xref: /openbmc/linux/arch/x86/kernel/traps.c (revision 6c1e0256fad84a843d915414e4b5973b7443d48d)
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/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/module.h>
25 #include <linux/ptrace.h>
26 #include <linux/string.h>
27 #include <linux/delay.h>
28 #include <linux/errno.h>
29 #include <linux/kexec.h>
30 #include <linux/sched.h>
31 #include <linux/timer.h>
32 #include <linux/init.h>
33 #include <linux/bug.h>
34 #include <linux/nmi.h>
35 #include <linux/mm.h>
36 #include <linux/smp.h>
37 #include <linux/io.h>
38 
39 #ifdef CONFIG_EISA
40 #include <linux/ioport.h>
41 #include <linux/eisa.h>
42 #endif
43 
44 #if defined(CONFIG_EDAC)
45 #include <linux/edac.h>
46 #endif
47 
48 #include <asm/kmemcheck.h>
49 #include <asm/stacktrace.h>
50 #include <asm/processor.h>
51 #include <asm/debugreg.h>
52 #include <linux/atomic.h>
53 #include <asm/ftrace.h>
54 #include <asm/traps.h>
55 #include <asm/desc.h>
56 #include <asm/i387.h>
57 #include <asm/fpu-internal.h>
58 #include <asm/mce.h>
59 #include <asm/mach_traps.h>
60 
61 #ifdef CONFIG_X86_64
62 #include <asm/x86_init.h>
63 #include <asm/pgalloc.h>
64 #include <asm/proto.h>
65 #else
66 #include <asm/processor-flags.h>
67 #include <asm/setup.h>
68 
69 asmlinkage int system_call(void);
70 
71 /*
72  * The IDT has to be page-aligned to simplify the Pentium
73  * F0 0F bug workaround.
74  */
75 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
76 #endif
77 
78 DECLARE_BITMAP(used_vectors, NR_VECTORS);
79 EXPORT_SYMBOL_GPL(used_vectors);
80 
81 static inline void conditional_sti(struct pt_regs *regs)
82 {
83 	if (regs->flags & X86_EFLAGS_IF)
84 		local_irq_enable();
85 }
86 
87 static inline void preempt_conditional_sti(struct pt_regs *regs)
88 {
89 	inc_preempt_count();
90 	if (regs->flags & X86_EFLAGS_IF)
91 		local_irq_enable();
92 }
93 
94 static inline void conditional_cli(struct pt_regs *regs)
95 {
96 	if (regs->flags & X86_EFLAGS_IF)
97 		local_irq_disable();
98 }
99 
100 static inline void preempt_conditional_cli(struct pt_regs *regs)
101 {
102 	if (regs->flags & X86_EFLAGS_IF)
103 		local_irq_disable();
104 	dec_preempt_count();
105 }
106 
107 static int __kprobes
108 do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
109 		  struct pt_regs *regs,	long error_code)
110 {
111 #ifdef CONFIG_X86_32
112 	if (regs->flags & X86_VM_MASK) {
113 		/*
114 		 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
115 		 * On nmi (interrupt 2), do_trap should not be called.
116 		 */
117 		if (trapnr < X86_TRAP_UD) {
118 			if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
119 						error_code, trapnr))
120 				return 0;
121 		}
122 		return -1;
123 	}
124 #endif
125 	if (!user_mode(regs)) {
126 		if (!fixup_exception(regs)) {
127 			tsk->thread.error_code = error_code;
128 			tsk->thread.trap_nr = trapnr;
129 			die(str, regs, error_code);
130 		}
131 		return 0;
132 	}
133 
134 	return -1;
135 }
136 
137 static void __kprobes
138 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
139 	long error_code, siginfo_t *info)
140 {
141 	struct task_struct *tsk = current;
142 
143 
144 	if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
145 		return;
146 	/*
147 	 * We want error_code and trap_nr set for userspace faults and
148 	 * kernelspace faults which result in die(), but not
149 	 * kernelspace faults which are fixed up.  die() gives the
150 	 * process no chance to handle the signal and notice the
151 	 * kernel fault information, so that won't result in polluting
152 	 * the information about previously queued, but not yet
153 	 * delivered, faults.  See also do_general_protection below.
154 	 */
155 	tsk->thread.error_code = error_code;
156 	tsk->thread.trap_nr = trapnr;
157 
158 #ifdef CONFIG_X86_64
159 	if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
160 	    printk_ratelimit()) {
161 		pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
162 			tsk->comm, tsk->pid, str,
163 			regs->ip, regs->sp, error_code);
164 		print_vma_addr(" in ", regs->ip);
165 		pr_cont("\n");
166 	}
167 #endif
168 
169 	if (info)
170 		force_sig_info(signr, info, tsk);
171 	else
172 		force_sig(signr, tsk);
173 }
174 
175 #define DO_ERROR(trapnr, signr, str, name)				\
176 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
177 {									\
178 	enum ctx_state prev_state;					\
179 									\
180 	prev_state = exception_enter();					\
181 	if (notify_die(DIE_TRAP, str, regs, error_code,			\
182 			trapnr, signr) == NOTIFY_STOP) {		\
183 		exception_exit(prev_state);				\
184 		return;							\
185 	}								\
186 	conditional_sti(regs);						\
187 	do_trap(trapnr, signr, str, regs, error_code, NULL);		\
188 	exception_exit(prev_state);					\
189 }
190 
191 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)		\
192 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
193 {									\
194 	siginfo_t info;							\
195 	enum ctx_state prev_state;					\
196 									\
197 	info.si_signo = signr;						\
198 	info.si_errno = 0;						\
199 	info.si_code = sicode;						\
200 	info.si_addr = (void __user *)siaddr;				\
201 	prev_state = exception_enter();					\
202 	if (notify_die(DIE_TRAP, str, regs, error_code,			\
203 			trapnr, signr) == NOTIFY_STOP) {		\
204 		exception_exit(prev_state);				\
205 		return;							\
206 	}								\
207 	conditional_sti(regs);						\
208 	do_trap(trapnr, signr, str, regs, error_code, &info);		\
209 	exception_exit(prev_state);					\
210 }
211 
212 DO_ERROR_INFO(X86_TRAP_DE, SIGFPE, "divide error", divide_error, FPE_INTDIV,
213 		regs->ip)
214 DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
215 DO_ERROR(X86_TRAP_BR, SIGSEGV, "bounds", bounds)
216 DO_ERROR_INFO(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN,
217 		regs->ip)
218 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",
219 		coprocessor_segment_overrun)
220 DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
221 DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
222 #ifdef CONFIG_X86_32
223 DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
224 #endif
225 DO_ERROR_INFO(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check,
226 		BUS_ADRALN, 0)
227 
228 #ifdef CONFIG_X86_64
229 /* Runs on IST stack */
230 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
231 {
232 	enum ctx_state prev_state;
233 
234 	prev_state = exception_enter();
235 	if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
236 		       X86_TRAP_SS, SIGBUS) != NOTIFY_STOP) {
237 		preempt_conditional_sti(regs);
238 		do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
239 		preempt_conditional_cli(regs);
240 	}
241 	exception_exit(prev_state);
242 }
243 
244 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
245 {
246 	static const char str[] = "double fault";
247 	struct task_struct *tsk = current;
248 
249 	exception_enter();
250 	/* Return not checked because double check cannot be ignored */
251 	notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
252 
253 	tsk->thread.error_code = error_code;
254 	tsk->thread.trap_nr = X86_TRAP_DF;
255 
256 	/*
257 	 * This is always a kernel trap and never fixable (and thus must
258 	 * never return).
259 	 */
260 	for (;;)
261 		die(str, regs, error_code);
262 }
263 #endif
264 
265 dotraplinkage void __kprobes
266 do_general_protection(struct pt_regs *regs, long error_code)
267 {
268 	struct task_struct *tsk;
269 	enum ctx_state prev_state;
270 
271 	prev_state = exception_enter();
272 	conditional_sti(regs);
273 
274 #ifdef CONFIG_X86_32
275 	if (regs->flags & X86_VM_MASK) {
276 		local_irq_enable();
277 		handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
278 		goto exit;
279 	}
280 #endif
281 
282 	tsk = current;
283 	if (!user_mode(regs)) {
284 		if (fixup_exception(regs))
285 			goto exit;
286 
287 		tsk->thread.error_code = error_code;
288 		tsk->thread.trap_nr = X86_TRAP_GP;
289 		if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
290 			       X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
291 			die("general protection fault", regs, error_code);
292 		goto exit;
293 	}
294 
295 	tsk->thread.error_code = error_code;
296 	tsk->thread.trap_nr = X86_TRAP_GP;
297 
298 	if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
299 			printk_ratelimit()) {
300 		pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
301 			tsk->comm, task_pid_nr(tsk),
302 			regs->ip, regs->sp, error_code);
303 		print_vma_addr(" in ", regs->ip);
304 		pr_cont("\n");
305 	}
306 
307 	force_sig(SIGSEGV, tsk);
308 exit:
309 	exception_exit(prev_state);
310 }
311 
312 /* May run on IST stack. */
313 dotraplinkage void __kprobes notrace do_int3(struct pt_regs *regs, long error_code)
314 {
315 	enum ctx_state prev_state;
316 
317 #ifdef CONFIG_DYNAMIC_FTRACE
318 	/*
319 	 * ftrace must be first, everything else may cause a recursive crash.
320 	 * See note by declaration of modifying_ftrace_code in ftrace.c
321 	 */
322 	if (unlikely(atomic_read(&modifying_ftrace_code)) &&
323 	    ftrace_int3_handler(regs))
324 		return;
325 #endif
326 	prev_state = exception_enter();
327 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
328 	if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
329 				SIGTRAP) == NOTIFY_STOP)
330 		goto exit;
331 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
332 
333 	if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
334 			SIGTRAP) == NOTIFY_STOP)
335 		goto exit;
336 
337 	/*
338 	 * Let others (NMI) know that the debug stack is in use
339 	 * as we may switch to the interrupt stack.
340 	 */
341 	debug_stack_usage_inc();
342 	preempt_conditional_sti(regs);
343 	do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
344 	preempt_conditional_cli(regs);
345 	debug_stack_usage_dec();
346 exit:
347 	exception_exit(prev_state);
348 }
349 
350 #ifdef CONFIG_X86_64
351 /*
352  * Help handler running on IST stack to switch back to user stack
353  * for scheduling or signal handling. The actual stack switch is done in
354  * entry.S
355  */
356 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
357 {
358 	struct pt_regs *regs = eregs;
359 	/* Did already sync */
360 	if (eregs == (struct pt_regs *)eregs->sp)
361 		;
362 	/* Exception from user space */
363 	else if (user_mode(eregs))
364 		regs = task_pt_regs(current);
365 	/*
366 	 * Exception from kernel and interrupts are enabled. Move to
367 	 * kernel process stack.
368 	 */
369 	else if (eregs->flags & X86_EFLAGS_IF)
370 		regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
371 	if (eregs != regs)
372 		*regs = *eregs;
373 	return regs;
374 }
375 #endif
376 
377 /*
378  * Our handling of the processor debug registers is non-trivial.
379  * We do not clear them on entry and exit from the kernel. Therefore
380  * it is possible to get a watchpoint trap here from inside the kernel.
381  * However, the code in ./ptrace.c has ensured that the user can
382  * only set watchpoints on userspace addresses. Therefore the in-kernel
383  * watchpoint trap can only occur in code which is reading/writing
384  * from user space. Such code must not hold kernel locks (since it
385  * can equally take a page fault), therefore it is safe to call
386  * force_sig_info even though that claims and releases locks.
387  *
388  * Code in ./signal.c ensures that the debug control register
389  * is restored before we deliver any signal, and therefore that
390  * user code runs with the correct debug control register even though
391  * we clear it here.
392  *
393  * Being careful here means that we don't have to be as careful in a
394  * lot of more complicated places (task switching can be a bit lazy
395  * about restoring all the debug state, and ptrace doesn't have to
396  * find every occurrence of the TF bit that could be saved away even
397  * by user code)
398  *
399  * May run on IST stack.
400  */
401 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
402 {
403 	struct task_struct *tsk = current;
404 	enum ctx_state prev_state;
405 	int user_icebp = 0;
406 	unsigned long dr6;
407 	int si_code;
408 
409 	prev_state = exception_enter();
410 
411 	get_debugreg(dr6, 6);
412 
413 	/* Filter out all the reserved bits which are preset to 1 */
414 	dr6 &= ~DR6_RESERVED;
415 
416 	/*
417 	 * If dr6 has no reason to give us about the origin of this trap,
418 	 * then it's very likely the result of an icebp/int01 trap.
419 	 * User wants a sigtrap for that.
420 	 */
421 	if (!dr6 && user_mode(regs))
422 		user_icebp = 1;
423 
424 	/* Catch kmemcheck conditions first of all! */
425 	if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
426 		goto exit;
427 
428 	/* DR6 may or may not be cleared by the CPU */
429 	set_debugreg(0, 6);
430 
431 	/*
432 	 * The processor cleared BTF, so don't mark that we need it set.
433 	 */
434 	clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
435 
436 	/* Store the virtualized DR6 value */
437 	tsk->thread.debugreg6 = dr6;
438 
439 	if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
440 							SIGTRAP) == NOTIFY_STOP)
441 		goto exit;
442 
443 	/*
444 	 * Let others (NMI) know that the debug stack is in use
445 	 * as we may switch to the interrupt stack.
446 	 */
447 	debug_stack_usage_inc();
448 
449 	/* It's safe to allow irq's after DR6 has been saved */
450 	preempt_conditional_sti(regs);
451 
452 	if (regs->flags & X86_VM_MASK) {
453 		handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
454 					X86_TRAP_DB);
455 		preempt_conditional_cli(regs);
456 		debug_stack_usage_dec();
457 		goto exit;
458 	}
459 
460 	/*
461 	 * Single-stepping through system calls: ignore any exceptions in
462 	 * kernel space, but re-enable TF when returning to user mode.
463 	 *
464 	 * We already checked v86 mode above, so we can check for kernel mode
465 	 * by just checking the CPL of CS.
466 	 */
467 	if ((dr6 & DR_STEP) && !user_mode(regs)) {
468 		tsk->thread.debugreg6 &= ~DR_STEP;
469 		set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
470 		regs->flags &= ~X86_EFLAGS_TF;
471 	}
472 	si_code = get_si_code(tsk->thread.debugreg6);
473 	if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
474 		send_sigtrap(tsk, regs, error_code, si_code);
475 	preempt_conditional_cli(regs);
476 	debug_stack_usage_dec();
477 
478 exit:
479 	exception_exit(prev_state);
480 }
481 
482 /*
483  * Note that we play around with the 'TS' bit in an attempt to get
484  * the correct behaviour even in the presence of the asynchronous
485  * IRQ13 behaviour
486  */
487 void math_error(struct pt_regs *regs, int error_code, int trapnr)
488 {
489 	struct task_struct *task = current;
490 	siginfo_t info;
491 	unsigned short err;
492 	char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
493 						"simd exception";
494 
495 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
496 		return;
497 	conditional_sti(regs);
498 
499 	if (!user_mode_vm(regs))
500 	{
501 		if (!fixup_exception(regs)) {
502 			task->thread.error_code = error_code;
503 			task->thread.trap_nr = trapnr;
504 			die(str, regs, error_code);
505 		}
506 		return;
507 	}
508 
509 	/*
510 	 * Save the info for the exception handler and clear the error.
511 	 */
512 	save_init_fpu(task);
513 	task->thread.trap_nr = trapnr;
514 	task->thread.error_code = error_code;
515 	info.si_signo = SIGFPE;
516 	info.si_errno = 0;
517 	info.si_addr = (void __user *)regs->ip;
518 	if (trapnr == X86_TRAP_MF) {
519 		unsigned short cwd, swd;
520 		/*
521 		 * (~cwd & swd) will mask out exceptions that are not set to unmasked
522 		 * status.  0x3f is the exception bits in these regs, 0x200 is the
523 		 * C1 reg you need in case of a stack fault, 0x040 is the stack
524 		 * fault bit.  We should only be taking one exception at a time,
525 		 * so if this combination doesn't produce any single exception,
526 		 * then we have a bad program that isn't synchronizing its FPU usage
527 		 * and it will suffer the consequences since we won't be able to
528 		 * fully reproduce the context of the exception
529 		 */
530 		cwd = get_fpu_cwd(task);
531 		swd = get_fpu_swd(task);
532 
533 		err = swd & ~cwd;
534 	} else {
535 		/*
536 		 * The SIMD FPU exceptions are handled a little differently, as there
537 		 * is only a single status/control register.  Thus, to determine which
538 		 * unmasked exception was caught we must mask the exception mask bits
539 		 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
540 		 */
541 		unsigned short mxcsr = get_fpu_mxcsr(task);
542 		err = ~(mxcsr >> 7) & mxcsr;
543 	}
544 
545 	if (err & 0x001) {	/* Invalid op */
546 		/*
547 		 * swd & 0x240 == 0x040: Stack Underflow
548 		 * swd & 0x240 == 0x240: Stack Overflow
549 		 * User must clear the SF bit (0x40) if set
550 		 */
551 		info.si_code = FPE_FLTINV;
552 	} else if (err & 0x004) { /* Divide by Zero */
553 		info.si_code = FPE_FLTDIV;
554 	} else if (err & 0x008) { /* Overflow */
555 		info.si_code = FPE_FLTOVF;
556 	} else if (err & 0x012) { /* Denormal, Underflow */
557 		info.si_code = FPE_FLTUND;
558 	} else if (err & 0x020) { /* Precision */
559 		info.si_code = FPE_FLTRES;
560 	} else {
561 		/*
562 		 * If we're using IRQ 13, or supposedly even some trap
563 		 * X86_TRAP_MF implementations, it's possible
564 		 * we get a spurious trap, which is not an error.
565 		 */
566 		return;
567 	}
568 	force_sig_info(SIGFPE, &info, task);
569 }
570 
571 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
572 {
573 	enum ctx_state prev_state;
574 
575 	prev_state = exception_enter();
576 	math_error(regs, error_code, X86_TRAP_MF);
577 	exception_exit(prev_state);
578 }
579 
580 dotraplinkage void
581 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
582 {
583 	enum ctx_state prev_state;
584 
585 	prev_state = exception_enter();
586 	math_error(regs, error_code, X86_TRAP_XF);
587 	exception_exit(prev_state);
588 }
589 
590 dotraplinkage void
591 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
592 {
593 	conditional_sti(regs);
594 #if 0
595 	/* No need to warn about this any longer. */
596 	pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
597 #endif
598 }
599 
600 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
601 {
602 }
603 
604 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
605 {
606 }
607 
608 /*
609  * 'math_state_restore()' saves the current math information in the
610  * old math state array, and gets the new ones from the current task
611  *
612  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
613  * Don't touch unless you *really* know how it works.
614  *
615  * Must be called with kernel preemption disabled (eg with local
616  * local interrupts as in the case of do_device_not_available).
617  */
618 void math_state_restore(void)
619 {
620 	struct task_struct *tsk = current;
621 
622 	if (!tsk_used_math(tsk)) {
623 		local_irq_enable();
624 		/*
625 		 * does a slab alloc which can sleep
626 		 */
627 		if (init_fpu(tsk)) {
628 			/*
629 			 * ran out of memory!
630 			 */
631 			do_group_exit(SIGKILL);
632 			return;
633 		}
634 		local_irq_disable();
635 	}
636 
637 	__thread_fpu_begin(tsk);
638 
639 	/*
640 	 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
641 	 */
642 	if (unlikely(restore_fpu_checking(tsk))) {
643 		drop_init_fpu(tsk);
644 		force_sig(SIGSEGV, tsk);
645 		return;
646 	}
647 
648 	tsk->fpu_counter++;
649 }
650 EXPORT_SYMBOL_GPL(math_state_restore);
651 
652 dotraplinkage void __kprobes
653 do_device_not_available(struct pt_regs *regs, long error_code)
654 {
655 	enum ctx_state prev_state;
656 
657 	prev_state = exception_enter();
658 	BUG_ON(use_eager_fpu());
659 
660 #ifdef CONFIG_MATH_EMULATION
661 	if (read_cr0() & X86_CR0_EM) {
662 		struct math_emu_info info = { };
663 
664 		conditional_sti(regs);
665 
666 		info.regs = regs;
667 		math_emulate(&info);
668 		exception_exit(prev_state);
669 		return;
670 	}
671 #endif
672 	math_state_restore(); /* interrupts still off */
673 #ifdef CONFIG_X86_32
674 	conditional_sti(regs);
675 #endif
676 	exception_exit(prev_state);
677 }
678 
679 #ifdef CONFIG_X86_32
680 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
681 {
682 	siginfo_t info;
683 	enum ctx_state prev_state;
684 
685 	prev_state = exception_enter();
686 	local_irq_enable();
687 
688 	info.si_signo = SIGILL;
689 	info.si_errno = 0;
690 	info.si_code = ILL_BADSTK;
691 	info.si_addr = NULL;
692 	if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
693 			X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
694 		do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
695 			&info);
696 	}
697 	exception_exit(prev_state);
698 }
699 #endif
700 
701 /* Set of traps needed for early debugging. */
702 void __init early_trap_init(void)
703 {
704 	set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
705 	/* int3 can be called from all */
706 	set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
707 #ifdef CONFIG_X86_32
708 	set_intr_gate(X86_TRAP_PF, &page_fault);
709 #endif
710 	load_idt(&idt_descr);
711 }
712 
713 void __init early_trap_pf_init(void)
714 {
715 #ifdef CONFIG_X86_64
716 	set_intr_gate(X86_TRAP_PF, &page_fault);
717 #endif
718 }
719 
720 void __init trap_init(void)
721 {
722 	int i;
723 
724 #ifdef CONFIG_EISA
725 	void __iomem *p = early_ioremap(0x0FFFD9, 4);
726 
727 	if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
728 		EISA_bus = 1;
729 	early_iounmap(p, 4);
730 #endif
731 
732 	set_intr_gate(X86_TRAP_DE, &divide_error);
733 	set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
734 	/* int4 can be called from all */
735 	set_system_intr_gate(X86_TRAP_OF, &overflow);
736 	set_intr_gate(X86_TRAP_BR, &bounds);
737 	set_intr_gate(X86_TRAP_UD, &invalid_op);
738 	set_intr_gate(X86_TRAP_NM, &device_not_available);
739 #ifdef CONFIG_X86_32
740 	set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
741 #else
742 	set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
743 #endif
744 	set_intr_gate(X86_TRAP_OLD_MF, &coprocessor_segment_overrun);
745 	set_intr_gate(X86_TRAP_TS, &invalid_TSS);
746 	set_intr_gate(X86_TRAP_NP, &segment_not_present);
747 	set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
748 	set_intr_gate(X86_TRAP_GP, &general_protection);
749 	set_intr_gate(X86_TRAP_SPURIOUS, &spurious_interrupt_bug);
750 	set_intr_gate(X86_TRAP_MF, &coprocessor_error);
751 	set_intr_gate(X86_TRAP_AC, &alignment_check);
752 #ifdef CONFIG_X86_MCE
753 	set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
754 #endif
755 	set_intr_gate(X86_TRAP_XF, &simd_coprocessor_error);
756 
757 	/* Reserve all the builtin and the syscall vector: */
758 	for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
759 		set_bit(i, used_vectors);
760 
761 #ifdef CONFIG_IA32_EMULATION
762 	set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
763 	set_bit(IA32_SYSCALL_VECTOR, used_vectors);
764 #endif
765 
766 #ifdef CONFIG_X86_32
767 	set_system_trap_gate(SYSCALL_VECTOR, &system_call);
768 	set_bit(SYSCALL_VECTOR, used_vectors);
769 #endif
770 
771 	/*
772 	 * Should be a barrier for any external CPU state:
773 	 */
774 	cpu_init();
775 
776 	x86_init.irqs.trap_init();
777 
778 #ifdef CONFIG_X86_64
779 	memcpy(&nmi_idt_table, &idt_table, IDT_ENTRIES * 16);
780 	set_nmi_gate(X86_TRAP_DB, &debug);
781 	set_nmi_gate(X86_TRAP_BP, &int3);
782 #endif
783 }
784