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>
40 #include <linux/hardirq.h>
41 #include <linux/atomic.h>
43 #include <asm/stacktrace.h>
44 #include <asm/processor.h>
45 #include <asm/debugreg.h>
46 #include <asm/text-patching.h>
47 #include <asm/ftrace.h>
48 #include <asm/traps.h>
50 #include <asm/fpu/internal.h>
52 #include <asm/cpu_entry_area.h>
54 #include <asm/fixmap.h>
55 #include <asm/mach_traps.h>
56 #include <asm/alternative.h>
57 #include <asm/fpu/xstate.h>
61 #include <asm/insn-eval.h>
64 #include <asm/x86_init.h>
65 #include <asm/pgalloc.h>
66 #include <asm/proto.h>
68 #include <asm/processor-flags.h>
69 #include <asm/setup.h>
70 #include <asm/proto.h>
73 DECLARE_BITMAP(system_vectors
, NR_VECTORS
);
75 static inline void cond_local_irq_enable(struct pt_regs
*regs
)
77 if (regs
->flags
& X86_EFLAGS_IF
)
81 static inline void cond_local_irq_disable(struct pt_regs
*regs
)
83 if (regs
->flags
& X86_EFLAGS_IF
)
87 int is_valid_bugaddr(unsigned long addr
)
91 if (addr
< TASK_SIZE_MAX
)
94 if (probe_kernel_address((unsigned short *)addr
, ud
))
97 return ud
== INSN_UD0
|| ud
== INSN_UD2
;
100 int fixup_bug(struct pt_regs
*regs
, int trapnr
)
102 if (trapnr
!= X86_TRAP_UD
)
105 switch (report_bug(regs
->ip
, regs
)) {
106 case BUG_TRAP_TYPE_NONE
:
107 case BUG_TRAP_TYPE_BUG
:
110 case BUG_TRAP_TYPE_WARN
:
118 static nokprobe_inline
int
119 do_trap_no_signal(struct task_struct
*tsk
, int trapnr
, const char *str
,
120 struct pt_regs
*regs
, long error_code
)
122 if (v8086_mode(regs
)) {
124 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
125 * On nmi (interrupt 2), do_trap should not be called.
127 if (trapnr
< X86_TRAP_UD
) {
128 if (!handle_vm86_trap((struct kernel_vm86_regs
*) regs
,
132 } else if (!user_mode(regs
)) {
133 if (fixup_exception(regs
, trapnr
, error_code
, 0))
136 tsk
->thread
.error_code
= error_code
;
137 tsk
->thread
.trap_nr
= trapnr
;
138 die(str
, regs
, error_code
);
142 * We want error_code and trap_nr set for userspace faults and
143 * kernelspace faults which result in die(), but not
144 * kernelspace faults which are fixed up. die() gives the
145 * process no chance to handle the signal and notice the
146 * kernel fault information, so that won't result in polluting
147 * the information about previously queued, but not yet
148 * delivered, faults. See also do_general_protection below.
150 tsk
->thread
.error_code
= error_code
;
151 tsk
->thread
.trap_nr
= trapnr
;
156 static void show_signal(struct task_struct
*tsk
, int signr
,
157 const char *type
, const char *desc
,
158 struct pt_regs
*regs
, long error_code
)
160 if (show_unhandled_signals
&& unhandled_signal(tsk
, signr
) &&
161 printk_ratelimit()) {
162 pr_info("%s[%d] %s%s ip:%lx sp:%lx error:%lx",
163 tsk
->comm
, task_pid_nr(tsk
), type
, desc
,
164 regs
->ip
, regs
->sp
, error_code
);
165 print_vma_addr(KERN_CONT
" in ", regs
->ip
);
171 do_trap(int trapnr
, int signr
, char *str
, struct pt_regs
*regs
,
172 long error_code
, int sicode
, void __user
*addr
)
174 struct task_struct
*tsk
= current
;
176 if (!do_trap_no_signal(tsk
, trapnr
, str
, regs
, error_code
))
179 show_signal(tsk
, signr
, "trap ", str
, regs
, error_code
);
184 force_sig_fault(signr
, sicode
, addr
);
186 NOKPROBE_SYMBOL(do_trap
);
188 static void do_error_trap(struct pt_regs
*regs
, long error_code
, char *str
,
189 unsigned long trapnr
, int signr
, int sicode
, void __user
*addr
)
191 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
194 * WARN*()s end up here; fix them up before we call the
197 if (!user_mode(regs
) && fixup_bug(regs
, trapnr
))
200 if (notify_die(DIE_TRAP
, str
, regs
, error_code
, trapnr
, signr
) !=
202 cond_local_irq_enable(regs
);
203 do_trap(trapnr
, signr
, str
, regs
, error_code
, sicode
, addr
);
207 #define IP ((void __user *)uprobe_get_trap_addr(regs))
208 #define DO_ERROR(trapnr, signr, sicode, addr, str, name) \
209 dotraplinkage void do_##name(struct pt_regs *regs, long error_code) \
211 do_error_trap(regs, error_code, str, trapnr, signr, sicode, addr); \
214 DO_ERROR(X86_TRAP_DE
, SIGFPE
, FPE_INTDIV
, IP
, "divide error", divide_error
)
215 DO_ERROR(X86_TRAP_OF
, SIGSEGV
, 0, NULL
, "overflow", overflow
)
216 DO_ERROR(X86_TRAP_UD
, SIGILL
, ILL_ILLOPN
, IP
, "invalid opcode", invalid_op
)
217 DO_ERROR(X86_TRAP_OLD_MF
, SIGFPE
, 0, NULL
, "coprocessor segment overrun", coprocessor_segment_overrun
)
218 DO_ERROR(X86_TRAP_TS
, SIGSEGV
, 0, NULL
, "invalid TSS", invalid_TSS
)
219 DO_ERROR(X86_TRAP_NP
, SIGBUS
, 0, NULL
, "segment not present", segment_not_present
)
220 DO_ERROR(X86_TRAP_SS
, SIGBUS
, 0, NULL
, "stack segment", stack_segment
)
223 dotraplinkage
void do_alignment_check(struct pt_regs
*regs
, long error_code
)
225 char *str
= "alignment check";
227 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
229 if (notify_die(DIE_TRAP
, str
, regs
, error_code
, X86_TRAP_AC
, SIGBUS
) == NOTIFY_STOP
)
232 if (!user_mode(regs
))
233 die("Split lock detected\n", regs
, error_code
);
237 if (handle_user_split_lock(regs
, error_code
))
240 do_trap(X86_TRAP_AC
, SIGBUS
, "alignment check", regs
,
241 error_code
, BUS_ADRALN
, NULL
);
244 #ifdef CONFIG_VMAP_STACK
245 __visible
void __noreturn
handle_stack_overflow(const char *message
,
246 struct pt_regs
*regs
,
247 unsigned long fault_address
)
249 printk(KERN_EMERG
"BUG: stack guard page was hit at %p (stack is %p..%p)\n",
250 (void *)fault_address
, current
->stack
,
251 (char *)current
->stack
+ THREAD_SIZE
- 1);
252 die(message
, regs
, 0);
254 /* Be absolutely certain we don't return. */
255 panic("%s", message
);
260 * Runs on an IST stack for x86_64 and on a special task stack for x86_32.
262 * On x86_64, this is more or less a normal kernel entry. Notwithstanding the
263 * SDM's warnings about double faults being unrecoverable, returning works as
264 * expected. Presumably what the SDM actually means is that the CPU may get
265 * the register state wrong on entry, so returning could be a bad idea.
267 * Various CPU engineers have promised that double faults due to an IRET fault
268 * while the stack is read-only are, in fact, recoverable.
270 * On x86_32, this is entered through a task gate, and regs are synthesized
271 * from the TSS. Returning is, in principle, okay, but changes to regs will
272 * be lost. If, for some reason, we need to return to a context with modified
273 * regs, the shim code could be adjusted to synchronize the registers.
275 dotraplinkage
void do_double_fault(struct pt_regs
*regs
, long error_code
, unsigned long cr2
)
277 static const char str
[] = "double fault";
278 struct task_struct
*tsk
= current
;
280 #ifdef CONFIG_X86_ESPFIX64
281 extern unsigned char native_irq_return_iret
[];
284 * If IRET takes a non-IST fault on the espfix64 stack, then we
285 * end up promoting it to a doublefault. In that case, take
286 * advantage of the fact that we're not using the normal (TSS.sp0)
287 * stack right now. We can write a fake #GP(0) frame at TSS.sp0
288 * and then modify our own IRET frame so that, when we return,
289 * we land directly at the #GP(0) vector with the stack already
290 * set up according to its expectations.
292 * The net result is that our #GP handler will think that we
293 * entered from usermode with the bad user context.
295 * No need for nmi_enter() here because we don't use RCU.
297 if (((long)regs
->sp
>> P4D_SHIFT
) == ESPFIX_PGD_ENTRY
&&
298 regs
->cs
== __KERNEL_CS
&&
299 regs
->ip
== (unsigned long)native_irq_return_iret
)
301 struct pt_regs
*gpregs
= (struct pt_regs
*)this_cpu_read(cpu_tss_rw
.x86_tss
.sp0
) - 1;
304 * regs->sp points to the failing IRET frame on the
305 * ESPFIX64 stack. Copy it to the entry stack. This fills
306 * in gpregs->ss through gpregs->ip.
309 memmove(&gpregs
->ip
, (void *)regs
->sp
, 5*8);
310 gpregs
->orig_ax
= 0; /* Missing (lost) #GP error code */
313 * Adjust our frame so that we return straight to the #GP
314 * vector with the expected RSP value. This is safe because
315 * we won't enable interupts or schedule before we invoke
316 * general_protection, so nothing will clobber the stack
317 * frame we just set up.
319 * We will enter general_protection with kernel GSBASE,
320 * which is what the stub expects, given that the faulting
321 * RIP will be the IRET instruction.
323 regs
->ip
= (unsigned long)general_protection
;
324 regs
->sp
= (unsigned long)&gpregs
->orig_ax
;
331 notify_die(DIE_TRAP
, str
, regs
, error_code
, X86_TRAP_DF
, SIGSEGV
);
333 tsk
->thread
.error_code
= error_code
;
334 tsk
->thread
.trap_nr
= X86_TRAP_DF
;
336 #ifdef CONFIG_VMAP_STACK
338 * If we overflow the stack into a guard page, the CPU will fail
339 * to deliver #PF and will send #DF instead. Similarly, if we
340 * take any non-IST exception while too close to the bottom of
341 * the stack, the processor will get a page fault while
342 * delivering the exception and will generate a double fault.
344 * According to the SDM (footnote in 6.15 under "Interrupt 14 -
345 * Page-Fault Exception (#PF):
347 * Processors update CR2 whenever a page fault is detected. If a
348 * second page fault occurs while an earlier page fault is being
349 * delivered, the faulting linear address of the second fault will
350 * overwrite the contents of CR2 (replacing the previous
351 * address). These updates to CR2 occur even if the page fault
352 * results in a double fault or occurs during the delivery of a
355 * The logic below has a small possibility of incorrectly diagnosing
356 * some errors as stack overflows. For example, if the IDT or GDT
357 * gets corrupted such that #GP delivery fails due to a bad descriptor
358 * causing #GP and we hit this condition while CR2 coincidentally
359 * points to the stack guard page, we'll think we overflowed the
360 * stack. Given that we're going to panic one way or another
361 * if this happens, this isn't necessarily worth fixing.
363 * If necessary, we could improve the test by only diagnosing
364 * a stack overflow if the saved RSP points within 47 bytes of
365 * the bottom of the stack: if RSP == tsk_stack + 48 and we
366 * take an exception, the stack is already aligned and there
367 * will be enough room SS, RSP, RFLAGS, CS, RIP, and a
368 * possible error code, so a stack overflow would *not* double
369 * fault. With any less space left, exception delivery could
370 * fail, and, as a practical matter, we've overflowed the
371 * stack even if the actual trigger for the double fault was
374 if ((unsigned long)task_stack_page(tsk
) - 1 - cr2
< PAGE_SIZE
)
375 handle_stack_overflow("kernel stack overflow (double-fault)", regs
, cr2
);
378 pr_emerg("PANIC: double fault, error_code: 0x%lx\n", error_code
);
379 die("double fault", regs
, error_code
);
380 panic("Machine halted.");
383 dotraplinkage
void do_bounds(struct pt_regs
*regs
, long error_code
)
385 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
386 if (notify_die(DIE_TRAP
, "bounds", regs
, error_code
,
387 X86_TRAP_BR
, SIGSEGV
) == NOTIFY_STOP
)
389 cond_local_irq_enable(regs
);
391 if (!user_mode(regs
))
392 die("bounds", regs
, error_code
);
394 do_trap(X86_TRAP_BR
, SIGSEGV
, "bounds", regs
, error_code
, 0, NULL
);
397 enum kernel_gp_hint
{
404 * When an uncaught #GP occurs, try to determine the memory address accessed by
405 * the instruction and return that address to the caller. Also, try to figure
406 * out whether any part of the access to that address was non-canonical.
408 static enum kernel_gp_hint
get_kernel_gp_address(struct pt_regs
*regs
,
411 u8 insn_buf
[MAX_INSN_SIZE
];
414 if (probe_kernel_read(insn_buf
, (void *)regs
->ip
, MAX_INSN_SIZE
))
417 kernel_insn_init(&insn
, insn_buf
, MAX_INSN_SIZE
);
418 insn_get_modrm(&insn
);
421 *addr
= (unsigned long)insn_get_addr_ref(&insn
, regs
);
428 * - the operand is not in the kernel half
429 * - the last byte of the operand is not in the user canonical half
431 if (*addr
< ~__VIRTUAL_MASK
&&
432 *addr
+ insn
.opnd_bytes
- 1 > __VIRTUAL_MASK
)
433 return GP_NON_CANONICAL
;
439 #define GPFSTR "general protection fault"
441 dotraplinkage
void do_general_protection(struct pt_regs
*regs
, long error_code
)
443 char desc
[sizeof(GPFSTR
) + 50 + 2*sizeof(unsigned long) + 1] = GPFSTR
;
444 enum kernel_gp_hint hint
= GP_NO_HINT
;
445 struct task_struct
*tsk
;
446 unsigned long gp_addr
;
449 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
450 cond_local_irq_enable(regs
);
452 if (static_cpu_has(X86_FEATURE_UMIP
)) {
453 if (user_mode(regs
) && fixup_umip_exception(regs
))
457 if (v8086_mode(regs
)) {
459 handle_vm86_fault((struct kernel_vm86_regs
*) regs
, error_code
);
465 if (user_mode(regs
)) {
466 tsk
->thread
.error_code
= error_code
;
467 tsk
->thread
.trap_nr
= X86_TRAP_GP
;
469 show_signal(tsk
, SIGSEGV
, "", desc
, regs
, error_code
);
475 if (fixup_exception(regs
, X86_TRAP_GP
, error_code
, 0))
478 tsk
->thread
.error_code
= error_code
;
479 tsk
->thread
.trap_nr
= X86_TRAP_GP
;
482 * To be potentially processing a kprobe fault and to trust the result
483 * from kprobe_running(), we have to be non-preemptible.
485 if (!preemptible() &&
487 kprobe_fault_handler(regs
, X86_TRAP_GP
))
490 ret
= notify_die(DIE_GPF
, desc
, regs
, error_code
, X86_TRAP_GP
, SIGSEGV
);
491 if (ret
== NOTIFY_STOP
)
495 snprintf(desc
, sizeof(desc
), "segment-related " GPFSTR
);
497 hint
= get_kernel_gp_address(regs
, &gp_addr
);
499 if (hint
!= GP_NO_HINT
)
500 snprintf(desc
, sizeof(desc
), GPFSTR
", %s 0x%lx",
501 (hint
== GP_NON_CANONICAL
) ? "probably for non-canonical address"
502 : "maybe for address",
506 * KASAN is interested only in the non-canonical case, clear it
509 if (hint
!= GP_NON_CANONICAL
)
512 die_addr(desc
, regs
, error_code
, gp_addr
);
515 NOKPROBE_SYMBOL(do_general_protection
);
517 dotraplinkage
void notrace
do_int3(struct pt_regs
*regs
, long error_code
)
519 if (poke_int3_handler(regs
))
523 * Unlike any other non-IST entry, we can be called from pretty much
524 * any location in the kernel through kprobes -- text_poke() will most
525 * likely be handled by poke_int3_handler() above. This means this
526 * handler is effectively NMI-like.
528 if (!user_mode(regs
))
531 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
532 if (kgdb_ll_trap(DIE_INT3
, "int3", regs
, error_code
, X86_TRAP_BP
,
533 SIGTRAP
) == NOTIFY_STOP
)
535 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
537 #ifdef CONFIG_KPROBES
538 if (kprobe_int3_handler(regs
))
542 if (notify_die(DIE_INT3
, "int3", regs
, error_code
, X86_TRAP_BP
,
543 SIGTRAP
) == NOTIFY_STOP
)
546 cond_local_irq_enable(regs
);
547 do_trap(X86_TRAP_BP
, SIGTRAP
, "int3", regs
, error_code
, 0, NULL
);
548 cond_local_irq_disable(regs
);
551 if (!user_mode(regs
))
554 NOKPROBE_SYMBOL(do_int3
);
558 * Help handler running on a per-cpu (IST or entry trampoline) stack
559 * to switch to the normal thread stack if the interrupted code was in
560 * user mode. The actual stack switch is done in entry_64.S
562 asmlinkage __visible notrace
struct pt_regs
*sync_regs(struct pt_regs
*eregs
)
564 struct pt_regs
*regs
= (struct pt_regs
*)this_cpu_read(cpu_current_top_of_stack
) - 1;
569 NOKPROBE_SYMBOL(sync_regs
);
571 struct bad_iret_stack
{
572 void *error_entry_ret
;
576 asmlinkage __visible notrace
577 struct bad_iret_stack
*fixup_bad_iret(struct bad_iret_stack
*s
)
580 * This is called from entry_64.S early in handling a fault
581 * caused by a bad iret to user mode. To handle the fault
582 * correctly, we want to move our stack frame to where it would
583 * be had we entered directly on the entry stack (rather than
584 * just below the IRET frame) and we want to pretend that the
585 * exception came from the IRET target.
587 struct bad_iret_stack
*new_stack
=
588 (struct bad_iret_stack
*)this_cpu_read(cpu_tss_rw
.x86_tss
.sp0
) - 1;
590 /* Copy the IRET target to the new stack. */
591 memmove(&new_stack
->regs
.ip
, (void *)s
->regs
.sp
, 5*8);
593 /* Copy the remainder of the stack from the current stack. */
594 memmove(new_stack
, s
, offsetof(struct bad_iret_stack
, regs
.ip
));
596 BUG_ON(!user_mode(&new_stack
->regs
));
599 NOKPROBE_SYMBOL(fixup_bad_iret
);
602 static bool is_sysenter_singlestep(struct pt_regs
*regs
)
605 * We don't try for precision here. If we're anywhere in the region of
606 * code that can be single-stepped in the SYSENTER entry path, then
607 * assume that this is a useless single-step trap due to SYSENTER
608 * being invoked with TF set. (We don't know in advance exactly
609 * which instructions will be hit because BTF could plausibly
613 return (regs
->ip
- (unsigned long)__begin_SYSENTER_singlestep_region
) <
614 (unsigned long)__end_SYSENTER_singlestep_region
-
615 (unsigned long)__begin_SYSENTER_singlestep_region
;
616 #elif defined(CONFIG_IA32_EMULATION)
617 return (regs
->ip
- (unsigned long)entry_SYSENTER_compat
) <
618 (unsigned long)__end_entry_SYSENTER_compat
-
619 (unsigned long)entry_SYSENTER_compat
;
626 * Our handling of the processor debug registers is non-trivial.
627 * We do not clear them on entry and exit from the kernel. Therefore
628 * it is possible to get a watchpoint trap here from inside the kernel.
629 * However, the code in ./ptrace.c has ensured that the user can
630 * only set watchpoints on userspace addresses. Therefore the in-kernel
631 * watchpoint trap can only occur in code which is reading/writing
632 * from user space. Such code must not hold kernel locks (since it
633 * can equally take a page fault), therefore it is safe to call
634 * force_sig_info even though that claims and releases locks.
636 * Code in ./signal.c ensures that the debug control register
637 * is restored before we deliver any signal, and therefore that
638 * user code runs with the correct debug control register even though
641 * Being careful here means that we don't have to be as careful in a
642 * lot of more complicated places (task switching can be a bit lazy
643 * about restoring all the debug state, and ptrace doesn't have to
644 * find every occurrence of the TF bit that could be saved away even
647 * May run on IST stack.
649 dotraplinkage
void do_debug(struct pt_regs
*regs
, long error_code
)
651 struct task_struct
*tsk
= current
;
658 get_debugreg(dr6
, 6);
660 * The Intel SDM says:
662 * Certain debug exceptions may clear bits 0-3. The remaining
663 * contents of the DR6 register are never cleared by the
664 * processor. To avoid confusion in identifying debug
665 * exceptions, debug handlers should clear the register before
666 * returning to the interrupted task.
668 * Keep it simple: clear DR6 immediately.
672 /* Filter out all the reserved bits which are preset to 1 */
673 dr6
&= ~DR6_RESERVED
;
676 * The SDM says "The processor clears the BTF flag when it
677 * generates a debug exception." Clear TIF_BLOCKSTEP to keep
678 * TIF_BLOCKSTEP in sync with the hardware BTF flag.
680 clear_tsk_thread_flag(tsk
, TIF_BLOCKSTEP
);
682 if (unlikely(!user_mode(regs
) && (dr6
& DR_STEP
) &&
683 is_sysenter_singlestep(regs
))) {
688 * else we might have gotten a single-step trap and hit a
689 * watchpoint at the same time, in which case we should fall
690 * through and handle the watchpoint.
695 * If dr6 has no reason to give us about the origin of this trap,
696 * then it's very likely the result of an icebp/int01 trap.
697 * User wants a sigtrap for that.
699 if (!dr6
&& user_mode(regs
))
702 /* Store the virtualized DR6 value */
703 tsk
->thread
.debugreg6
= dr6
;
705 #ifdef CONFIG_KPROBES
706 if (kprobe_debug_handler(regs
))
710 if (notify_die(DIE_DEBUG
, "debug", regs
, (long)&dr6
, error_code
,
711 SIGTRAP
) == NOTIFY_STOP
)
715 * Let others (NMI) know that the debug stack is in use
716 * as we may switch to the interrupt stack.
718 debug_stack_usage_inc();
720 /* It's safe to allow irq's after DR6 has been saved */
721 cond_local_irq_enable(regs
);
723 if (v8086_mode(regs
)) {
724 handle_vm86_trap((struct kernel_vm86_regs
*) regs
, error_code
,
726 cond_local_irq_disable(regs
);
727 debug_stack_usage_dec();
731 if (WARN_ON_ONCE((dr6
& DR_STEP
) && !user_mode(regs
))) {
733 * Historical junk that used to handle SYSENTER single-stepping.
734 * This should be unreachable now. If we survive for a while
735 * without anyone hitting this warning, we'll turn this into
738 tsk
->thread
.debugreg6
&= ~DR_STEP
;
739 set_tsk_thread_flag(tsk
, TIF_SINGLESTEP
);
740 regs
->flags
&= ~X86_EFLAGS_TF
;
742 si_code
= get_si_code(tsk
->thread
.debugreg6
);
743 if (tsk
->thread
.debugreg6
& (DR_STEP
| DR_TRAP_BITS
) || user_icebp
)
744 send_sigtrap(regs
, error_code
, si_code
);
745 cond_local_irq_disable(regs
);
746 debug_stack_usage_dec();
751 NOKPROBE_SYMBOL(do_debug
);
754 * Note that we play around with the 'TS' bit in an attempt to get
755 * the correct behaviour even in the presence of the asynchronous
758 static void math_error(struct pt_regs
*regs
, int error_code
, int trapnr
)
760 struct task_struct
*task
= current
;
761 struct fpu
*fpu
= &task
->thread
.fpu
;
763 char *str
= (trapnr
== X86_TRAP_MF
) ? "fpu exception" :
766 cond_local_irq_enable(regs
);
768 if (!user_mode(regs
)) {
769 if (fixup_exception(regs
, trapnr
, error_code
, 0))
772 task
->thread
.error_code
= error_code
;
773 task
->thread
.trap_nr
= trapnr
;
775 if (notify_die(DIE_TRAP
, str
, regs
, error_code
,
776 trapnr
, SIGFPE
) != NOTIFY_STOP
)
777 die(str
, regs
, error_code
);
782 * Save the info for the exception handler and clear the error.
786 task
->thread
.trap_nr
= trapnr
;
787 task
->thread
.error_code
= error_code
;
789 si_code
= fpu__exception_code(fpu
, trapnr
);
790 /* Retry when we get spurious exceptions: */
794 force_sig_fault(SIGFPE
, si_code
,
795 (void __user
*)uprobe_get_trap_addr(regs
));
798 dotraplinkage
void do_coprocessor_error(struct pt_regs
*regs
, long error_code
)
800 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
801 math_error(regs
, error_code
, X86_TRAP_MF
);
805 do_simd_coprocessor_error(struct pt_regs
*regs
, long error_code
)
807 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
808 math_error(regs
, error_code
, X86_TRAP_XF
);
812 do_spurious_interrupt_bug(struct pt_regs
*regs
, long error_code
)
815 * This addresses a Pentium Pro Erratum:
817 * PROBLEM: If the APIC subsystem is configured in mixed mode with
818 * Virtual Wire mode implemented through the local APIC, an
819 * interrupt vector of 0Fh (Intel reserved encoding) may be
820 * generated by the local APIC (Int 15). This vector may be
821 * generated upon receipt of a spurious interrupt (an interrupt
822 * which is removed before the system receives the INTA sequence)
823 * instead of the programmed 8259 spurious interrupt vector.
825 * IMPLICATION: The spurious interrupt vector programmed in the
826 * 8259 is normally handled by an operating system's spurious
827 * interrupt handler. However, a vector of 0Fh is unknown to some
828 * operating systems, which would crash if this erratum occurred.
830 * In theory this could be limited to 32bit, but the handler is not
831 * hurting and who knows which other CPUs suffer from this.
836 do_device_not_available(struct pt_regs
*regs
, long error_code
)
838 unsigned long cr0
= read_cr0();
840 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
842 #ifdef CONFIG_MATH_EMULATION
843 if (!boot_cpu_has(X86_FEATURE_FPU
) && (cr0
& X86_CR0_EM
)) {
844 struct math_emu_info info
= { };
846 cond_local_irq_enable(regs
);
854 /* This should not happen. */
855 if (WARN(cr0
& X86_CR0_TS
, "CR0.TS was set")) {
856 /* Try to fix it up and carry on. */
857 write_cr0(cr0
& ~X86_CR0_TS
);
860 * Something terrible happened, and we're better off trying
861 * to kill the task than getting stuck in a never-ending
862 * loop of #NM faults.
864 die("unexpected #NM exception", regs
, error_code
);
867 NOKPROBE_SYMBOL(do_device_not_available
);
870 dotraplinkage
void do_iret_error(struct pt_regs
*regs
, long error_code
)
872 RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
875 if (notify_die(DIE_TRAP
, "iret exception", regs
, error_code
,
876 X86_TRAP_IRET
, SIGILL
) != NOTIFY_STOP
) {
877 do_trap(X86_TRAP_IRET
, SIGILL
, "iret exception", regs
, error_code
,
878 ILL_BADSTK
, (void __user
*)NULL
);
883 void __init
trap_init(void)
885 /* Init cpu_entry_area before IST entries are set up */
886 setup_cpu_entry_areas();
891 * Set the IDT descriptor to a fixed read-only location, so that the
892 * "sidt" instruction will not leak the location of the kernel, and
893 * to defend the IDT against arbitrary memory write vulnerabilities.
894 * It will be reloaded in cpu_init() */
895 cea_set_pte(CPU_ENTRY_AREA_RO_IDT_VADDR
, __pa_symbol(idt_table
),
897 idt_descr
.address
= CPU_ENTRY_AREA_RO_IDT
;
900 * Should be a barrier for any external CPU state:
904 idt_setup_ist_traps();
906 idt_setup_debugidt_traps();