2 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
9 * 'Traps.c' handles hardware traps and faults after we have saved some
12 #include <linux/sched.h>
13 #include <linux/kernel.h>
14 #include <linux/string.h>
15 #include <linux/errno.h>
16 #include <linux/timer.h>
18 #include <linux/init.h>
19 #include <linux/delay.h>
20 #include <linux/spinlock.h>
21 #include <linux/interrupt.h>
22 #include <linux/highmem.h>
23 #include <linux/kallsyms.h>
24 #include <linux/ptrace.h>
25 #include <linux/utsname.h>
26 #include <linux/kprobes.h>
27 #include <linux/kexec.h>
28 #include <linux/unwind.h>
29 #include <linux/uaccess.h>
30 #include <linux/nmi.h>
31 #include <linux/bug.h>
34 #include <linux/ioport.h>
35 #include <linux/eisa.h>
39 #include <linux/mca.h>
42 #if defined(CONFIG_EDAC)
43 #include <linux/edac.h>
46 #include <asm/processor.h>
47 #include <asm/system.h>
49 #include <asm/atomic.h>
50 #include <asm/debugreg.h>
54 #include <asm/unwind.h>
56 #include <asm/arch_hooks.h>
57 #include <linux/kdebug.h>
58 #include <asm/stacktrace.h>
60 #include <linux/module.h>
62 #include "mach_traps.h"
64 int panic_on_unrecovered_nmi
;
66 DECLARE_BITMAP(used_vectors
, NR_VECTORS
);
67 EXPORT_SYMBOL_GPL(used_vectors
);
69 asmlinkage
int system_call(void);
71 /* Do we ignore FPU interrupts ? */
72 char ignore_fpu_irq
= 0;
75 * The IDT has to be page-aligned to simplify the Pentium
76 * F0 0F bug workaround.. We have a special link segment
79 gate_desc idt_table
[256]
80 __attribute__((__section__(".data.idt"))) = { { { { 0, 0 } } }, };
82 asmlinkage
void divide_error(void);
83 asmlinkage
void debug(void);
84 asmlinkage
void nmi(void);
85 asmlinkage
void int3(void);
86 asmlinkage
void overflow(void);
87 asmlinkage
void bounds(void);
88 asmlinkage
void invalid_op(void);
89 asmlinkage
void device_not_available(void);
90 asmlinkage
void coprocessor_segment_overrun(void);
91 asmlinkage
void invalid_TSS(void);
92 asmlinkage
void segment_not_present(void);
93 asmlinkage
void stack_segment(void);
94 asmlinkage
void general_protection(void);
95 asmlinkage
void page_fault(void);
96 asmlinkage
void coprocessor_error(void);
97 asmlinkage
void simd_coprocessor_error(void);
98 asmlinkage
void alignment_check(void);
99 asmlinkage
void spurious_interrupt_bug(void);
100 asmlinkage
void machine_check(void);
102 int kstack_depth_to_print
= 24;
103 static unsigned int code_bytes
= 64;
105 static inline int valid_stack_ptr(struct thread_info
*tinfo
, void *p
, unsigned size
)
107 return p
> (void *)tinfo
&&
108 p
<= (void *)tinfo
+ THREAD_SIZE
- size
;
111 /* The form of the top of the frame on the stack */
113 struct stack_frame
*next_frame
;
114 unsigned long return_address
;
117 static inline unsigned long print_context_stack(struct thread_info
*tinfo
,
118 unsigned long *stack
, unsigned long bp
,
119 const struct stacktrace_ops
*ops
, void *data
)
121 #ifdef CONFIG_FRAME_POINTER
122 struct stack_frame
*frame
= (struct stack_frame
*)bp
;
123 while (valid_stack_ptr(tinfo
, frame
, sizeof(*frame
))) {
124 struct stack_frame
*next
;
127 addr
= frame
->return_address
;
128 ops
->address(data
, addr
);
130 * break out of recursive entries (such as
131 * end_of_stack_stop_unwind_function). Also,
132 * we can never allow a frame pointer to
135 next
= frame
->next_frame
;
141 while (valid_stack_ptr(tinfo
, stack
, sizeof(*stack
))) {
145 if (__kernel_text_address(addr
))
146 ops
->address(data
, addr
);
152 #define MSG(msg) ops->warning(data, msg)
154 void dump_trace(struct task_struct
*task
, struct pt_regs
*regs
,
155 unsigned long *stack
,
156 const struct stacktrace_ops
*ops
, void *data
)
158 unsigned long bp
= 0;
167 stack
= (unsigned long *)task
->thread
.sp
;
170 #ifdef CONFIG_FRAME_POINTER
172 if (task
== current
) {
173 /* Grab bp right from our regs */
174 asm ("movl %%ebp, %0" : "=r" (bp
) : );
176 /* bp is the last reg pushed by switch_to */
177 bp
= *(unsigned long *) task
->thread
.sp
;
183 struct thread_info
*context
;
184 context
= (struct thread_info
*)
185 ((unsigned long)stack
& (~(THREAD_SIZE
- 1)));
186 bp
= print_context_stack(context
, stack
, bp
, ops
, data
);
187 /* Should be after the line below, but somewhere
188 in early boot context comes out corrupted and we
189 can't reference it -AK */
190 if (ops
->stack(data
, "IRQ") < 0)
192 stack
= (unsigned long*)context
->previous_esp
;
195 touch_nmi_watchdog();
198 EXPORT_SYMBOL(dump_trace
);
201 print_trace_warning_symbol(void *data
, char *msg
, unsigned long symbol
)
204 print_symbol(msg
, symbol
);
208 static void print_trace_warning(void *data
, char *msg
)
210 printk("%s%s\n", (char *)data
, msg
);
213 static int print_trace_stack(void *data
, char *name
)
219 * Print one address/symbol entries per line.
221 static void print_trace_address(void *data
, unsigned long addr
)
223 printk("%s [<%08lx>] ", (char *)data
, addr
);
224 print_symbol("%s\n", addr
);
225 touch_nmi_watchdog();
228 static const struct stacktrace_ops print_trace_ops
= {
229 .warning
= print_trace_warning
,
230 .warning_symbol
= print_trace_warning_symbol
,
231 .stack
= print_trace_stack
,
232 .address
= print_trace_address
,
236 show_trace_log_lvl(struct task_struct
*task
, struct pt_regs
*regs
,
237 unsigned long * stack
, char *log_lvl
)
239 dump_trace(task
, regs
, stack
, &print_trace_ops
, log_lvl
);
240 printk("%s =======================\n", log_lvl
);
243 void show_trace(struct task_struct
*task
, struct pt_regs
*regs
,
244 unsigned long * stack
)
246 show_trace_log_lvl(task
, regs
, stack
, "");
249 static void show_stack_log_lvl(struct task_struct
*task
, struct pt_regs
*regs
,
250 unsigned long *sp
, char *log_lvl
)
252 unsigned long *stack
;
257 sp
= (unsigned long*)task
->thread
.sp
;
259 sp
= (unsigned long *)&sp
;
263 for(i
= 0; i
< kstack_depth_to_print
; i
++) {
264 if (kstack_end(stack
))
266 if (i
&& ((i
% 8) == 0))
267 printk("\n%s ", log_lvl
);
268 printk("%08lx ", *stack
++);
270 printk("\n%sCall Trace:\n", log_lvl
);
271 show_trace_log_lvl(task
, regs
, sp
, log_lvl
);
274 void show_stack(struct task_struct
*task
, unsigned long *sp
)
277 show_stack_log_lvl(task
, NULL
, sp
, "");
281 * The architecture-independent dump_stack generator
283 void dump_stack(void)
287 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
288 current
->pid
, current
->comm
, print_tainted(),
289 init_utsname()->release
,
290 (int)strcspn(init_utsname()->version
, " "),
291 init_utsname()->version
);
292 show_trace(current
, NULL
, &stack
);
295 EXPORT_SYMBOL(dump_stack
);
297 void show_registers(struct pt_regs
*regs
)
302 __show_registers(regs
, 0);
303 printk(KERN_EMERG
"Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
304 TASK_COMM_LEN
, current
->comm
, task_pid_nr(current
),
305 current_thread_info(), current
, task_thread_info(current
));
307 * When in-kernel, we also print out the stack and code at the
308 * time of the fault..
310 if (!user_mode_vm(regs
)) {
312 unsigned int code_prologue
= code_bytes
* 43 / 64;
313 unsigned int code_len
= code_bytes
;
316 printk("\n" KERN_EMERG
"Stack: ");
317 show_stack_log_lvl(NULL
, regs
, ®s
->sp
, KERN_EMERG
);
319 printk(KERN_EMERG
"Code: ");
321 ip
= (u8
*)regs
->ip
- code_prologue
;
322 if (ip
< (u8
*)PAGE_OFFSET
||
323 probe_kernel_address(ip
, c
)) {
324 /* try starting at EIP */
326 code_len
= code_len
- code_prologue
+ 1;
328 for (i
= 0; i
< code_len
; i
++, ip
++) {
329 if (ip
< (u8
*)PAGE_OFFSET
||
330 probe_kernel_address(ip
, c
)) {
331 printk(" Bad EIP value.");
334 if (ip
== (u8
*)regs
->ip
)
335 printk("<%02x> ", c
);
343 int is_valid_bugaddr(unsigned long ip
)
347 if (ip
< PAGE_OFFSET
)
349 if (probe_kernel_address((unsigned short *)ip
, ud2
))
352 return ud2
== 0x0b0f;
355 static int die_counter
;
357 int __kprobes
__die(const char * str
, struct pt_regs
* regs
, long err
)
362 printk(KERN_EMERG
"%s: %04lx [#%d] ", str
, err
& 0xffff, ++die_counter
);
363 #ifdef CONFIG_PREEMPT
369 #ifdef CONFIG_DEBUG_PAGEALLOC
370 printk("DEBUG_PAGEALLOC");
374 if (notify_die(DIE_OOPS
, str
, regs
, err
,
375 current
->thread
.trap_no
, SIGSEGV
) !=
377 show_registers(regs
);
378 /* Executive summary in case the oops scrolled away */
379 sp
= (unsigned long) (®s
->sp
);
381 if (user_mode(regs
)) {
383 ss
= regs
->ss
& 0xffff;
385 printk(KERN_EMERG
"EIP: [<%08lx>] ", regs
->ip
);
386 print_symbol("%s", regs
->ip
);
387 printk(" SS:ESP %04x:%08lx\n", ss
, sp
);
395 * This is gone through when something in the kernel has done something bad and
396 * is about to be terminated.
398 void die(const char * str
, struct pt_regs
* regs
, long err
)
403 int lock_owner_depth
;
405 .lock
= __RAW_SPIN_LOCK_UNLOCKED
,
407 .lock_owner_depth
= 0
413 if (die
.lock_owner
!= raw_smp_processor_id()) {
415 raw_local_irq_save(flags
);
416 __raw_spin_lock(&die
.lock
);
417 die
.lock_owner
= smp_processor_id();
418 die
.lock_owner_depth
= 0;
421 raw_local_irq_save(flags
);
423 if (++die
.lock_owner_depth
< 3) {
424 report_bug(regs
->ip
, regs
);
426 if (__die(str
, regs
, err
))
429 printk(KERN_EMERG
"Recursive die() failure, output suppressed\n");
434 add_taint(TAINT_DIE
);
435 __raw_spin_unlock(&die
.lock
);
436 raw_local_irq_restore(flags
);
441 if (kexec_should_crash(current
))
445 panic("Fatal exception in interrupt");
448 panic("Fatal exception");
454 static inline void die_if_kernel(const char * str
, struct pt_regs
* regs
, long err
)
456 if (!user_mode_vm(regs
))
460 static void __kprobes
do_trap(int trapnr
, int signr
, char *str
, int vm86
,
461 struct pt_regs
* regs
, long error_code
,
464 struct task_struct
*tsk
= current
;
466 if (regs
->flags
& VM_MASK
) {
472 if (!user_mode(regs
))
477 * We want error_code and trap_no set for userspace faults and
478 * kernelspace faults which result in die(), but not
479 * kernelspace faults which are fixed up. die() gives the
480 * process no chance to handle the signal and notice the
481 * kernel fault information, so that won't result in polluting
482 * the information about previously queued, but not yet
483 * delivered, faults. See also do_general_protection below.
485 tsk
->thread
.error_code
= error_code
;
486 tsk
->thread
.trap_no
= trapnr
;
489 force_sig_info(signr
, info
, tsk
);
491 force_sig(signr
, tsk
);
496 if (!fixup_exception(regs
)) {
497 tsk
->thread
.error_code
= error_code
;
498 tsk
->thread
.trap_no
= trapnr
;
499 die(str
, regs
, error_code
);
505 int ret
= handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
, trapnr
);
506 if (ret
) goto trap_signal
;
511 #define DO_ERROR(trapnr, signr, str, name) \
512 void do_##name(struct pt_regs * regs, long error_code) \
514 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
517 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
520 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
521 void do_##name(struct pt_regs * regs, long error_code) \
525 local_irq_enable(); \
526 info.si_signo = signr; \
528 info.si_code = sicode; \
529 info.si_addr = (void __user *)siaddr; \
530 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
533 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
536 #define DO_VM86_ERROR(trapnr, signr, str, name) \
537 void do_##name(struct pt_regs * regs, long error_code) \
539 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
542 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
545 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
546 void do_##name(struct pt_regs * regs, long error_code) \
549 info.si_signo = signr; \
551 info.si_code = sicode; \
552 info.si_addr = (void __user *)siaddr; \
553 trace_hardirqs_fixup(); \
554 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
557 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
560 DO_VM86_ERROR_INFO( 0, SIGFPE
, "divide error", divide_error
, FPE_INTDIV
, regs
->ip
)
561 #ifndef CONFIG_KPROBES
562 DO_VM86_ERROR( 3, SIGTRAP
, "int3", int3
)
564 DO_VM86_ERROR( 4, SIGSEGV
, "overflow", overflow
)
565 DO_VM86_ERROR( 5, SIGSEGV
, "bounds", bounds
)
566 DO_ERROR_INFO( 6, SIGILL
, "invalid opcode", invalid_op
, ILL_ILLOPN
, regs
->ip
, 0)
567 DO_ERROR( 9, SIGFPE
, "coprocessor segment overrun", coprocessor_segment_overrun
)
568 DO_ERROR(10, SIGSEGV
, "invalid TSS", invalid_TSS
)
569 DO_ERROR(11, SIGBUS
, "segment not present", segment_not_present
)
570 DO_ERROR(12, SIGBUS
, "stack segment", stack_segment
)
571 DO_ERROR_INFO(17, SIGBUS
, "alignment check", alignment_check
, BUS_ADRALN
, 0, 0)
572 DO_ERROR_INFO(32, SIGSEGV
, "iret exception", iret_error
, ILL_BADSTK
, 0, 1)
574 void __kprobes
do_general_protection(struct pt_regs
* regs
,
578 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
579 struct thread_struct
*thread
= ¤t
->thread
;
582 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
583 * invalid offset set (the LAZY one) and the faulting thread has
584 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
585 * and we set the offset field correctly. Then we let the CPU to
586 * restart the faulting instruction.
588 if (tss
->x86_tss
.io_bitmap_base
== INVALID_IO_BITMAP_OFFSET_LAZY
&&
589 thread
->io_bitmap_ptr
) {
590 memcpy(tss
->io_bitmap
, thread
->io_bitmap_ptr
,
591 thread
->io_bitmap_max
);
593 * If the previously set map was extending to higher ports
594 * than the current one, pad extra space with 0xff (no access).
596 if (thread
->io_bitmap_max
< tss
->io_bitmap_max
)
597 memset((char *) tss
->io_bitmap
+
598 thread
->io_bitmap_max
, 0xff,
599 tss
->io_bitmap_max
- thread
->io_bitmap_max
);
600 tss
->io_bitmap_max
= thread
->io_bitmap_max
;
601 tss
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
602 tss
->io_bitmap_owner
= thread
;
608 if (regs
->flags
& VM_MASK
)
611 if (!user_mode(regs
))
614 current
->thread
.error_code
= error_code
;
615 current
->thread
.trap_no
= 13;
616 if (show_unhandled_signals
&& unhandled_signal(current
, SIGSEGV
) &&
619 "%s[%d] general protection ip:%lx sp:%lx error:%lx\n",
620 current
->comm
, task_pid_nr(current
),
621 regs
->ip
, regs
->sp
, error_code
);
623 force_sig(SIGSEGV
, current
);
628 handle_vm86_fault((struct kernel_vm86_regs
*) regs
, error_code
);
632 if (!fixup_exception(regs
)) {
633 current
->thread
.error_code
= error_code
;
634 current
->thread
.trap_no
= 13;
635 if (notify_die(DIE_GPF
, "general protection fault", regs
,
636 error_code
, 13, SIGSEGV
) == NOTIFY_STOP
)
638 die("general protection fault", regs
, error_code
);
642 static __kprobes
void
643 mem_parity_error(unsigned char reason
, struct pt_regs
* regs
)
645 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x on "
646 "CPU %d.\n", reason
, smp_processor_id());
647 printk(KERN_EMERG
"You have some hardware problem, likely on the PCI bus.\n");
649 #if defined(CONFIG_EDAC)
650 if(edac_handler_set()) {
651 edac_atomic_assert_error();
656 if (panic_on_unrecovered_nmi
)
657 panic("NMI: Not continuing");
659 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
661 /* Clear and disable the memory parity error line. */
662 clear_mem_error(reason
);
665 static __kprobes
void
666 io_check_error(unsigned char reason
, struct pt_regs
* regs
)
670 printk(KERN_EMERG
"NMI: IOCK error (debug interrupt?)\n");
671 show_registers(regs
);
673 /* Re-enable the IOCK line, wait for a few seconds */
674 reason
= (reason
& 0xf) | 8;
677 while (--i
) udelay(1000);
682 static __kprobes
void
683 unknown_nmi_error(unsigned char reason
, struct pt_regs
* regs
)
686 /* Might actually be able to figure out what the guilty party
693 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x on "
694 "CPU %d.\n", reason
, smp_processor_id());
695 printk(KERN_EMERG
"Do you have a strange power saving mode enabled?\n");
696 if (panic_on_unrecovered_nmi
)
697 panic("NMI: Not continuing");
699 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
702 static DEFINE_SPINLOCK(nmi_print_lock
);
704 void __kprobes
die_nmi(struct pt_regs
*regs
, const char *msg
)
706 if (notify_die(DIE_NMIWATCHDOG
, msg
, regs
, 0, 2, SIGINT
) ==
710 spin_lock(&nmi_print_lock
);
712 * We are in trouble anyway, lets at least try
713 * to get a message out.
716 printk(KERN_EMERG
"%s", msg
);
717 printk(" on CPU%d, ip %08lx, registers:\n",
718 smp_processor_id(), regs
->ip
);
719 show_registers(regs
);
721 spin_unlock(&nmi_print_lock
);
724 /* If we are in kernel we are probably nested up pretty bad
725 * and might aswell get out now while we still can.
727 if (!user_mode_vm(regs
)) {
728 current
->thread
.trap_no
= 2;
735 static __kprobes
void default_do_nmi(struct pt_regs
* regs
)
737 unsigned char reason
= 0;
739 /* Only the BSP gets external NMIs from the system. */
740 if (!smp_processor_id())
741 reason
= get_nmi_reason();
743 if (!(reason
& 0xc0)) {
744 if (notify_die(DIE_NMI_IPI
, "nmi_ipi", regs
, reason
, 2, SIGINT
)
747 #ifdef CONFIG_X86_LOCAL_APIC
749 * Ok, so this is none of the documented NMI sources,
750 * so it must be the NMI watchdog.
752 if (nmi_watchdog_tick(regs
, reason
))
754 if (!do_nmi_callback(regs
, smp_processor_id()))
756 unknown_nmi_error(reason
, regs
);
760 if (notify_die(DIE_NMI
, "nmi", regs
, reason
, 2, SIGINT
) == NOTIFY_STOP
)
763 mem_parity_error(reason
, regs
);
765 io_check_error(reason
, regs
);
767 * Reassert NMI in case it became active meanwhile
768 * as it's edge-triggered.
773 static int ignore_nmis
;
775 __kprobes
void do_nmi(struct pt_regs
* regs
, long error_code
)
781 cpu
= smp_processor_id();
786 default_do_nmi(regs
);
797 void restart_nmi(void)
803 #ifdef CONFIG_KPROBES
804 void __kprobes
do_int3(struct pt_regs
*regs
, long error_code
)
806 trace_hardirqs_fixup();
808 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, 3, SIGTRAP
)
811 /* This is an interrupt gate, because kprobes wants interrupts
812 disabled. Normal trap handlers don't. */
813 restore_interrupts(regs
);
814 do_trap(3, SIGTRAP
, "int3", 1, regs
, error_code
, NULL
);
819 * Our handling of the processor debug registers is non-trivial.
820 * We do not clear them on entry and exit from the kernel. Therefore
821 * it is possible to get a watchpoint trap here from inside the kernel.
822 * However, the code in ./ptrace.c has ensured that the user can
823 * only set watchpoints on userspace addresses. Therefore the in-kernel
824 * watchpoint trap can only occur in code which is reading/writing
825 * from user space. Such code must not hold kernel locks (since it
826 * can equally take a page fault), therefore it is safe to call
827 * force_sig_info even though that claims and releases locks.
829 * Code in ./signal.c ensures that the debug control register
830 * is restored before we deliver any signal, and therefore that
831 * user code runs with the correct debug control register even though
834 * Being careful here means that we don't have to be as careful in a
835 * lot of more complicated places (task switching can be a bit lazy
836 * about restoring all the debug state, and ptrace doesn't have to
837 * find every occurrence of the TF bit that could be saved away even
840 void __kprobes
do_debug(struct pt_regs
* regs
, long error_code
)
842 unsigned int condition
;
843 struct task_struct
*tsk
= current
;
845 trace_hardirqs_fixup();
847 get_debugreg(condition
, 6);
850 * The processor cleared BTF, so don't mark that we need it set.
852 clear_tsk_thread_flag(tsk
, TIF_DEBUGCTLMSR
);
853 tsk
->thread
.debugctlmsr
= 0;
855 if (notify_die(DIE_DEBUG
, "debug", regs
, condition
, error_code
,
856 SIGTRAP
) == NOTIFY_STOP
)
858 /* It's safe to allow irq's after DR6 has been saved */
859 if (regs
->flags
& X86_EFLAGS_IF
)
862 /* Mask out spurious debug traps due to lazy DR7 setting */
863 if (condition
& (DR_TRAP0
|DR_TRAP1
|DR_TRAP2
|DR_TRAP3
)) {
864 if (!tsk
->thread
.debugreg7
)
868 if (regs
->flags
& VM_MASK
)
871 /* Save debug status register where ptrace can see it */
872 tsk
->thread
.debugreg6
= condition
;
875 * Single-stepping through TF: make sure we ignore any events in
876 * kernel space (but re-enable TF when returning to user mode).
878 if (condition
& DR_STEP
) {
880 * We already checked v86 mode above, so we can
881 * check for kernel mode by just checking the CPL
884 if (!user_mode(regs
))
885 goto clear_TF_reenable
;
888 /* Ok, finally something we can handle */
889 send_sigtrap(tsk
, regs
, error_code
);
891 /* Disable additional traps. They'll be re-enabled when
892 * the signal is delivered.
899 handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
, 1);
903 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
904 regs
->flags
&= ~TF_MASK
;
909 * Note that we play around with the 'TS' bit in an attempt to get
910 * the correct behaviour even in the presence of the asynchronous
913 void math_error(void __user
*ip
)
915 struct task_struct
* task
;
917 unsigned short cwd
, swd
;
920 * Save the info for the exception handler and clear the error.
924 task
->thread
.trap_no
= 16;
925 task
->thread
.error_code
= 0;
926 info
.si_signo
= SIGFPE
;
928 info
.si_code
= __SI_FAULT
;
931 * (~cwd & swd) will mask out exceptions that are not set to unmasked
932 * status. 0x3f is the exception bits in these regs, 0x200 is the
933 * C1 reg you need in case of a stack fault, 0x040 is the stack
934 * fault bit. We should only be taking one exception at a time,
935 * so if this combination doesn't produce any single exception,
936 * then we have a bad program that isn't syncronizing its FPU usage
937 * and it will suffer the consequences since we won't be able to
938 * fully reproduce the context of the exception
940 cwd
= get_fpu_cwd(task
);
941 swd
= get_fpu_swd(task
);
942 switch (swd
& ~cwd
& 0x3f) {
943 case 0x000: /* No unmasked exception */
945 default: /* Multiple exceptions */
947 case 0x001: /* Invalid Op */
949 * swd & 0x240 == 0x040: Stack Underflow
950 * swd & 0x240 == 0x240: Stack Overflow
951 * User must clear the SF bit (0x40) if set
953 info
.si_code
= FPE_FLTINV
;
955 case 0x002: /* Denormalize */
956 case 0x010: /* Underflow */
957 info
.si_code
= FPE_FLTUND
;
959 case 0x004: /* Zero Divide */
960 info
.si_code
= FPE_FLTDIV
;
962 case 0x008: /* Overflow */
963 info
.si_code
= FPE_FLTOVF
;
965 case 0x020: /* Precision */
966 info
.si_code
= FPE_FLTRES
;
969 force_sig_info(SIGFPE
, &info
, task
);
972 void do_coprocessor_error(struct pt_regs
* regs
, long error_code
)
975 math_error((void __user
*)regs
->ip
);
978 static void simd_math_error(void __user
*ip
)
980 struct task_struct
* task
;
982 unsigned short mxcsr
;
985 * Save the info for the exception handler and clear the error.
989 task
->thread
.trap_no
= 19;
990 task
->thread
.error_code
= 0;
991 info
.si_signo
= SIGFPE
;
993 info
.si_code
= __SI_FAULT
;
996 * The SIMD FPU exceptions are handled a little differently, as there
997 * is only a single status/control register. Thus, to determine which
998 * unmasked exception was caught we must mask the exception mask bits
999 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
1001 mxcsr
= get_fpu_mxcsr(task
);
1002 switch (~((mxcsr
& 0x1f80) >> 7) & (mxcsr
& 0x3f)) {
1006 case 0x001: /* Invalid Op */
1007 info
.si_code
= FPE_FLTINV
;
1009 case 0x002: /* Denormalize */
1010 case 0x010: /* Underflow */
1011 info
.si_code
= FPE_FLTUND
;
1013 case 0x004: /* Zero Divide */
1014 info
.si_code
= FPE_FLTDIV
;
1016 case 0x008: /* Overflow */
1017 info
.si_code
= FPE_FLTOVF
;
1019 case 0x020: /* Precision */
1020 info
.si_code
= FPE_FLTRES
;
1023 force_sig_info(SIGFPE
, &info
, task
);
1026 void do_simd_coprocessor_error(struct pt_regs
* regs
,
1030 /* Handle SIMD FPU exceptions on PIII+ processors. */
1032 simd_math_error((void __user
*)regs
->ip
);
1035 * Handle strange cache flush from user space exception
1036 * in all other cases. This is undocumented behaviour.
1038 if (regs
->flags
& VM_MASK
) {
1039 handle_vm86_fault((struct kernel_vm86_regs
*)regs
,
1043 current
->thread
.trap_no
= 19;
1044 current
->thread
.error_code
= error_code
;
1045 die_if_kernel("cache flush denied", regs
, error_code
);
1046 force_sig(SIGSEGV
, current
);
1050 void do_spurious_interrupt_bug(struct pt_regs
* regs
,
1054 /* No need to warn about this any longer. */
1055 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1059 unsigned long patch_espfix_desc(unsigned long uesp
,
1062 struct desc_struct
*gdt
= __get_cpu_var(gdt_page
).gdt
;
1063 unsigned long base
= (kesp
- uesp
) & -THREAD_SIZE
;
1064 unsigned long new_kesp
= kesp
- base
;
1065 unsigned long lim_pages
= (new_kesp
| (THREAD_SIZE
- 1)) >> PAGE_SHIFT
;
1066 __u64 desc
= *(__u64
*)&gdt
[GDT_ENTRY_ESPFIX_SS
];
1067 /* Set up base for espfix segment */
1068 desc
&= 0x00f0ff0000000000ULL
;
1069 desc
|= ((((__u64
)base
) << 16) & 0x000000ffffff0000ULL
) |
1070 ((((__u64
)base
) << 32) & 0xff00000000000000ULL
) |
1071 ((((__u64
)lim_pages
) << 32) & 0x000f000000000000ULL
) |
1072 (lim_pages
& 0xffff);
1073 *(__u64
*)&gdt
[GDT_ENTRY_ESPFIX_SS
] = desc
;
1078 * 'math_state_restore()' saves the current math information in the
1079 * old math state array, and gets the new ones from the current task
1081 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1082 * Don't touch unless you *really* know how it works.
1084 * Must be called with kernel preemption disabled (in this case,
1085 * local interrupts are disabled at the call-site in entry.S).
1087 asmlinkage
void math_state_restore(void)
1089 struct thread_info
*thread
= current_thread_info();
1090 struct task_struct
*tsk
= thread
->task
;
1092 clts(); /* Allow maths ops (or we recurse) */
1093 if (!tsk_used_math(tsk
))
1096 thread
->status
|= TS_USEDFPU
; /* So we fnsave on switch_to() */
1099 EXPORT_SYMBOL_GPL(math_state_restore
);
1101 #ifndef CONFIG_MATH_EMULATION
1103 asmlinkage
void math_emulate(long arg
)
1105 printk(KERN_EMERG
"math-emulation not enabled and no coprocessor found.\n");
1106 printk(KERN_EMERG
"killing %s.\n",current
->comm
);
1107 force_sig(SIGFPE
,current
);
1111 #endif /* CONFIG_MATH_EMULATION */
1114 void __init
trap_init(void)
1119 void __iomem
*p
= ioremap(0x0FFFD9, 4);
1120 if (readl(p
) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1126 #ifdef CONFIG_X86_LOCAL_APIC
1127 init_apic_mappings();
1130 set_trap_gate(0,÷_error
);
1131 set_intr_gate(1,&debug
);
1132 set_intr_gate(2,&nmi
);
1133 set_system_intr_gate(3, &int3
); /* int3/4 can be called from all */
1134 set_system_gate(4,&overflow
);
1135 set_trap_gate(5,&bounds
);
1136 set_trap_gate(6,&invalid_op
);
1137 set_trap_gate(7,&device_not_available
);
1138 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS
);
1139 set_trap_gate(9,&coprocessor_segment_overrun
);
1140 set_trap_gate(10,&invalid_TSS
);
1141 set_trap_gate(11,&segment_not_present
);
1142 set_trap_gate(12,&stack_segment
);
1143 set_trap_gate(13,&general_protection
);
1144 set_intr_gate(14,&page_fault
);
1145 set_trap_gate(15,&spurious_interrupt_bug
);
1146 set_trap_gate(16,&coprocessor_error
);
1147 set_trap_gate(17,&alignment_check
);
1148 #ifdef CONFIG_X86_MCE
1149 set_trap_gate(18,&machine_check
);
1151 set_trap_gate(19,&simd_coprocessor_error
);
1155 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1156 * Generates a compile-time "error: zero width for bit-field" if
1157 * the alignment is wrong.
1159 struct fxsrAlignAssert
{
1160 int _
:!(offsetof(struct task_struct
,
1161 thread
.i387
.fxsave
) & 15);
1164 printk(KERN_INFO
"Enabling fast FPU save and restore... ");
1165 set_in_cr4(X86_CR4_OSFXSR
);
1169 printk(KERN_INFO
"Enabling unmasked SIMD FPU exception "
1171 set_in_cr4(X86_CR4_OSXMMEXCPT
);
1175 set_system_gate(SYSCALL_VECTOR
,&system_call
);
1177 /* Reserve all the builtin and the syscall vector. */
1178 for (i
= 0; i
< FIRST_EXTERNAL_VECTOR
; i
++)
1179 set_bit(i
, used_vectors
);
1180 set_bit(SYSCALL_VECTOR
, used_vectors
);
1183 * Should be a barrier for any external CPU state.
1190 static int __init
kstack_setup(char *s
)
1192 kstack_depth_to_print
= simple_strtoul(s
, NULL
, 0);
1195 __setup("kstack=", kstack_setup
);
1197 static int __init
code_bytes_setup(char *s
)
1199 code_bytes
= simple_strtoul(s
, NULL
, 0);
1200 if (code_bytes
> 8192)
1205 __setup("code_bytes=", code_bytes_setup
);