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
); /* used modular by kdb */
105 int unregister_die_notifier(struct notifier_block
*nb
)
107 return atomic_notifier_chain_unregister(&i386die_chain
, nb
);
109 EXPORT_SYMBOL(unregister_die_notifier
); /* used modular by kdb */
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
);
190 if (unw_ret
> 0 && !arch_unw_user_mode(&info
)) {
191 #ifdef CONFIG_STACK_UNWIND
192 print_symbol("DWARF2 unwinder stuck at %s\n",
194 if (call_trace
== 1) {
195 printk("Leftover inexact backtrace:\n");
197 stack
= (void *)UNW_SP(&info
);
198 } else if (call_trace
> 1)
201 printk("Full inexact backtrace again:\n");
203 printk("Inexact backtrace:\n");
208 if (task
== current
) {
209 /* Grab ebp right from our regs */
210 asm ("movl %%ebp, %0" : "=r" (ebp
) : );
212 /* ebp is the last reg pushed by switch_to */
213 ebp
= *(unsigned long *) task
->thread
.esp
;
217 struct thread_info
*context
;
218 context
= (struct thread_info
*)
219 ((unsigned long)stack
& (~(THREAD_SIZE
- 1)));
220 ebp
= print_context_stack(context
, stack
, ebp
, log_lvl
);
221 stack
= (unsigned long*)context
->previous_esp
;
224 printk("%s =======================\n", log_lvl
);
228 void show_trace(struct task_struct
*task
, struct pt_regs
*regs
, unsigned long * stack
)
230 show_trace_log_lvl(task
, regs
, stack
, "");
233 static void show_stack_log_lvl(struct task_struct
*task
, struct pt_regs
*regs
,
234 unsigned long *esp
, char *log_lvl
)
236 unsigned long *stack
;
241 esp
= (unsigned long*)task
->thread
.esp
;
243 esp
= (unsigned long *)&esp
;
247 for(i
= 0; i
< kstack_depth_to_print
; i
++) {
248 if (kstack_end(stack
))
250 if (i
&& ((i
% 8) == 0))
251 printk("\n%s ", log_lvl
);
252 printk("%08lx ", *stack
++);
254 printk("\n%sCall Trace:\n", log_lvl
);
255 show_trace_log_lvl(task
, regs
, esp
, log_lvl
);
258 void show_stack(struct task_struct
*task
, unsigned long *esp
)
261 show_stack_log_lvl(task
, NULL
, esp
, "");
265 * The architecture-independent dump_stack generator
267 void dump_stack(void)
271 show_trace(current
, NULL
, &stack
);
274 EXPORT_SYMBOL(dump_stack
);
276 void show_registers(struct pt_regs
*regs
)
283 esp
= (unsigned long) (®s
->esp
);
285 if (user_mode_vm(regs
)) {
288 ss
= regs
->xss
& 0xffff;
291 printk(KERN_EMERG
"CPU: %d\nEIP: %04x:[<%08lx>] %s VLI\n"
292 "EFLAGS: %08lx (%s %.*s) \n",
293 smp_processor_id(), 0xffff & regs
->xcs
, regs
->eip
,
294 print_tainted(), regs
->eflags
, system_utsname
.release
,
295 (int)strcspn(system_utsname
.version
, " "),
296 system_utsname
.version
);
297 print_symbol(KERN_EMERG
"EIP is at %s\n", regs
->eip
);
298 printk(KERN_EMERG
"eax: %08lx ebx: %08lx ecx: %08lx edx: %08lx\n",
299 regs
->eax
, regs
->ebx
, regs
->ecx
, regs
->edx
);
300 printk(KERN_EMERG
"esi: %08lx edi: %08lx ebp: %08lx esp: %08lx\n",
301 regs
->esi
, regs
->edi
, regs
->ebp
, esp
);
302 printk(KERN_EMERG
"ds: %04x es: %04x ss: %04x\n",
303 regs
->xds
& 0xffff, regs
->xes
& 0xffff, ss
);
304 printk(KERN_EMERG
"Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
305 TASK_COMM_LEN
, current
->comm
, current
->pid
,
306 current_thread_info(), current
, current
->thread_info
);
308 * When in-kernel, we also print out the stack and code at the
309 * time of the fault..
314 printk("\n" KERN_EMERG
"Stack: ");
315 show_stack_log_lvl(NULL
, regs
, (unsigned long *)esp
, KERN_EMERG
);
317 printk(KERN_EMERG
"Code: ");
319 eip
= (u8 __user
*)regs
->eip
- 43;
320 for (i
= 0; i
< 64; i
++, eip
++) {
323 if (eip
< (u8 __user
*)PAGE_OFFSET
|| __get_user(c
, eip
)) {
324 printk(" Bad EIP value.");
327 if (eip
== (u8 __user
*)regs
->eip
)
328 printk("<%02x> ", c
);
336 static void handle_BUG(struct pt_regs
*regs
)
338 unsigned long eip
= regs
->eip
;
341 if (eip
< PAGE_OFFSET
)
343 if (__get_user(ud2
, (unsigned short __user
*)eip
))
348 printk(KERN_EMERG
"------------[ cut here ]------------\n");
350 #ifdef CONFIG_DEBUG_BUGVERBOSE
356 if (__get_user(line
, (unsigned short __user
*)(eip
+ 2)))
358 if (__get_user(file
, (char * __user
*)(eip
+ 4)) ||
359 (unsigned long)file
< PAGE_OFFSET
|| __get_user(c
, file
))
360 file
= "<bad filename>";
362 printk(KERN_EMERG
"kernel BUG at %s:%d!\n", file
, line
);
366 printk(KERN_EMERG
"Kernel BUG at [verbose debug info unavailable]\n");
369 /* This is gone through when something in the kernel
370 * has done something bad and is about to be terminated.
372 void die(const char * str
, struct pt_regs
* regs
, long err
)
377 int lock_owner_depth
;
379 .lock
= SPIN_LOCK_UNLOCKED
,
381 .lock_owner_depth
= 0
383 static int die_counter
;
388 if (die
.lock_owner
!= raw_smp_processor_id()) {
390 spin_lock_irqsave(&die
.lock
, flags
);
391 die
.lock_owner
= smp_processor_id();
392 die
.lock_owner_depth
= 0;
396 local_save_flags(flags
);
398 if (++die
.lock_owner_depth
< 3) {
404 printk(KERN_EMERG
"%s: %04lx [#%d]\n", str
, err
& 0xffff, ++die_counter
);
405 #ifdef CONFIG_PREEMPT
406 printk(KERN_EMERG
"PREEMPT ");
415 #ifdef CONFIG_DEBUG_PAGEALLOC
418 printk("DEBUG_PAGEALLOC");
423 if (notify_die(DIE_OOPS
, str
, regs
, err
,
424 current
->thread
.trap_no
, SIGSEGV
) !=
426 show_registers(regs
);
427 /* Executive summary in case the oops scrolled away */
428 esp
= (unsigned long) (®s
->esp
);
430 if (user_mode(regs
)) {
432 ss
= regs
->xss
& 0xffff;
434 printk(KERN_EMERG
"EIP: [<%08lx>] ", regs
->eip
);
435 print_symbol("%s", regs
->eip
);
436 printk(" SS:ESP %04x:%08lx\n", ss
, esp
);
441 printk(KERN_EMERG
"Recursive die() failure, output suppressed\n");
445 spin_unlock_irqrestore(&die
.lock
, flags
);
450 if (kexec_should_crash(current
))
454 panic("Fatal exception in interrupt");
457 panic("Fatal exception: panic_on_oops");
463 static inline void die_if_kernel(const char * str
, struct pt_regs
* regs
, long err
)
465 if (!user_mode_vm(regs
))
469 static void __kprobes
do_trap(int trapnr
, int signr
, char *str
, int vm86
,
470 struct pt_regs
* regs
, long error_code
,
473 struct task_struct
*tsk
= current
;
474 tsk
->thread
.error_code
= error_code
;
475 tsk
->thread
.trap_no
= trapnr
;
477 if (regs
->eflags
& VM_MASK
) {
483 if (!user_mode(regs
))
488 force_sig_info(signr
, info
, tsk
);
490 force_sig(signr
, tsk
);
495 if (!fixup_exception(regs
))
496 die(str
, regs
, error_code
);
501 int ret
= handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
, trapnr
);
502 if (ret
) goto trap_signal
;
507 #define DO_ERROR(trapnr, signr, str, name) \
508 fastcall void do_##name(struct pt_regs * regs, long error_code) \
510 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
513 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
516 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
517 fastcall void do_##name(struct pt_regs * regs, long error_code) \
520 info.si_signo = signr; \
522 info.si_code = sicode; \
523 info.si_addr = (void __user *)siaddr; \
524 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
527 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
530 #define DO_VM86_ERROR(trapnr, signr, str, name) \
531 fastcall void do_##name(struct pt_regs * regs, long error_code) \
533 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
536 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
539 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
540 fastcall void do_##name(struct pt_regs * regs, long error_code) \
543 info.si_signo = signr; \
545 info.si_code = sicode; \
546 info.si_addr = (void __user *)siaddr; \
547 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
550 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
553 DO_VM86_ERROR_INFO( 0, SIGFPE
, "divide error", divide_error
, FPE_INTDIV
, regs
->eip
)
554 #ifndef CONFIG_KPROBES
555 DO_VM86_ERROR( 3, SIGTRAP
, "int3", int3
)
557 DO_VM86_ERROR( 4, SIGSEGV
, "overflow", overflow
)
558 DO_VM86_ERROR( 5, SIGSEGV
, "bounds", bounds
)
559 DO_ERROR_INFO( 6, SIGILL
, "invalid opcode", invalid_op
, ILL_ILLOPN
, regs
->eip
)
560 DO_ERROR( 9, SIGFPE
, "coprocessor segment overrun", coprocessor_segment_overrun
)
561 DO_ERROR(10, SIGSEGV
, "invalid TSS", invalid_TSS
)
562 DO_ERROR(11, SIGBUS
, "segment not present", segment_not_present
)
563 DO_ERROR(12, SIGBUS
, "stack segment", stack_segment
)
564 DO_ERROR_INFO(17, SIGBUS
, "alignment check", alignment_check
, BUS_ADRALN
, 0)
565 DO_ERROR_INFO(32, SIGSEGV
, "iret exception", iret_error
, ILL_BADSTK
, 0)
567 fastcall
void __kprobes
do_general_protection(struct pt_regs
* regs
,
571 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
572 struct thread_struct
*thread
= ¤t
->thread
;
575 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
576 * invalid offset set (the LAZY one) and the faulting thread has
577 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
578 * and we set the offset field correctly. Then we let the CPU to
579 * restart the faulting instruction.
581 if (tss
->io_bitmap_base
== INVALID_IO_BITMAP_OFFSET_LAZY
&&
582 thread
->io_bitmap_ptr
) {
583 memcpy(tss
->io_bitmap
, thread
->io_bitmap_ptr
,
584 thread
->io_bitmap_max
);
586 * If the previously set map was extending to higher ports
587 * than the current one, pad extra space with 0xff (no access).
589 if (thread
->io_bitmap_max
< tss
->io_bitmap_max
)
590 memset((char *) tss
->io_bitmap
+
591 thread
->io_bitmap_max
, 0xff,
592 tss
->io_bitmap_max
- thread
->io_bitmap_max
);
593 tss
->io_bitmap_max
= thread
->io_bitmap_max
;
594 tss
->io_bitmap_base
= IO_BITMAP_OFFSET
;
595 tss
->io_bitmap_owner
= thread
;
601 current
->thread
.error_code
= error_code
;
602 current
->thread
.trap_no
= 13;
604 if (regs
->eflags
& VM_MASK
)
607 if (!user_mode(regs
))
610 current
->thread
.error_code
= error_code
;
611 current
->thread
.trap_no
= 13;
612 force_sig(SIGSEGV
, current
);
617 handle_vm86_fault((struct kernel_vm86_regs
*) regs
, error_code
);
621 if (!fixup_exception(regs
)) {
622 if (notify_die(DIE_GPF
, "general protection fault", regs
,
623 error_code
, 13, SIGSEGV
) == NOTIFY_STOP
)
625 die("general protection fault", regs
, error_code
);
629 static void mem_parity_error(unsigned char reason
, struct pt_regs
* regs
)
631 printk(KERN_EMERG
"Uhhuh. NMI received. Dazed and confused, but trying "
633 printk(KERN_EMERG
"You probably have a hardware problem with your RAM "
636 /* Clear and disable the memory parity error line. */
637 clear_mem_error(reason
);
640 static void io_check_error(unsigned char reason
, struct pt_regs
* regs
)
644 printk(KERN_EMERG
"NMI: IOCK error (debug interrupt?)\n");
645 show_registers(regs
);
647 /* Re-enable the IOCK line, wait for a few seconds */
648 reason
= (reason
& 0xf) | 8;
651 while (--i
) udelay(1000);
656 static void unknown_nmi_error(unsigned char reason
, struct pt_regs
* regs
)
659 /* Might actually be able to figure out what the guilty party
666 printk("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
667 reason
, smp_processor_id());
668 printk("Dazed and confused, but trying to continue\n");
669 printk("Do you have a strange power saving mode enabled?\n");
672 static DEFINE_SPINLOCK(nmi_print_lock
);
674 void die_nmi (struct pt_regs
*regs
, const char *msg
)
676 if (notify_die(DIE_NMIWATCHDOG
, msg
, regs
, 0, 2, SIGINT
) ==
680 spin_lock(&nmi_print_lock
);
682 * We are in trouble anyway, lets at least try
683 * to get a message out.
686 printk(KERN_EMERG
"%s", msg
);
687 printk(" on CPU%d, eip %08lx, registers:\n",
688 smp_processor_id(), regs
->eip
);
689 show_registers(regs
);
690 printk(KERN_EMERG
"console shuts up ...\n");
692 spin_unlock(&nmi_print_lock
);
695 /* If we are in kernel we are probably nested up pretty bad
696 * and might aswell get out now while we still can.
698 if (!user_mode_vm(regs
)) {
699 current
->thread
.trap_no
= 2;
706 static void default_do_nmi(struct pt_regs
* regs
)
708 unsigned char reason
= 0;
710 /* Only the BSP gets external NMIs from the system. */
711 if (!smp_processor_id())
712 reason
= get_nmi_reason();
714 if (!(reason
& 0xc0)) {
715 if (notify_die(DIE_NMI_IPI
, "nmi_ipi", regs
, reason
, 2, SIGINT
)
718 #ifdef CONFIG_X86_LOCAL_APIC
720 * Ok, so this is none of the documented NMI sources,
721 * so it must be the NMI watchdog.
724 nmi_watchdog_tick(regs
);
728 unknown_nmi_error(reason
, regs
);
731 if (notify_die(DIE_NMI
, "nmi", regs
, reason
, 2, SIGINT
) == NOTIFY_STOP
)
734 mem_parity_error(reason
, regs
);
736 io_check_error(reason
, regs
);
738 * Reassert NMI in case it became active meanwhile
739 * as it's edge-triggered.
744 static int dummy_nmi_callback(struct pt_regs
* regs
, int cpu
)
749 static nmi_callback_t nmi_callback
= dummy_nmi_callback
;
751 fastcall
void do_nmi(struct pt_regs
* regs
, long error_code
)
757 cpu
= smp_processor_id();
761 if (!rcu_dereference(nmi_callback
)(regs
, cpu
))
762 default_do_nmi(regs
);
767 void set_nmi_callback(nmi_callback_t callback
)
770 rcu_assign_pointer(nmi_callback
, callback
);
772 EXPORT_SYMBOL_GPL(set_nmi_callback
);
774 void unset_nmi_callback(void)
776 nmi_callback
= dummy_nmi_callback
;
778 EXPORT_SYMBOL_GPL(unset_nmi_callback
);
780 #ifdef CONFIG_KPROBES
781 fastcall
void __kprobes
do_int3(struct pt_regs
*regs
, long error_code
)
783 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, 3, SIGTRAP
)
786 /* This is an interrupt gate, because kprobes wants interrupts
787 disabled. Normal trap handlers don't. */
788 restore_interrupts(regs
);
789 do_trap(3, SIGTRAP
, "int3", 1, regs
, error_code
, NULL
);
794 * Our handling of the processor debug registers is non-trivial.
795 * We do not clear them on entry and exit from the kernel. Therefore
796 * it is possible to get a watchpoint trap here from inside the kernel.
797 * However, the code in ./ptrace.c has ensured that the user can
798 * only set watchpoints on userspace addresses. Therefore the in-kernel
799 * watchpoint trap can only occur in code which is reading/writing
800 * from user space. Such code must not hold kernel locks (since it
801 * can equally take a page fault), therefore it is safe to call
802 * force_sig_info even though that claims and releases locks.
804 * Code in ./signal.c ensures that the debug control register
805 * is restored before we deliver any signal, and therefore that
806 * user code runs with the correct debug control register even though
809 * Being careful here means that we don't have to be as careful in a
810 * lot of more complicated places (task switching can be a bit lazy
811 * about restoring all the debug state, and ptrace doesn't have to
812 * find every occurrence of the TF bit that could be saved away even
815 fastcall
void __kprobes
do_debug(struct pt_regs
* regs
, long error_code
)
817 unsigned int condition
;
818 struct task_struct
*tsk
= current
;
820 get_debugreg(condition
, 6);
822 if (notify_die(DIE_DEBUG
, "debug", regs
, condition
, error_code
,
823 SIGTRAP
) == NOTIFY_STOP
)
825 /* It's safe to allow irq's after DR6 has been saved */
826 if (regs
->eflags
& X86_EFLAGS_IF
)
829 /* Mask out spurious debug traps due to lazy DR7 setting */
830 if (condition
& (DR_TRAP0
|DR_TRAP1
|DR_TRAP2
|DR_TRAP3
)) {
831 if (!tsk
->thread
.debugreg
[7])
835 if (regs
->eflags
& VM_MASK
)
838 /* Save debug status register where ptrace can see it */
839 tsk
->thread
.debugreg
[6] = condition
;
842 * Single-stepping through TF: make sure we ignore any events in
843 * kernel space (but re-enable TF when returning to user mode).
845 if (condition
& DR_STEP
) {
847 * We already checked v86 mode above, so we can
848 * check for kernel mode by just checking the CPL
851 if (!user_mode(regs
))
852 goto clear_TF_reenable
;
855 /* Ok, finally something we can handle */
856 send_sigtrap(tsk
, regs
, error_code
);
858 /* Disable additional traps. They'll be re-enabled when
859 * the signal is delivered.
866 handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
, 1);
870 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
871 regs
->eflags
&= ~TF_MASK
;
876 * Note that we play around with the 'TS' bit in an attempt to get
877 * the correct behaviour even in the presence of the asynchronous
880 void math_error(void __user
*eip
)
882 struct task_struct
* task
;
884 unsigned short cwd
, swd
;
887 * Save the info for the exception handler and clear the error.
891 task
->thread
.trap_no
= 16;
892 task
->thread
.error_code
= 0;
893 info
.si_signo
= SIGFPE
;
895 info
.si_code
= __SI_FAULT
;
898 * (~cwd & swd) will mask out exceptions that are not set to unmasked
899 * status. 0x3f is the exception bits in these regs, 0x200 is the
900 * C1 reg you need in case of a stack fault, 0x040 is the stack
901 * fault bit. We should only be taking one exception at a time,
902 * so if this combination doesn't produce any single exception,
903 * then we have a bad program that isn't syncronizing its FPU usage
904 * and it will suffer the consequences since we won't be able to
905 * fully reproduce the context of the exception
907 cwd
= get_fpu_cwd(task
);
908 swd
= get_fpu_swd(task
);
909 switch (swd
& ~cwd
& 0x3f) {
910 case 0x000: /* No unmasked exception */
912 default: /* Multiple exceptions */
914 case 0x001: /* Invalid Op */
916 * swd & 0x240 == 0x040: Stack Underflow
917 * swd & 0x240 == 0x240: Stack Overflow
918 * User must clear the SF bit (0x40) if set
920 info
.si_code
= FPE_FLTINV
;
922 case 0x002: /* Denormalize */
923 case 0x010: /* Underflow */
924 info
.si_code
= FPE_FLTUND
;
926 case 0x004: /* Zero Divide */
927 info
.si_code
= FPE_FLTDIV
;
929 case 0x008: /* Overflow */
930 info
.si_code
= FPE_FLTOVF
;
932 case 0x020: /* Precision */
933 info
.si_code
= FPE_FLTRES
;
936 force_sig_info(SIGFPE
, &info
, task
);
939 fastcall
void do_coprocessor_error(struct pt_regs
* regs
, long error_code
)
942 math_error((void __user
*)regs
->eip
);
945 static void simd_math_error(void __user
*eip
)
947 struct task_struct
* task
;
949 unsigned short mxcsr
;
952 * Save the info for the exception handler and clear the error.
956 task
->thread
.trap_no
= 19;
957 task
->thread
.error_code
= 0;
958 info
.si_signo
= SIGFPE
;
960 info
.si_code
= __SI_FAULT
;
963 * The SIMD FPU exceptions are handled a little differently, as there
964 * is only a single status/control register. Thus, to determine which
965 * unmasked exception was caught we must mask the exception mask bits
966 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
968 mxcsr
= get_fpu_mxcsr(task
);
969 switch (~((mxcsr
& 0x1f80) >> 7) & (mxcsr
& 0x3f)) {
973 case 0x001: /* Invalid Op */
974 info
.si_code
= FPE_FLTINV
;
976 case 0x002: /* Denormalize */
977 case 0x010: /* Underflow */
978 info
.si_code
= FPE_FLTUND
;
980 case 0x004: /* Zero Divide */
981 info
.si_code
= FPE_FLTDIV
;
983 case 0x008: /* Overflow */
984 info
.si_code
= FPE_FLTOVF
;
986 case 0x020: /* Precision */
987 info
.si_code
= FPE_FLTRES
;
990 force_sig_info(SIGFPE
, &info
, task
);
993 fastcall
void do_simd_coprocessor_error(struct pt_regs
* regs
,
997 /* Handle SIMD FPU exceptions on PIII+ processors. */
999 simd_math_error((void __user
*)regs
->eip
);
1002 * Handle strange cache flush from user space exception
1003 * in all other cases. This is undocumented behaviour.
1005 if (regs
->eflags
& VM_MASK
) {
1006 handle_vm86_fault((struct kernel_vm86_regs
*)regs
,
1010 current
->thread
.trap_no
= 19;
1011 current
->thread
.error_code
= error_code
;
1012 die_if_kernel("cache flush denied", regs
, error_code
);
1013 force_sig(SIGSEGV
, current
);
1017 fastcall
void do_spurious_interrupt_bug(struct pt_regs
* regs
,
1021 /* No need to warn about this any longer. */
1022 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1026 fastcall
void setup_x86_bogus_stack(unsigned char * stk
)
1028 unsigned long *switch16_ptr
, *switch32_ptr
;
1029 struct pt_regs
*regs
;
1030 unsigned long stack_top
, stack_bot
;
1031 unsigned short iret_frame16_off
;
1032 int cpu
= smp_processor_id();
1033 /* reserve the space on 32bit stack for the magic switch16 pointer */
1034 memmove(stk
, stk
+ 8, sizeof(struct pt_regs
));
1035 switch16_ptr
= (unsigned long *)(stk
+ sizeof(struct pt_regs
));
1036 regs
= (struct pt_regs
*)stk
;
1037 /* now the switch32 on 16bit stack */
1038 stack_bot
= (unsigned long)&per_cpu(cpu_16bit_stack
, cpu
);
1039 stack_top
= stack_bot
+ CPU_16BIT_STACK_SIZE
;
1040 switch32_ptr
= (unsigned long *)(stack_top
- 8);
1041 iret_frame16_off
= CPU_16BIT_STACK_SIZE
- 8 - 20;
1042 /* copy iret frame on 16bit stack */
1043 memcpy((void *)(stack_bot
+ iret_frame16_off
), ®s
->eip
, 20);
1044 /* fill in the switch pointers */
1045 switch16_ptr
[0] = (regs
->esp
& 0xffff0000) | iret_frame16_off
;
1046 switch16_ptr
[1] = __ESPFIX_SS
;
1047 switch32_ptr
[0] = (unsigned long)stk
+ sizeof(struct pt_regs
) +
1048 8 - CPU_16BIT_STACK_SIZE
;
1049 switch32_ptr
[1] = __KERNEL_DS
;
1052 fastcall
unsigned char * fixup_x86_bogus_stack(unsigned short sp
)
1054 unsigned long *switch32_ptr
;
1055 unsigned char *stack16
, *stack32
;
1056 unsigned long stack_top
, stack_bot
;
1058 int cpu
= smp_processor_id();
1059 stack_bot
= (unsigned long)&per_cpu(cpu_16bit_stack
, cpu
);
1060 stack_top
= stack_bot
+ CPU_16BIT_STACK_SIZE
;
1061 switch32_ptr
= (unsigned long *)(stack_top
- 8);
1062 /* copy the data from 16bit stack to 32bit stack */
1063 len
= CPU_16BIT_STACK_SIZE
- 8 - sp
;
1064 stack16
= (unsigned char *)(stack_bot
+ sp
);
1065 stack32
= (unsigned char *)
1066 (switch32_ptr
[0] + CPU_16BIT_STACK_SIZE
- 8 - len
);
1067 memcpy(stack32
, stack16
, len
);
1072 * 'math_state_restore()' saves the current math information in the
1073 * old math state array, and gets the new ones from the current task
1075 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1076 * Don't touch unless you *really* know how it works.
1078 * Must be called with kernel preemption disabled (in this case,
1079 * local interrupts are disabled at the call-site in entry.S).
1081 asmlinkage
void math_state_restore(struct pt_regs regs
)
1083 struct thread_info
*thread
= current_thread_info();
1084 struct task_struct
*tsk
= thread
->task
;
1086 clts(); /* Allow maths ops (or we recurse) */
1087 if (!tsk_used_math(tsk
))
1090 thread
->status
|= TS_USEDFPU
; /* So we fnsave on switch_to() */
1093 #ifndef CONFIG_MATH_EMULATION
1095 asmlinkage
void math_emulate(long arg
)
1097 printk(KERN_EMERG
"math-emulation not enabled and no coprocessor found.\n");
1098 printk(KERN_EMERG
"killing %s.\n",current
->comm
);
1099 force_sig(SIGFPE
,current
);
1103 #endif /* CONFIG_MATH_EMULATION */
1105 #ifdef CONFIG_X86_F00F_BUG
1106 void __init
trap_init_f00f_bug(void)
1108 __set_fixmap(FIX_F00F_IDT
, __pa(&idt_table
), PAGE_KERNEL_RO
);
1111 * Update the IDT descriptor and reload the IDT so that
1112 * it uses the read-only mapped virtual address.
1114 idt_descr
.address
= fix_to_virt(FIX_F00F_IDT
);
1115 load_idt(&idt_descr
);
1119 #define _set_gate(gate_addr,type,dpl,addr,seg) \
1122 __asm__ __volatile__ ("movw %%dx,%%ax\n\t" \
1123 "movw %4,%%dx\n\t" \
1124 "movl %%eax,%0\n\t" \
1126 :"=m" (*((long *) (gate_addr))), \
1127 "=m" (*(1+(long *) (gate_addr))), "=&a" (__d0), "=&d" (__d1) \
1128 :"i" ((short) (0x8000+(dpl<<13)+(type<<8))), \
1129 "3" ((char *) (addr)),"2" ((seg) << 16)); \
1134 * This needs to use 'idt_table' rather than 'idt', and
1135 * thus use the _nonmapped_ version of the IDT, as the
1136 * Pentium F0 0F bugfix can have resulted in the mapped
1137 * IDT being write-protected.
1139 void set_intr_gate(unsigned int n
, void *addr
)
1141 _set_gate(idt_table
+n
,14,0,addr
,__KERNEL_CS
);
1145 * This routine sets up an interrupt gate at directory privilege level 3.
1147 static inline void set_system_intr_gate(unsigned int n
, void *addr
)
1149 _set_gate(idt_table
+n
, 14, 3, addr
, __KERNEL_CS
);
1152 static void __init
set_trap_gate(unsigned int n
, void *addr
)
1154 _set_gate(idt_table
+n
,15,0,addr
,__KERNEL_CS
);
1157 static void __init
set_system_gate(unsigned int n
, void *addr
)
1159 _set_gate(idt_table
+n
,15,3,addr
,__KERNEL_CS
);
1162 static void __init
set_task_gate(unsigned int n
, unsigned int gdt_entry
)
1164 _set_gate(idt_table
+n
,5,0,0,(gdt_entry
<<3));
1168 void __init
trap_init(void)
1171 void __iomem
*p
= ioremap(0x0FFFD9, 4);
1172 if (readl(p
) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1178 #ifdef CONFIG_X86_LOCAL_APIC
1179 init_apic_mappings();
1182 set_trap_gate(0,÷_error
);
1183 set_intr_gate(1,&debug
);
1184 set_intr_gate(2,&nmi
);
1185 set_system_intr_gate(3, &int3
); /* int3/4 can be called from all */
1186 set_system_gate(4,&overflow
);
1187 set_trap_gate(5,&bounds
);
1188 set_trap_gate(6,&invalid_op
);
1189 set_trap_gate(7,&device_not_available
);
1190 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS
);
1191 set_trap_gate(9,&coprocessor_segment_overrun
);
1192 set_trap_gate(10,&invalid_TSS
);
1193 set_trap_gate(11,&segment_not_present
);
1194 set_trap_gate(12,&stack_segment
);
1195 set_trap_gate(13,&general_protection
);
1196 set_intr_gate(14,&page_fault
);
1197 set_trap_gate(15,&spurious_interrupt_bug
);
1198 set_trap_gate(16,&coprocessor_error
);
1199 set_trap_gate(17,&alignment_check
);
1200 #ifdef CONFIG_X86_MCE
1201 set_trap_gate(18,&machine_check
);
1203 set_trap_gate(19,&simd_coprocessor_error
);
1207 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1208 * Generates a compile-time "error: zero width for bit-field" if
1209 * the alignment is wrong.
1211 struct fxsrAlignAssert
{
1212 int _
:!(offsetof(struct task_struct
,
1213 thread
.i387
.fxsave
) & 15);
1216 printk(KERN_INFO
"Enabling fast FPU save and restore... ");
1217 set_in_cr4(X86_CR4_OSFXSR
);
1221 printk(KERN_INFO
"Enabling unmasked SIMD FPU exception "
1223 set_in_cr4(X86_CR4_OSXMMEXCPT
);
1227 set_system_gate(SYSCALL_VECTOR
,&system_call
);
1230 * Should be a barrier for any external CPU state.
1237 static int __init
kstack_setup(char *s
)
1239 kstack_depth_to_print
= simple_strtoul(s
, NULL
, 0);
1242 __setup("kstack=", kstack_setup
);
1244 static int __init
call_trace_setup(char *s
)
1246 if (strcmp(s
, "old") == 0)
1248 else if (strcmp(s
, "both") == 0)
1250 else if (strcmp(s
, "newfallback") == 0)
1252 else if (strcmp(s
, "new") == 2)
1256 __setup("call_trace=", call_trace_setup
);