2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
5 * Pentium III FXSR, SSE support
6 * Gareth Hughes <gareth@valinux.com>, May 2000
10 * Handle hardware traps and faults.
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/context_tracking.h>
16 #include <linux/interrupt.h>
17 #include <linux/kallsyms.h>
18 #include <linux/spinlock.h>
19 #include <linux/kprobes.h>
20 #include <linux/uaccess.h>
21 #include <linux/kdebug.h>
22 #include <linux/kgdb.h>
23 #include <linux/kernel.h>
24 #include <linux/export.h>
25 #include <linux/ptrace.h>
26 #include <linux/uprobes.h>
27 #include <linux/string.h>
28 #include <linux/delay.h>
29 #include <linux/errno.h>
30 #include <linux/kexec.h>
31 #include <linux/sched.h>
32 #include <linux/sched/task_stack.h>
33 #include <linux/timer.h>
34 #include <linux/init.h>
35 #include <linux/bug.h>
36 #include <linux/nmi.h>
38 #include <linux/smp.h>
42 #include <linux/ioport.h>
43 #include <linux/eisa.h>
46 #if defined(CONFIG_EDAC)
47 #include <linux/edac.h>
50 #include <asm/kmemcheck.h>
51 #include <asm/stacktrace.h>
52 #include <asm/processor.h>
53 #include <asm/debugreg.h>
54 #include <linux/atomic.h>
55 #include <asm/text-patching.h>
56 #include <asm/ftrace.h>
57 #include <asm/traps.h>
59 #include <asm/fpu/internal.h>
61 #include <asm/fixmap.h>
62 #include <asm/mach_traps.h>
63 #include <asm/alternative.h>
64 #include <asm/fpu/xstate.h>
65 #include <asm/trace/mpx.h>
70 #include <asm/x86_init.h>
71 #include <asm/pgalloc.h>
72 #include <asm/proto.h>
74 /* No need to be aligned, but done to keep all IDTs defined the same way. */
75 gate_desc debug_idt_table
[NR_VECTORS
] __page_aligned_bss
;
77 #include <asm/processor-flags.h>
78 #include <asm/setup.h>
79 #include <asm/proto.h>
82 /* Must be page-aligned because the real IDT is used in a fixmap. */
83 gate_desc idt_table
[NR_VECTORS
] __page_aligned_bss
;
85 DECLARE_BITMAP(used_vectors
, NR_VECTORS
);
86 EXPORT_SYMBOL_GPL(used_vectors
);
88 static inline void cond_local_irq_enable(struct pt_regs
*regs
)
90 if (regs
->flags
& X86_EFLAGS_IF
)
94 static inline void cond_local_irq_disable(struct pt_regs
*regs
)
96 if (regs
->flags
& X86_EFLAGS_IF
)
101 * In IST context, we explicitly disable preemption. This serves two
102 * purposes: it makes it much less likely that we would accidentally
103 * schedule in IST context and it will force a warning if we somehow
104 * manage to schedule by accident.
106 void ist_enter(struct pt_regs
*regs
)
108 if (user_mode(regs
)) {
109 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
112 * We might have interrupted pretty much anything. In
113 * fact, if we're a machine check, we can even interrupt
114 * NMI processing. We don't want in_nmi() to return true,
115 * but we need to notify RCU.
122 /* This code is a bit fragile. Test it. */
123 RCU_LOCKDEP_WARN(!rcu_is_watching(), "ist_enter didn't work");
126 void ist_exit(struct pt_regs
*regs
)
128 preempt_enable_no_resched();
130 if (!user_mode(regs
))
135 * ist_begin_non_atomic() - begin a non-atomic section in an IST exception
136 * @regs: regs passed to the IST exception handler
138 * IST exception handlers normally cannot schedule. As a special
139 * exception, if the exception interrupted userspace code (i.e.
140 * user_mode(regs) would return true) and the exception was not
141 * a double fault, it can be safe to schedule. ist_begin_non_atomic()
142 * begins a non-atomic section within an ist_enter()/ist_exit() region.
143 * Callers are responsible for enabling interrupts themselves inside
144 * the non-atomic section, and callers must call ist_end_non_atomic()
147 void ist_begin_non_atomic(struct pt_regs
*regs
)
149 BUG_ON(!user_mode(regs
));
152 * Sanity check: we need to be on the normal thread stack. This
153 * will catch asm bugs and any attempt to use ist_preempt_enable
156 BUG_ON(!on_thread_stack());
158 preempt_enable_no_resched();
162 * ist_end_non_atomic() - begin a non-atomic section in an IST exception
164 * Ends a non-atomic section started with ist_begin_non_atomic().
166 void ist_end_non_atomic(void)
171 int is_valid_bugaddr(unsigned long addr
)
175 if (addr
< TASK_SIZE_MAX
)
178 if (probe_kernel_address((unsigned short *)addr
, ud
))
181 return ud
== INSN_UD0
|| ud
== INSN_UD2
;
184 int fixup_bug(struct pt_regs
*regs
, int trapnr
)
186 if (trapnr
!= X86_TRAP_UD
)
189 switch (report_bug(regs
->ip
, regs
)) {
190 case BUG_TRAP_TYPE_NONE
:
191 case BUG_TRAP_TYPE_BUG
:
194 case BUG_TRAP_TYPE_WARN
:
202 static nokprobe_inline
int
203 do_trap_no_signal(struct task_struct
*tsk
, int trapnr
, char *str
,
204 struct pt_regs
*regs
, long error_code
)
206 if (v8086_mode(regs
)) {
208 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
209 * On nmi (interrupt 2), do_trap should not be called.
211 if (trapnr
< X86_TRAP_UD
) {
212 if (!handle_vm86_trap((struct kernel_vm86_regs
*) regs
,
219 if (!user_mode(regs
)) {
220 if (fixup_exception(regs
, trapnr
))
223 tsk
->thread
.error_code
= error_code
;
224 tsk
->thread
.trap_nr
= trapnr
;
225 die(str
, regs
, error_code
);
231 static siginfo_t
*fill_trap_info(struct pt_regs
*regs
, int signr
, int trapnr
,
234 unsigned long siaddr
;
239 return SEND_SIG_PRIV
;
243 siaddr
= uprobe_get_trap_addr(regs
);
247 siaddr
= uprobe_get_trap_addr(regs
);
255 info
->si_signo
= signr
;
257 info
->si_code
= sicode
;
258 info
->si_addr
= (void __user
*)siaddr
;
263 do_trap(int trapnr
, int signr
, char *str
, struct pt_regs
*regs
,
264 long error_code
, siginfo_t
*info
)
266 struct task_struct
*tsk
= current
;
269 if (!do_trap_no_signal(tsk
, trapnr
, str
, regs
, error_code
))
272 * We want error_code and trap_nr set for userspace faults and
273 * kernelspace faults which result in die(), but not
274 * kernelspace faults which are fixed up. die() gives the
275 * process no chance to handle the signal and notice the
276 * kernel fault information, so that won't result in polluting
277 * the information about previously queued, but not yet
278 * delivered, faults. See also do_general_protection below.
280 tsk
->thread
.error_code
= error_code
;
281 tsk
->thread
.trap_nr
= trapnr
;
283 if (show_unhandled_signals
&& unhandled_signal(tsk
, signr
) &&
284 printk_ratelimit()) {
285 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
286 tsk
->comm
, tsk
->pid
, str
,
287 regs
->ip
, regs
->sp
, error_code
);
288 print_vma_addr(KERN_CONT
" in ", regs
->ip
);
292 force_sig_info(signr
, info
?: SEND_SIG_PRIV
, tsk
);
294 NOKPROBE_SYMBOL(do_trap
);
296 static void do_error_trap(struct pt_regs
*regs
, long error_code
, char *str
,
297 unsigned long trapnr
, int signr
)
301 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
304 * WARN*()s end up here; fix them up before we call the
307 if (!user_mode(regs
) && fixup_bug(regs
, trapnr
))
310 if (notify_die(DIE_TRAP
, str
, regs
, error_code
, trapnr
, signr
) !=
312 cond_local_irq_enable(regs
);
313 do_trap(trapnr
, signr
, str
, regs
, error_code
,
314 fill_trap_info(regs
, signr
, trapnr
, &info
));
318 #define DO_ERROR(trapnr, signr, str, name) \
319 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
321 do_error_trap(regs, error_code, str, trapnr, signr); \
324 DO_ERROR(X86_TRAP_DE
, SIGFPE
, "divide error", divide_error
)
325 DO_ERROR(X86_TRAP_OF
, SIGSEGV
, "overflow", overflow
)
326 DO_ERROR(X86_TRAP_UD
, SIGILL
, "invalid opcode", invalid_op
)
327 DO_ERROR(X86_TRAP_OLD_MF
, SIGFPE
, "coprocessor segment overrun",coprocessor_segment_overrun
)
328 DO_ERROR(X86_TRAP_TS
, SIGSEGV
, "invalid TSS", invalid_TSS
)
329 DO_ERROR(X86_TRAP_NP
, SIGBUS
, "segment not present", segment_not_present
)
330 DO_ERROR(X86_TRAP_SS
, SIGBUS
, "stack segment", stack_segment
)
331 DO_ERROR(X86_TRAP_AC
, SIGBUS
, "alignment check", alignment_check
)
333 #ifdef CONFIG_VMAP_STACK
334 __visible
void __noreturn
handle_stack_overflow(const char *message
,
335 struct pt_regs
*regs
,
336 unsigned long fault_address
)
338 printk(KERN_EMERG
"BUG: stack guard page was hit at %p (stack is %p..%p)\n",
339 (void *)fault_address
, current
->stack
,
340 (char *)current
->stack
+ THREAD_SIZE
- 1);
341 die(message
, regs
, 0);
343 /* Be absolutely certain we don't return. */
349 /* Runs on IST stack */
350 dotraplinkage
void do_double_fault(struct pt_regs
*regs
, long error_code
)
352 static const char str
[] = "double fault";
353 struct task_struct
*tsk
= current
;
354 #ifdef CONFIG_VMAP_STACK
358 #ifdef CONFIG_X86_ESPFIX64
359 extern unsigned char native_irq_return_iret
[];
362 * If IRET takes a non-IST fault on the espfix64 stack, then we
363 * end up promoting it to a doublefault. In that case, take
364 * advantage of the fact that we're not using the normal (TSS.sp0)
365 * stack right now. We can write a fake #GP(0) frame at TSS.sp0
366 * and then modify our own IRET frame so that, when we return,
367 * we land directly at the #GP(0) vector with the stack already
368 * set up according to its expectations.
370 * The net result is that our #GP handler will think that we
371 * entered from usermode with the bad user context.
373 * No need for ist_enter here because we don't use RCU.
375 if (((long)regs
->sp
>> PGDIR_SHIFT
) == ESPFIX_PGD_ENTRY
&&
376 regs
->cs
== __KERNEL_CS
&&
377 regs
->ip
== (unsigned long)native_irq_return_iret
)
379 struct pt_regs
*gpregs
= (struct pt_regs
*)this_cpu_read(cpu_tss
.x86_tss
.sp0
) - 1;
382 * regs->sp points to the failing IRET frame on the
383 * ESPFIX64 stack. Copy it to the entry stack. This fills
384 * in gpregs->ss through gpregs->ip.
387 memmove(&gpregs
->ip
, (void *)regs
->sp
, 5*8);
388 gpregs
->orig_ax
= 0; /* Missing (lost) #GP error code */
391 * Adjust our frame so that we return straight to the #GP
392 * vector with the expected RSP value. This is safe because
393 * we won't enable interupts or schedule before we invoke
394 * general_protection, so nothing will clobber the stack
395 * frame we just set up.
397 regs
->ip
= (unsigned long)general_protection
;
398 regs
->sp
= (unsigned long)&gpregs
->orig_ax
;
405 notify_die(DIE_TRAP
, str
, regs
, error_code
, X86_TRAP_DF
, SIGSEGV
);
407 tsk
->thread
.error_code
= error_code
;
408 tsk
->thread
.trap_nr
= X86_TRAP_DF
;
410 #ifdef CONFIG_VMAP_STACK
412 * If we overflow the stack into a guard page, the CPU will fail
413 * to deliver #PF and will send #DF instead. Similarly, if we
414 * take any non-IST exception while too close to the bottom of
415 * the stack, the processor will get a page fault while
416 * delivering the exception and will generate a double fault.
418 * According to the SDM (footnote in 6.15 under "Interrupt 14 -
419 * Page-Fault Exception (#PF):
421 * Processors update CR2 whenever a page fault is detected. If a
422 * second page fault occurs while an earlier page fault is being
423 * delivered, the faulting linear address of the second fault will
424 * overwrite the contents of CR2 (replacing the previous
425 * address). These updates to CR2 occur even if the page fault
426 * results in a double fault or occurs during the delivery of a
429 * The logic below has a small possibility of incorrectly diagnosing
430 * some errors as stack overflows. For example, if the IDT or GDT
431 * gets corrupted such that #GP delivery fails due to a bad descriptor
432 * causing #GP and we hit this condition while CR2 coincidentally
433 * points to the stack guard page, we'll think we overflowed the
434 * stack. Given that we're going to panic one way or another
435 * if this happens, this isn't necessarily worth fixing.
437 * If necessary, we could improve the test by only diagnosing
438 * a stack overflow if the saved RSP points within 47 bytes of
439 * the bottom of the stack: if RSP == tsk_stack + 48 and we
440 * take an exception, the stack is already aligned and there
441 * will be enough room SS, RSP, RFLAGS, CS, RIP, and a
442 * possible error code, so a stack overflow would *not* double
443 * fault. With any less space left, exception delivery could
444 * fail, and, as a practical matter, we've overflowed the
445 * stack even if the actual trigger for the double fault was
449 if ((unsigned long)task_stack_page(tsk
) - 1 - cr2
< PAGE_SIZE
)
450 handle_stack_overflow("kernel stack overflow (double-fault)", regs
, cr2
);
453 #ifdef CONFIG_DOUBLEFAULT
454 df_debug(regs
, error_code
);
457 * This is always a kernel trap and never fixable (and thus must
461 die(str
, regs
, error_code
);
465 dotraplinkage
void do_bounds(struct pt_regs
*regs
, long error_code
)
467 const struct mpx_bndcsr
*bndcsr
;
470 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
471 if (notify_die(DIE_TRAP
, "bounds", regs
, error_code
,
472 X86_TRAP_BR
, SIGSEGV
) == NOTIFY_STOP
)
474 cond_local_irq_enable(regs
);
476 if (!user_mode(regs
))
477 die("bounds", regs
, error_code
);
479 if (!cpu_feature_enabled(X86_FEATURE_MPX
)) {
480 /* The exception is not from Intel MPX */
485 * We need to look at BNDSTATUS to resolve this exception.
486 * A NULL here might mean that it is in its 'init state',
487 * which is all zeros which indicates MPX was not
488 * responsible for the exception.
490 bndcsr
= get_xsave_field_ptr(XFEATURE_MASK_BNDCSR
);
494 trace_bounds_exception_mpx(bndcsr
);
496 * The error code field of the BNDSTATUS register communicates status
497 * information of a bound range exception #BR or operation involving
500 switch (bndcsr
->bndstatus
& MPX_BNDSTA_ERROR_CODE
) {
501 case 2: /* Bound directory has invalid entry. */
502 if (mpx_handle_bd_fault())
504 break; /* Success, it was handled */
505 case 1: /* Bound violation. */
506 info
= mpx_generate_siginfo(regs
);
509 * We failed to decode the MPX instruction. Act as if
510 * the exception was not caused by MPX.
515 * Success, we decoded the instruction and retrieved
516 * an 'info' containing the address being accessed
517 * which caused the exception. This information
518 * allows and application to possibly handle the
519 * #BR exception itself.
521 do_trap(X86_TRAP_BR
, SIGSEGV
, "bounds", regs
, error_code
, info
);
524 case 0: /* No exception caused by Intel MPX operations. */
527 die("bounds", regs
, error_code
);
534 * This path out is for all the cases where we could not
535 * handle the exception in some way (like allocating a
536 * table or telling userspace about it. We will also end
537 * up here if the kernel has MPX turned off at compile
540 do_trap(X86_TRAP_BR
, SIGSEGV
, "bounds", regs
, error_code
, NULL
);
544 do_general_protection(struct pt_regs
*regs
, long error_code
)
546 struct task_struct
*tsk
;
548 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
549 cond_local_irq_enable(regs
);
551 if (v8086_mode(regs
)) {
553 handle_vm86_fault((struct kernel_vm86_regs
*) regs
, error_code
);
558 if (!user_mode(regs
)) {
559 if (fixup_exception(regs
, X86_TRAP_GP
))
562 tsk
->thread
.error_code
= error_code
;
563 tsk
->thread
.trap_nr
= X86_TRAP_GP
;
564 if (notify_die(DIE_GPF
, "general protection fault", regs
, error_code
,
565 X86_TRAP_GP
, SIGSEGV
) != NOTIFY_STOP
)
566 die("general protection fault", regs
, error_code
);
570 tsk
->thread
.error_code
= error_code
;
571 tsk
->thread
.trap_nr
= X86_TRAP_GP
;
573 if (show_unhandled_signals
&& unhandled_signal(tsk
, SIGSEGV
) &&
574 printk_ratelimit()) {
575 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
576 tsk
->comm
, task_pid_nr(tsk
),
577 regs
->ip
, regs
->sp
, error_code
);
578 print_vma_addr(KERN_CONT
" in ", regs
->ip
);
582 force_sig_info(SIGSEGV
, SEND_SIG_PRIV
, tsk
);
584 NOKPROBE_SYMBOL(do_general_protection
);
586 /* May run on IST stack. */
587 dotraplinkage
void notrace
do_int3(struct pt_regs
*regs
, long error_code
)
589 #ifdef CONFIG_DYNAMIC_FTRACE
591 * ftrace must be first, everything else may cause a recursive crash.
592 * See note by declaration of modifying_ftrace_code in ftrace.c
594 if (unlikely(atomic_read(&modifying_ftrace_code
)) &&
595 ftrace_int3_handler(regs
))
598 if (poke_int3_handler(regs
))
602 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
603 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
604 if (kgdb_ll_trap(DIE_INT3
, "int3", regs
, error_code
, X86_TRAP_BP
,
605 SIGTRAP
) == NOTIFY_STOP
)
607 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
609 #ifdef CONFIG_KPROBES
610 if (kprobe_int3_handler(regs
))
614 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, X86_TRAP_BP
,
615 SIGTRAP
) == NOTIFY_STOP
)
619 * Let others (NMI) know that the debug stack is in use
620 * as we may switch to the interrupt stack.
622 debug_stack_usage_inc();
623 cond_local_irq_enable(regs
);
624 do_trap(X86_TRAP_BP
, SIGTRAP
, "int3", regs
, error_code
, NULL
);
625 cond_local_irq_disable(regs
);
626 debug_stack_usage_dec();
630 NOKPROBE_SYMBOL(do_int3
);
634 * Help handler running on IST stack to switch off the IST stack if the
635 * interrupted code was in user mode. The actual stack switch is done in
638 asmlinkage __visible notrace
struct pt_regs
*sync_regs(struct pt_regs
*eregs
)
640 struct pt_regs
*regs
= task_pt_regs(current
);
644 NOKPROBE_SYMBOL(sync_regs
);
646 struct bad_iret_stack
{
647 void *error_entry_ret
;
651 asmlinkage __visible notrace
652 struct bad_iret_stack
*fixup_bad_iret(struct bad_iret_stack
*s
)
655 * This is called from entry_64.S early in handling a fault
656 * caused by a bad iret to user mode. To handle the fault
657 * correctly, we want move our stack frame to task_pt_regs
658 * and we want to pretend that the exception came from the
661 struct bad_iret_stack
*new_stack
=
662 container_of(task_pt_regs(current
),
663 struct bad_iret_stack
, regs
);
665 /* Copy the IRET target to the new stack. */
666 memmove(&new_stack
->regs
.ip
, (void *)s
->regs
.sp
, 5*8);
668 /* Copy the remainder of the stack from the current stack. */
669 memmove(new_stack
, s
, offsetof(struct bad_iret_stack
, regs
.ip
));
671 BUG_ON(!user_mode(&new_stack
->regs
));
674 NOKPROBE_SYMBOL(fixup_bad_iret
);
677 static bool is_sysenter_singlestep(struct pt_regs
*regs
)
680 * We don't try for precision here. If we're anywhere in the region of
681 * code that can be single-stepped in the SYSENTER entry path, then
682 * assume that this is a useless single-step trap due to SYSENTER
683 * being invoked with TF set. (We don't know in advance exactly
684 * which instructions will be hit because BTF could plausibly
688 return (regs
->ip
- (unsigned long)__begin_SYSENTER_singlestep_region
) <
689 (unsigned long)__end_SYSENTER_singlestep_region
-
690 (unsigned long)__begin_SYSENTER_singlestep_region
;
691 #elif defined(CONFIG_IA32_EMULATION)
692 return (regs
->ip
- (unsigned long)entry_SYSENTER_compat
) <
693 (unsigned long)__end_entry_SYSENTER_compat
-
694 (unsigned long)entry_SYSENTER_compat
;
701 * Our handling of the processor debug registers is non-trivial.
702 * We do not clear them on entry and exit from the kernel. Therefore
703 * it is possible to get a watchpoint trap here from inside the kernel.
704 * However, the code in ./ptrace.c has ensured that the user can
705 * only set watchpoints on userspace addresses. Therefore the in-kernel
706 * watchpoint trap can only occur in code which is reading/writing
707 * from user space. Such code must not hold kernel locks (since it
708 * can equally take a page fault), therefore it is safe to call
709 * force_sig_info even though that claims and releases locks.
711 * Code in ./signal.c ensures that the debug control register
712 * is restored before we deliver any signal, and therefore that
713 * user code runs with the correct debug control register even though
716 * Being careful here means that we don't have to be as careful in a
717 * lot of more complicated places (task switching can be a bit lazy
718 * about restoring all the debug state, and ptrace doesn't have to
719 * find every occurrence of the TF bit that could be saved away even
722 * May run on IST stack.
724 dotraplinkage
void do_debug(struct pt_regs
*regs
, long error_code
)
726 struct task_struct
*tsk
= current
;
733 get_debugreg(dr6
, 6);
735 * The Intel SDM says:
737 * Certain debug exceptions may clear bits 0-3. The remaining
738 * contents of the DR6 register are never cleared by the
739 * processor. To avoid confusion in identifying debug
740 * exceptions, debug handlers should clear the register before
741 * returning to the interrupted task.
743 * Keep it simple: clear DR6 immediately.
747 /* Filter out all the reserved bits which are preset to 1 */
748 dr6
&= ~DR6_RESERVED
;
751 * The SDM says "The processor clears the BTF flag when it
752 * generates a debug exception." Clear TIF_BLOCKSTEP to keep
753 * TIF_BLOCKSTEP in sync with the hardware BTF flag.
755 clear_tsk_thread_flag(tsk
, TIF_BLOCKSTEP
);
757 if (unlikely(!user_mode(regs
) && (dr6
& DR_STEP
) &&
758 is_sysenter_singlestep(regs
))) {
763 * else we might have gotten a single-step trap and hit a
764 * watchpoint at the same time, in which case we should fall
765 * through and handle the watchpoint.
770 * If dr6 has no reason to give us about the origin of this trap,
771 * then it's very likely the result of an icebp/int01 trap.
772 * User wants a sigtrap for that.
774 if (!dr6
&& user_mode(regs
))
777 /* Catch kmemcheck conditions! */
778 if ((dr6
& DR_STEP
) && kmemcheck_trap(regs
))
781 /* Store the virtualized DR6 value */
782 tsk
->thread
.debugreg6
= dr6
;
784 #ifdef CONFIG_KPROBES
785 if (kprobe_debug_handler(regs
))
789 if (notify_die(DIE_DEBUG
, "debug", regs
, (long)&dr6
, error_code
,
790 SIGTRAP
) == NOTIFY_STOP
)
794 * Let others (NMI) know that the debug stack is in use
795 * as we may switch to the interrupt stack.
797 debug_stack_usage_inc();
799 /* It's safe to allow irq's after DR6 has been saved */
800 cond_local_irq_enable(regs
);
802 if (v8086_mode(regs
)) {
803 handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
,
805 cond_local_irq_disable(regs
);
806 debug_stack_usage_dec();
810 if (WARN_ON_ONCE((dr6
& DR_STEP
) && !user_mode(regs
))) {
812 * Historical junk that used to handle SYSENTER single-stepping.
813 * This should be unreachable now. If we survive for a while
814 * without anyone hitting this warning, we'll turn this into
817 tsk
->thread
.debugreg6
&= ~DR_STEP
;
818 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
819 regs
->flags
&= ~X86_EFLAGS_TF
;
821 si_code
= get_si_code(tsk
->thread
.debugreg6
);
822 if (tsk
->thread
.debugreg6
& (DR_STEP
| DR_TRAP_BITS
) || user_icebp
)
823 send_sigtrap(tsk
, regs
, error_code
, si_code
);
824 cond_local_irq_disable(regs
);
825 debug_stack_usage_dec();
829 * This is the most likely code path that involves non-trivial use
830 * of the SYSENTER stack. Check that we haven't overrun it.
832 WARN(this_cpu_read(cpu_tss
.SYSENTER_stack_canary
) != STACK_END_MAGIC
,
833 "Overran or corrupted SYSENTER stack\n");
837 NOKPROBE_SYMBOL(do_debug
);
840 * Note that we play around with the 'TS' bit in an attempt to get
841 * the correct behaviour even in the presence of the asynchronous
844 static void math_error(struct pt_regs
*regs
, int error_code
, int trapnr
)
846 struct task_struct
*task
= current
;
847 struct fpu
*fpu
= &task
->thread
.fpu
;
849 char *str
= (trapnr
== X86_TRAP_MF
) ? "fpu exception" :
852 if (notify_die(DIE_TRAP
, str
, regs
, error_code
, trapnr
, SIGFPE
) == NOTIFY_STOP
)
854 cond_local_irq_enable(regs
);
856 if (!user_mode(regs
)) {
857 if (!fixup_exception(regs
, trapnr
)) {
858 task
->thread
.error_code
= error_code
;
859 task
->thread
.trap_nr
= trapnr
;
860 die(str
, regs
, error_code
);
866 * Save the info for the exception handler and clear the error.
870 task
->thread
.trap_nr
= trapnr
;
871 task
->thread
.error_code
= error_code
;
872 info
.si_signo
= SIGFPE
;
874 info
.si_addr
= (void __user
*)uprobe_get_trap_addr(regs
);
876 info
.si_code
= fpu__exception_code(fpu
, trapnr
);
878 /* Retry when we get spurious exceptions: */
882 force_sig_info(SIGFPE
, &info
, task
);
885 dotraplinkage
void do_coprocessor_error(struct pt_regs
*regs
, long error_code
)
887 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
888 math_error(regs
, error_code
, X86_TRAP_MF
);
892 do_simd_coprocessor_error(struct pt_regs
*regs
, long error_code
)
894 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
895 math_error(regs
, error_code
, X86_TRAP_XF
);
899 do_spurious_interrupt_bug(struct pt_regs
*regs
, long error_code
)
901 cond_local_irq_enable(regs
);
905 do_device_not_available(struct pt_regs
*regs
, long error_code
)
909 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
911 #ifdef CONFIG_MATH_EMULATION
912 if (!boot_cpu_has(X86_FEATURE_FPU
) && (read_cr0() & X86_CR0_EM
)) {
913 struct math_emu_info info
= { };
915 cond_local_irq_enable(regs
);
923 /* This should not happen. */
925 if (WARN(cr0
& X86_CR0_TS
, "CR0.TS was set")) {
926 /* Try to fix it up and carry on. */
927 write_cr0(cr0
& ~X86_CR0_TS
);
930 * Something terrible happened, and we're better off trying
931 * to kill the task than getting stuck in a never-ending
932 * loop of #NM faults.
934 die("unexpected #NM exception", regs
, error_code
);
937 NOKPROBE_SYMBOL(do_device_not_available
);
940 dotraplinkage
void do_iret_error(struct pt_regs
*regs
, long error_code
)
944 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
947 info
.si_signo
= SIGILL
;
949 info
.si_code
= ILL_BADSTK
;
951 if (notify_die(DIE_TRAP
, "iret exception", regs
, error_code
,
952 X86_TRAP_IRET
, SIGILL
) != NOTIFY_STOP
) {
953 do_trap(X86_TRAP_IRET
, SIGILL
, "iret exception", regs
, error_code
,
959 /* Set of traps needed for early debugging. */
960 void __init
early_trap_init(void)
963 * Don't use IST to set DEBUG_STACK as it doesn't work until TSS
964 * is ready in cpu_init() <-- trap_init(). Before trap_init(),
965 * CPU runs at ring 0 so it is impossible to hit an invalid
966 * stack. Using the original stack works well enough at this
967 * early stage. DEBUG_STACK will be equipped after cpu_init() in
970 * We don't need to set trace_idt_table like set_intr_gate(),
971 * since we don't have trace_debug and it will be reset to
972 * 'debug' in trap_init() by set_intr_gate_ist().
974 set_intr_gate_notrace(X86_TRAP_DB
, debug
);
975 /* int3 can be called from all */
976 set_system_intr_gate(X86_TRAP_BP
, &int3
);
978 set_intr_gate(X86_TRAP_PF
, page_fault
);
980 load_idt(&idt_descr
);
983 void __init
early_trap_pf_init(void)
986 set_intr_gate(X86_TRAP_PF
, page_fault
);
990 void __init
trap_init(void)
995 void __iomem
*p
= early_ioremap(0x0FFFD9, 4);
997 if (readl(p
) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
1002 set_intr_gate(X86_TRAP_DE
, divide_error
);
1003 set_intr_gate_ist(X86_TRAP_NMI
, &nmi
, NMI_STACK
);
1004 /* int4 can be called from all */
1005 set_system_intr_gate(X86_TRAP_OF
, &overflow
);
1006 set_intr_gate(X86_TRAP_BR
, bounds
);
1007 set_intr_gate(X86_TRAP_UD
, invalid_op
);
1008 set_intr_gate(X86_TRAP_NM
, device_not_available
);
1009 #ifdef CONFIG_X86_32
1010 set_task_gate(X86_TRAP_DF
, GDT_ENTRY_DOUBLEFAULT_TSS
);
1012 set_intr_gate_ist(X86_TRAP_DF
, &double_fault
, DOUBLEFAULT_STACK
);
1014 set_intr_gate(X86_TRAP_OLD_MF
, coprocessor_segment_overrun
);
1015 set_intr_gate(X86_TRAP_TS
, invalid_TSS
);
1016 set_intr_gate(X86_TRAP_NP
, segment_not_present
);
1017 set_intr_gate(X86_TRAP_SS
, stack_segment
);
1018 set_intr_gate(X86_TRAP_GP
, general_protection
);
1019 set_intr_gate(X86_TRAP_SPURIOUS
, spurious_interrupt_bug
);
1020 set_intr_gate(X86_TRAP_MF
, coprocessor_error
);
1021 set_intr_gate(X86_TRAP_AC
, alignment_check
);
1022 #ifdef CONFIG_X86_MCE
1023 set_intr_gate_ist(X86_TRAP_MC
, &machine_check
, MCE_STACK
);
1025 set_intr_gate(X86_TRAP_XF
, simd_coprocessor_error
);
1027 /* Reserve all the builtin and the syscall vector: */
1028 for (i
= 0; i
< FIRST_EXTERNAL_VECTOR
; i
++)
1029 set_bit(i
, used_vectors
);
1031 #ifdef CONFIG_IA32_EMULATION
1032 set_system_intr_gate(IA32_SYSCALL_VECTOR
, entry_INT80_compat
);
1033 set_bit(IA32_SYSCALL_VECTOR
, used_vectors
);
1036 #ifdef CONFIG_X86_32
1037 set_system_intr_gate(IA32_SYSCALL_VECTOR
, entry_INT80_32
);
1038 set_bit(IA32_SYSCALL_VECTOR
, used_vectors
);
1042 * Set the IDT descriptor to a fixed read-only location, so that the
1043 * "sidt" instruction will not leak the location of the kernel, and
1044 * to defend the IDT against arbitrary memory write vulnerabilities.
1045 * It will be reloaded in cpu_init() */
1046 __set_fixmap(FIX_RO_IDT
, __pa_symbol(idt_table
), PAGE_KERNEL_RO
);
1047 idt_descr
.address
= fix_to_virt(FIX_RO_IDT
);
1050 * Should be a barrier for any external CPU state:
1055 * X86_TRAP_DB and X86_TRAP_BP have been set
1056 * in early_trap_init(). However, ITS works only after
1057 * cpu_init() loads TSS. See comments in early_trap_init().
1059 set_intr_gate_ist(X86_TRAP_DB
, &debug
, DEBUG_STACK
);
1060 /* int3 can be called from all */
1061 set_system_intr_gate_ist(X86_TRAP_BP
, &int3
, DEBUG_STACK
);
1063 x86_init
.irqs
.trap_init();
1065 #ifdef CONFIG_X86_64
1066 memcpy(&debug_idt_table
, &idt_table
, IDT_ENTRIES
* 16);
1067 set_nmi_gate(X86_TRAP_DB
, &debug
);
1068 set_nmi_gate(X86_TRAP_BP
, &int3
);