xref: /openbmc/linux/arch/x86/kernel/traps.c (revision 861e10be)
1 /*
2  *  Copyright (C) 1991, 1992  Linus Torvalds
3  *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
4  *
5  *  Pentium III FXSR, SSE support
6  *	Gareth Hughes <gareth@valinux.com>, May 2000
7  */
8 
9 /*
10  * Handle hardware traps and faults.
11  */
12 
13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 
15 #include <linux/interrupt.h>
16 #include <linux/kallsyms.h>
17 #include <linux/spinlock.h>
18 #include <linux/kprobes.h>
19 #include <linux/uaccess.h>
20 #include <linux/kdebug.h>
21 #include <linux/kgdb.h>
22 #include <linux/kernel.h>
23 #include <linux/module.h>
24 #include <linux/ptrace.h>
25 #include <linux/string.h>
26 #include <linux/delay.h>
27 #include <linux/errno.h>
28 #include <linux/kexec.h>
29 #include <linux/sched.h>
30 #include <linux/timer.h>
31 #include <linux/init.h>
32 #include <linux/bug.h>
33 #include <linux/nmi.h>
34 #include <linux/mm.h>
35 #include <linux/smp.h>
36 #include <linux/io.h>
37 
38 #ifdef CONFIG_EISA
39 #include <linux/ioport.h>
40 #include <linux/eisa.h>
41 #endif
42 
43 #if defined(CONFIG_EDAC)
44 #include <linux/edac.h>
45 #endif
46 
47 #include <asm/kmemcheck.h>
48 #include <asm/stacktrace.h>
49 #include <asm/processor.h>
50 #include <asm/debugreg.h>
51 #include <linux/atomic.h>
52 #include <asm/ftrace.h>
53 #include <asm/traps.h>
54 #include <asm/desc.h>
55 #include <asm/i387.h>
56 #include <asm/fpu-internal.h>
57 #include <asm/mce.h>
58 #include <asm/context_tracking.h>
59 
60 #include <asm/mach_traps.h>
61 
62 #ifdef CONFIG_X86_64
63 #include <asm/x86_init.h>
64 #include <asm/pgalloc.h>
65 #include <asm/proto.h>
66 #else
67 #include <asm/processor-flags.h>
68 #include <asm/setup.h>
69 
70 asmlinkage int system_call(void);
71 
72 /*
73  * The IDT has to be page-aligned to simplify the Pentium
74  * F0 0F bug workaround.
75  */
76 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
77 #endif
78 
79 DECLARE_BITMAP(used_vectors, NR_VECTORS);
80 EXPORT_SYMBOL_GPL(used_vectors);
81 
82 static inline void conditional_sti(struct pt_regs *regs)
83 {
84 	if (regs->flags & X86_EFLAGS_IF)
85 		local_irq_enable();
86 }
87 
88 static inline void preempt_conditional_sti(struct pt_regs *regs)
89 {
90 	inc_preempt_count();
91 	if (regs->flags & X86_EFLAGS_IF)
92 		local_irq_enable();
93 }
94 
95 static inline void conditional_cli(struct pt_regs *regs)
96 {
97 	if (regs->flags & X86_EFLAGS_IF)
98 		local_irq_disable();
99 }
100 
101 static inline void preempt_conditional_cli(struct pt_regs *regs)
102 {
103 	if (regs->flags & X86_EFLAGS_IF)
104 		local_irq_disable();
105 	dec_preempt_count();
106 }
107 
108 static int __kprobes
109 do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
110 		  struct pt_regs *regs,	long error_code)
111 {
112 #ifdef CONFIG_X86_32
113 	if (regs->flags & X86_VM_MASK) {
114 		/*
115 		 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
116 		 * On nmi (interrupt 2), do_trap should not be called.
117 		 */
118 		if (trapnr < X86_TRAP_UD) {
119 			if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
120 						error_code, trapnr))
121 				return 0;
122 		}
123 		return -1;
124 	}
125 #endif
126 	if (!user_mode(regs)) {
127 		if (!fixup_exception(regs)) {
128 			tsk->thread.error_code = error_code;
129 			tsk->thread.trap_nr = trapnr;
130 			die(str, regs, error_code);
131 		}
132 		return 0;
133 	}
134 
135 	return -1;
136 }
137 
138 static void __kprobes
139 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
140 	long error_code, siginfo_t *info)
141 {
142 	struct task_struct *tsk = current;
143 
144 
145 	if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
146 		return;
147 	/*
148 	 * We want error_code and trap_nr set for userspace faults and
149 	 * kernelspace faults which result in die(), but not
150 	 * kernelspace faults which are fixed up.  die() gives the
151 	 * process no chance to handle the signal and notice the
152 	 * kernel fault information, so that won't result in polluting
153 	 * the information about previously queued, but not yet
154 	 * delivered, faults.  See also do_general_protection below.
155 	 */
156 	tsk->thread.error_code = error_code;
157 	tsk->thread.trap_nr = trapnr;
158 
159 #ifdef CONFIG_X86_64
160 	if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
161 	    printk_ratelimit()) {
162 		pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
163 			tsk->comm, tsk->pid, str,
164 			regs->ip, regs->sp, error_code);
165 		print_vma_addr(" in ", regs->ip);
166 		pr_cont("\n");
167 	}
168 #endif
169 
170 	if (info)
171 		force_sig_info(signr, info, tsk);
172 	else
173 		force_sig(signr, tsk);
174 }
175 
176 #define DO_ERROR(trapnr, signr, str, name)				\
177 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
178 {									\
179 	exception_enter(regs);						\
180 	if (notify_die(DIE_TRAP, str, regs, error_code,			\
181 			trapnr, signr) == NOTIFY_STOP) {		\
182 		exception_exit(regs);					\
183 		return;							\
184 	}								\
185 	conditional_sti(regs);						\
186 	do_trap(trapnr, signr, str, regs, error_code, NULL);		\
187 	exception_exit(regs);						\
188 }
189 
190 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)		\
191 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
192 {									\
193 	siginfo_t info;							\
194 	info.si_signo = signr;						\
195 	info.si_errno = 0;						\
196 	info.si_code = sicode;						\
197 	info.si_addr = (void __user *)siaddr;				\
198 	exception_enter(regs);						\
199 	if (notify_die(DIE_TRAP, str, regs, error_code,			\
200 			trapnr, signr) == NOTIFY_STOP) {		\
201 		exception_exit(regs);					\
202 		return;							\
203 	}								\
204 	conditional_sti(regs);						\
205 	do_trap(trapnr, signr, str, regs, error_code, &info);		\
206 	exception_exit(regs);						\
207 }
208 
209 DO_ERROR_INFO(X86_TRAP_DE, SIGFPE, "divide error", divide_error, FPE_INTDIV,
210 		regs->ip)
211 DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
212 DO_ERROR(X86_TRAP_BR, SIGSEGV, "bounds", bounds)
213 DO_ERROR_INFO(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN,
214 		regs->ip)
215 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",
216 		coprocessor_segment_overrun)
217 DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
218 DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
219 #ifdef CONFIG_X86_32
220 DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
221 #endif
222 DO_ERROR_INFO(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check,
223 		BUS_ADRALN, 0)
224 
225 #ifdef CONFIG_X86_64
226 /* Runs on IST stack */
227 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
228 {
229 	exception_enter(regs);
230 	if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
231 		       X86_TRAP_SS, SIGBUS) != NOTIFY_STOP) {
232 		preempt_conditional_sti(regs);
233 		do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
234 		preempt_conditional_cli(regs);
235 	}
236 	exception_exit(regs);
237 }
238 
239 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
240 {
241 	static const char str[] = "double fault";
242 	struct task_struct *tsk = current;
243 
244 	exception_enter(regs);
245 	/* Return not checked because double check cannot be ignored */
246 	notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
247 
248 	tsk->thread.error_code = error_code;
249 	tsk->thread.trap_nr = X86_TRAP_DF;
250 
251 	/*
252 	 * This is always a kernel trap and never fixable (and thus must
253 	 * never return).
254 	 */
255 	for (;;)
256 		die(str, regs, error_code);
257 }
258 #endif
259 
260 dotraplinkage void __kprobes
261 do_general_protection(struct pt_regs *regs, long error_code)
262 {
263 	struct task_struct *tsk;
264 
265 	exception_enter(regs);
266 	conditional_sti(regs);
267 
268 #ifdef CONFIG_X86_32
269 	if (regs->flags & X86_VM_MASK) {
270 		local_irq_enable();
271 		handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
272 		goto exit;
273 	}
274 #endif
275 
276 	tsk = current;
277 	if (!user_mode(regs)) {
278 		if (fixup_exception(regs))
279 			goto exit;
280 
281 		tsk->thread.error_code = error_code;
282 		tsk->thread.trap_nr = X86_TRAP_GP;
283 		if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
284 			       X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
285 			die("general protection fault", regs, error_code);
286 		goto exit;
287 	}
288 
289 	tsk->thread.error_code = error_code;
290 	tsk->thread.trap_nr = X86_TRAP_GP;
291 
292 	if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
293 			printk_ratelimit()) {
294 		pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
295 			tsk->comm, task_pid_nr(tsk),
296 			regs->ip, regs->sp, error_code);
297 		print_vma_addr(" in ", regs->ip);
298 		pr_cont("\n");
299 	}
300 
301 	force_sig(SIGSEGV, tsk);
302 exit:
303 	exception_exit(regs);
304 }
305 
306 /* May run on IST stack. */
307 dotraplinkage void __kprobes notrace do_int3(struct pt_regs *regs, long error_code)
308 {
309 #ifdef CONFIG_DYNAMIC_FTRACE
310 	/*
311 	 * ftrace must be first, everything else may cause a recursive crash.
312 	 * See note by declaration of modifying_ftrace_code in ftrace.c
313 	 */
314 	if (unlikely(atomic_read(&modifying_ftrace_code)) &&
315 	    ftrace_int3_handler(regs))
316 		return;
317 #endif
318 	exception_enter(regs);
319 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
320 	if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
321 				SIGTRAP) == NOTIFY_STOP)
322 		goto exit;
323 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
324 
325 	if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
326 			SIGTRAP) == NOTIFY_STOP)
327 		goto exit;
328 
329 	/*
330 	 * Let others (NMI) know that the debug stack is in use
331 	 * as we may switch to the interrupt stack.
332 	 */
333 	debug_stack_usage_inc();
334 	preempt_conditional_sti(regs);
335 	do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
336 	preempt_conditional_cli(regs);
337 	debug_stack_usage_dec();
338 exit:
339 	exception_exit(regs);
340 }
341 
342 #ifdef CONFIG_X86_64
343 /*
344  * Help handler running on IST stack to switch back to user stack
345  * for scheduling or signal handling. The actual stack switch is done in
346  * entry.S
347  */
348 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
349 {
350 	struct pt_regs *regs = eregs;
351 	/* Did already sync */
352 	if (eregs == (struct pt_regs *)eregs->sp)
353 		;
354 	/* Exception from user space */
355 	else if (user_mode(eregs))
356 		regs = task_pt_regs(current);
357 	/*
358 	 * Exception from kernel and interrupts are enabled. Move to
359 	 * kernel process stack.
360 	 */
361 	else if (eregs->flags & X86_EFLAGS_IF)
362 		regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
363 	if (eregs != regs)
364 		*regs = *eregs;
365 	return regs;
366 }
367 #endif
368 
369 /*
370  * Our handling of the processor debug registers is non-trivial.
371  * We do not clear them on entry and exit from the kernel. Therefore
372  * it is possible to get a watchpoint trap here from inside the kernel.
373  * However, the code in ./ptrace.c has ensured that the user can
374  * only set watchpoints on userspace addresses. Therefore the in-kernel
375  * watchpoint trap can only occur in code which is reading/writing
376  * from user space. Such code must not hold kernel locks (since it
377  * can equally take a page fault), therefore it is safe to call
378  * force_sig_info even though that claims and releases locks.
379  *
380  * Code in ./signal.c ensures that the debug control register
381  * is restored before we deliver any signal, and therefore that
382  * user code runs with the correct debug control register even though
383  * we clear it here.
384  *
385  * Being careful here means that we don't have to be as careful in a
386  * lot of more complicated places (task switching can be a bit lazy
387  * about restoring all the debug state, and ptrace doesn't have to
388  * find every occurrence of the TF bit that could be saved away even
389  * by user code)
390  *
391  * May run on IST stack.
392  */
393 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
394 {
395 	struct task_struct *tsk = current;
396 	int user_icebp = 0;
397 	unsigned long dr6;
398 	int si_code;
399 
400 	exception_enter(regs);
401 
402 	get_debugreg(dr6, 6);
403 
404 	/* Filter out all the reserved bits which are preset to 1 */
405 	dr6 &= ~DR6_RESERVED;
406 
407 	/*
408 	 * If dr6 has no reason to give us about the origin of this trap,
409 	 * then it's very likely the result of an icebp/int01 trap.
410 	 * User wants a sigtrap for that.
411 	 */
412 	if (!dr6 && user_mode(regs))
413 		user_icebp = 1;
414 
415 	/* Catch kmemcheck conditions first of all! */
416 	if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
417 		goto exit;
418 
419 	/* DR6 may or may not be cleared by the CPU */
420 	set_debugreg(0, 6);
421 
422 	/*
423 	 * The processor cleared BTF, so don't mark that we need it set.
424 	 */
425 	clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
426 
427 	/* Store the virtualized DR6 value */
428 	tsk->thread.debugreg6 = dr6;
429 
430 	if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
431 							SIGTRAP) == NOTIFY_STOP)
432 		goto exit;
433 
434 	/*
435 	 * Let others (NMI) know that the debug stack is in use
436 	 * as we may switch to the interrupt stack.
437 	 */
438 	debug_stack_usage_inc();
439 
440 	/* It's safe to allow irq's after DR6 has been saved */
441 	preempt_conditional_sti(regs);
442 
443 	if (regs->flags & X86_VM_MASK) {
444 		handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
445 					X86_TRAP_DB);
446 		preempt_conditional_cli(regs);
447 		debug_stack_usage_dec();
448 		goto exit;
449 	}
450 
451 	/*
452 	 * Single-stepping through system calls: ignore any exceptions in
453 	 * kernel space, but re-enable TF when returning to user mode.
454 	 *
455 	 * We already checked v86 mode above, so we can check for kernel mode
456 	 * by just checking the CPL of CS.
457 	 */
458 	if ((dr6 & DR_STEP) && !user_mode(regs)) {
459 		tsk->thread.debugreg6 &= ~DR_STEP;
460 		set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
461 		regs->flags &= ~X86_EFLAGS_TF;
462 	}
463 	si_code = get_si_code(tsk->thread.debugreg6);
464 	if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
465 		send_sigtrap(tsk, regs, error_code, si_code);
466 	preempt_conditional_cli(regs);
467 	debug_stack_usage_dec();
468 
469 exit:
470 	exception_exit(regs);
471 }
472 
473 /*
474  * Note that we play around with the 'TS' bit in an attempt to get
475  * the correct behaviour even in the presence of the asynchronous
476  * IRQ13 behaviour
477  */
478 void math_error(struct pt_regs *regs, int error_code, int trapnr)
479 {
480 	struct task_struct *task = current;
481 	siginfo_t info;
482 	unsigned short err;
483 	char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
484 						"simd exception";
485 
486 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
487 		return;
488 	conditional_sti(regs);
489 
490 	if (!user_mode_vm(regs))
491 	{
492 		if (!fixup_exception(regs)) {
493 			task->thread.error_code = error_code;
494 			task->thread.trap_nr = trapnr;
495 			die(str, regs, error_code);
496 		}
497 		return;
498 	}
499 
500 	/*
501 	 * Save the info for the exception handler and clear the error.
502 	 */
503 	save_init_fpu(task);
504 	task->thread.trap_nr = trapnr;
505 	task->thread.error_code = error_code;
506 	info.si_signo = SIGFPE;
507 	info.si_errno = 0;
508 	info.si_addr = (void __user *)regs->ip;
509 	if (trapnr == X86_TRAP_MF) {
510 		unsigned short cwd, swd;
511 		/*
512 		 * (~cwd & swd) will mask out exceptions that are not set to unmasked
513 		 * status.  0x3f is the exception bits in these regs, 0x200 is the
514 		 * C1 reg you need in case of a stack fault, 0x040 is the stack
515 		 * fault bit.  We should only be taking one exception at a time,
516 		 * so if this combination doesn't produce any single exception,
517 		 * then we have a bad program that isn't synchronizing its FPU usage
518 		 * and it will suffer the consequences since we won't be able to
519 		 * fully reproduce the context of the exception
520 		 */
521 		cwd = get_fpu_cwd(task);
522 		swd = get_fpu_swd(task);
523 
524 		err = swd & ~cwd;
525 	} else {
526 		/*
527 		 * The SIMD FPU exceptions are handled a little differently, as there
528 		 * is only a single status/control register.  Thus, to determine which
529 		 * unmasked exception was caught we must mask the exception mask bits
530 		 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
531 		 */
532 		unsigned short mxcsr = get_fpu_mxcsr(task);
533 		err = ~(mxcsr >> 7) & mxcsr;
534 	}
535 
536 	if (err & 0x001) {	/* Invalid op */
537 		/*
538 		 * swd & 0x240 == 0x040: Stack Underflow
539 		 * swd & 0x240 == 0x240: Stack Overflow
540 		 * User must clear the SF bit (0x40) if set
541 		 */
542 		info.si_code = FPE_FLTINV;
543 	} else if (err & 0x004) { /* Divide by Zero */
544 		info.si_code = FPE_FLTDIV;
545 	} else if (err & 0x008) { /* Overflow */
546 		info.si_code = FPE_FLTOVF;
547 	} else if (err & 0x012) { /* Denormal, Underflow */
548 		info.si_code = FPE_FLTUND;
549 	} else if (err & 0x020) { /* Precision */
550 		info.si_code = FPE_FLTRES;
551 	} else {
552 		/*
553 		 * If we're using IRQ 13, or supposedly even some trap
554 		 * X86_TRAP_MF implementations, it's possible
555 		 * we get a spurious trap, which is not an error.
556 		 */
557 		return;
558 	}
559 	force_sig_info(SIGFPE, &info, task);
560 }
561 
562 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
563 {
564 	exception_enter(regs);
565 	math_error(regs, error_code, X86_TRAP_MF);
566 	exception_exit(regs);
567 }
568 
569 dotraplinkage void
570 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
571 {
572 	exception_enter(regs);
573 	math_error(regs, error_code, X86_TRAP_XF);
574 	exception_exit(regs);
575 }
576 
577 dotraplinkage void
578 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
579 {
580 	conditional_sti(regs);
581 #if 0
582 	/* No need to warn about this any longer. */
583 	pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
584 #endif
585 }
586 
587 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
588 {
589 }
590 
591 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
592 {
593 }
594 
595 /*
596  * 'math_state_restore()' saves the current math information in the
597  * old math state array, and gets the new ones from the current task
598  *
599  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
600  * Don't touch unless you *really* know how it works.
601  *
602  * Must be called with kernel preemption disabled (eg with local
603  * local interrupts as in the case of do_device_not_available).
604  */
605 void math_state_restore(void)
606 {
607 	struct task_struct *tsk = current;
608 
609 	if (!tsk_used_math(tsk)) {
610 		local_irq_enable();
611 		/*
612 		 * does a slab alloc which can sleep
613 		 */
614 		if (init_fpu(tsk)) {
615 			/*
616 			 * ran out of memory!
617 			 */
618 			do_group_exit(SIGKILL);
619 			return;
620 		}
621 		local_irq_disable();
622 	}
623 
624 	__thread_fpu_begin(tsk);
625 
626 	/*
627 	 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
628 	 */
629 	if (unlikely(restore_fpu_checking(tsk))) {
630 		drop_init_fpu(tsk);
631 		force_sig(SIGSEGV, tsk);
632 		return;
633 	}
634 
635 	tsk->fpu_counter++;
636 }
637 EXPORT_SYMBOL_GPL(math_state_restore);
638 
639 dotraplinkage void __kprobes
640 do_device_not_available(struct pt_regs *regs, long error_code)
641 {
642 	exception_enter(regs);
643 	BUG_ON(use_eager_fpu());
644 
645 #ifdef CONFIG_MATH_EMULATION
646 	if (read_cr0() & X86_CR0_EM) {
647 		struct math_emu_info info = { };
648 
649 		conditional_sti(regs);
650 
651 		info.regs = regs;
652 		math_emulate(&info);
653 		exception_exit(regs);
654 		return;
655 	}
656 #endif
657 	math_state_restore(); /* interrupts still off */
658 #ifdef CONFIG_X86_32
659 	conditional_sti(regs);
660 #endif
661 	exception_exit(regs);
662 }
663 
664 #ifdef CONFIG_X86_32
665 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
666 {
667 	siginfo_t info;
668 
669 	exception_enter(regs);
670 	local_irq_enable();
671 
672 	info.si_signo = SIGILL;
673 	info.si_errno = 0;
674 	info.si_code = ILL_BADSTK;
675 	info.si_addr = NULL;
676 	if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
677 			X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
678 		do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
679 			&info);
680 	}
681 	exception_exit(regs);
682 }
683 #endif
684 
685 /* Set of traps needed for early debugging. */
686 void __init early_trap_init(void)
687 {
688 	set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
689 	/* int3 can be called from all */
690 	set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
691 	set_intr_gate(X86_TRAP_PF, &page_fault);
692 	load_idt(&idt_descr);
693 }
694 
695 void __init trap_init(void)
696 {
697 	int i;
698 
699 #ifdef CONFIG_EISA
700 	void __iomem *p = early_ioremap(0x0FFFD9, 4);
701 
702 	if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
703 		EISA_bus = 1;
704 	early_iounmap(p, 4);
705 #endif
706 
707 	set_intr_gate(X86_TRAP_DE, &divide_error);
708 	set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
709 	/* int4 can be called from all */
710 	set_system_intr_gate(X86_TRAP_OF, &overflow);
711 	set_intr_gate(X86_TRAP_BR, &bounds);
712 	set_intr_gate(X86_TRAP_UD, &invalid_op);
713 	set_intr_gate(X86_TRAP_NM, &device_not_available);
714 #ifdef CONFIG_X86_32
715 	set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
716 #else
717 	set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
718 #endif
719 	set_intr_gate(X86_TRAP_OLD_MF, &coprocessor_segment_overrun);
720 	set_intr_gate(X86_TRAP_TS, &invalid_TSS);
721 	set_intr_gate(X86_TRAP_NP, &segment_not_present);
722 	set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
723 	set_intr_gate(X86_TRAP_GP, &general_protection);
724 	set_intr_gate(X86_TRAP_SPURIOUS, &spurious_interrupt_bug);
725 	set_intr_gate(X86_TRAP_MF, &coprocessor_error);
726 	set_intr_gate(X86_TRAP_AC, &alignment_check);
727 #ifdef CONFIG_X86_MCE
728 	set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
729 #endif
730 	set_intr_gate(X86_TRAP_XF, &simd_coprocessor_error);
731 
732 	/* Reserve all the builtin and the syscall vector: */
733 	for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
734 		set_bit(i, used_vectors);
735 
736 #ifdef CONFIG_IA32_EMULATION
737 	set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
738 	set_bit(IA32_SYSCALL_VECTOR, used_vectors);
739 #endif
740 
741 #ifdef CONFIG_X86_32
742 	set_system_trap_gate(SYSCALL_VECTOR, &system_call);
743 	set_bit(SYSCALL_VECTOR, used_vectors);
744 #endif
745 
746 	/*
747 	 * Should be a barrier for any external CPU state:
748 	 */
749 	cpu_init();
750 
751 	x86_init.irqs.trap_init();
752 
753 #ifdef CONFIG_X86_64
754 	memcpy(&nmi_idt_table, &idt_table, IDT_ENTRIES * 16);
755 	set_nmi_gate(X86_TRAP_DB, &debug);
756 	set_nmi_gate(X86_TRAP_BP, &int3);
757 #endif
758 }
759