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>
41 #if defined(CONFIG_EDAC)
42 #include <linux/edac.h>
45 #include <asm/kmemcheck.h>
46 #include <asm/stacktrace.h>
47 #include <asm/processor.h>
48 #include <asm/debugreg.h>
49 #include <linux/atomic.h>
50 #include <asm/text-patching.h>
51 #include <asm/ftrace.h>
52 #include <asm/traps.h>
54 #include <asm/fpu/internal.h>
56 #include <asm/fixmap.h>
57 #include <asm/mach_traps.h>
58 #include <asm/alternative.h>
59 #include <asm/fpu/xstate.h>
60 #include <asm/trace/mpx.h>
65 #include <asm/x86_init.h>
66 #include <asm/pgalloc.h>
67 #include <asm/proto.h>
69 #include <asm/processor-flags.h>
70 #include <asm/setup.h>
71 #include <asm/proto.h>
74 DECLARE_BITMAP(system_vectors
, NR_VECTORS
);
76 static inline void cond_local_irq_enable(struct pt_regs
*regs
)
78 if (regs
->flags
& X86_EFLAGS_IF
)
82 static inline void cond_local_irq_disable(struct pt_regs
*regs
)
84 if (regs
->flags
& X86_EFLAGS_IF
)
89 * In IST context, we explicitly disable preemption. This serves two
90 * purposes: it makes it much less likely that we would accidentally
91 * schedule in IST context and it will force a warning if we somehow
92 * manage to schedule by accident.
94 void ist_enter(struct pt_regs
*regs
)
96 if (user_mode(regs
)) {
97 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
100 * We might have interrupted pretty much anything. In
101 * fact, if we're a machine check, we can even interrupt
102 * NMI processing. We don't want in_nmi() to return true,
103 * but we need to notify RCU.
110 /* This code is a bit fragile. Test it. */
111 RCU_LOCKDEP_WARN(!rcu_is_watching(), "ist_enter didn't work");
114 void ist_exit(struct pt_regs
*regs
)
116 preempt_enable_no_resched();
118 if (!user_mode(regs
))
123 * ist_begin_non_atomic() - begin a non-atomic section in an IST exception
124 * @regs: regs passed to the IST exception handler
126 * IST exception handlers normally cannot schedule. As a special
127 * exception, if the exception interrupted userspace code (i.e.
128 * user_mode(regs) would return true) and the exception was not
129 * a double fault, it can be safe to schedule. ist_begin_non_atomic()
130 * begins a non-atomic section within an ist_enter()/ist_exit() region.
131 * Callers are responsible for enabling interrupts themselves inside
132 * the non-atomic section, and callers must call ist_end_non_atomic()
135 void ist_begin_non_atomic(struct pt_regs
*regs
)
137 BUG_ON(!user_mode(regs
));
140 * Sanity check: we need to be on the normal thread stack. This
141 * will catch asm bugs and any attempt to use ist_preempt_enable
144 BUG_ON((unsigned long)(current_top_of_stack() -
145 current_stack_pointer()) >= THREAD_SIZE
);
147 preempt_enable_no_resched();
151 * ist_end_non_atomic() - begin a non-atomic section in an IST exception
153 * Ends a non-atomic section started with ist_begin_non_atomic().
155 void ist_end_non_atomic(void)
160 int is_valid_bugaddr(unsigned long addr
)
164 if (addr
< TASK_SIZE_MAX
)
167 if (probe_kernel_address((unsigned short *)addr
, ud
))
170 return ud
== INSN_UD0
|| ud
== INSN_UD2
;
173 int fixup_bug(struct pt_regs
*regs
, int trapnr
)
175 if (trapnr
!= X86_TRAP_UD
)
178 switch (report_bug(regs
->ip
, regs
)) {
179 case BUG_TRAP_TYPE_NONE
:
180 case BUG_TRAP_TYPE_BUG
:
183 case BUG_TRAP_TYPE_WARN
:
191 static nokprobe_inline
int
192 do_trap_no_signal(struct task_struct
*tsk
, int trapnr
, char *str
,
193 struct pt_regs
*regs
, long error_code
)
195 if (v8086_mode(regs
)) {
197 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
198 * On nmi (interrupt 2), do_trap should not be called.
200 if (trapnr
< X86_TRAP_UD
) {
201 if (!handle_vm86_trap((struct kernel_vm86_regs
*) regs
,
208 if (!user_mode(regs
)) {
209 if (fixup_exception(regs
, trapnr
))
212 if (fixup_bug(regs
, trapnr
))
215 tsk
->thread
.error_code
= error_code
;
216 tsk
->thread
.trap_nr
= trapnr
;
217 die(str
, regs
, error_code
);
223 static siginfo_t
*fill_trap_info(struct pt_regs
*regs
, int signr
, int trapnr
,
226 unsigned long siaddr
;
231 return SEND_SIG_PRIV
;
235 siaddr
= uprobe_get_trap_addr(regs
);
239 siaddr
= uprobe_get_trap_addr(regs
);
247 info
->si_signo
= signr
;
249 info
->si_code
= sicode
;
250 info
->si_addr
= (void __user
*)siaddr
;
255 do_trap(int trapnr
, int signr
, char *str
, struct pt_regs
*regs
,
256 long error_code
, siginfo_t
*info
)
258 struct task_struct
*tsk
= current
;
261 if (!do_trap_no_signal(tsk
, trapnr
, str
, regs
, error_code
))
264 * We want error_code and trap_nr set for userspace faults and
265 * kernelspace faults which result in die(), but not
266 * kernelspace faults which are fixed up. die() gives the
267 * process no chance to handle the signal and notice the
268 * kernel fault information, so that won't result in polluting
269 * the information about previously queued, but not yet
270 * delivered, faults. See also do_general_protection below.
272 tsk
->thread
.error_code
= error_code
;
273 tsk
->thread
.trap_nr
= trapnr
;
275 if (show_unhandled_signals
&& unhandled_signal(tsk
, signr
) &&
276 printk_ratelimit()) {
277 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
278 tsk
->comm
, tsk
->pid
, str
,
279 regs
->ip
, regs
->sp
, error_code
);
280 print_vma_addr(KERN_CONT
" in ", regs
->ip
);
284 force_sig_info(signr
, info
?: SEND_SIG_PRIV
, tsk
);
286 NOKPROBE_SYMBOL(do_trap
);
288 static void do_error_trap(struct pt_regs
*regs
, long error_code
, char *str
,
289 unsigned long trapnr
, int signr
)
293 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
295 if (notify_die(DIE_TRAP
, str
, regs
, error_code
, trapnr
, signr
) !=
297 cond_local_irq_enable(regs
);
298 do_trap(trapnr
, signr
, str
, regs
, error_code
,
299 fill_trap_info(regs
, signr
, trapnr
, &info
));
303 #define DO_ERROR(trapnr, signr, str, name) \
304 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
306 do_error_trap(regs, error_code, str, trapnr, signr); \
309 DO_ERROR(X86_TRAP_DE
, SIGFPE
, "divide error", divide_error
)
310 DO_ERROR(X86_TRAP_OF
, SIGSEGV
, "overflow", overflow
)
311 DO_ERROR(X86_TRAP_UD
, SIGILL
, "invalid opcode", invalid_op
)
312 DO_ERROR(X86_TRAP_OLD_MF
, SIGFPE
, "coprocessor segment overrun",coprocessor_segment_overrun
)
313 DO_ERROR(X86_TRAP_TS
, SIGSEGV
, "invalid TSS", invalid_TSS
)
314 DO_ERROR(X86_TRAP_NP
, SIGBUS
, "segment not present", segment_not_present
)
315 DO_ERROR(X86_TRAP_SS
, SIGBUS
, "stack segment", stack_segment
)
316 DO_ERROR(X86_TRAP_AC
, SIGBUS
, "alignment check", alignment_check
)
318 #ifdef CONFIG_VMAP_STACK
319 __visible
void __noreturn
handle_stack_overflow(const char *message
,
320 struct pt_regs
*regs
,
321 unsigned long fault_address
)
323 printk(KERN_EMERG
"BUG: stack guard page was hit at %p (stack is %p..%p)\n",
324 (void *)fault_address
, current
->stack
,
325 (char *)current
->stack
+ THREAD_SIZE
- 1);
326 die(message
, regs
, 0);
328 /* Be absolutely certain we don't return. */
334 /* Runs on IST stack */
335 dotraplinkage
void do_double_fault(struct pt_regs
*regs
, long error_code
)
337 static const char str
[] = "double fault";
338 struct task_struct
*tsk
= current
;
339 #ifdef CONFIG_VMAP_STACK
343 #ifdef CONFIG_X86_ESPFIX64
344 extern unsigned char native_irq_return_iret
[];
347 * If IRET takes a non-IST fault on the espfix64 stack, then we
348 * end up promoting it to a doublefault. In that case, modify
349 * the stack to make it look like we just entered the #GP
350 * handler from user space, similar to bad_iret.
352 * No need for ist_enter here because we don't use RCU.
354 if (((long)regs
->sp
>> PGDIR_SHIFT
) == ESPFIX_PGD_ENTRY
&&
355 regs
->cs
== __KERNEL_CS
&&
356 regs
->ip
== (unsigned long)native_irq_return_iret
)
358 struct pt_regs
*normal_regs
= task_pt_regs(current
);
360 /* Fake a #GP(0) from userspace. */
361 memmove(&normal_regs
->ip
, (void *)regs
->sp
, 5*8);
362 normal_regs
->orig_ax
= 0; /* Missing (lost) #GP error code */
363 regs
->ip
= (unsigned long)general_protection
;
364 regs
->sp
= (unsigned long)&normal_regs
->orig_ax
;
371 notify_die(DIE_TRAP
, str
, regs
, error_code
, X86_TRAP_DF
, SIGSEGV
);
373 tsk
->thread
.error_code
= error_code
;
374 tsk
->thread
.trap_nr
= X86_TRAP_DF
;
376 #ifdef CONFIG_VMAP_STACK
378 * If we overflow the stack into a guard page, the CPU will fail
379 * to deliver #PF and will send #DF instead. Similarly, if we
380 * take any non-IST exception while too close to the bottom of
381 * the stack, the processor will get a page fault while
382 * delivering the exception and will generate a double fault.
384 * According to the SDM (footnote in 6.15 under "Interrupt 14 -
385 * Page-Fault Exception (#PF):
387 * Processors update CR2 whenever a page fault is detected. If a
388 * second page fault occurs while an earlier page fault is being
389 * deliv- ered, the faulting linear address of the second fault will
390 * overwrite the contents of CR2 (replacing the previous
391 * address). These updates to CR2 occur even if the page fault
392 * results in a double fault or occurs during the delivery of a
395 * The logic below has a small possibility of incorrectly diagnosing
396 * some errors as stack overflows. For example, if the IDT or GDT
397 * gets corrupted such that #GP delivery fails due to a bad descriptor
398 * causing #GP and we hit this condition while CR2 coincidentally
399 * points to the stack guard page, we'll think we overflowed the
400 * stack. Given that we're going to panic one way or another
401 * if this happens, this isn't necessarily worth fixing.
403 * If necessary, we could improve the test by only diagnosing
404 * a stack overflow if the saved RSP points within 47 bytes of
405 * the bottom of the stack: if RSP == tsk_stack + 48 and we
406 * take an exception, the stack is already aligned and there
407 * will be enough room SS, RSP, RFLAGS, CS, RIP, and a
408 * possible error code, so a stack overflow would *not* double
409 * fault. With any less space left, exception delivery could
410 * fail, and, as a practical matter, we've overflowed the
411 * stack even if the actual trigger for the double fault was
415 if ((unsigned long)task_stack_page(tsk
) - 1 - cr2
< PAGE_SIZE
)
416 handle_stack_overflow("kernel stack overflow (double-fault)", regs
, cr2
);
419 #ifdef CONFIG_DOUBLEFAULT
420 df_debug(regs
, error_code
);
423 * This is always a kernel trap and never fixable (and thus must
427 die(str
, regs
, error_code
);
431 dotraplinkage
void do_bounds(struct pt_regs
*regs
, long error_code
)
433 const struct mpx_bndcsr
*bndcsr
;
436 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
437 if (notify_die(DIE_TRAP
, "bounds", regs
, error_code
,
438 X86_TRAP_BR
, SIGSEGV
) == NOTIFY_STOP
)
440 cond_local_irq_enable(regs
);
442 if (!user_mode(regs
))
443 die("bounds", regs
, error_code
);
445 if (!cpu_feature_enabled(X86_FEATURE_MPX
)) {
446 /* The exception is not from Intel MPX */
451 * We need to look at BNDSTATUS to resolve this exception.
452 * A NULL here might mean that it is in its 'init state',
453 * which is all zeros which indicates MPX was not
454 * responsible for the exception.
456 bndcsr
= get_xsave_field_ptr(XFEATURE_MASK_BNDCSR
);
460 trace_bounds_exception_mpx(bndcsr
);
462 * The error code field of the BNDSTATUS register communicates status
463 * information of a bound range exception #BR or operation involving
466 switch (bndcsr
->bndstatus
& MPX_BNDSTA_ERROR_CODE
) {
467 case 2: /* Bound directory has invalid entry. */
468 if (mpx_handle_bd_fault())
470 break; /* Success, it was handled */
471 case 1: /* Bound violation. */
472 info
= mpx_generate_siginfo(regs
);
475 * We failed to decode the MPX instruction. Act as if
476 * the exception was not caused by MPX.
481 * Success, we decoded the instruction and retrieved
482 * an 'info' containing the address being accessed
483 * which caused the exception. This information
484 * allows and application to possibly handle the
485 * #BR exception itself.
487 do_trap(X86_TRAP_BR
, SIGSEGV
, "bounds", regs
, error_code
, info
);
490 case 0: /* No exception caused by Intel MPX operations. */
493 die("bounds", regs
, error_code
);
500 * This path out is for all the cases where we could not
501 * handle the exception in some way (like allocating a
502 * table or telling userspace about it. We will also end
503 * up here if the kernel has MPX turned off at compile
506 do_trap(X86_TRAP_BR
, SIGSEGV
, "bounds", regs
, error_code
, NULL
);
510 do_general_protection(struct pt_regs
*regs
, long error_code
)
512 struct task_struct
*tsk
;
514 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
515 cond_local_irq_enable(regs
);
517 if (v8086_mode(regs
)) {
519 handle_vm86_fault((struct kernel_vm86_regs
*) regs
, error_code
);
524 if (!user_mode(regs
)) {
525 if (fixup_exception(regs
, X86_TRAP_GP
))
528 tsk
->thread
.error_code
= error_code
;
529 tsk
->thread
.trap_nr
= X86_TRAP_GP
;
530 if (notify_die(DIE_GPF
, "general protection fault", regs
, error_code
,
531 X86_TRAP_GP
, SIGSEGV
) != NOTIFY_STOP
)
532 die("general protection fault", regs
, error_code
);
536 tsk
->thread
.error_code
= error_code
;
537 tsk
->thread
.trap_nr
= X86_TRAP_GP
;
539 if (show_unhandled_signals
&& unhandled_signal(tsk
, SIGSEGV
) &&
540 printk_ratelimit()) {
541 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
542 tsk
->comm
, task_pid_nr(tsk
),
543 regs
->ip
, regs
->sp
, error_code
);
544 print_vma_addr(KERN_CONT
" in ", regs
->ip
);
548 force_sig_info(SIGSEGV
, SEND_SIG_PRIV
, tsk
);
550 NOKPROBE_SYMBOL(do_general_protection
);
552 /* May run on IST stack. */
553 dotraplinkage
void notrace
do_int3(struct pt_regs
*regs
, long error_code
)
555 #ifdef CONFIG_DYNAMIC_FTRACE
557 * ftrace must be first, everything else may cause a recursive crash.
558 * See note by declaration of modifying_ftrace_code in ftrace.c
560 if (unlikely(atomic_read(&modifying_ftrace_code
)) &&
561 ftrace_int3_handler(regs
))
564 if (poke_int3_handler(regs
))
568 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
569 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
570 if (kgdb_ll_trap(DIE_INT3
, "int3", regs
, error_code
, X86_TRAP_BP
,
571 SIGTRAP
) == NOTIFY_STOP
)
573 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
575 #ifdef CONFIG_KPROBES
576 if (kprobe_int3_handler(regs
))
580 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, X86_TRAP_BP
,
581 SIGTRAP
) == NOTIFY_STOP
)
585 * Let others (NMI) know that the debug stack is in use
586 * as we may switch to the interrupt stack.
588 debug_stack_usage_inc();
589 cond_local_irq_enable(regs
);
590 do_trap(X86_TRAP_BP
, SIGTRAP
, "int3", regs
, error_code
, NULL
);
591 cond_local_irq_disable(regs
);
592 debug_stack_usage_dec();
596 NOKPROBE_SYMBOL(do_int3
);
600 * Help handler running on IST stack to switch off the IST stack if the
601 * interrupted code was in user mode. The actual stack switch is done in
604 asmlinkage __visible notrace
struct pt_regs
*sync_regs(struct pt_regs
*eregs
)
606 struct pt_regs
*regs
= task_pt_regs(current
);
610 NOKPROBE_SYMBOL(sync_regs
);
612 struct bad_iret_stack
{
613 void *error_entry_ret
;
617 asmlinkage __visible notrace
618 struct bad_iret_stack
*fixup_bad_iret(struct bad_iret_stack
*s
)
621 * This is called from entry_64.S early in handling a fault
622 * caused by a bad iret to user mode. To handle the fault
623 * correctly, we want move our stack frame to task_pt_regs
624 * and we want to pretend that the exception came from the
627 struct bad_iret_stack
*new_stack
=
628 container_of(task_pt_regs(current
),
629 struct bad_iret_stack
, regs
);
631 /* Copy the IRET target to the new stack. */
632 memmove(&new_stack
->regs
.ip
, (void *)s
->regs
.sp
, 5*8);
634 /* Copy the remainder of the stack from the current stack. */
635 memmove(new_stack
, s
, offsetof(struct bad_iret_stack
, regs
.ip
));
637 BUG_ON(!user_mode(&new_stack
->regs
));
640 NOKPROBE_SYMBOL(fixup_bad_iret
);
643 static bool is_sysenter_singlestep(struct pt_regs
*regs
)
646 * We don't try for precision here. If we're anywhere in the region of
647 * code that can be single-stepped in the SYSENTER entry path, then
648 * assume that this is a useless single-step trap due to SYSENTER
649 * being invoked with TF set. (We don't know in advance exactly
650 * which instructions will be hit because BTF could plausibly
654 return (regs
->ip
- (unsigned long)__begin_SYSENTER_singlestep_region
) <
655 (unsigned long)__end_SYSENTER_singlestep_region
-
656 (unsigned long)__begin_SYSENTER_singlestep_region
;
657 #elif defined(CONFIG_IA32_EMULATION)
658 return (regs
->ip
- (unsigned long)entry_SYSENTER_compat
) <
659 (unsigned long)__end_entry_SYSENTER_compat
-
660 (unsigned long)entry_SYSENTER_compat
;
667 * Our handling of the processor debug registers is non-trivial.
668 * We do not clear them on entry and exit from the kernel. Therefore
669 * it is possible to get a watchpoint trap here from inside the kernel.
670 * However, the code in ./ptrace.c has ensured that the user can
671 * only set watchpoints on userspace addresses. Therefore the in-kernel
672 * watchpoint trap can only occur in code which is reading/writing
673 * from user space. Such code must not hold kernel locks (since it
674 * can equally take a page fault), therefore it is safe to call
675 * force_sig_info even though that claims and releases locks.
677 * Code in ./signal.c ensures that the debug control register
678 * is restored before we deliver any signal, and therefore that
679 * user code runs with the correct debug control register even though
682 * Being careful here means that we don't have to be as careful in a
683 * lot of more complicated places (task switching can be a bit lazy
684 * about restoring all the debug state, and ptrace doesn't have to
685 * find every occurrence of the TF bit that could be saved away even
688 * May run on IST stack.
690 dotraplinkage
void do_debug(struct pt_regs
*regs
, long error_code
)
692 struct task_struct
*tsk
= current
;
699 get_debugreg(dr6
, 6);
701 * The Intel SDM says:
703 * Certain debug exceptions may clear bits 0-3. The remaining
704 * contents of the DR6 register are never cleared by the
705 * processor. To avoid confusion in identifying debug
706 * exceptions, debug handlers should clear the register before
707 * returning to the interrupted task.
709 * Keep it simple: clear DR6 immediately.
713 /* Filter out all the reserved bits which are preset to 1 */
714 dr6
&= ~DR6_RESERVED
;
717 * The SDM says "The processor clears the BTF flag when it
718 * generates a debug exception." Clear TIF_BLOCKSTEP to keep
719 * TIF_BLOCKSTEP in sync with the hardware BTF flag.
721 clear_tsk_thread_flag(tsk
, TIF_BLOCKSTEP
);
723 if (unlikely(!user_mode(regs
) && (dr6
& DR_STEP
) &&
724 is_sysenter_singlestep(regs
))) {
729 * else we might have gotten a single-step trap and hit a
730 * watchpoint at the same time, in which case we should fall
731 * through and handle the watchpoint.
736 * If dr6 has no reason to give us about the origin of this trap,
737 * then it's very likely the result of an icebp/int01 trap.
738 * User wants a sigtrap for that.
740 if (!dr6
&& user_mode(regs
))
743 /* Catch kmemcheck conditions! */
744 if ((dr6
& DR_STEP
) && kmemcheck_trap(regs
))
747 /* Store the virtualized DR6 value */
748 tsk
->thread
.debugreg6
= dr6
;
750 #ifdef CONFIG_KPROBES
751 if (kprobe_debug_handler(regs
))
755 if (notify_die(DIE_DEBUG
, "debug", regs
, (long)&dr6
, error_code
,
756 SIGTRAP
) == NOTIFY_STOP
)
760 * Let others (NMI) know that the debug stack is in use
761 * as we may switch to the interrupt stack.
763 debug_stack_usage_inc();
765 /* It's safe to allow irq's after DR6 has been saved */
766 cond_local_irq_enable(regs
);
768 if (v8086_mode(regs
)) {
769 handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
,
771 cond_local_irq_disable(regs
);
772 debug_stack_usage_dec();
776 if (WARN_ON_ONCE((dr6
& DR_STEP
) && !user_mode(regs
))) {
778 * Historical junk that used to handle SYSENTER single-stepping.
779 * This should be unreachable now. If we survive for a while
780 * without anyone hitting this warning, we'll turn this into
783 tsk
->thread
.debugreg6
&= ~DR_STEP
;
784 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
785 regs
->flags
&= ~X86_EFLAGS_TF
;
787 si_code
= get_si_code(tsk
->thread
.debugreg6
);
788 if (tsk
->thread
.debugreg6
& (DR_STEP
| DR_TRAP_BITS
) || user_icebp
)
789 send_sigtrap(tsk
, regs
, error_code
, si_code
);
790 cond_local_irq_disable(regs
);
791 debug_stack_usage_dec();
794 #if defined(CONFIG_X86_32)
796 * This is the most likely code path that involves non-trivial use
797 * of the SYSENTER stack. Check that we haven't overrun it.
799 WARN(this_cpu_read(cpu_tss
.SYSENTER_stack_canary
) != STACK_END_MAGIC
,
800 "Overran or corrupted SYSENTER stack\n");
804 NOKPROBE_SYMBOL(do_debug
);
807 * Note that we play around with the 'TS' bit in an attempt to get
808 * the correct behaviour even in the presence of the asynchronous
811 static void math_error(struct pt_regs
*regs
, int error_code
, int trapnr
)
813 struct task_struct
*task
= current
;
814 struct fpu
*fpu
= &task
->thread
.fpu
;
816 char *str
= (trapnr
== X86_TRAP_MF
) ? "fpu exception" :
819 if (notify_die(DIE_TRAP
, str
, regs
, error_code
, trapnr
, SIGFPE
) == NOTIFY_STOP
)
821 cond_local_irq_enable(regs
);
823 if (!user_mode(regs
)) {
824 if (!fixup_exception(regs
, trapnr
)) {
825 task
->thread
.error_code
= error_code
;
826 task
->thread
.trap_nr
= trapnr
;
827 die(str
, regs
, error_code
);
833 * Save the info for the exception handler and clear the error.
837 task
->thread
.trap_nr
= trapnr
;
838 task
->thread
.error_code
= error_code
;
839 info
.si_signo
= SIGFPE
;
841 info
.si_addr
= (void __user
*)uprobe_get_trap_addr(regs
);
843 info
.si_code
= fpu__exception_code(fpu
, trapnr
);
845 /* Retry when we get spurious exceptions: */
849 force_sig_info(SIGFPE
, &info
, task
);
852 dotraplinkage
void do_coprocessor_error(struct pt_regs
*regs
, long error_code
)
854 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
855 math_error(regs
, error_code
, X86_TRAP_MF
);
859 do_simd_coprocessor_error(struct pt_regs
*regs
, long error_code
)
861 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
862 math_error(regs
, error_code
, X86_TRAP_XF
);
866 do_spurious_interrupt_bug(struct pt_regs
*regs
, long error_code
)
868 cond_local_irq_enable(regs
);
872 do_device_not_available(struct pt_regs
*regs
, long error_code
)
876 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
878 #ifdef CONFIG_MATH_EMULATION
879 if (!boot_cpu_has(X86_FEATURE_FPU
) && (read_cr0() & X86_CR0_EM
)) {
880 struct math_emu_info info
= { };
882 cond_local_irq_enable(regs
);
890 /* This should not happen. */
892 if (WARN(cr0
& X86_CR0_TS
, "CR0.TS was set")) {
893 /* Try to fix it up and carry on. */
894 write_cr0(cr0
& ~X86_CR0_TS
);
897 * Something terrible happened, and we're better off trying
898 * to kill the task than getting stuck in a never-ending
899 * loop of #NM faults.
901 die("unexpected #NM exception", regs
, error_code
);
904 NOKPROBE_SYMBOL(do_device_not_available
);
907 dotraplinkage
void do_iret_error(struct pt_regs
*regs
, long error_code
)
911 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
914 info
.si_signo
= SIGILL
;
916 info
.si_code
= ILL_BADSTK
;
918 if (notify_die(DIE_TRAP
, "iret exception", regs
, error_code
,
919 X86_TRAP_IRET
, SIGILL
) != NOTIFY_STOP
) {
920 do_trap(X86_TRAP_IRET
, SIGILL
, "iret exception", regs
, error_code
,
926 void __init
trap_init(void)
931 * Set the IDT descriptor to a fixed read-only location, so that the
932 * "sidt" instruction will not leak the location of the kernel, and
933 * to defend the IDT against arbitrary memory write vulnerabilities.
934 * It will be reloaded in cpu_init() */
935 __set_fixmap(FIX_RO_IDT
, __pa_symbol(idt_table
), PAGE_KERNEL_RO
);
936 idt_descr
.address
= fix_to_virt(FIX_RO_IDT
);
939 * Should be a barrier for any external CPU state:
943 idt_setup_ist_traps();
945 x86_init
.irqs
.trap_init();
947 idt_setup_debugidt_traps();