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 #include <linux/interrupt.h>
13 #include <linux/kallsyms.h>
14 #include <linux/spinlock.h>
15 #include <linux/kprobes.h>
16 #include <linux/uaccess.h>
17 #include <linux/kdebug.h>
18 #include <linux/kgdb.h>
19 #include <linux/kernel.h>
20 #include <linux/module.h>
21 #include <linux/ptrace.h>
22 #include <linux/string.h>
23 #include <linux/delay.h>
24 #include <linux/errno.h>
25 #include <linux/kexec.h>
26 #include <linux/sched.h>
27 #include <linux/timer.h>
28 #include <linux/init.h>
29 #include <linux/bug.h>
30 #include <linux/nmi.h>
31 #include <linux/mm.h>
32 #include <linux/smp.h>
33 #include <linux/io.h>
34 
35 #ifdef CONFIG_EISA
36 #include <linux/ioport.h>
37 #include <linux/eisa.h>
38 #endif
39 
40 #ifdef CONFIG_MCA
41 #include <linux/mca.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/traps.h>
54 #include <asm/desc.h>
55 #include <asm/i387.h>
56 #include <asm/fpu-internal.h>
57 #include <asm/mce.h>
58 
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 /* Do we ignore FPU interrupts ? */
72 char ignore_fpu_irq;
73 
74 /*
75  * The IDT has to be page-aligned to simplify the Pentium
76  * F0 0F bug workaround.
77  */
78 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
79 #endif
80 
81 DECLARE_BITMAP(used_vectors, NR_VECTORS);
82 EXPORT_SYMBOL_GPL(used_vectors);
83 
conditional_sti(struct pt_regs * regs)84 static inline void conditional_sti(struct pt_regs *regs)
85 {
86 	if (regs->flags & X86_EFLAGS_IF)
87 		local_irq_enable();
88 }
89 
preempt_conditional_sti(struct pt_regs * regs)90 static inline void preempt_conditional_sti(struct pt_regs *regs)
91 {
92 	inc_preempt_count();
93 	if (regs->flags & X86_EFLAGS_IF)
94 		local_irq_enable();
95 }
96 
conditional_cli(struct pt_regs * regs)97 static inline void conditional_cli(struct pt_regs *regs)
98 {
99 	if (regs->flags & X86_EFLAGS_IF)
100 		local_irq_disable();
101 }
102 
preempt_conditional_cli(struct pt_regs * regs)103 static inline void preempt_conditional_cli(struct pt_regs *regs)
104 {
105 	if (regs->flags & X86_EFLAGS_IF)
106 		local_irq_disable();
107 	dec_preempt_count();
108 }
109 
110 static void __kprobes
do_trap(int trapnr,int signr,char * str,struct pt_regs * regs,long error_code,siginfo_t * info)111 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
112 	long error_code, siginfo_t *info)
113 {
114 	struct task_struct *tsk = current;
115 
116 #ifdef CONFIG_X86_32
117 	if (regs->flags & X86_VM_MASK) {
118 		/*
119 		 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
120 		 * On nmi (interrupt 2), do_trap should not be called.
121 		 */
122 		if (trapnr < X86_TRAP_UD)
123 			goto vm86_trap;
124 		goto trap_signal;
125 	}
126 #endif
127 
128 	if (!user_mode(regs))
129 		goto kernel_trap;
130 
131 #ifdef CONFIG_X86_32
132 trap_signal:
133 #endif
134 	/*
135 	 * We want error_code and trap_nr set for userspace faults and
136 	 * kernelspace faults which result in die(), but not
137 	 * kernelspace faults which are fixed up.  die() gives the
138 	 * process no chance to handle the signal and notice the
139 	 * kernel fault information, so that won't result in polluting
140 	 * the information about previously queued, but not yet
141 	 * delivered, faults.  See also do_general_protection below.
142 	 */
143 	tsk->thread.error_code = error_code;
144 	tsk->thread.trap_nr = trapnr;
145 
146 #ifdef CONFIG_X86_64
147 	if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
148 	    printk_ratelimit()) {
149 		printk(KERN_INFO
150 		       "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
151 		       tsk->comm, tsk->pid, str,
152 		       regs->ip, regs->sp, error_code);
153 		print_vma_addr(" in ", regs->ip);
154 		printk("\n");
155 	}
156 #endif
157 
158 	if (info)
159 		force_sig_info(signr, info, tsk);
160 	else
161 		force_sig(signr, tsk);
162 	return;
163 
164 kernel_trap:
165 	if (!fixup_exception(regs)) {
166 		tsk->thread.error_code = error_code;
167 		tsk->thread.trap_nr = trapnr;
168 		die(str, regs, error_code);
169 	}
170 	return;
171 
172 #ifdef CONFIG_X86_32
173 vm86_trap:
174 	if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
175 						error_code, trapnr))
176 		goto trap_signal;
177 	return;
178 #endif
179 }
180 
181 #define DO_ERROR(trapnr, signr, str, name)				\
182 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
183 {									\
184 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)	\
185 							== NOTIFY_STOP)	\
186 		return;							\
187 	conditional_sti(regs);						\
188 	do_trap(trapnr, signr, str, regs, error_code, NULL);		\
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 	info.si_signo = signr;						\
196 	info.si_errno = 0;						\
197 	info.si_code = sicode;						\
198 	info.si_addr = (void __user *)siaddr;				\
199 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)	\
200 							== NOTIFY_STOP)	\
201 		return;							\
202 	conditional_sti(regs);						\
203 	do_trap(trapnr, signr, str, regs, error_code, &info);		\
204 }
205 
206 DO_ERROR_INFO(X86_TRAP_DE, SIGFPE, "divide error", divide_error, FPE_INTDIV,
207 		regs->ip)
208 DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
209 DO_ERROR(X86_TRAP_BR, SIGSEGV, "bounds", bounds)
210 DO_ERROR_INFO(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN,
211 		regs->ip)
212 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",
213 		coprocessor_segment_overrun)
214 DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
215 DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
216 #ifdef CONFIG_X86_32
217 DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
218 #endif
219 DO_ERROR_INFO(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check,
220 		BUS_ADRALN, 0)
221 
222 #ifdef CONFIG_X86_64
223 /* Runs on IST stack */
do_stack_segment(struct pt_regs * regs,long error_code)224 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
225 {
226 	if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
227 			X86_TRAP_SS, SIGBUS) == NOTIFY_STOP)
228 		return;
229 	preempt_conditional_sti(regs);
230 	do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
231 	preempt_conditional_cli(regs);
232 }
233 
do_double_fault(struct pt_regs * regs,long error_code)234 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
235 {
236 	static const char str[] = "double fault";
237 	struct task_struct *tsk = current;
238 
239 	/* Return not checked because double check cannot be ignored */
240 	notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
241 
242 	tsk->thread.error_code = error_code;
243 	tsk->thread.trap_nr = X86_TRAP_DF;
244 
245 	/*
246 	 * This is always a kernel trap and never fixable (and thus must
247 	 * never return).
248 	 */
249 	for (;;)
250 		die(str, regs, error_code);
251 }
252 #endif
253 
254 dotraplinkage void __kprobes
do_general_protection(struct pt_regs * regs,long error_code)255 do_general_protection(struct pt_regs *regs, long error_code)
256 {
257 	struct task_struct *tsk;
258 
259 	conditional_sti(regs);
260 
261 #ifdef CONFIG_X86_32
262 	if (regs->flags & X86_VM_MASK)
263 		goto gp_in_vm86;
264 #endif
265 
266 	tsk = current;
267 	if (!user_mode(regs))
268 		goto gp_in_kernel;
269 
270 	tsk->thread.error_code = error_code;
271 	tsk->thread.trap_nr = X86_TRAP_GP;
272 
273 	if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
274 			printk_ratelimit()) {
275 		printk(KERN_INFO
276 			"%s[%d] general protection ip:%lx sp:%lx error:%lx",
277 			tsk->comm, task_pid_nr(tsk),
278 			regs->ip, regs->sp, error_code);
279 		print_vma_addr(" in ", regs->ip);
280 		printk("\n");
281 	}
282 
283 	force_sig(SIGSEGV, tsk);
284 	return;
285 
286 #ifdef CONFIG_X86_32
287 gp_in_vm86:
288 	local_irq_enable();
289 	handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
290 	return;
291 #endif
292 
293 gp_in_kernel:
294 	if (fixup_exception(regs))
295 		return;
296 
297 	tsk->thread.error_code = error_code;
298 	tsk->thread.trap_nr = X86_TRAP_GP;
299 	if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
300 			X86_TRAP_GP, SIGSEGV) == NOTIFY_STOP)
301 		return;
302 	die("general protection fault", regs, error_code);
303 }
304 
305 /* May run on IST stack. */
do_int3(struct pt_regs * regs,long error_code)306 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
307 {
308 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
309 	if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
310 				SIGTRAP) == NOTIFY_STOP)
311 		return;
312 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
313 
314 	if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
315 			SIGTRAP) == NOTIFY_STOP)
316 		return;
317 
318 	/*
319 	 * Let others (NMI) know that the debug stack is in use
320 	 * as we may switch to the interrupt stack.
321 	 */
322 	debug_stack_usage_inc();
323 	preempt_conditional_sti(regs);
324 	do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
325 	preempt_conditional_cli(regs);
326 	debug_stack_usage_dec();
327 }
328 
329 #ifdef CONFIG_X86_64
330 /*
331  * Help handler running on IST stack to switch back to user stack
332  * for scheduling or signal handling. The actual stack switch is done in
333  * entry.S
334  */
sync_regs(struct pt_regs * eregs)335 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
336 {
337 	struct pt_regs *regs = eregs;
338 	/* Did already sync */
339 	if (eregs == (struct pt_regs *)eregs->sp)
340 		;
341 	/* Exception from user space */
342 	else if (user_mode(eregs))
343 		regs = task_pt_regs(current);
344 	/*
345 	 * Exception from kernel and interrupts are enabled. Move to
346 	 * kernel process stack.
347 	 */
348 	else if (eregs->flags & X86_EFLAGS_IF)
349 		regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
350 	if (eregs != regs)
351 		*regs = *eregs;
352 	return regs;
353 }
354 #endif
355 
356 /*
357  * Our handling of the processor debug registers is non-trivial.
358  * We do not clear them on entry and exit from the kernel. Therefore
359  * it is possible to get a watchpoint trap here from inside the kernel.
360  * However, the code in ./ptrace.c has ensured that the user can
361  * only set watchpoints on userspace addresses. Therefore the in-kernel
362  * watchpoint trap can only occur in code which is reading/writing
363  * from user space. Such code must not hold kernel locks (since it
364  * can equally take a page fault), therefore it is safe to call
365  * force_sig_info even though that claims and releases locks.
366  *
367  * Code in ./signal.c ensures that the debug control register
368  * is restored before we deliver any signal, and therefore that
369  * user code runs with the correct debug control register even though
370  * we clear it here.
371  *
372  * Being careful here means that we don't have to be as careful in a
373  * lot of more complicated places (task switching can be a bit lazy
374  * about restoring all the debug state, and ptrace doesn't have to
375  * find every occurrence of the TF bit that could be saved away even
376  * by user code)
377  *
378  * May run on IST stack.
379  */
do_debug(struct pt_regs * regs,long error_code)380 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
381 {
382 	struct task_struct *tsk = current;
383 	int user_icebp = 0;
384 	unsigned long dr6;
385 	int si_code;
386 
387 	get_debugreg(dr6, 6);
388 
389 	/* Filter out all the reserved bits which are preset to 1 */
390 	dr6 &= ~DR6_RESERVED;
391 
392 	/*
393 	 * If dr6 has no reason to give us about the origin of this trap,
394 	 * then it's very likely the result of an icebp/int01 trap.
395 	 * User wants a sigtrap for that.
396 	 */
397 	if (!dr6 && user_mode(regs))
398 		user_icebp = 1;
399 
400 	/* Catch kmemcheck conditions first of all! */
401 	if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
402 		return;
403 
404 	/* DR6 may or may not be cleared by the CPU */
405 	set_debugreg(0, 6);
406 
407 	/*
408 	 * The processor cleared BTF, so don't mark that we need it set.
409 	 */
410 	clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
411 
412 	/* Store the virtualized DR6 value */
413 	tsk->thread.debugreg6 = dr6;
414 
415 	if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
416 							SIGTRAP) == NOTIFY_STOP)
417 		return;
418 
419 	/*
420 	 * Let others (NMI) know that the debug stack is in use
421 	 * as we may switch to the interrupt stack.
422 	 */
423 	debug_stack_usage_inc();
424 
425 	/* It's safe to allow irq's after DR6 has been saved */
426 	preempt_conditional_sti(regs);
427 
428 	if (regs->flags & X86_VM_MASK) {
429 		handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
430 					X86_TRAP_DB);
431 		preempt_conditional_cli(regs);
432 		debug_stack_usage_dec();
433 		return;
434 	}
435 
436 	/*
437 	 * Single-stepping through system calls: ignore any exceptions in
438 	 * kernel space, but re-enable TF when returning to user mode.
439 	 *
440 	 * We already checked v86 mode above, so we can check for kernel mode
441 	 * by just checking the CPL of CS.
442 	 */
443 	if ((dr6 & DR_STEP) && !user_mode(regs)) {
444 		tsk->thread.debugreg6 &= ~DR_STEP;
445 		set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
446 		regs->flags &= ~X86_EFLAGS_TF;
447 	}
448 	si_code = get_si_code(tsk->thread.debugreg6);
449 	if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
450 		send_sigtrap(tsk, regs, error_code, si_code);
451 	preempt_conditional_cli(regs);
452 	debug_stack_usage_dec();
453 
454 	return;
455 }
456 
457 /*
458  * Note that we play around with the 'TS' bit in an attempt to get
459  * the correct behaviour even in the presence of the asynchronous
460  * IRQ13 behaviour
461  */
math_error(struct pt_regs * regs,int error_code,int trapnr)462 void math_error(struct pt_regs *regs, int error_code, int trapnr)
463 {
464 	struct task_struct *task = current;
465 	siginfo_t info;
466 	unsigned short err;
467 	char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
468 						"simd exception";
469 
470 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
471 		return;
472 	conditional_sti(regs);
473 
474 	if (!user_mode_vm(regs))
475 	{
476 		if (!fixup_exception(regs)) {
477 			task->thread.error_code = error_code;
478 			task->thread.trap_nr = trapnr;
479 			die(str, regs, error_code);
480 		}
481 		return;
482 	}
483 
484 	/*
485 	 * Save the info for the exception handler and clear the error.
486 	 */
487 	save_init_fpu(task);
488 	task->thread.trap_nr = trapnr;
489 	task->thread.error_code = error_code;
490 	info.si_signo = SIGFPE;
491 	info.si_errno = 0;
492 	info.si_addr = (void __user *)regs->ip;
493 	if (trapnr == X86_TRAP_MF) {
494 		unsigned short cwd, swd;
495 		/*
496 		 * (~cwd & swd) will mask out exceptions that are not set to unmasked
497 		 * status.  0x3f is the exception bits in these regs, 0x200 is the
498 		 * C1 reg you need in case of a stack fault, 0x040 is the stack
499 		 * fault bit.  We should only be taking one exception at a time,
500 		 * so if this combination doesn't produce any single exception,
501 		 * then we have a bad program that isn't synchronizing its FPU usage
502 		 * and it will suffer the consequences since we won't be able to
503 		 * fully reproduce the context of the exception
504 		 */
505 		cwd = get_fpu_cwd(task);
506 		swd = get_fpu_swd(task);
507 
508 		err = swd & ~cwd;
509 	} else {
510 		/*
511 		 * The SIMD FPU exceptions are handled a little differently, as there
512 		 * is only a single status/control register.  Thus, to determine which
513 		 * unmasked exception was caught we must mask the exception mask bits
514 		 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
515 		 */
516 		unsigned short mxcsr = get_fpu_mxcsr(task);
517 		err = ~(mxcsr >> 7) & mxcsr;
518 	}
519 
520 	if (err & 0x001) {	/* Invalid op */
521 		/*
522 		 * swd & 0x240 == 0x040: Stack Underflow
523 		 * swd & 0x240 == 0x240: Stack Overflow
524 		 * User must clear the SF bit (0x40) if set
525 		 */
526 		info.si_code = FPE_FLTINV;
527 	} else if (err & 0x004) { /* Divide by Zero */
528 		info.si_code = FPE_FLTDIV;
529 	} else if (err & 0x008) { /* Overflow */
530 		info.si_code = FPE_FLTOVF;
531 	} else if (err & 0x012) { /* Denormal, Underflow */
532 		info.si_code = FPE_FLTUND;
533 	} else if (err & 0x020) { /* Precision */
534 		info.si_code = FPE_FLTRES;
535 	} else {
536 		/*
537 		 * If we're using IRQ 13, or supposedly even some trap
538 		 * X86_TRAP_MF implementations, it's possible
539 		 * we get a spurious trap, which is not an error.
540 		 */
541 		return;
542 	}
543 	force_sig_info(SIGFPE, &info, task);
544 }
545 
do_coprocessor_error(struct pt_regs * regs,long error_code)546 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
547 {
548 #ifdef CONFIG_X86_32
549 	ignore_fpu_irq = 1;
550 #endif
551 
552 	math_error(regs, error_code, X86_TRAP_MF);
553 }
554 
555 dotraplinkage void
do_simd_coprocessor_error(struct pt_regs * regs,long error_code)556 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
557 {
558 	math_error(regs, error_code, X86_TRAP_XF);
559 }
560 
561 dotraplinkage void
do_spurious_interrupt_bug(struct pt_regs * regs,long error_code)562 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
563 {
564 	conditional_sti(regs);
565 #if 0
566 	/* No need to warn about this any longer. */
567 	printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
568 #endif
569 }
570 
smp_thermal_interrupt(void)571 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
572 {
573 }
574 
smp_threshold_interrupt(void)575 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
576 {
577 }
578 
579 /*
580  * 'math_state_restore()' saves the current math information in the
581  * old math state array, and gets the new ones from the current task
582  *
583  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
584  * Don't touch unless you *really* know how it works.
585  *
586  * Must be called with kernel preemption disabled (eg with local
587  * local interrupts as in the case of do_device_not_available).
588  */
math_state_restore(void)589 void math_state_restore(void)
590 {
591 	struct task_struct *tsk = current;
592 
593 	if (!tsk_used_math(tsk)) {
594 		local_irq_enable();
595 		/*
596 		 * does a slab alloc which can sleep
597 		 */
598 		if (init_fpu(tsk)) {
599 			/*
600 			 * ran out of memory!
601 			 */
602 			do_group_exit(SIGKILL);
603 			return;
604 		}
605 		local_irq_disable();
606 	}
607 
608 	__thread_fpu_begin(tsk);
609 	/*
610 	 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
611 	 */
612 	if (unlikely(restore_fpu_checking(tsk))) {
613 		__thread_fpu_end(tsk);
614 		force_sig(SIGSEGV, tsk);
615 		return;
616 	}
617 
618 	tsk->fpu_counter++;
619 }
620 EXPORT_SYMBOL_GPL(math_state_restore);
621 
622 dotraplinkage void __kprobes
do_device_not_available(struct pt_regs * regs,long error_code)623 do_device_not_available(struct pt_regs *regs, long error_code)
624 {
625 #ifdef CONFIG_MATH_EMULATION
626 	if (read_cr0() & X86_CR0_EM) {
627 		struct math_emu_info info = { };
628 
629 		conditional_sti(regs);
630 
631 		info.regs = regs;
632 		math_emulate(&info);
633 		return;
634 	}
635 #endif
636 	math_state_restore(); /* interrupts still off */
637 #ifdef CONFIG_X86_32
638 	conditional_sti(regs);
639 #endif
640 }
641 
642 #ifdef CONFIG_X86_32
do_iret_error(struct pt_regs * regs,long error_code)643 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
644 {
645 	siginfo_t info;
646 	local_irq_enable();
647 
648 	info.si_signo = SIGILL;
649 	info.si_errno = 0;
650 	info.si_code = ILL_BADSTK;
651 	info.si_addr = NULL;
652 	if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
653 			X86_TRAP_IRET, SIGILL) == NOTIFY_STOP)
654 		return;
655 	do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
656 		&info);
657 }
658 #endif
659 
660 /* Set of traps needed for early debugging. */
early_trap_init(void)661 void __init early_trap_init(void)
662 {
663 	set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
664 	/* int3 can be called from all */
665 	set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
666 	set_intr_gate(X86_TRAP_PF, &page_fault);
667 	load_idt(&idt_descr);
668 }
669 
trap_init(void)670 void __init trap_init(void)
671 {
672 	int i;
673 
674 #ifdef CONFIG_EISA
675 	void __iomem *p = early_ioremap(0x0FFFD9, 4);
676 
677 	if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
678 		EISA_bus = 1;
679 	early_iounmap(p, 4);
680 #endif
681 
682 	set_intr_gate(X86_TRAP_DE, &divide_error);
683 	set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
684 	/* int4 can be called from all */
685 	set_system_intr_gate(X86_TRAP_OF, &overflow);
686 	set_intr_gate(X86_TRAP_BR, &bounds);
687 	set_intr_gate(X86_TRAP_UD, &invalid_op);
688 	set_intr_gate(X86_TRAP_NM, &device_not_available);
689 #ifdef CONFIG_X86_32
690 	set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
691 #else
692 	set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
693 #endif
694 	set_intr_gate(X86_TRAP_OLD_MF, &coprocessor_segment_overrun);
695 	set_intr_gate(X86_TRAP_TS, &invalid_TSS);
696 	set_intr_gate(X86_TRAP_NP, &segment_not_present);
697 	set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
698 	set_intr_gate(X86_TRAP_GP, &general_protection);
699 	set_intr_gate(X86_TRAP_SPURIOUS, &spurious_interrupt_bug);
700 	set_intr_gate(X86_TRAP_MF, &coprocessor_error);
701 	set_intr_gate(X86_TRAP_AC, &alignment_check);
702 #ifdef CONFIG_X86_MCE
703 	set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
704 #endif
705 	set_intr_gate(X86_TRAP_XF, &simd_coprocessor_error);
706 
707 	/* Reserve all the builtin and the syscall vector: */
708 	for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
709 		set_bit(i, used_vectors);
710 
711 #ifdef CONFIG_IA32_EMULATION
712 	set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
713 	set_bit(IA32_SYSCALL_VECTOR, used_vectors);
714 #endif
715 
716 #ifdef CONFIG_X86_32
717 	set_system_trap_gate(SYSCALL_VECTOR, &system_call);
718 	set_bit(SYSCALL_VECTOR, used_vectors);
719 #endif
720 
721 	/*
722 	 * Should be a barrier for any external CPU state:
723 	 */
724 	cpu_init();
725 
726 	x86_init.irqs.trap_init();
727 
728 #ifdef CONFIG_X86_64
729 	memcpy(&nmi_idt_table, &idt_table, IDT_ENTRIES * 16);
730 	set_nmi_gate(X86_TRAP_DB, &debug);
731 	set_nmi_gate(X86_TRAP_BP, &int3);
732 #endif
733 }
734