * fault bit. We should only be taking one exception at a time,
1049 * so if this combination doesn't produce any single exception,
1050 * then we have a bad program that isn't synchronizing its FPU usage
1051 * and it will suffer the consequences since we won't be able to
1052 * fully reproduce the context of the exception
1054 cwd
= get_fpu_cwd(task
);
1055 swd
= get_fpu_swd(task
);
1056 switch (swd
& ~cwd
& 0x3f) {
1057 case 0x000: /* No unmasked exception */
1059 default: /* Multiple exceptions */
1061 case 0x001: /* Invalid Op */
1063 * swd & 0x240 == 0x040: Stack Underflow
1064 * swd & 0x240 == 0x240: Stack Overflow
1065 * User must clear the SF bit (0x40) if set
1067 info
.si_code
= FPE_FLTINV
;
1069 case 0x002: /* Denormalize */
1070 case 0x010: /* Underflow */
1071 info
.si_code
= FPE_FLTUND
;
1073 case 0x004: /* Zero Divide */
1074 info
.si_code
= FPE_FLTDIV
;
1076 case 0x008: /* Overflow */
1077 info
.si_code
= FPE_FLTOVF
;
1079 case 0x020: /* Precision */
1080 info
.si_code
= FPE_FLTRES
;
1083 force_sig_info(SIGFPE
, &info
, task
);
1086 void do_coprocessor_error(struct pt_regs
*regs
, long error_code
)
1089 math_error((void __user
*)regs
->ip
);
1092 static void simd_math_error(void __user
*ip
)
1094 struct task_struct
*task
;
1096 unsigned short mxcsr
;
1099 * Save the info for the exception handler and clear the error.
1102 save_init_fpu(task
);
1103 task
->thread
.trap_no
= 19;
1104 task
->thread
.error_code
= 0;
1105 info
.si_signo
= SIGFPE
;
1107 info
.si_code
= __SI_FAULT
;
1110 * The SIMD FPU exceptions are handled a little differently, as there
1111 * is only a single status/control register. Thus, to determine which
1112 * unmasked exception was caught we must mask the exception mask bits
1113 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1115 mxcsr
= get_fpu_mxcsr(task
);
1116 switch (~((mxcsr
& 0x1f80) >> 7) & (mxcsr
& 0x3f)) {
1120 case 0x001: /* Invalid Op */
1121 info
.si_code
= FPE_FLTINV
;
1123 case 0x002: /* Denormalize */
1124 case 0x010: /* Underflow */
1125 info
.si_code
= FPE_FLTUND
;
1127 case 0x004: /* Zero Divide */
1128 info
.si_code
= FPE_FLTDIV
;
1130 case 0x008: /* Overflow */
1131 info
.si_code
= FPE_FLTOVF
;
1133 case 0x020: /* Precision */
1134 info
.si_code
= FPE_FLTRES
;
1137 force_sig_info(SIGFPE
, &info
, task
);
1140 void do_simd_coprocessor_error(struct pt_regs
*regs
, long error_code
)
1143 /* Handle SIMD FPU exceptions on PIII+ processors. */
1145 simd_math_error((void __user
*)regs
->ip
);
1149 * Handle strange cache flush from user space exception
1150 * in all other cases. This is undocumented behaviour.
1152 if (regs
->flags
& X86_VM_MASK
) {
1153 handle_vm86_fault((struct kernel_vm86_regs
*)regs
, error_code
);
1156 current
->thread
.trap_no
= 19;
1157 current
->thread
.error_code
= error_code
;
1158 die_if_kernel("cache flush denied", regs
, error_code
);
1159 force_sig(SIGSEGV
, current
);
1162 void do_spurious_interrupt_bug(struct pt_regs
*regs
, long error_code
)
1165 /* No need to warn about this any longer. */
1166 printk(KERN_INFO
"Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1170 unsigned long patch_espfix_desc(unsigned long uesp
, unsigned long kesp
)
1172 struct desc_struct
*gdt
= get_cpu_gdt_table(smp_processor_id());
1173 unsigned long base
= (kesp
- uesp
) & -THREAD_SIZE
;
1174 unsigned long new_kesp
= kesp
- base
;
1175 unsigned long lim_pages
= (new_kesp
| (THREAD_SIZE
- 1)) >> PAGE_SHIFT
;
1176 __u64 desc
= *(__u64
*)&gdt
[GDT_ENTRY_ESPFIX_SS
];
1178 /* Set up base for espfix segment */
1179 desc
&= 0x00f0ff0000000000ULL
;
1180 desc
|= ((((__u64
)base
) << 16) & 0x000000ffffff0000ULL
) |
1181 ((((__u64
)base
) << 32) & 0xff00000000000000ULL
) |
1182 ((((__u64
)lim_pages
) << 32) & 0x000f000000000000ULL
) |
1183 (lim_pages
& 0xffff);
1184 *(__u64
*)&gdt
[GDT_ENTRY_ESPFIX_SS
] = desc
;
1190 * 'math_state_restore()' saves the current math information in the
1191 * old math state array, and gets the new ones from the current task
1193 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1194 * Don't touch unless you *really* know how it works.
1196 * Must be called with kernel preemption disabled (in this case,
1197 * local interrupts are disabled at the call-site in entry.S).
1199 asmlinkage
void math_state_restore(void)
1201 struct thread_info
*thread
= current_thread_info();
1202 struct task_struct
*tsk
= thread
->task
;
1204 if (!tsk_used_math(tsk
)) {
1207 * does a slab alloc which can sleep
1209 if (init_fpu(tsk
)) {
1211 * ran out of memory!
1213 do_group_exit(SIGKILL
);
1216 local_irq_disable();
1219 clts(); /* Allow maths ops (or we recurse) */
1221 thread
->status
|= TS_USEDFPU
; /* So we fnsave on switch_to() */
1224 EXPORT_SYMBOL_GPL(math_state_restore
);
1226 #ifndef CONFIG_MATH_EMULATION
1228 asmlinkage
void math_emulate(long arg
)
1231 "math-emulation not enabled and no coprocessor found.\n");
1232 printk(KERN_EMERG
"killing %s.\n", current
->comm
);
1233 force_sig(SIGFPE
, current
);
1237 #endif /* CONFIG_MATH_EMULATION */
1239 void __init
trap_init(void)
1244 void __iomem
*p
= early_ioremap(0x0FFFD9, 4);
1246 if (readl(p
) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
1248 early_iounmap(p
, 4);
1251 set_intr_gate(0, ÷_error
);
1252 set_intr_gate(1, &debug
);
1253 set_intr_gate(2, &nmi
);
1254 set_system_intr_gate(3, &int3
); /* int3 can be called from all */
1255 set_system_intr_gate(4, &overflow
); /* int4 can be called from all */
1256 set_intr_gate(5, &bounds
);
1257 set_trap_gate(6, &invalid_op
);
1258 set_trap_gate(7, &device_not_available
);
1259 set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS
);
1260 set_trap_gate(9, &coprocessor_segment_overrun
);
1261 set_trap_gate(10, &invalid_TSS
);
1262 set_trap_gate(11, &segment_not_present
);
1263 set_trap_gate(12, &stack_segment
);
1264 set_trap_gate(13, &general_protection
);
1265 set_intr_gate(14, &page_fault
);
1266 set_trap_gate(15, &spurious_interrupt_bug
);
1267 set_trap_gate(16, &coprocessor_error
);
1268 set_trap_gate(17, &alignment_check
);
1269 #ifdef CONFIG_X86_MCE
1270 set_trap_gate(18, &machine_check
);
1272 set_trap_gate(19, &simd_coprocessor_error
);
1275 printk(KERN_INFO
"Enabling fast FPU save and restore... ");
1276 set_in_cr4(X86_CR4_OSFXSR
);
1281 "Enabling unmasked SIMD FPU exception support... ");
1282 set_in_cr4(X86_CR4_OSXMMEXCPT
);
1286 set_system_gate(SYSCALL_VECTOR
, &system_call
);
1288 /* Reserve all the builtin and the syscall vector: */
1289 for (i
= 0; i
< FIRST_EXTERNAL_VECTOR
; i
++)
1290 set_bit(i
, used_vectors
);
1292 set_bit(SYSCALL_VECTOR
, used_vectors
);
1295 * Should be a barrier for any external CPU state:
1302 static int __init
kstack_setup(char *s
)
1304 kstack_depth_to_print
= simple_strtoul(s
, NULL
, 0);
1308 __setup("kstack=", kstack_setup
);
1310 static int __init
code_bytes_setup(char *s
)
1312 code_bytes
= simple_strtoul(s
, NULL
, 0);
1313 if (code_bytes
> 8192)
1318 __setup("code_bytes=", code_bytes_setup
);