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 struct desc_struct idt_table
[256] __attribute__((__section__(".data.idt"))) = { {0, 0}, };
81 asmlinkage
void divide_error(void);
82 asmlinkage
void debug(void);
83 asmlinkage
void nmi(void);
84 asmlinkage
void int3(void);
85 asmlinkage
void overflow(void);
86 asmlinkage
void bounds(void);
87 asmlinkage
void invalid_op(void);
88 asmlinkage
void device_not_available(void);
89 asmlinkage
void coprocessor_segment_overrun(void);
90 asmlinkage
void invalid_TSS(void);
91 asmlinkage
void segment_not_present(void);
92 asmlinkage
void stack_segment(void);
93 asmlinkage
void general_protection(void);
94 asmlinkage
void page_fault(void);
95 asmlinkage
void coprocessor_error(void);
96 asmlinkage
void simd_coprocessor_error(void);
97 asmlinkage
void alignment_check(void);
98 asmlinkage
void spurious_interrupt_bug(void);
99 asmlinkage
void machine_check(void);
101 int kstack_depth_to_print
= 24;
102 static unsigned int code_bytes
= 64;
104 static inline int valid_stack_ptr(struct thread_info
*tinfo
, void *p
, unsigned size
)
106 return p
> (void *)tinfo
&&
107 p
<= (void *)tinfo
+ THREAD_SIZE
- size
;
110 /* The form of the top of the frame on the stack */
112 struct stack_frame
*next_frame
;
113 unsigned long return_address
;
116 static inline unsigned long print_context_stack(struct thread_info
*tinfo
,
117 unsigned long *stack
, unsigned long ebp
,
118 const struct stacktrace_ops
*ops
, void *data
)
120 #ifdef CONFIG_FRAME_POINTER
121 struct stack_frame
*frame
= (struct stack_frame
*)ebp
;
122 while (valid_stack_ptr(tinfo
, frame
, sizeof(*frame
))) {
123 struct stack_frame
*next
;
126 addr
= frame
->return_address
;
127 ops
->address(data
, addr
);
129 * break out of recursive entries (such as
130 * end_of_stack_stop_unwind_function). Also,
131 * we can never allow a frame pointer to
134 next
= frame
->next_frame
;
140 while (valid_stack_ptr(tinfo
, stack
, sizeof(*stack
))) {
144 if (__kernel_text_address(addr
))
145 ops
->address(data
, addr
);
151 #define MSG(msg) ops->warning(data, msg)
153 void dump_trace(struct task_struct
*task
, struct pt_regs
*regs
,
154 unsigned long *stack
,
155 const struct stacktrace_ops
*ops
, void *data
)
157 unsigned long ebp
= 0;
166 stack
= (unsigned long *)task
->thread
.esp
;
169 #ifdef CONFIG_FRAME_POINTER
171 if (task
== current
) {
172 /* Grab ebp right from our regs */
173 asm ("movl %%ebp, %0" : "=r" (ebp
) : );
175 /* ebp is the last reg pushed by switch_to */
176 ebp
= *(unsigned long *) task
->thread
.esp
;
182 struct thread_info
*context
;
183 context
= (struct thread_info
*)
184 ((unsigned long)stack
& (~(THREAD_SIZE
- 1)));
185 ebp
= print_context_stack(context
, stack
, ebp
, ops
, data
);
186 /* Should be after the line below, but somewhere
187 in early boot context comes out corrupted and we
188 can't reference it -AK */
189 if (ops
->stack(data
, "IRQ") < 0)
191 stack
= (unsigned long*)context
->previous_esp
;
194 touch_nmi_watchdog();
197 EXPORT_SYMBOL(dump_trace
);
200 print_trace_warning_symbol(void *data
, char *msg
, unsigned long symbol
)
203 print_symbol(msg
, symbol
);
207 static void print_trace_warning(void *data
, char *msg
)
209 printk("%s%s\n", (char *)data
, msg
);
212 static int print_trace_stack(void *data
, char *name
)
218 * Print one address/symbol entries per line.
220 static void print_trace_address(void *data
, unsigned long addr
)
222 printk("%s [<%08lx>] ", (char *)data
, addr
);
223 print_symbol("%s\n", addr
);
224 touch_nmi_watchdog();
227 static const struct stacktrace_ops print_trace_ops
= {
228 .warning
= print_trace_warning
,
229 .warning_symbol
= print_trace_warning_symbol
,
230 .stack
= print_trace_stack
,
231 .address
= print_trace_address
,
235 show_trace_log_lvl(struct task_struct
*task
, struct pt_regs
*regs
,
236 unsigned long * stack
, char *log_lvl
)
238 dump_trace(task
, regs
, stack
, &print_trace_ops
, log_lvl
);
239 printk("%s =======================\n", log_lvl
);
242 void show_trace(struct task_struct
*task
, struct pt_regs
*regs
,
243 unsigned long * stack
)
245 show_trace_log_lvl(task
, regs
, stack
, "");
248 static void show_stack_log_lvl(struct task_struct
*task
, struct pt_regs
*regs
,
249 unsigned long *esp
, char *log_lvl
)
251 unsigned long *stack
;
256 esp
= (unsigned long*)task
->thread
.esp
;
258 esp
= (unsigned long *)&esp
;
262 for(i
= 0; i
< kstack_depth_to_print
; i
++) {
263 if (kstack_end(stack
))
265 if (i
&& ((i
% 8) == 0))
266 printk("\n%s ", log_lvl
);
267 printk("%08lx ", *stack
++);
269 printk("\n%sCall Trace:\n", log_lvl
);
270 show_trace_log_lvl(task
, regs
, esp
, log_lvl
);
273 void show_stack(struct task_struct
*task
, unsigned long *esp
)
276 show_stack_log_lvl(task
, NULL
, esp
, "");
280 * The architecture-independent dump_stack generator
282 void dump_stack(void)
286 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
287 current
->pid
, current
->comm
, print_tainted(),
288 init_utsname()->release
,
289 (int)strcspn(init_utsname()->version
, " "),
290 init_utsname()->version
);
291 show_trace(current
, NULL
, &stack
);
294 EXPORT_SYMBOL(dump_stack
);
296 void show_registers(struct pt_regs
*regs
)
301 __show_registers(regs
, 0);
302 printk(KERN_EMERG
"Process %.*s (pid: %d, ti=%p task=%p task.ti=%p)",
303 TASK_COMM_LEN
, current
->comm
, task_pid_nr(current
),
304 current_thread_info(), current
, task_thread_info(current
));
306 * When in-kernel, we also print out the stack and code at the
307 * time of the fault..
309 if (!user_mode_vm(regs
)) {
311 unsigned int code_prologue
= code_bytes
* 43 / 64;
312 unsigned int code_len
= code_bytes
;
315 printk("\n" KERN_EMERG
"Stack: ");
316 show_stack_log_lvl(NULL
, regs
, ®s
->esp
, KERN_EMERG
);
318 printk(KERN_EMERG
"Code: ");
320 eip
= (u8
*)regs
->eip
- code_prologue
;
321 if (eip
< (u8
*)PAGE_OFFSET
||
322 probe_kernel_address(eip
, c
)) {
323 /* try starting at EIP */
324 eip
= (u8
*)regs
->eip
;
325 code_len
= code_len
- code_prologue
+ 1;
327 for (i
= 0; i
< code_len
; i
++, eip
++) {
328 if (eip
< (u8
*)PAGE_OFFSET
||
329 probe_kernel_address(eip
, c
)) {
330 printk(" Bad EIP value.");
333 if (eip
== (u8
*)regs
->eip
)
334 printk("<%02x> ", c
);
342 int is_valid_bugaddr(unsigned long eip
)
346 if (eip
< PAGE_OFFSET
)
348 if (probe_kernel_address((unsigned short *)eip
, ud2
))
351 return ud2
== 0x0b0f;
355 * This is gone through when something in the kernel has done something bad and
356 * is about to be terminated.
358 void die(const char * str
, struct pt_regs
* regs
, long err
)
363 int lock_owner_depth
;
365 .lock
= __RAW_SPIN_LOCK_UNLOCKED
,
367 .lock_owner_depth
= 0
369 static int die_counter
;
374 if (die
.lock_owner
!= raw_smp_processor_id()) {
376 raw_local_irq_save(flags
);
377 __raw_spin_lock(&die
.lock
);
378 die
.lock_owner
= smp_processor_id();
379 die
.lock_owner_depth
= 0;
382 raw_local_irq_save(flags
);
384 if (++die
.lock_owner_depth
< 3) {
388 report_bug(regs
->eip
, regs
);
390 printk(KERN_EMERG
"%s: %04lx [#%d] ", str
, err
& 0xffff,
392 #ifdef CONFIG_PREEMPT
398 #ifdef CONFIG_DEBUG_PAGEALLOC
399 printk("DEBUG_PAGEALLOC");
403 if (notify_die(DIE_OOPS
, str
, regs
, err
,
404 current
->thread
.trap_no
, SIGSEGV
) !=
406 show_registers(regs
);
407 /* Executive summary in case the oops scrolled away */
408 esp
= (unsigned long) (®s
->esp
);
410 if (user_mode(regs
)) {
412 ss
= regs
->xss
& 0xffff;
414 printk(KERN_EMERG
"EIP: [<%08lx>] ", regs
->eip
);
415 print_symbol("%s", regs
->eip
);
416 printk(" SS:ESP %04x:%08lx\n", ss
, esp
);
421 printk(KERN_EMERG
"Recursive die() failure, output suppressed\n");
425 add_taint(TAINT_DIE
);
426 __raw_spin_unlock(&die
.lock
);
427 raw_local_irq_restore(flags
);
432 if (kexec_should_crash(current
))
436 panic("Fatal exception in interrupt");
439 panic("Fatal exception");
445 static inline void die_if_kernel(const char * str
, struct pt_regs
* regs
, long err
)
447 if (!user_mode_vm(regs
))
451 static void __kprobes
do_trap(int trapnr
, int signr
, char *str
, int vm86
,
452 struct pt_regs
* regs
, long error_code
,
455 struct task_struct
*tsk
= current
;
457 if (regs
->eflags
& VM_MASK
) {
463 if (!user_mode(regs
))
468 * We want error_code and trap_no set for userspace faults and
469 * kernelspace faults which result in die(), but not
470 * kernelspace faults which are fixed up. die() gives the
471 * process no chance to handle the signal and notice the
472 * kernel fault information, so that won't result in polluting
473 * the information about previously queued, but not yet
474 * delivered, faults. See also do_general_protection below.
476 tsk
->thread
.error_code
= error_code
;
477 tsk
->thread
.trap_no
= trapnr
;
480 force_sig_info(signr
, info
, tsk
);
482 force_sig(signr
, tsk
);
487 if (!fixup_exception(regs
)) {
488 tsk
->thread
.error_code
= error_code
;
489 tsk
->thread
.trap_no
= trapnr
;
490 die(str
, regs
, error_code
);
496 int ret
= handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
, trapnr
);
497 if (ret
) goto trap_signal
;
502 #define DO_ERROR(trapnr, signr, str, name) \
503 fastcall void do_##name(struct pt_regs * regs, long error_code) \
505 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
508 do_trap(trapnr, signr, str, 0, regs, error_code, NULL); \
511 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr, irq) \
512 fastcall void do_##name(struct pt_regs * regs, long error_code) \
516 local_irq_enable(); \
517 info.si_signo = signr; \
519 info.si_code = sicode; \
520 info.si_addr = (void __user *)siaddr; \
521 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
524 do_trap(trapnr, signr, str, 0, regs, error_code, &info); \
527 #define DO_VM86_ERROR(trapnr, signr, str, name) \
528 fastcall void do_##name(struct pt_regs * regs, long error_code) \
530 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
533 do_trap(trapnr, signr, str, 1, regs, error_code, NULL); \
536 #define DO_VM86_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr) \
537 fastcall void do_##name(struct pt_regs * regs, long error_code) \
540 info.si_signo = signr; \
542 info.si_code = sicode; \
543 info.si_addr = (void __user *)siaddr; \
544 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) \
547 do_trap(trapnr, signr, str, 1, regs, error_code, &info); \
550 DO_VM86_ERROR_INFO( 0, SIGFPE
, "divide error", divide_error
, FPE_INTDIV
, regs
->eip
)
551 #ifndef CONFIG_KPROBES
552 DO_VM86_ERROR( 3, SIGTRAP
, "int3", int3
)
554 DO_VM86_ERROR( 4, SIGSEGV
, "overflow", overflow
)
555 DO_VM86_ERROR( 5, SIGSEGV
, "bounds", bounds
)
556 DO_ERROR_INFO( 6, SIGILL
, "invalid opcode", invalid_op
, ILL_ILLOPN
, regs
->eip
, 0)
557 DO_ERROR( 9, SIGFPE
, "coprocessor segment overrun", coprocessor_segment_overrun
)
558 DO_ERROR(10, SIGSEGV
, "invalid TSS", invalid_TSS
)
559 DO_ERROR(11, SIGBUS
, "segment not present", segment_not_present
)
560 DO_ERROR(12, SIGBUS
, "stack segment", stack_segment
)
561 DO_ERROR_INFO(17, SIGBUS
, "alignment check", alignment_check
, BUS_ADRALN
, 0, 0)
562 DO_ERROR_INFO(32, SIGSEGV
, "iret exception", iret_error
, ILL_BADSTK
, 0, 1)
564 fastcall
void __kprobes
do_general_protection(struct pt_regs
* regs
,
568 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
569 struct thread_struct
*thread
= ¤t
->thread
;
572 * Perform the lazy TSS's I/O bitmap copy. If the TSS has an
573 * invalid offset set (the LAZY one) and the faulting thread has
574 * a valid I/O bitmap pointer, we copy the I/O bitmap in the TSS
575 * and we set the offset field correctly. Then we let the CPU to
576 * restart the faulting instruction.
578 if (tss
->x86_tss
.io_bitmap_base
== INVALID_IO_BITMAP_OFFSET_LAZY
&&
579 thread
->io_bitmap_ptr
) {
580 memcpy(tss
->io_bitmap
, thread
->io_bitmap_ptr
,
581 thread
->io_bitmap_max
);
583 * If the previously set map was extending to higher ports
584 * than the current one, pad extra space with 0xff (no access).
586 if (thread
->io_bitmap_max
< tss
->io_bitmap_max
)
587 memset((char *) tss
->io_bitmap
+
588 thread
->io_bitmap_max
, 0xff,
589 tss
->io_bitmap_max
- thread
->io_bitmap_max
);
590 tss
->io_bitmap_max
= thread
->io_bitmap_max
;
591 tss
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
592 tss
->io_bitmap_owner
= thread
;
598 if (regs
->eflags
& VM_MASK
)
601 if (!user_mode(regs
))
604 current
->thread
.error_code
= error_code
;
605 current
->thread
.trap_no
= 13;
606 if (show_unhandled_signals
&& unhandled_signal(current
, SIGSEGV
) &&
609 "%s[%d] general protection eip:%lx esp:%lx error:%lx\n",
610 current
->comm
, task_pid_nr(current
),
611 regs
->eip
, regs
->esp
, error_code
);
613 force_sig(SIGSEGV
, current
);
618 handle_vm86_fault((struct kernel_vm86_regs
*) regs
, error_code
);
622 if (!fixup_exception(regs
)) {
623 current
->thread
.error_code
= error_code
;
624 current
->thread
.trap_no
= 13;
625 if (notify_die(DIE_GPF
, "general protection fault", regs
,
626 error_code
, 13, SIGSEGV
) == NOTIFY_STOP
)
628 die("general protection fault", regs
, error_code
);
632 static __kprobes
void
633 mem_parity_error(unsigned char reason
, struct pt_regs
* regs
)
635 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x on "
636 "CPU %d.\n", reason
, smp_processor_id());
637 printk(KERN_EMERG
"You have some hardware problem, likely on the PCI bus.\n");
639 #if defined(CONFIG_EDAC)
640 if(edac_handler_set()) {
641 edac_atomic_assert_error();
646 if (panic_on_unrecovered_nmi
)
647 panic("NMI: Not continuing");
649 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
651 /* Clear and disable the memory parity error line. */
652 clear_mem_error(reason
);
655 static __kprobes
void
656 io_check_error(unsigned char reason
, struct pt_regs
* regs
)
660 printk(KERN_EMERG
"NMI: IOCK error (debug interrupt?)\n");
661 show_registers(regs
);
663 /* Re-enable the IOCK line, wait for a few seconds */
664 reason
= (reason
& 0xf) | 8;
667 while (--i
) udelay(1000);
672 static __kprobes
void
673 unknown_nmi_error(unsigned char reason
, struct pt_regs
* regs
)
676 /* Might actually be able to figure out what the guilty party
683 printk(KERN_EMERG
"Uhhuh. NMI received for unknown reason %02x on "
684 "CPU %d.\n", reason
, smp_processor_id());
685 printk(KERN_EMERG
"Do you have a strange power saving mode enabled?\n");
686 if (panic_on_unrecovered_nmi
)
687 panic("NMI: Not continuing");
689 printk(KERN_EMERG
"Dazed and confused, but trying to continue\n");
692 static DEFINE_SPINLOCK(nmi_print_lock
);
694 void __kprobes
die_nmi(struct pt_regs
*regs
, const char *msg
)
696 if (notify_die(DIE_NMIWATCHDOG
, msg
, regs
, 0, 2, SIGINT
) ==
700 spin_lock(&nmi_print_lock
);
702 * We are in trouble anyway, lets at least try
703 * to get a message out.
706 printk(KERN_EMERG
"%s", msg
);
707 printk(" on CPU%d, eip %08lx, registers:\n",
708 smp_processor_id(), regs
->eip
);
709 show_registers(regs
);
711 spin_unlock(&nmi_print_lock
);
714 /* If we are in kernel we are probably nested up pretty bad
715 * and might aswell get out now while we still can.
717 if (!user_mode_vm(regs
)) {
718 current
->thread
.trap_no
= 2;
725 static __kprobes
void default_do_nmi(struct pt_regs
* regs
)
727 unsigned char reason
= 0;
729 /* Only the BSP gets external NMIs from the system. */
730 if (!smp_processor_id())
731 reason
= get_nmi_reason();
733 if (!(reason
& 0xc0)) {
734 if (notify_die(DIE_NMI_IPI
, "nmi_ipi", regs
, reason
, 2, SIGINT
)
737 #ifdef CONFIG_X86_LOCAL_APIC
739 * Ok, so this is none of the documented NMI sources,
740 * so it must be the NMI watchdog.
742 if (nmi_watchdog_tick(regs
, reason
))
744 if (!do_nmi_callback(regs
, smp_processor_id()))
746 unknown_nmi_error(reason
, regs
);
750 if (notify_die(DIE_NMI
, "nmi", regs
, reason
, 2, SIGINT
) == NOTIFY_STOP
)
753 mem_parity_error(reason
, regs
);
755 io_check_error(reason
, regs
);
757 * Reassert NMI in case it became active meanwhile
758 * as it's edge-triggered.
763 static int ignore_nmis
;
765 fastcall __kprobes
void do_nmi(struct pt_regs
* regs
, long error_code
)
771 cpu
= smp_processor_id();
776 default_do_nmi(regs
);
787 void restart_nmi(void)
793 #ifdef CONFIG_KPROBES
794 fastcall
void __kprobes
do_int3(struct pt_regs
*regs
, long error_code
)
796 trace_hardirqs_fixup();
798 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, 3, SIGTRAP
)
801 /* This is an interrupt gate, because kprobes wants interrupts
802 disabled. Normal trap handlers don't. */
803 restore_interrupts(regs
);
804 do_trap(3, SIGTRAP
, "int3", 1, regs
, error_code
, NULL
);
809 * Our handling of the processor debug registers is non-trivial.
810 * We do not clear them on entry and exit from the kernel. Therefore
811 * it is possible to get a watchpoint trap here from inside the kernel.
812 * However, the code in ./ptrace.c has ensured that the user can
813 * only set watchpoints on userspace addresses. Therefore the in-kernel
814 * watchpoint trap can only occur in code which is reading/writing
815 * from user space. Such code must not hold kernel locks (since it
816 * can equally take a page fault), therefore it is safe to call
817 * force_sig_info even though that claims and releases locks.
819 * Code in ./signal.c ensures that the debug control register
820 * is restored before we deliver any signal, and therefore that
821 * user code runs with the correct debug control register even though
824 * Being careful here means that we don't have to be as careful in a
825 * lot of more complicated places (task switching can be a bit lazy
826 * about restoring all the debug state, and ptrace doesn't have to
827 * find every occurrence of the TF bit that could be saved away even
830 fastcall
void __kprobes
do_debug(struct pt_regs
* regs
, long error_code
)
832 unsigned int condition
;
833 struct task_struct
*tsk
= current
;
835 trace_hardirqs_fixup();
837 get_debugreg(condition
, 6);
839 if (notify_die(DIE_DEBUG
, "debug", regs
, condition
, error_code
,
840 SIGTRAP
) == NOTIFY_STOP
)
842 /* It's safe to allow irq's after DR6 has been saved */
843 if (regs
->eflags
& X86_EFLAGS_IF
)
846 /* Mask out spurious debug traps due to lazy DR7 setting */
847 if (condition
& (DR_TRAP0
|DR_TRAP1
|DR_TRAP2
|DR_TRAP3
)) {
848 if (!tsk
->thread
.debugreg
[7])
852 if (regs
->eflags
& VM_MASK
)
855 /* Save debug status register where ptrace can see it */
856 tsk
->thread
.debugreg
[6] = condition
;
859 * Single-stepping through TF: make sure we ignore any events in
860 * kernel space (but re-enable TF when returning to user mode).
862 if (condition
& DR_STEP
) {
864 * We already checked v86 mode above, so we can
865 * check for kernel mode by just checking the CPL
868 if (!user_mode(regs
))
869 goto clear_TF_reenable
;
872 /* Ok, finally something we can handle */
873 send_sigtrap(tsk
, regs
, error_code
);
875 /* Disable additional traps. They'll be re-enabled when
876 * the signal is delivered.
883 handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
, 1);
887 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
888 regs
->eflags
&= ~TF_MASK
;
893 * Note that we play around with the 'TS' bit in an attempt to get
894 * the correct behaviour even in the presence of the asynchronous
897 void math_error(void __user
*eip
)
899 struct task_struct
* task
;
901 unsigned short cwd
, swd
;
904 * Save the info for the exception handler and clear the error.
908 task
->thread
.trap_no
= 16;
909 task
->thread
.error_code
= 0;
910 info
.si_signo
= SIGFPE
;
912 info
.si_code
= __SI_FAULT
;
915 * (~cwd & swd) will mask out exceptions that are not set to unmasked
916 * status. 0x3f is the exception bits in these regs, 0x200 is the
917 * C1 reg you need in case of a stack fault, 0x040 is the stack
918 * fault bit. We should only be taking one exception at a time,
919 * so if this combination doesn't produce any single exception,
920 * then we have a bad program that isn't syncronizing its FPU usage
921 * and it will suffer the consequences since we won't be able to
922 * fully reproduce the context of the exception
924 cwd
= get_fpu_cwd(task
);
925 swd
= get_fpu_swd(task
);
926 switch (swd
& ~cwd
& 0x3f) {
927 case 0x000: /* No unmasked exception */
929 default: /* Multiple exceptions */
931 case 0x001: /* Invalid Op */
933 * swd & 0x240 == 0x040: Stack Underflow
934 * swd & 0x240 == 0x240: Stack Overflow
935 * User must clear the SF bit (0x40) if set
937 info
.si_code
= FPE_FLTINV
;
939 case 0x002: /* Denormalize */
940 case 0x010: /* Underflow */
941 info
.si_code
= FPE_FLTUND
;
943 case 0x004: /* Zero Divide */
944 info
.si_code
= FPE_FLTDIV
;
946 case 0x008: /* Overflow */
947 info
.si_code
= FPE_FLTOVF
;
949 case 0x020: /* Precision */
950 info
.si_code
= FPE_FLTRES
;
953 force_sig_info(SIGFPE
, &info
, task
);
956 fastcall
void do_coprocessor_error(struct pt_regs
* regs
, long error_code
)
959 math_error((void __user
*)regs
->eip
);
962 static void simd_math_error(void __user
*eip
)
964 struct task_struct
* task
;
966 unsigned short mxcsr
;
969 * Save the info for the exception handler and clear the error.
973 task
->thread
.trap_no
= 19;
974 task
->thread
.error_code
= 0;
975 info
.si_signo
= SIGFPE
;
977 info
.si_code
= __SI_FAULT
;
980 * The SIMD FPU exceptions are handled a little differently, as there
981 * is only a single status/control register. Thus, to determine which
982 * unmasked exception was caught we must mask the exception mask bits
983 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
985 mxcsr
= get_fpu_mxcsr(task
);
986 switch (~((mxcsr
& 0x1f80) >> 7) & (mxcsr
& 0x3f)) {
990 case 0x001: /* Invalid Op */
991 info
.si_code
= FPE_FLTINV
;
993 case 0x002: /* Denormalize */
994 case 0x010: /* Underflow */
995 info
.si_code
= FPE_FLTUND
;
997 case 0x004: /* Zero Divide */
998 info
.si_code
= FPE_FLTDIV
;
1000 case 0x008: /* Overflow */
1001 info
.si_code
= FPE_FLTOVF
;
1003 case 0x020: /* Precision */
1004 info
.si_code
= FPE_FLTRES
;
1007 force_sig_info(SIGFPE
, &info
, task
);
1010 fastcall
void do_simd_coprocessor_error(struct pt_regs
* regs
,
1014 /* Handle SIMD FPU exceptions on PIII+ processors. */
1016 simd_math_error((void __user
*)regs
->eip
);
1019 * Handle strange cache flush from user space exception
1020 * in all other cases. This is undocumented behaviour.
1022 if (regs
->eflags
& VM_MASK
) {
1023 handle_vm86_fault((struct kernel_vm86_regs
*)regs
,
1027 current
->thread
.trap_no
= 19;
1028 current
->thread
.error_code
= error_code
;
1029 die_if_kernel("cache flush denied", regs
, error_code
);
1030 force_sig(SIGSEGV
, current
);
1034 fastcall
void do_spurious_interrupt_bug(struct pt_regs
* regs
,
1038 /* No need to warn about this any longer. */
1039 printk("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
1043 fastcall
unsigned long patch_espfix_desc(unsigned long uesp
,
1046 struct desc_struct
*gdt
= __get_cpu_var(gdt_page
).gdt
;
1047 unsigned long base
= (kesp
- uesp
) & -THREAD_SIZE
;
1048 unsigned long new_kesp
= kesp
- base
;
1049 unsigned long lim_pages
= (new_kesp
| (THREAD_SIZE
- 1)) >> PAGE_SHIFT
;
1050 __u64 desc
= *(__u64
*)&gdt
[GDT_ENTRY_ESPFIX_SS
];
1051 /* Set up base for espfix segment */
1052 desc
&= 0x00f0ff0000000000ULL
;
1053 desc
|= ((((__u64
)base
) << 16) & 0x000000ffffff0000ULL
) |
1054 ((((__u64
)base
) << 32) & 0xff00000000000000ULL
) |
1055 ((((__u64
)lim_pages
) << 32) & 0x000f000000000000ULL
) |
1056 (lim_pages
& 0xffff);
1057 *(__u64
*)&gdt
[GDT_ENTRY_ESPFIX_SS
] = desc
;
1062 * 'math_state_restore()' saves the current math information in the
1063 * old math state array, and gets the new ones from the current task
1065 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
1066 * Don't touch unless you *really* know how it works.
1068 * Must be called with kernel preemption disabled (in this case,
1069 * local interrupts are disabled at the call-site in entry.S).
1071 asmlinkage
void math_state_restore(void)
1073 struct thread_info
*thread
= current_thread_info();
1074 struct task_struct
*tsk
= thread
->task
;
1076 clts(); /* Allow maths ops (or we recurse) */
1077 if (!tsk_used_math(tsk
))
1080 thread
->status
|= TS_USEDFPU
; /* So we fnsave on switch_to() */
1083 EXPORT_SYMBOL_GPL(math_state_restore
);
1085 #ifndef CONFIG_MATH_EMULATION
1087 asmlinkage
void math_emulate(long arg
)
1089 printk(KERN_EMERG
"math-emulation not enabled and no coprocessor found.\n");
1090 printk(KERN_EMERG
"killing %s.\n",current
->comm
);
1091 force_sig(SIGFPE
,current
);
1095 #endif /* CONFIG_MATH_EMULATION */
1098 * This needs to use 'idt_table' rather than 'idt', and
1099 * thus use the _nonmapped_ version of the IDT, as the
1100 * Pentium F0 0F bugfix can have resulted in the mapped
1101 * IDT being write-protected.
1103 void set_intr_gate(unsigned int n
, void *addr
)
1105 _set_gate(n
, DESCTYPE_INT
, addr
, __KERNEL_CS
);
1109 * This routine sets up an interrupt gate at directory privilege level 3.
1111 static inline void set_system_intr_gate(unsigned int n
, void *addr
)
1113 _set_gate(n
, DESCTYPE_INT
| DESCTYPE_DPL3
, addr
, __KERNEL_CS
);
1116 static void __init
set_trap_gate(unsigned int n
, void *addr
)
1118 _set_gate(n
, DESCTYPE_TRAP
, addr
, __KERNEL_CS
);
1121 static void __init
set_system_gate(unsigned int n
, void *addr
)
1123 _set_gate(n
, DESCTYPE_TRAP
| DESCTYPE_DPL3
, addr
, __KERNEL_CS
);
1126 static void __init
set_task_gate(unsigned int n
, unsigned int gdt_entry
)
1128 _set_gate(n
, DESCTYPE_TASK
, (void *)0, (gdt_entry
<<3));
1132 void __init
trap_init(void)
1137 void __iomem
*p
= ioremap(0x0FFFD9, 4);
1138 if (readl(p
) == 'E'+('I'<<8)+('S'<<16)+('A'<<24)) {
1144 #ifdef CONFIG_X86_LOCAL_APIC
1145 init_apic_mappings();
1148 set_trap_gate(0,÷_error
);
1149 set_intr_gate(1,&debug
);
1150 set_intr_gate(2,&nmi
);
1151 set_system_intr_gate(3, &int3
); /* int3/4 can be called from all */
1152 set_system_gate(4,&overflow
);
1153 set_trap_gate(5,&bounds
);
1154 set_trap_gate(6,&invalid_op
);
1155 set_trap_gate(7,&device_not_available
);
1156 set_task_gate(8,GDT_ENTRY_DOUBLEFAULT_TSS
);
1157 set_trap_gate(9,&coprocessor_segment_overrun
);
1158 set_trap_gate(10,&invalid_TSS
);
1159 set_trap_gate(11,&segment_not_present
);
1160 set_trap_gate(12,&stack_segment
);
1161 set_trap_gate(13,&general_protection
);
1162 set_intr_gate(14,&page_fault
);
1163 set_trap_gate(15,&spurious_interrupt_bug
);
1164 set_trap_gate(16,&coprocessor_error
);
1165 set_trap_gate(17,&alignment_check
);
1166 #ifdef CONFIG_X86_MCE
1167 set_trap_gate(18,&machine_check
);
1169 set_trap_gate(19,&simd_coprocessor_error
);
1173 * Verify that the FXSAVE/FXRSTOR data will be 16-byte aligned.
1174 * Generates a compile-time "error: zero width for bit-field" if
1175 * the alignment is wrong.
1177 struct fxsrAlignAssert
{
1178 int _
:!(offsetof(struct task_struct
,
1179 thread
.i387
.fxsave
) & 15);
1182 printk(KERN_INFO
"Enabling fast FPU save and restore... ");
1183 set_in_cr4(X86_CR4_OSFXSR
);
1187 printk(KERN_INFO
"Enabling unmasked SIMD FPU exception "
1189 set_in_cr4(X86_CR4_OSXMMEXCPT
);
1193 set_system_gate(SYSCALL_VECTOR
,&system_call
);
1195 /* Reserve all the builtin and the syscall vector. */
1196 for (i
= 0; i
< FIRST_EXTERNAL_VECTOR
; i
++)
1197 set_bit(i
, used_vectors
);
1198 set_bit(SYSCALL_VECTOR
, used_vectors
);
1201 * Should be a barrier for any external CPU state.
1208 static int __init
kstack_setup(char *s
)
1210 kstack_depth_to_print
= simple_strtoul(s
, NULL
, 0);
1213 __setup("kstack=", kstack_setup
);
1215 static int __init
code_bytes_setup(char *s
)
1217 code_bytes
= simple_strtoul(s
, NULL
, 0);
1218 if (code_bytes
> 8192)
1223 __setup("code_bytes=", code_bytes_setup
);