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(used_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(!on_thread_stack());
146 preempt_enable_no_resched();
150 * ist_end_non_atomic() - begin a non-atomic section in an IST exception
152 * Ends a non-atomic section started with ist_begin_non_atomic().
154 void ist_end_non_atomic(void)
159 int is_valid_bugaddr(unsigned long addr)
163 if (addr < TASK_SIZE_MAX)
166 if (probe_kernel_address((unsigned short *)addr, ud))
169 return ud == INSN_UD0 || ud == INSN_UD2;
172 int fixup_bug(struct pt_regs *regs, int trapnr)
174 if (trapnr != X86_TRAP_UD)
177 switch (report_bug(regs->ip, regs)) {
178 case BUG_TRAP_TYPE_NONE:
179 case BUG_TRAP_TYPE_BUG:
182 case BUG_TRAP_TYPE_WARN:
190 static nokprobe_inline int
191 do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
192 struct pt_regs *regs, long error_code)
194 if (v8086_mode(regs)) {
196 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
197 * On nmi (interrupt 2), do_trap should not be called.
199 if (trapnr < X86_TRAP_UD) {
200 if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
207 if (!user_mode(regs)) {
208 if (fixup_exception(regs, trapnr))
211 tsk->thread.error_code = error_code;
212 tsk->thread.trap_nr = trapnr;
213 die(str, regs, error_code);
219 static siginfo_t *fill_trap_info(struct pt_regs *regs, int signr, int trapnr,
222 unsigned long siaddr;
227 return SEND_SIG_PRIV;
231 siaddr = uprobe_get_trap_addr(regs);
235 siaddr = uprobe_get_trap_addr(regs);
243 info->si_signo = signr;
245 info->si_code = sicode;
246 info->si_addr = (void __user *)siaddr;
251 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
252 long error_code, siginfo_t *info)
254 struct task_struct *tsk = current;
257 if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
260 * We want error_code and trap_nr set for userspace faults and
261 * kernelspace faults which result in die(), but not
262 * kernelspace faults which are fixed up. die() gives the
263 * process no chance to handle the signal and notice the
264 * kernel fault information, so that won't result in polluting
265 * the information about previously queued, but not yet
266 * delivered, faults. See also do_general_protection below.
268 tsk->thread.error_code = error_code;
269 tsk->thread.trap_nr = trapnr;
271 if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
272 printk_ratelimit()) {
273 pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
274 tsk->comm, tsk->pid, str,
275 regs->ip, regs->sp, error_code);
276 print_vma_addr(KERN_CONT " in ", regs->ip);
280 force_sig_info(signr, info ?: SEND_SIG_PRIV, tsk);
282 NOKPROBE_SYMBOL(do_trap);
284 static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
285 unsigned long trapnr, int signr)
289 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
292 * WARN*()s end up here; fix them up before we call the
295 if (!user_mode(regs) && fixup_bug(regs, trapnr))
298 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
300 cond_local_irq_enable(regs);
301 do_trap(trapnr, signr, str, regs, error_code,
302 fill_trap_info(regs, signr, trapnr, &info));
306 #define DO_ERROR(trapnr, signr, str, name) \
307 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
309 do_error_trap(regs, error_code, str, trapnr, signr); \
312 DO_ERROR(X86_TRAP_DE, SIGFPE, "divide error", divide_error)
313 DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
314 DO_ERROR(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op)
315 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",coprocessor_segment_overrun)
316 DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
317 DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
318 DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
319 DO_ERROR(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check)
321 #ifdef CONFIG_VMAP_STACK
322 __visible void __noreturn handle_stack_overflow(const char *message,
323 struct pt_regs *regs,
324 unsigned long fault_address)
326 printk(KERN_EMERG "BUG: stack guard page was hit at %p (stack is %p..%p)\n",
327 (void *)fault_address, current->stack,
328 (char *)current->stack + THREAD_SIZE - 1);
329 die(message, regs, 0);
331 /* Be absolutely certain we don't return. */
337 /* Runs on IST stack */
338 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
340 static const char str[] = "double fault";
341 struct task_struct *tsk = current;
342 #ifdef CONFIG_VMAP_STACK
346 #ifdef CONFIG_X86_ESPFIX64
347 extern unsigned char native_irq_return_iret[];
350 * If IRET takes a non-IST fault on the espfix64 stack, then we
351 * end up promoting it to a doublefault. In that case, take
352 * advantage of the fact that we're not using the normal (TSS.sp0)
353 * stack right now. We can write a fake #GP(0) frame at TSS.sp0
354 * and then modify our own IRET frame so that, when we return,
355 * we land directly at the #GP(0) vector with the stack already
356 * set up according to its expectations.
358 * The net result is that our #GP handler will think that we
359 * entered from usermode with the bad user context.
361 * No need for ist_enter here because we don't use RCU.
363 if (((long)regs->sp >> PGDIR_SHIFT) == ESPFIX_PGD_ENTRY &&
364 regs->cs == __KERNEL_CS &&
365 regs->ip == (unsigned long)native_irq_return_iret)
367 struct pt_regs *gpregs = (struct pt_regs *)this_cpu_read(cpu_tss.x86_tss.sp0) - 1;
370 * regs->sp points to the failing IRET frame on the
371 * ESPFIX64 stack. Copy it to the entry stack. This fills
372 * in gpregs->ss through gpregs->ip.
375 memmove(&gpregs->ip, (void *)regs->sp, 5*8);
376 gpregs->orig_ax = 0; /* Missing (lost) #GP error code */
379 * Adjust our frame so that we return straight to the #GP
380 * vector with the expected RSP value. This is safe because
381 * we won't enable interupts or schedule before we invoke
382 * general_protection, so nothing will clobber the stack
383 * frame we just set up.
385 regs->ip = (unsigned long)general_protection;
386 regs->sp = (unsigned long)&gpregs->orig_ax;
393 notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
395 tsk->thread.error_code = error_code;
396 tsk->thread.trap_nr = X86_TRAP_DF;
398 #ifdef CONFIG_VMAP_STACK
400 * If we overflow the stack into a guard page, the CPU will fail
401 * to deliver #PF and will send #DF instead. Similarly, if we
402 * take any non-IST exception while too close to the bottom of
403 * the stack, the processor will get a page fault while
404 * delivering the exception and will generate a double fault.
406 * According to the SDM (footnote in 6.15 under "Interrupt 14 -
407 * Page-Fault Exception (#PF):
409 * Processors update CR2 whenever a page fault is detected. If a
410 * second page fault occurs while an earlier page fault is being
411 * delivered, the faulting linear address of the second fault will
412 * overwrite the contents of CR2 (replacing the previous
413 * address). These updates to CR2 occur even if the page fault
414 * results in a double fault or occurs during the delivery of a
417 * The logic below has a small possibility of incorrectly diagnosing
418 * some errors as stack overflows. For example, if the IDT or GDT
419 * gets corrupted such that #GP delivery fails due to a bad descriptor
420 * causing #GP and we hit this condition while CR2 coincidentally
421 * points to the stack guard page, we'll think we overflowed the
422 * stack. Given that we're going to panic one way or another
423 * if this happens, this isn't necessarily worth fixing.
425 * If necessary, we could improve the test by only diagnosing
426 * a stack overflow if the saved RSP points within 47 bytes of
427 * the bottom of the stack: if RSP == tsk_stack + 48 and we
428 * take an exception, the stack is already aligned and there
429 * will be enough room SS, RSP, RFLAGS, CS, RIP, and a
430 * possible error code, so a stack overflow would *not* double
431 * fault. With any less space left, exception delivery could
432 * fail, and, as a practical matter, we've overflowed the
433 * stack even if the actual trigger for the double fault was
437 if ((unsigned long)task_stack_page(tsk) - 1 - cr2 < PAGE_SIZE)
438 handle_stack_overflow("kernel stack overflow (double-fault)", regs, cr2);
441 #ifdef CONFIG_DOUBLEFAULT
442 df_debug(regs, error_code);
445 * This is always a kernel trap and never fixable (and thus must
449 die(str, regs, error_code);
453 dotraplinkage void do_bounds(struct pt_regs *regs, long error_code)
455 const struct mpx_bndcsr *bndcsr;
458 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
459 if (notify_die(DIE_TRAP, "bounds", regs, error_code,
460 X86_TRAP_BR, SIGSEGV) == NOTIFY_STOP)
462 cond_local_irq_enable(regs);
464 if (!user_mode(regs))
465 die("bounds", regs, error_code);
467 if (!cpu_feature_enabled(X86_FEATURE_MPX)) {
468 /* The exception is not from Intel MPX */
473 * We need to look at BNDSTATUS to resolve this exception.
474 * A NULL here might mean that it is in its 'init state',
475 * which is all zeros which indicates MPX was not
476 * responsible for the exception.
478 bndcsr = get_xsave_field_ptr(XFEATURE_MASK_BNDCSR);
482 trace_bounds_exception_mpx(bndcsr);
484 * The error code field of the BNDSTATUS register communicates status
485 * information of a bound range exception #BR or operation involving
488 switch (bndcsr->bndstatus & MPX_BNDSTA_ERROR_CODE) {
489 case 2: /* Bound directory has invalid entry. */
490 if (mpx_handle_bd_fault())
492 break; /* Success, it was handled */
493 case 1: /* Bound violation. */
494 info = mpx_generate_siginfo(regs);
497 * We failed to decode the MPX instruction. Act as if
498 * the exception was not caused by MPX.
503 * Success, we decoded the instruction and retrieved
504 * an 'info' containing the address being accessed
505 * which caused the exception. This information
506 * allows and application to possibly handle the
507 * #BR exception itself.
509 do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, info);
512 case 0: /* No exception caused by Intel MPX operations. */
515 die("bounds", regs, error_code);
522 * This path out is for all the cases where we could not
523 * handle the exception in some way (like allocating a
524 * table or telling userspace about it. We will also end
525 * up here if the kernel has MPX turned off at compile
528 do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, NULL);
532 do_general_protection(struct pt_regs *regs, long error_code)
534 struct task_struct *tsk;
536 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
537 cond_local_irq_enable(regs);
539 if (v8086_mode(regs)) {
541 handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
546 if (!user_mode(regs)) {
547 if (fixup_exception(regs, X86_TRAP_GP))
550 tsk->thread.error_code = error_code;
551 tsk->thread.trap_nr = X86_TRAP_GP;
552 if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
553 X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
554 die("general protection fault", regs, error_code);
558 tsk->thread.error_code = error_code;
559 tsk->thread.trap_nr = X86_TRAP_GP;
561 if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
562 printk_ratelimit()) {
563 pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
564 tsk->comm, task_pid_nr(tsk),
565 regs->ip, regs->sp, error_code);
566 print_vma_addr(KERN_CONT " in ", regs->ip);
570 force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
572 NOKPROBE_SYMBOL(do_general_protection);
574 /* May run on IST stack. */
575 dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code)
577 #ifdef CONFIG_DYNAMIC_FTRACE
579 * ftrace must be first, everything else may cause a recursive crash.
580 * See note by declaration of modifying_ftrace_code in ftrace.c
582 if (unlikely(atomic_read(&modifying_ftrace_code)) &&
583 ftrace_int3_handler(regs))
586 if (poke_int3_handler(regs))
590 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
591 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
592 if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
593 SIGTRAP) == NOTIFY_STOP)
595 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
597 #ifdef CONFIG_KPROBES
598 if (kprobe_int3_handler(regs))
602 if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
603 SIGTRAP) == NOTIFY_STOP)
607 * Let others (NMI) know that the debug stack is in use
608 * as we may switch to the interrupt stack.
610 debug_stack_usage_inc();
611 cond_local_irq_enable(regs);
612 do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
613 cond_local_irq_disable(regs);
614 debug_stack_usage_dec();
618 NOKPROBE_SYMBOL(do_int3);
622 * Help handler running on a per-cpu (IST or entry trampoline) stack
623 * to switch to the normal thread stack if the interrupted code was in
624 * user mode. The actual stack switch is done in entry_64.S
626 asmlinkage __visible notrace struct pt_regs *sync_regs(struct pt_regs *eregs)
628 struct pt_regs *regs = (struct pt_regs *)this_cpu_read(cpu_current_top_of_stack) - 1;
633 NOKPROBE_SYMBOL(sync_regs);
635 struct bad_iret_stack {
636 void *error_entry_ret;
640 asmlinkage __visible notrace
641 struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
644 * This is called from entry_64.S early in handling a fault
645 * caused by a bad iret to user mode. To handle the fault
646 * correctly, we want to move our stack frame to where it would
647 * be had we entered directly on the entry stack (rather than
648 * just below the IRET frame) and we want to pretend that the
649 * exception came from the IRET target.
651 struct bad_iret_stack *new_stack =
652 (struct bad_iret_stack *)this_cpu_read(cpu_tss.x86_tss.sp0) - 1;
654 /* Copy the IRET target to the new stack. */
655 memmove(&new_stack->regs.ip, (void *)s->regs.sp, 5*8);
657 /* Copy the remainder of the stack from the current stack. */
658 memmove(new_stack, s, offsetof(struct bad_iret_stack, regs.ip));
660 BUG_ON(!user_mode(&new_stack->regs));
663 NOKPROBE_SYMBOL(fixup_bad_iret);
666 static bool is_sysenter_singlestep(struct pt_regs *regs)
669 * We don't try for precision here. If we're anywhere in the region of
670 * code that can be single-stepped in the SYSENTER entry path, then
671 * assume that this is a useless single-step trap due to SYSENTER
672 * being invoked with TF set. (We don't know in advance exactly
673 * which instructions will be hit because BTF could plausibly
677 return (regs->ip - (unsigned long)__begin_SYSENTER_singlestep_region) <
678 (unsigned long)__end_SYSENTER_singlestep_region -
679 (unsigned long)__begin_SYSENTER_singlestep_region;
680 #elif defined(CONFIG_IA32_EMULATION)
681 return (regs->ip - (unsigned long)entry_SYSENTER_compat) <
682 (unsigned long)__end_entry_SYSENTER_compat -
683 (unsigned long)entry_SYSENTER_compat;
690 * Our handling of the processor debug registers is non-trivial.
691 * We do not clear them on entry and exit from the kernel. Therefore
692 * it is possible to get a watchpoint trap here from inside the kernel.
693 * However, the code in ./ptrace.c has ensured that the user can
694 * only set watchpoints on userspace addresses. Therefore the in-kernel
695 * watchpoint trap can only occur in code which is reading/writing
696 * from user space. Such code must not hold kernel locks (since it
697 * can equally take a page fault), therefore it is safe to call
698 * force_sig_info even though that claims and releases locks.
700 * Code in ./signal.c ensures that the debug control register
701 * is restored before we deliver any signal, and therefore that
702 * user code runs with the correct debug control register even though
705 * Being careful here means that we don't have to be as careful in a
706 * lot of more complicated places (task switching can be a bit lazy
707 * about restoring all the debug state, and ptrace doesn't have to
708 * find every occurrence of the TF bit that could be saved away even
711 * May run on IST stack.
713 dotraplinkage void do_debug(struct pt_regs *regs, long error_code)
715 struct task_struct *tsk = current;
722 get_debugreg(dr6, 6);
724 * The Intel SDM says:
726 * Certain debug exceptions may clear bits 0-3. The remaining
727 * contents of the DR6 register are never cleared by the
728 * processor. To avoid confusion in identifying debug
729 * exceptions, debug handlers should clear the register before
730 * returning to the interrupted task.
732 * Keep it simple: clear DR6 immediately.
736 /* Filter out all the reserved bits which are preset to 1 */
737 dr6 &= ~DR6_RESERVED;
740 * The SDM says "The processor clears the BTF flag when it
741 * generates a debug exception." Clear TIF_BLOCKSTEP to keep
742 * TIF_BLOCKSTEP in sync with the hardware BTF flag.
744 clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
746 if (unlikely(!user_mode(regs) && (dr6 & DR_STEP) &&
747 is_sysenter_singlestep(regs))) {
752 * else we might have gotten a single-step trap and hit a
753 * watchpoint at the same time, in which case we should fall
754 * through and handle the watchpoint.
759 * If dr6 has no reason to give us about the origin of this trap,
760 * then it's very likely the result of an icebp/int01 trap.
761 * User wants a sigtrap for that.
763 if (!dr6 && user_mode(regs))
766 /* Catch kmemcheck conditions! */
767 if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
770 /* Store the virtualized DR6 value */
771 tsk->thread.debugreg6 = dr6;
773 #ifdef CONFIG_KPROBES
774 if (kprobe_debug_handler(regs))
778 if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
779 SIGTRAP) == NOTIFY_STOP)
783 * Let others (NMI) know that the debug stack is in use
784 * as we may switch to the interrupt stack.
786 debug_stack_usage_inc();
788 /* It's safe to allow irq's after DR6 has been saved */
789 cond_local_irq_enable(regs);
791 if (v8086_mode(regs)) {
792 handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
794 cond_local_irq_disable(regs);
795 debug_stack_usage_dec();
799 if (WARN_ON_ONCE((dr6 & DR_STEP) && !user_mode(regs))) {
801 * Historical junk that used to handle SYSENTER single-stepping.
802 * This should be unreachable now. If we survive for a while
803 * without anyone hitting this warning, we'll turn this into
806 tsk->thread.debugreg6 &= ~DR_STEP;
807 set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
808 regs->flags &= ~X86_EFLAGS_TF;
810 si_code = get_si_code(tsk->thread.debugreg6);
811 if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
812 send_sigtrap(tsk, regs, error_code, si_code);
813 cond_local_irq_disable(regs);
814 debug_stack_usage_dec();
818 * This is the most likely code path that involves non-trivial use
819 * of the SYSENTER stack. Check that we haven't overrun it.
821 WARN(this_cpu_read(cpu_tss.SYSENTER_stack_canary) != STACK_END_MAGIC,
822 "Overran or corrupted SYSENTER stack\n");
826 NOKPROBE_SYMBOL(do_debug);
829 * Note that we play around with the 'TS' bit in an attempt to get
830 * the correct behaviour even in the presence of the asynchronous
833 static void math_error(struct pt_regs *regs, int error_code, int trapnr)
835 struct task_struct *task = current;
836 struct fpu *fpu = &task->thread.fpu;
838 char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
841 if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
843 cond_local_irq_enable(regs);
845 if (!user_mode(regs)) {
846 if (!fixup_exception(regs, trapnr)) {
847 task->thread.error_code = error_code;
848 task->thread.trap_nr = trapnr;
849 die(str, regs, error_code);
855 * Save the info for the exception handler and clear the error.
859 task->thread.trap_nr = trapnr;
860 task->thread.error_code = error_code;
861 info.si_signo = SIGFPE;
863 info.si_addr = (void __user *)uprobe_get_trap_addr(regs);
865 info.si_code = fpu__exception_code(fpu, trapnr);
867 /* Retry when we get spurious exceptions: */
871 force_sig_info(SIGFPE, &info, task);
874 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
876 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
877 math_error(regs, error_code, X86_TRAP_MF);
881 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
883 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
884 math_error(regs, error_code, X86_TRAP_XF);
888 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
890 cond_local_irq_enable(regs);
894 do_device_not_available(struct pt_regs *regs, long error_code)
898 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
900 #ifdef CONFIG_MATH_EMULATION
901 if (!boot_cpu_has(X86_FEATURE_FPU) && (read_cr0() & X86_CR0_EM)) {
902 struct math_emu_info info = { };
904 cond_local_irq_enable(regs);
912 /* This should not happen. */
914 if (WARN(cr0 & X86_CR0_TS, "CR0.TS was set")) {
915 /* Try to fix it up and carry on. */
916 write_cr0(cr0 & ~X86_CR0_TS);
919 * Something terrible happened, and we're better off trying
920 * to kill the task than getting stuck in a never-ending
921 * loop of #NM faults.
923 die("unexpected #NM exception", regs, error_code);
926 NOKPROBE_SYMBOL(do_device_not_available);
929 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
933 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
936 info.si_signo = SIGILL;
938 info.si_code = ILL_BADSTK;
940 if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
941 X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
942 do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
948 void __init trap_init(void)
953 * Set the IDT descriptor to a fixed read-only location, so that the
954 * "sidt" instruction will not leak the location of the kernel, and
955 * to defend the IDT against arbitrary memory write vulnerabilities.
956 * It will be reloaded in cpu_init() */
957 __set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
958 idt_descr.address = fix_to_virt(FIX_RO_IDT);
961 * Should be a barrier for any external CPU state:
965 idt_setup_ist_traps();
967 x86_init.irqs.trap_init();
969 idt_setup_debugidt_traps();