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>
33 #include <linux/ioport.h>
34 #include <linux/eisa.h>
38 #include <linux/mca.h>
41 #include <asm/processor.h>
42 #include <asm/system.h>
43 #include <asm/uaccess.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>
55 #include <linux/module.h>
57 #include "mach_traps.h"
59 asmlinkage
int system_call(void);
61 struct desc_struct default_ldt
[] = { { 0, 0 }, { 0, 0 }, { 0, 0 },
64 /* Do we ignore FPU interrupts ? */
65 char ignore_fpu_irq
= 0;
68 * The IDT has to be page-aligned to simplify the Pentium
69 * F0 0F bug workaround.. We have a special link segment
72 struct desc_struct idt_table
[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
74 asmlinkage
void divide_error(void);
75 asmlinkage
void debug(void);
76 asmlinkage
void nmi(void);
77 asmlinkage
void int3(void);
78 asmlinkage
void overflow(void);
79 asmlinkage
void bounds(void);
80 asmlinkage
void invalid_op(void);
81 asmlinkage
void device_not_available(void);
82 asmlinkage
void coprocessor_segment_overrun(void);
83 asmlinkage
void invalid_TSS(void);
84 asmlinkage
void segment_not_present(void);
85 asmlinkage
void stack_segment(void);
86 asmlinkage
void general_protection(void);
87 asmlinkage
void page_fault(void);
88 asmlinkage
void coprocessor_error(void);
89 asmlinkage
void simd_coprocessor_error(void);
90 asmlinkage
void alignment_check(void);
91 asmlinkage
void spurious_interrupt_bug(void);
92 asmlinkage
void machine_check(void);
94 static int kstack_depth_to_print
= 24;
95 static int call_trace
= 1;
96 ATOMIC_NOTIFIER_HEAD(i386die_chain
);
98 int register_die_notifier(struct notifier_block
*nb
)
101 return atomic_notifier_chain_register(&i386die_chain
, nb
);
103 EXPORT_SYMBOL(register_die_notifier
);
105 int unregister_die_notifier(struct notifier_block
*nb
)
107 return atomic_notifier_chain_unregister(&i386die_chain
, nb
);
109 EXPORT_SYMBOL(unregister_die_notifier
);
111 static inline int valid_stack_ptr(struct thread_info
*tinfo
, void *p
)
113 return p
> (void *)tinfo
&&
114 p
< (void *)tinfo
+ THREAD_SIZE
- 3;
118 * Print one address/symbol entries per line.
120 static inline void print_addr_and_symbol(unsigned long addr
, char *log_lvl
)
122 printk(" [<%08lx>] ", addr
);
124 print_symbol("%s\n", addr
);
127 static inline unsigned long print_context_stack(struct thread_info
*tinfo
,
128 unsigned long *stack
, unsigned long ebp
,
133 #ifdef CONFIG_FRAME_POINTER
134 while (valid_stack_ptr(tinfo
, (void *)ebp
)) {
135 addr
= *(unsigned long *)(ebp
+ 4);
136 print_addr_and_symbol(addr
, log_lvl
);
138 * break out of recursive entries (such as
139 * end_of_stack_stop_unwind_function):
141 if (ebp
== *(unsigned long *)ebp
)
143 ebp
= *(unsigned long *)ebp
;
146 while (valid_stack_ptr(tinfo
, stack
)) {
148 if (__kernel_text_address(addr
))
149 print_addr_and_symbol(addr
, log_lvl
);
155 static asmlinkage
int
156 show_trace_unwind(struct unwind_frame_info
*info
, void *log_lvl
)
160 while (unwind(info
) == 0 && UNW_PC(info
)) {
162 print_addr_and_symbol(UNW_PC(info
), log_lvl
);
163 if (arch_unw_user_mode(info
))
169 static void show_trace_log_lvl(struct task_struct
*task
, struct pt_regs
*regs
,
170 unsigned long *stack
, char *log_lvl
)
177 if (call_trace
>= 0) {
179 struct unwind_frame_info info
;
182 if (unwind_init_frame_info(&info
, task
, regs
) == 0)
183 unw_ret
= show_trace_unwind(&info
, log_lvl
);
184 } else if (task
== current
)
185 unw_ret
= unwind_init_running(&info
, show_trace_unwind
, log_lvl
);
187 if (unwind_init_blocked(&info
, task
) == 0)
188 unw_ret
= show_trace_unwind(&info
, log_lvl
);
193 printk("%sLegacy call trace:\n", log_lvl
);
197 if (task
== current
) {
198 /* Grab ebp right from our regs */
199 asm ("movl %%ebp, %0" : "=r" (ebp
) : );
201 /* ebp is the last reg pushed by switch_to */
202 ebp
= *(unsigned long *) task
->thread
.esp
;
206 struct thread_info
*context
;
207 context
= (struct thread_info
*)
208 ((unsigned long)stack
& (~(THREAD_SIZE
- 1)));
209 ebp
= print_context_stack(context
, stack
, ebp
, log_lvl
);
210 stack
= (unsigned long*)context
->previous_esp
;
213 printk("%s =======================\n", log_lvl
);
217 void show_trace(struct task_struct
*task
, struct pt_regs
*regs
, unsigned long * stack
)
219 show_trace_log_lvl(task
, regs
, stack
, "");
222 static void show_stack_log_lvl(struct task_struct
*task
, struct pt_regs
*regs
,
223 unsigned long *esp
, char *log_lvl
)
225 unsigned long *stack
;
230 esp
= (unsigned long*)task
->thread
.esp
;
232 esp
= (unsigned long *)&esp
;
236 for(i
= 0; i
< kstack_depth_to_print
; i
++) {
237 if (kstack_end(stack
))
239 if (i
&& ((i
% 8) == 0))
240 printk("\n%s ", log_lvl
);
241 printk("%08lx ", *stack
++);
243 printk("\n%sCall Trace:\n", log_lvl
);
244 show_trace_log_lvl(task
, regs
, esp
, log_lvl
);
247 void show_stack(struct task_struct
*task
, unsigned long *esp
)
250 show_stack_log_lvl(task
, NULL
, esp
, "");
254 * The architecture-independent dump_stack generator
256 void dump_stack(void)
260 show_trace(current
, NULL
, &stack
);
263 EXPORT_SYMBOL(dump_stack
);
265 void show_registers(struct pt_regs
*regs
)
272 esp
= (unsigned long) (®s
->esp
);
274 if (user_mode_vm(regs
)) {
277 ss
= regs
->xss
& 0xffff;
280 printk(KERN_EMERG
"CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\n"
281 "EFLAGS: %08lx (%s %.*s) \n",
282 smp_processor_id(), 0xffff & regs
->xcs
, regs
->eip
,
283 print_tainted(), regs
->eflags
, system_utsname
.release
,
284 (int)strcspn(system_utsname
.version
, " "),
285 system_utsname
.version
);
286 print_symbol(KERN_EMERG
"EIP is at %s\n", regs
->eip
);
287 printk(KERN_EMERG
"eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
288 regs
->eax
, regs
->ebx
, regs
->ecx
, regs
->edx
);
289 printk(KERN_EMERG
"esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
290 regs
->esi
, regs
->edi
, regs
->ebp
, esp
);
291 printk(KERN_EMERG
"ds: %04x es: %04x ss: %04x\n",
292 regs
->xds
& 0xffff, regs
->xes
& 0xffff, ss
);
293 printk(KERN_EMERG
"Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
294 TASK_COMM_LEN
, current
->comm
, current
->pid
,
295 current_thread_info(), current
, current
->thread_info
);
297 * When in-kernel, we also print out the stack and code at the
298 * time of the fault..
303 printk("\n" KERN_EMERG
"Stack: ");
304 show_stack_log_lvl(NULL
, regs
, (unsigned long *)esp
, KERN_EMERG
);
306 printk(KERN_EMERG
"Code: ");
308 eip
= (u8 __user
*)regs
->eip
- 43;
309 for (i
= 0; i
< 64; i
++, eip
++) {
312 if (eip
< (u8 __user
*)PAGE_OFFSET
|| __get_user(c
, eip
)) {
313 printk(" Bad EIP value.");
316 if (eip
== (u8 __user
*)regs
->eip
)
317 printk("<%02x> ", c
);
325 static void handle_BUG(struct pt_regs
*regs
)
335 if (eip
< PAGE_OFFSET
)
337 if (__get_user(ud2
, (unsigned short __user
*)eip
))
341 if (__get_user(line
, (unsigned short __user
*)(eip
+ 2)))
343 if (__get_user(file
, (char * __user
*)(eip
+ 4)) ||
344 (unsigned long)file
< PAGE_OFFSET
|| __get_user(c
, file
))
345 file
= "<bad filename>";
347 printk(KERN_EMERG
"------------[ cut here ]------------\n");
348 printk(KERN_EMERG
"kernel BUG at %s:%d!\n", file
, line
);
353 /* Here we know it was a BUG but file-n-line is unavailable */
355 printk(KERN_EMERG
"Kernel BUG\n");
358 /* This is gone through when something in the kernel
359 * has done something bad and is about to be terminated.
361 void die(const char * str
, struct pt_regs
* regs
, long err
)
366 int lock_owner_depth
;
368 .lock
= SPIN_LOCK_UNLOCKED
,
370 .lock_owner_depth
= 0
372 static int die_counter
;
377 if (die
.lock_owner
!= raw_smp_processor_id()) {
379 spin_lock_irqsave(&die
.lock
, flags
);
380 die
.lock_owner
= smp_processor_id();
381 die
.lock_owner_depth
= 0;
385 local_save_flags(flags
);
387 if (++die
.lock_owner_depth
< 3) {
393 printk(KERN_EMERG
"%s: %04lx [#%d]\n", str
, err
& 0xffff, ++die_counter
);
394 #ifdef CONFIG_PREEMPT
395 printk(KERN_EMERG
"PREEMPT ");
404 #ifdef CONFIG_DEBUG_PAGEALLOC
407 printk("DEBUG_PAGEALLOC");
412 if (notify_die(DIE_OOPS
, str
, regs
, err
,
413 current
->thread
.trap_no
, SIGSEGV
) !=
415 show_registers(regs
);
416 /* Executive summary in case the oops scrolled away */
417 esp
= (unsigned long) (®s
->esp
);
419 if (user_mode(regs
)) {
421 ss
= regs
->xss
& 0xffff;
423 printk(KERN_EMERG
"EIP: [<%08lx>] ", regs
->eip
);
424 print_symbol("%s", regs
->eip
);
425 printk(" SS:ESP %04x:%08lx\n", ss
, esp
);
430 printk(KERN_EMERG
"Recursive die() failure, output suppressed\n");
434 spin_unlock_irqrestore(&die
.lock
, flags
);
439 if (kexec_should_crash(current
))
443 panic("Fatal exception in interrupt");
446 printk(KERN_EMERG
"Fatal exception: panic in 5 seconds\n");
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
;
465 tsk
->thread
.error_code
= error_code
;
466 tsk
->thread
.trap_no
= trapnr
;
468 if (regs
->eflags
& VM_MASK
) {
474 if (!user_mode(regs
))
479 force_sig_info(signr
, info
, tsk
);
481 force_sig(signr
, tsk
);
486 if (!fixup_exception(regs
))
487 die(str
, regs
, error_code
);
492 int ret
= handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
, trapnr
);
493 if (ret
) goto trap_signal
;
498 #define DO_ERROR(trapnr, signr, str, name) \
499 fastcall void do_##name(struct pt_regs * regs, long error_code) \
501 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
504 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
507 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
508 fastcall void do_##name(struct pt_regs * regs, long error_code) \
511 info.si_signo = signr; \
513 info.si_code = sicode; \
514 info.si_addr = (void __user *)siaddr; \
515 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
518 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
521 #define DO_VM86_ERROR(trapnr, signr, str, name) \
522 fastcall void do_##name(struct pt_regs * regs, long error_code) \
524 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
527 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
530 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
531 fastcall void do_##name(struct pt_regs * regs, long error_code) \
534 info.si_signo = signr; \
536 info.si_code = sicode; \
537 info.si_addr = (void __user *)siaddr; \
538 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
541 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
544 DO_VM86_ERROR_INFO( 0, SIGFPE
, "divide error", divide_error
, FPE_INTDIV
, regs
->eip
)
545 #ifndef CONFIG_KPROBES
546 DO_VM86_ERROR( 3, SIGTRAP
, "int3", int3
)
548 DO_VM86_ERROR( 4, SIGSEGV
, "overflow", overflow
)
549 DO_VM86_ERROR( 5, SIGSEGV
, "bounds", bounds
)
550 DO_ERROR_INFO( 6, SIGILL
, "invalid opcode", invalid_op
, ILL_ILLOPN
, regs
->eip
)
551 DO_ERROR( 9, SIGFPE
, "coprocessor segment overrun", coprocessor_segment_overrun
)
552 DO_ERROR(10, SIGSEGV
, "invalid TSS", invalid_TSS
)
553 DO_ERROR(11, SIGBUS
, "segment not present", segment_not_present
)
554 DO_ERROR(12, SIGBUS
, "stack segment", stack_segment
)
555 DO_ERROR_INFO(17, SIGBUS
, "alignment check", alignment_check
, BUS_ADRALN
, 0)
556 DO_ERROR_INFO(32, SIGSEGV
, "iret exception", iret_error
, ILL_BADSTK
, 0)
558 fastcall
void __kprobes
do_general_protection(struct pt_regs
* regs
,
562 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
563 struct thread_struct
*thread
= ¤t
->thread
;
566 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
567 * invalid offset set (the LAZY one) and the faulting thread has
568 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
569 * and we set the offset field correctly. Then we let the CPU to
570 * restart the faulting instruction.
572 if (tss
->io_bitmap_base
== INVALID_IO_BITMAP_OFFSET_LAZY
&&
573 thread
->io_bitmap_ptr
) {
574 memcpy(tss
->io_bitmap
, thread
->io_bitmap_ptr
,
575 thread
->io_bitmap_max
);
577 * If the previously set map was extending to higher ports
578 * than the current one, pad extra space with 0xff (no access).
580 if (thread
->io_bitmap_max
< tss
->io_bitmap_max
)
581 memset((char *) tss
->io_bitmap
+
582 thread
->io_bitmap_max
, 0xff,
583 tss
->io_bitmap_max
- thread
->io_bitmap_max
);
584 tss
->io_bitmap_max
= thread
->io_bitmap_max
;
585 tss
->io_bitmap_base
= IO_BITMAP_OFFSET
;
586 tss
->io_bitmap_owner
= thread
;
592 current
->thread
.error_code
= error_code
;
593 current
->thread
.trap_no
= 13;
595 if (regs
->eflags
& VM_MASK
)
598 if (!user_mode(regs
))
601 current
->thread
.error_code
= error_code
;
602 current
->thread
.trap_no
= 13;
603 force_sig(SIGSEGV
, current
);
608 handle_vm86_fault((struct kernel_vm86_regs
*) regs
, error_code
);
612 if (!fixup_exception(regs
)) {
613 if (notify_die(DIE_GPF
, "general protection fault", regs
,
614 error_code
, 13, SIGSEGV
) == NOTIFY_STOP
)
616 die("general protection fault", regs
, error_code
);
620 static void mem_parity_error(unsigned char reason
, struct pt_regs
* regs
)
622 printk(KERN_EMERG
"Uhhuh. NMI received. Dazed and confused, but trying "
624 printk(KERN_EMERG
"You probably have a hardware problem with your RAM "
627 /* Clear and disable the memory parity error line. */
628 clear_mem_error(reason
);
631 static void io_check_error(unsigned char reason
, struct pt_regs
* regs
)
635 printk(KERN_EMERG
"NMI: IOCK error (debug interrupt?)\n");
636 show_registers(regs
);
638 /* Re-enable the IOCK line, wait for a few seconds */
639 reason
= (reason
& 0xf) | 8;
642 while (--i
) udelay(1000);
647 static void unknown_nmi_error(unsigned char reason
, struct pt_regs
* regs
)
650 /* Might actually be able to figure out what the guilty party
657 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
658 reason
, smp_processor_id());
659 printk("Dazed and confused, but trying to continue\n");
660 printk("Do you have a strange power saving mode enabled?\n");
663 static DEFINE_SPINLOCK(nmi_print_lock
);
665 void die_nmi (struct pt_regs
*regs
, const char *msg
)
667 if (notify_die(DIE_NMIWATCHDOG
, msg
, regs
, 0, 2, SIGINT
) ==
671 spin_lock(&nmi_print_lock
);
673 * We are in trouble anyway, lets at least try
674 * to get a message out.
677 printk(KERN_EMERG
"%s", msg
);
678 printk(" on CPU%d, eip %08lx, registers:\n",
679 smp_processor_id(), regs
->eip
);
680 show_registers(regs
);
681 printk(KERN_EMERG
"console shuts up ...\n");
683 spin_unlock(&nmi_print_lock
);
686 /* If we are in kernel we are probably nested up pretty bad
687 * and might aswell get out now while we still can.
689 if (!user_mode_vm(regs
)) {
690 current
->thread
.trap_no
= 2;
697 static void default_do_nmi(struct pt_regs
* regs
)
699 unsigned char reason
= 0;
701 /* Only the BSP gets external NMIs from the system. */
702 if (!smp_processor_id())
703 reason
= get_nmi_reason();
705 if (!(reason
& 0xc0)) {
706 if (notify_die(DIE_NMI_IPI
, "nmi_ipi", regs
, reason
, 2, SIGINT
)
709 #ifdef CONFIG_X86_LOCAL_APIC
711 * Ok, so this is none of the documented NMI sources,
712 * so it must be the NMI watchdog.
715 nmi_watchdog_tick(regs
);
719 unknown_nmi_error(reason
, regs
);
722 if (notify_die(DIE_NMI
, "nmi", regs
, reason
, 2, SIGINT
) == NOTIFY_STOP
)
725 mem_parity_error(reason
, regs
);
727 io_check_error(reason
, regs
);
729 * Reassert NMI in case it became active meanwhile
730 * as it's edge-triggered.
735 static int dummy_nmi_callback(struct pt_regs
* regs
, int cpu
)
740 static nmi_callback_t nmi_callback
= dummy_nmi_callback
;
742 fastcall
void do_nmi(struct pt_regs
* regs
, long error_code
)
748 cpu
= smp_processor_id();
752 if (!rcu_dereference(nmi_callback
)(regs
, cpu
))
753 default_do_nmi(regs
);
758 void set_nmi_callback(nmi_callback_t callback
)
761 rcu_assign_pointer(nmi_callback
, callback
);
763 EXPORT_SYMBOL_GPL(set_nmi_callback
);
765 void unset_nmi_callback(void)
767 nmi_callback
= dummy_nmi_callback
;
769 EXPORT_SYMBOL_GPL(unset_nmi_callback
);
771 #ifdef CONFIG_KPROBES
772 fastcall
void __kprobes
do_int3(struct pt_regs
*regs
, long error_code
)
774 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, 3, SIGTRAP
)
777 /* This is an interrupt gate, because kprobes wants interrupts
778 disabled. Normal trap handlers don't. */
779 restore_interrupts(regs
);
780 do_trap(3, SIGTRAP
, "int3", 1, regs
, error_code
, NULL
);
785 * Our handling of the processor debug registers is non-trivial.
786 * We do not clear them on entry and exit from the kernel. Therefore
787 * it is possible to get a watchpoint trap here from inside the kernel.
788 * However, the code in ./ptrace.c has ensured that the user can
789 * only set watchpoints on userspace addresses. Therefore the in-kernel
790 * watchpoint trap can only occur in code which is reading/writing
791 * from user space. Such code must not hold kernel locks (since it
792 * can equally take a page fault), therefore it is safe to call
793 * force_sig_info even though that claims and releases locks.
795 * Code in ./signal.c ensures that the debug control register
796 * is restored before we deliver any signal, and therefore that
797 * user code runs with the correct debug control register even though
800 * Being careful here means that we don't have to be as careful in a
801 * lot of more complicated places (task switching can be a bit lazy
802 * about restoring all the debug state, and ptrace doesn't have to
803 * find every occurrence of the TF bit that could be saved away even
806 fastcall
void __kprobes
do_debug(struct pt_regs
* regs
, long error_code
)
808 unsigned int condition
;
809 struct task_struct
*tsk
= current
;
811 get_debugreg(condition
, 6);
813 if (notify_die(DIE_DEBUG
, "debug", regs
, condition
, error_code
,
814 SIGTRAP
) == NOTIFY_STOP
)
816 /* It's safe to allow irq's after DR6 has been saved */
817 if (regs
->eflags
& X86_EFLAGS_IF
)
820 /* Mask out spurious debug traps due to lazy DR7 setting */
821 if (condition
& (DR_TRAP0
|DR_TRAP1
|DR_TRAP2
|DR_TRAP3
)) {
822 if (!tsk
->thread
.debugreg
[7])
826 if (regs
->eflags
& VM_MASK
)
829 /* Save debug status register where ptrace can see it */
830 tsk
->thread
.debugreg
[6] = condition
;
833 * Single-stepping through TF: make sure we ignore any events in
834 * kernel space (but re-enable TF when returning to user mode).
836 if (condition
& DR_STEP
) {
838 * We already checked v86 mode above, so we can
839 * check for kernel mode by just checking the CPL
842 if (!user_mode(regs
))
843 goto clear_TF_reenable
;
846 /* Ok, finally something we can handle */
847 send_sigtrap(tsk
, regs
, error_code
);
849 /* Disable additional traps. They'll be re-enabled when
850 * the signal is delivered.
857 handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
, 1);
861 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
862 regs
->eflags
&= ~TF_MASK
;
867 * Note that we play around with the 'TS' bit in an attempt to get
868 * the correct behaviour even in the presence of the asynchronous
871 void math_error(void __user
*eip
)
873 struct task_struct
* task
;
875 unsigned short cwd
, swd
;
878 * Save the info for the exception handler and clear the error.
882 task
->thread
.trap_no
= 16;
883 task
->thread
.error_code
= 0;
884 info
.si_signo
= SIGFPE
;
886 info
.si_code
= __SI_FAULT
;
889 * (~cwd & swd) will mask out exceptions that are not set to unmasked
890 * status. 0x3f is the exception bits in these regs, 0x200 is the
891 * C1 reg you need in case of a stack fault, 0x040 is the stack
892 * fault bit. We should only be taking one exception at a time,
893 * so if this combination doesn't produce any single exception,
894 * then we have a bad program that isn't syncronizing its FPU usage
895 * and it will suffer the consequences since we won't be able to
896 * fully reproduce the context of the exception
898 cwd
= get_fpu_cwd(task
);
899 swd
= get_fpu_swd(task
);
900 switch (swd
& ~cwd
& 0x3f) {
901 case 0x000: /* No unmasked exception */
903 default: /* Multiple exceptions */
905 case 0x001: /* Invalid Op */
907 * swd & 0x240 == 0x040: Stack Underflow
908 * swd & 0x240 == 0x240: Stack Overflow
909 * User must clear the SF bit (0x40) if set
911 info
.si_code
= FPE_FLTINV
;
913 case 0x002: /* Denormalize */
914 case 0x010: /* Underflow */
915 info
.si_code
= FPE_FLTUND
;
917 case 0x004: /* Zero Divide */
918 info
.si_code
= FPE_FLTDIV
;
920 case 0x008: /* Overflow */
921 info
.si_code
= FPE_FLTOVF
;
923 case 0x020: /* Precision */
924 info
.si_code
= FPE_FLTRES
;
927 force_sig_info(SIGFPE
, &info
, task
);
930 fastcall
void do_coprocessor_error(struct pt_regs
* regs
, long error_code
)
933 math_error((void __user
*)regs
->eip
);
936 static void simd_math_error(void __user
*eip
)
938 struct task_struct
* task
;
940 unsigned short mxcsr
;
943 * Save the info for the exception handler and clear the error.
947 task
->thread
.trap_no
= 19;
948 task
->thread
.error_code
= 0;
949 info
.si_signo
= SIGFPE
;
951 info
.si_code
= __SI_FAULT
;
954 * The SIMD FPU exceptions are handled a little differently, as there
955 * is only a single status/control register. Thus, to determine which
956 * unmasked exception was caught we must mask the exception mask bits
957 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
959 mxcsr
= get_fpu_mxcsr(task
);
960 switch (~((mxcsr
& 0x1f80) >> 7) & (mxcsr
& 0x3f)) {
964 case 0x001: /* Invalid Op */
965 info
.si_code
= FPE_FLTINV
;
967 case 0x002: /* Denormalize */
968 case 0x010: /* Underflow */
969 info
.si_code
= FPE_FLTUND
;
971 case 0x004: /* Zero Divide */
972 info
.si_code
= FPE_FLTDIV
;
974 case 0x008: /* Overflow */
975 info
.si_code
= FPE_FLTOVF
;
977 case 0x020: /* Precision */
978 info
.si_code
= FPE_FLTRES
;
981 force_sig_info(SIGFPE
, &info
, task
);
984 fastcall
void do_simd_coprocessor_error(struct pt_regs
* regs
,
988 /* Handle SIMD FPU exceptions on PIII+ processors. */
990 simd_math_error((void __user
*)regs
->eip
);
993 * Handle strange cache flush from user space exception
994 * in all other cases. This is undocumented behaviour.
996 if (regs
->eflags
& VM_MASK
) {
997 handle_vm86_fault((struct kernel_vm86_regs
*)regs
,
1001 current
->thread
.trap_no
= 19;
1002 current
->thread
.error_code
= error_code
;
1003 die_if_kernel("cache flush denied", regs
, error_code
);
1004 force_sig(SIGSEGV
, current
);
1008 fastcall
void do_spurious_interrupt_bug(struct pt_regs
* regs
,
1012 /* No need to warn about this any longer. */
1013 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1017 fastcall
void setup_x86_bogus_stack(unsigned char * stk
)
1019 unsigned long *switch16_ptr
, *switch32_ptr
;
1020 struct pt_regs
*regs
;
1021 unsigned long stack_top
, stack_bot
;
1022 unsigned short iret_frame16_off
;
1023 int cpu
= smp_processor_id();
1024 /* reserve the space on 32bit stack for the magic switch16 pointer */
1025 memmove(stk
, stk
+ 8, sizeof(struct pt_regs
));
1026 switch16_ptr
= (unsigned long *)(stk
+ sizeof(struct pt_regs
));
1027 regs
= (struct pt_regs
*)stk
;
1028 /* now the switch32 on 16bit stack */
1029 stack_bot
= (unsigned long)&per_cpu(cpu_16bit_stack
, cpu
);
1030 stack_top
= stack_bot
+ CPU_16BIT_STACK_SIZE
;
1031 switch32_ptr
= (unsigned long *)(stack_top
- 8);
1032 iret_frame16_off
= CPU_16BIT_STACK_SIZE
- 8 - 20;
1033 /* copy iret frame on 16bit stack */
1034 memcpy((void *)(stack_bot
+ iret_frame16_off
), ®s
->eip
, 20);
1035 /* fill in the switch pointers */
1036 switch16_ptr
[0] = (regs
->esp
& 0xffff0000) | iret_frame16_off
;
1037 switch16_ptr
[1] = __ESPFIX_SS
;
1038 switch32_ptr
[0] = (unsigned long)stk
+ sizeof(struct pt_regs
) +
1039 8 - CPU_16BIT_STACK_SIZE
;
1040 switch32_ptr
[1] = __KERNEL_DS
;
1043 fastcall
unsigned char * fixup_x86_bogus_stack(unsigned short sp
)
1045 unsigned long *switch32_ptr
;
1046 unsigned char *stack16
, *stack32
;
1047 unsigned long stack_top
, stack_bot
;
1049 int cpu
= smp_processor_id();
1050 stack_bot
= (unsigned long)&per_cpu(cpu_16bit_stack
, cpu
);
1051 stack_top
= stack_bot
+ CPU_16BIT_STACK_SIZE
;
1052 switch32_ptr
= (unsigned long *)(stack_top
- 8);
1053 /* copy the data from 16bit stack to 32bit stack */
1054 len
= CPU_16BIT_STACK_SIZE
- 8 - sp
;
1055 stack16
= (unsigned char *)(stack_bot
+ sp
);
1056 stack32
= (unsigned char *)
1057 (switch32_ptr
[0] + CPU_16BIT_STACK_SIZE
- 8 - len
);
1058 memcpy(stack32
, stack16
, len
);
1063 * 'math_state_restore()' saves the current math information in the
1064 * old math state array, and gets the new ones from the current task
1066 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1067 * Don't touch unless you *really* know how it works.
1069 * Must be called with kernel preemption disabled (in this case,
1070 * local interrupts are disabled at the call-site in entry.S).
1072 asmlinkage
void math_state_restore(struct pt_regs regs
)
1074 struct thread_info
*thread
= current_thread_info();
1075 struct task_struct
*tsk
= thread
->task
;
1077 clts(); /* Allow maths ops (or we recurse) */
1078 if (!tsk_used_math(tsk
))
1081 thread
->status
|= TS_USEDFPU
; /* So we fnsave on switch_to() */
1084 #ifndef CONFIG_MATH_EMULATION
1086 asmlinkage
void math_emulate(long arg
)
1088 printk(KERN_EMERG
"math-emulation not enabled and no coprocessor found.\n");
1089 printk(KERN_EMERG
"killing %s.\n",current
->comm
);
1090 force_sig(SIGFPE
,current
);
1094 #endif /* CONFIG_MATH_EMULATION */
1096 #ifdef CONFIG_X86_F00F_BUG
1097 void __init
trap_init_f00f_bug(void)
1099 __set_fixmap(FIX_F00F_IDT
, __pa(&idt_table
), PAGE_KERNEL_RO
);
1102 * Update the IDT descriptor and reload the IDT so that
1103 * it uses the read-only mapped virtual address.
1105 idt_descr
.address
= fix_to_virt(FIX_F00F_IDT
);
1106 load_idt(&idt_descr
);
1110 #define _set_gate(gate_addr,type,dpl,addr,seg) \
1113 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1114 "movw %4,%%dx\n\t" \
1115 "movl %%eax,%0\n\t" \
1117 :"=m" (*((long *) (gate_addr))), \
1118 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1119 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1120 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1125 * This needs to use 'idt_table' rather than 'idt', and
1126 * thus use the _nonmapped_ version of the IDT, as the
1127 * Pentium F0 0F bugfix can have resulted in the mapped
1128 * IDT being write-protected.
1130 void set_intr_gate(unsigned int n
, void *addr
)
1132 _set_gate(idt_table
+n
,14,0,addr
,__KERNEL_CS
);
1136 * This routine sets up an interrupt gate at directory privilege level 3.
1138 static inline void set_system_intr_gate(unsigned int n
, void *addr
)
1140 _set_gate(idt_table
+n
, 14, 3, addr
, __KERNEL_CS
);
1143 static void __init
set_trap_gate(unsigned int n
, void *addr
)
1145 _set_gate(idt_table
+n
,15,0,addr
,__KERNEL_CS
);
1148 static void __init
set_system_gate(unsigned int n
, void *addr
)
1150 _set_gate(idt_table
+n
,15,3,addr
,__KERNEL_CS
);
1153 static void __init
set_task_gate(unsigned int n
, unsigned int gdt_entry
)
1155 _set_gate(idt_table
+n
,5,0,0,(gdt_entry
<<3));
1159 void __init
trap_init(void)
1162 void __iomem
*p
= ioremap(0x0FFFD9, 4);
1163 if (readl(p
) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1169 #ifdef CONFIG_X86_LOCAL_APIC
1170 init_apic_mappings();
1173 set_trap_gate(0,÷_error
);
1174 set_intr_gate(1,&debug
);
1175 set_intr_gate(2,&nmi
);
1176 set_system_intr_gate(3, &int3
); /* int3/4 can be called from all */
1177 set_system_gate(4,&overflow
);
1178 set_trap_gate(5,&bounds
);
1179 set_trap_gate(6,&invalid_op
);
1180 set_trap_gate(7,&device_not_available
);
1181 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS
);
1182 set_trap_gate(9,&coprocessor_segment_overrun
);
1183 set_trap_gate(10,&invalid_TSS
);
1184 set_trap_gate(11,&segment_not_present
);
1185 set_trap_gate(12,&stack_segment
);
1186 set_trap_gate(13,&general_protection
);
1187 set_intr_gate(14,&page_fault
);
1188 set_trap_gate(15,&spurious_interrupt_bug
);
1189 set_trap_gate(16,&coprocessor_error
);
1190 set_trap_gate(17,&alignment_check
);
1191 #ifdef CONFIG_X86_MCE
1192 set_trap_gate(18,&machine_check
);
1194 set_trap_gate(19,&simd_coprocessor_error
);
1198 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1199 * Generates a compile-time "error: zero width for bit-field" if
1200 * the alignment is wrong.
1202 struct fxsrAlignAssert
{
1203 int _
:!(offsetof(struct task_struct
,
1204 thread
.i387
.fxsave
) & 15);
1207 printk(KERN_INFO
"Enabling fast FPU save and restore... ");
1208 set_in_cr4(X86_CR4_OSFXSR
);
1212 printk(KERN_INFO
"Enabling unmasked SIMD FPU exception "
1214 set_in_cr4(X86_CR4_OSXMMEXCPT
);
1218 set_system_gate(SYSCALL_VECTOR
,&system_call
);
1221 * Should be a barrier for any external CPU state.
1228 static int __init
kstack_setup(char *s
)
1230 kstack_depth_to_print
= simple_strtoul(s
, NULL
, 0);
1233 __setup("kstack=", kstack_setup
);
1235 static int __init
call_trace_setup(char *s
)
1237 if (strcmp(s
, "old") == 0)
1239 else if (strcmp(s
, "both") == 0)
1241 else if (strcmp(s
, "new") == 0)
1245 __setup("call_trace=", call_trace_setup
);