xref: /linux/arch/x86/kernel/traps.c (revision c4ee0af3fa0dc65f690fc908f02b8355f9576ea0)
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/fixmap.h>
60 #include <asm/mach_traps.h>
61 #include <asm/alternative.h>
62 
63 #ifdef CONFIG_X86_64
64 #include <asm/x86_init.h>
65 #include <asm/pgalloc.h>
66 #include <asm/proto.h>
67 
68 /* No need to be aligned, but done to keep all IDTs defined the same way. */
69 gate_desc debug_idt_table[NR_VECTORS] __page_aligned_bss;
70 #else
71 #include <asm/processor-flags.h>
72 #include <asm/setup.h>
73 
74 asmlinkage int system_call(void);
75 #endif
76 
77 /* Must be page-aligned because the real IDT is used in a fixmap. */
78 gate_desc idt_table[NR_VECTORS] __page_aligned_bss;
79 
80 DECLARE_BITMAP(used_vectors, NR_VECTORS);
81 EXPORT_SYMBOL_GPL(used_vectors);
82 
83 static inline void conditional_sti(struct pt_regs *regs)
84 {
85 	if (regs->flags & X86_EFLAGS_IF)
86 		local_irq_enable();
87 }
88 
89 static inline void preempt_conditional_sti(struct pt_regs *regs)
90 {
91 	preempt_count_inc();
92 	if (regs->flags & X86_EFLAGS_IF)
93 		local_irq_enable();
94 }
95 
96 static inline void conditional_cli(struct pt_regs *regs)
97 {
98 	if (regs->flags & X86_EFLAGS_IF)
99 		local_irq_disable();
100 }
101 
102 static inline void preempt_conditional_cli(struct pt_regs *regs)
103 {
104 	if (regs->flags & X86_EFLAGS_IF)
105 		local_irq_disable();
106 	preempt_count_dec();
107 }
108 
109 static int __kprobes
110 do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
111 		  struct pt_regs *regs,	long error_code)
112 {
113 #ifdef CONFIG_X86_32
114 	if (regs->flags & X86_VM_MASK) {
115 		/*
116 		 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
117 		 * On nmi (interrupt 2), do_trap should not be called.
118 		 */
119 		if (trapnr < X86_TRAP_UD) {
120 			if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
121 						error_code, trapnr))
122 				return 0;
123 		}
124 		return -1;
125 	}
126 #endif
127 	if (!user_mode(regs)) {
128 		if (!fixup_exception(regs)) {
129 			tsk->thread.error_code = error_code;
130 			tsk->thread.trap_nr = trapnr;
131 			die(str, regs, error_code);
132 		}
133 		return 0;
134 	}
135 
136 	return -1;
137 }
138 
139 static void __kprobes
140 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
141 	long error_code, siginfo_t *info)
142 {
143 	struct task_struct *tsk = current;
144 
145 
146 	if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
147 		return;
148 	/*
149 	 * We want error_code and trap_nr set for userspace faults and
150 	 * kernelspace faults which result in die(), but not
151 	 * kernelspace faults which are fixed up.  die() gives the
152 	 * process no chance to handle the signal and notice the
153 	 * kernel fault information, so that won't result in polluting
154 	 * the information about previously queued, but not yet
155 	 * delivered, faults.  See also do_general_protection below.
156 	 */
157 	tsk->thread.error_code = error_code;
158 	tsk->thread.trap_nr = trapnr;
159 
160 #ifdef CONFIG_X86_64
161 	if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
162 	    printk_ratelimit()) {
163 		pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
164 			tsk->comm, tsk->pid, str,
165 			regs->ip, regs->sp, error_code);
166 		print_vma_addr(" in ", regs->ip);
167 		pr_cont("\n");
168 	}
169 #endif
170 
171 	if (info)
172 		force_sig_info(signr, info, tsk);
173 	else
174 		force_sig(signr, tsk);
175 }
176 
177 #define DO_ERROR(trapnr, signr, str, name)				\
178 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
179 {									\
180 	enum ctx_state prev_state;					\
181 									\
182 	prev_state = exception_enter();					\
183 	if (notify_die(DIE_TRAP, str, regs, error_code,			\
184 			trapnr, signr) == NOTIFY_STOP) {		\
185 		exception_exit(prev_state);				\
186 		return;							\
187 	}								\
188 	conditional_sti(regs);						\
189 	do_trap(trapnr, signr, str, regs, error_code, NULL);		\
190 	exception_exit(prev_state);					\
191 }
192 
193 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)		\
194 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
195 {									\
196 	siginfo_t info;							\
197 	enum ctx_state prev_state;					\
198 									\
199 	info.si_signo = signr;						\
200 	info.si_errno = 0;						\
201 	info.si_code = sicode;						\
202 	info.si_addr = (void __user *)siaddr;				\
203 	prev_state = exception_enter();					\
204 	if (notify_die(DIE_TRAP, str, regs, error_code,			\
205 			trapnr, signr) == NOTIFY_STOP) {		\
206 		exception_exit(prev_state);				\
207 		return;							\
208 	}								\
209 	conditional_sti(regs);						\
210 	do_trap(trapnr, signr, str, regs, error_code, &info);		\
211 	exception_exit(prev_state);					\
212 }
213 
214 DO_ERROR_INFO(X86_TRAP_DE, SIGFPE, "divide error", divide_error, FPE_INTDIV,
215 		regs->ip)
216 DO_ERROR(X86_TRAP_OF, SIGSEGV, "overflow", overflow)
217 DO_ERROR(X86_TRAP_BR, SIGSEGV, "bounds", bounds)
218 DO_ERROR_INFO(X86_TRAP_UD, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN,
219 		regs->ip)
220 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE, "coprocessor segment overrun",
221 		coprocessor_segment_overrun)
222 DO_ERROR(X86_TRAP_TS, SIGSEGV, "invalid TSS", invalid_TSS)
223 DO_ERROR(X86_TRAP_NP, SIGBUS, "segment not present", segment_not_present)
224 #ifdef CONFIG_X86_32
225 DO_ERROR(X86_TRAP_SS, SIGBUS, "stack segment", stack_segment)
226 #endif
227 DO_ERROR_INFO(X86_TRAP_AC, SIGBUS, "alignment check", alignment_check,
228 		BUS_ADRALN, 0)
229 
230 #ifdef CONFIG_X86_64
231 /* Runs on IST stack */
232 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
233 {
234 	enum ctx_state prev_state;
235 
236 	prev_state = exception_enter();
237 	if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
238 		       X86_TRAP_SS, SIGBUS) != NOTIFY_STOP) {
239 		preempt_conditional_sti(regs);
240 		do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
241 		preempt_conditional_cli(regs);
242 	}
243 	exception_exit(prev_state);
244 }
245 
246 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
247 {
248 	static const char str[] = "double fault";
249 	struct task_struct *tsk = current;
250 
251 	exception_enter();
252 	/* Return not checked because double check cannot be ignored */
253 	notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
254 
255 	tsk->thread.error_code = error_code;
256 	tsk->thread.trap_nr = X86_TRAP_DF;
257 
258 #ifdef CONFIG_DOUBLEFAULT
259 	df_debug(regs, error_code);
260 #endif
261 	/*
262 	 * This is always a kernel trap and never fixable (and thus must
263 	 * never return).
264 	 */
265 	for (;;)
266 		die(str, regs, error_code);
267 }
268 #endif
269 
270 dotraplinkage void __kprobes
271 do_general_protection(struct pt_regs *regs, long error_code)
272 {
273 	struct task_struct *tsk;
274 	enum ctx_state prev_state;
275 
276 	prev_state = exception_enter();
277 	conditional_sti(regs);
278 
279 #ifdef CONFIG_X86_32
280 	if (regs->flags & X86_VM_MASK) {
281 		local_irq_enable();
282 		handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
283 		goto exit;
284 	}
285 #endif
286 
287 	tsk = current;
288 	if (!user_mode(regs)) {
289 		if (fixup_exception(regs))
290 			goto exit;
291 
292 		tsk->thread.error_code = error_code;
293 		tsk->thread.trap_nr = X86_TRAP_GP;
294 		if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
295 			       X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
296 			die("general protection fault", regs, error_code);
297 		goto exit;
298 	}
299 
300 	tsk->thread.error_code = error_code;
301 	tsk->thread.trap_nr = X86_TRAP_GP;
302 
303 	if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
304 			printk_ratelimit()) {
305 		pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
306 			tsk->comm, task_pid_nr(tsk),
307 			regs->ip, regs->sp, error_code);
308 		print_vma_addr(" in ", regs->ip);
309 		pr_cont("\n");
310 	}
311 
312 	force_sig(SIGSEGV, tsk);
313 exit:
314 	exception_exit(prev_state);
315 }
316 
317 /* May run on IST stack. */
318 dotraplinkage void __kprobes notrace do_int3(struct pt_regs *regs, long error_code)
319 {
320 	enum ctx_state prev_state;
321 
322 #ifdef CONFIG_DYNAMIC_FTRACE
323 	/*
324 	 * ftrace must be first, everything else may cause a recursive crash.
325 	 * See note by declaration of modifying_ftrace_code in ftrace.c
326 	 */
327 	if (unlikely(atomic_read(&modifying_ftrace_code)) &&
328 	    ftrace_int3_handler(regs))
329 		return;
330 #endif
331 	if (poke_int3_handler(regs))
332 		return;
333 
334 	prev_state = exception_enter();
335 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
336 	if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
337 				SIGTRAP) == NOTIFY_STOP)
338 		goto exit;
339 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
340 
341 	if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
342 			SIGTRAP) == NOTIFY_STOP)
343 		goto exit;
344 
345 	/*
346 	 * Let others (NMI) know that the debug stack is in use
347 	 * as we may switch to the interrupt stack.
348 	 */
349 	debug_stack_usage_inc();
350 	preempt_conditional_sti(regs);
351 	do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
352 	preempt_conditional_cli(regs);
353 	debug_stack_usage_dec();
354 exit:
355 	exception_exit(prev_state);
356 }
357 
358 #ifdef CONFIG_X86_64
359 /*
360  * Help handler running on IST stack to switch back to user stack
361  * for scheduling or signal handling. The actual stack switch is done in
362  * entry.S
363  */
364 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
365 {
366 	struct pt_regs *regs = eregs;
367 	/* Did already sync */
368 	if (eregs == (struct pt_regs *)eregs->sp)
369 		;
370 	/* Exception from user space */
371 	else if (user_mode(eregs))
372 		regs = task_pt_regs(current);
373 	/*
374 	 * Exception from kernel and interrupts are enabled. Move to
375 	 * kernel process stack.
376 	 */
377 	else if (eregs->flags & X86_EFLAGS_IF)
378 		regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
379 	if (eregs != regs)
380 		*regs = *eregs;
381 	return regs;
382 }
383 #endif
384 
385 /*
386  * Our handling of the processor debug registers is non-trivial.
387  * We do not clear them on entry and exit from the kernel. Therefore
388  * it is possible to get a watchpoint trap here from inside the kernel.
389  * However, the code in ./ptrace.c has ensured that the user can
390  * only set watchpoints on userspace addresses. Therefore the in-kernel
391  * watchpoint trap can only occur in code which is reading/writing
392  * from user space. Such code must not hold kernel locks (since it
393  * can equally take a page fault), therefore it is safe to call
394  * force_sig_info even though that claims and releases locks.
395  *
396  * Code in ./signal.c ensures that the debug control register
397  * is restored before we deliver any signal, and therefore that
398  * user code runs with the correct debug control register even though
399  * we clear it here.
400  *
401  * Being careful here means that we don't have to be as careful in a
402  * lot of more complicated places (task switching can be a bit lazy
403  * about restoring all the debug state, and ptrace doesn't have to
404  * find every occurrence of the TF bit that could be saved away even
405  * by user code)
406  *
407  * May run on IST stack.
408  */
409 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
410 {
411 	struct task_struct *tsk = current;
412 	enum ctx_state prev_state;
413 	int user_icebp = 0;
414 	unsigned long dr6;
415 	int si_code;
416 
417 	prev_state = exception_enter();
418 
419 	get_debugreg(dr6, 6);
420 
421 	/* Filter out all the reserved bits which are preset to 1 */
422 	dr6 &= ~DR6_RESERVED;
423 
424 	/*
425 	 * If dr6 has no reason to give us about the origin of this trap,
426 	 * then it's very likely the result of an icebp/int01 trap.
427 	 * User wants a sigtrap for that.
428 	 */
429 	if (!dr6 && user_mode(regs))
430 		user_icebp = 1;
431 
432 	/* Catch kmemcheck conditions first of all! */
433 	if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
434 		goto exit;
435 
436 	/* DR6 may or may not be cleared by the CPU */
437 	set_debugreg(0, 6);
438 
439 	/*
440 	 * The processor cleared BTF, so don't mark that we need it set.
441 	 */
442 	clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
443 
444 	/* Store the virtualized DR6 value */
445 	tsk->thread.debugreg6 = dr6;
446 
447 	if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
448 							SIGTRAP) == NOTIFY_STOP)
449 		goto exit;
450 
451 	/*
452 	 * Let others (NMI) know that the debug stack is in use
453 	 * as we may switch to the interrupt stack.
454 	 */
455 	debug_stack_usage_inc();
456 
457 	/* It's safe to allow irq's after DR6 has been saved */
458 	preempt_conditional_sti(regs);
459 
460 	if (regs->flags & X86_VM_MASK) {
461 		handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
462 					X86_TRAP_DB);
463 		preempt_conditional_cli(regs);
464 		debug_stack_usage_dec();
465 		goto exit;
466 	}
467 
468 	/*
469 	 * Single-stepping through system calls: ignore any exceptions in
470 	 * kernel space, but re-enable TF when returning to user mode.
471 	 *
472 	 * We already checked v86 mode above, so we can check for kernel mode
473 	 * by just checking the CPL of CS.
474 	 */
475 	if ((dr6 & DR_STEP) && !user_mode(regs)) {
476 		tsk->thread.debugreg6 &= ~DR_STEP;
477 		set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
478 		regs->flags &= ~X86_EFLAGS_TF;
479 	}
480 	si_code = get_si_code(tsk->thread.debugreg6);
481 	if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
482 		send_sigtrap(tsk, regs, error_code, si_code);
483 	preempt_conditional_cli(regs);
484 	debug_stack_usage_dec();
485 
486 exit:
487 	exception_exit(prev_state);
488 }
489 
490 /*
491  * Note that we play around with the 'TS' bit in an attempt to get
492  * the correct behaviour even in the presence of the asynchronous
493  * IRQ13 behaviour
494  */
495 void math_error(struct pt_regs *regs, int error_code, int trapnr)
496 {
497 	struct task_struct *task = current;
498 	siginfo_t info;
499 	unsigned short err;
500 	char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
501 						"simd exception";
502 
503 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
504 		return;
505 	conditional_sti(regs);
506 
507 	if (!user_mode_vm(regs))
508 	{
509 		if (!fixup_exception(regs)) {
510 			task->thread.error_code = error_code;
511 			task->thread.trap_nr = trapnr;
512 			die(str, regs, error_code);
513 		}
514 		return;
515 	}
516 
517 	/*
518 	 * Save the info for the exception handler and clear the error.
519 	 */
520 	save_init_fpu(task);
521 	task->thread.trap_nr = trapnr;
522 	task->thread.error_code = error_code;
523 	info.si_signo = SIGFPE;
524 	info.si_errno = 0;
525 	info.si_addr = (void __user *)regs->ip;
526 	if (trapnr == X86_TRAP_MF) {
527 		unsigned short cwd, swd;
528 		/*
529 		 * (~cwd & swd) will mask out exceptions that are not set to unmasked
530 		 * status.  0x3f is the exception bits in these regs, 0x200 is the
531 		 * C1 reg you need in case of a stack fault, 0x040 is the stack
532 		 * fault bit.  We should only be taking one exception at a time,
533 		 * so if this combination doesn't produce any single exception,
534 		 * then we have a bad program that isn't synchronizing its FPU usage
535 		 * and it will suffer the consequences since we won't be able to
536 		 * fully reproduce the context of the exception
537 		 */
538 		cwd = get_fpu_cwd(task);
539 		swd = get_fpu_swd(task);
540 
541 		err = swd & ~cwd;
542 	} else {
543 		/*
544 		 * The SIMD FPU exceptions are handled a little differently, as there
545 		 * is only a single status/control register.  Thus, to determine which
546 		 * unmasked exception was caught we must mask the exception mask bits
547 		 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
548 		 */
549 		unsigned short mxcsr = get_fpu_mxcsr(task);
550 		err = ~(mxcsr >> 7) & mxcsr;
551 	}
552 
553 	if (err & 0x001) {	/* Invalid op */
554 		/*
555 		 * swd & 0x240 == 0x040: Stack Underflow
556 		 * swd & 0x240 == 0x240: Stack Overflow
557 		 * User must clear the SF bit (0x40) if set
558 		 */
559 		info.si_code = FPE_FLTINV;
560 	} else if (err & 0x004) { /* Divide by Zero */
561 		info.si_code = FPE_FLTDIV;
562 	} else if (err & 0x008) { /* Overflow */
563 		info.si_code = FPE_FLTOVF;
564 	} else if (err & 0x012) { /* Denormal, Underflow */
565 		info.si_code = FPE_FLTUND;
566 	} else if (err & 0x020) { /* Precision */
567 		info.si_code = FPE_FLTRES;
568 	} else {
569 		/*
570 		 * If we're using IRQ 13, or supposedly even some trap
571 		 * X86_TRAP_MF implementations, it's possible
572 		 * we get a spurious trap, which is not an error.
573 		 */
574 		return;
575 	}
576 	force_sig_info(SIGFPE, &info, task);
577 }
578 
579 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
580 {
581 	enum ctx_state prev_state;
582 
583 	prev_state = exception_enter();
584 	math_error(regs, error_code, X86_TRAP_MF);
585 	exception_exit(prev_state);
586 }
587 
588 dotraplinkage void
589 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
590 {
591 	enum ctx_state prev_state;
592 
593 	prev_state = exception_enter();
594 	math_error(regs, error_code, X86_TRAP_XF);
595 	exception_exit(prev_state);
596 }
597 
598 dotraplinkage void
599 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
600 {
601 	conditional_sti(regs);
602 #if 0
603 	/* No need to warn about this any longer. */
604 	pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
605 #endif
606 }
607 
608 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
609 {
610 }
611 
612 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
613 {
614 }
615 
616 /*
617  * 'math_state_restore()' saves the current math information in the
618  * old math state array, and gets the new ones from the current task
619  *
620  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
621  * Don't touch unless you *really* know how it works.
622  *
623  * Must be called with kernel preemption disabled (eg with local
624  * local interrupts as in the case of do_device_not_available).
625  */
626 void math_state_restore(void)
627 {
628 	struct task_struct *tsk = current;
629 
630 	if (!tsk_used_math(tsk)) {
631 		local_irq_enable();
632 		/*
633 		 * does a slab alloc which can sleep
634 		 */
635 		if (init_fpu(tsk)) {
636 			/*
637 			 * ran out of memory!
638 			 */
639 			do_group_exit(SIGKILL);
640 			return;
641 		}
642 		local_irq_disable();
643 	}
644 
645 	__thread_fpu_begin(tsk);
646 
647 	/*
648 	 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
649 	 */
650 	if (unlikely(restore_fpu_checking(tsk))) {
651 		drop_init_fpu(tsk);
652 		force_sig(SIGSEGV, tsk);
653 		return;
654 	}
655 
656 	tsk->thread.fpu_counter++;
657 }
658 EXPORT_SYMBOL_GPL(math_state_restore);
659 
660 dotraplinkage void __kprobes
661 do_device_not_available(struct pt_regs *regs, long error_code)
662 {
663 	enum ctx_state prev_state;
664 
665 	prev_state = exception_enter();
666 	BUG_ON(use_eager_fpu());
667 
668 #ifdef CONFIG_MATH_EMULATION
669 	if (read_cr0() & X86_CR0_EM) {
670 		struct math_emu_info info = { };
671 
672 		conditional_sti(regs);
673 
674 		info.regs = regs;
675 		math_emulate(&info);
676 		exception_exit(prev_state);
677 		return;
678 	}
679 #endif
680 	math_state_restore(); /* interrupts still off */
681 #ifdef CONFIG_X86_32
682 	conditional_sti(regs);
683 #endif
684 	exception_exit(prev_state);
685 }
686 
687 #ifdef CONFIG_X86_32
688 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
689 {
690 	siginfo_t info;
691 	enum ctx_state prev_state;
692 
693 	prev_state = exception_enter();
694 	local_irq_enable();
695 
696 	info.si_signo = SIGILL;
697 	info.si_errno = 0;
698 	info.si_code = ILL_BADSTK;
699 	info.si_addr = NULL;
700 	if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
701 			X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
702 		do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
703 			&info);
704 	}
705 	exception_exit(prev_state);
706 }
707 #endif
708 
709 /* Set of traps needed for early debugging. */
710 void __init early_trap_init(void)
711 {
712 	set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
713 	/* int3 can be called from all */
714 	set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
715 #ifdef CONFIG_X86_32
716 	set_intr_gate(X86_TRAP_PF, page_fault);
717 #endif
718 	load_idt(&idt_descr);
719 }
720 
721 void __init early_trap_pf_init(void)
722 {
723 #ifdef CONFIG_X86_64
724 	set_intr_gate(X86_TRAP_PF, page_fault);
725 #endif
726 }
727 
728 void __init trap_init(void)
729 {
730 	int i;
731 
732 #ifdef CONFIG_EISA
733 	void __iomem *p = early_ioremap(0x0FFFD9, 4);
734 
735 	if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
736 		EISA_bus = 1;
737 	early_iounmap(p, 4);
738 #endif
739 
740 	set_intr_gate(X86_TRAP_DE, divide_error);
741 	set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
742 	/* int4 can be called from all */
743 	set_system_intr_gate(X86_TRAP_OF, &overflow);
744 	set_intr_gate(X86_TRAP_BR, bounds);
745 	set_intr_gate(X86_TRAP_UD, invalid_op);
746 	set_intr_gate(X86_TRAP_NM, device_not_available);
747 #ifdef CONFIG_X86_32
748 	set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
749 #else
750 	set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
751 #endif
752 	set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
753 	set_intr_gate(X86_TRAP_TS, invalid_TSS);
754 	set_intr_gate(X86_TRAP_NP, segment_not_present);
755 	set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
756 	set_intr_gate(X86_TRAP_GP, general_protection);
757 	set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
758 	set_intr_gate(X86_TRAP_MF, coprocessor_error);
759 	set_intr_gate(X86_TRAP_AC, alignment_check);
760 #ifdef CONFIG_X86_MCE
761 	set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
762 #endif
763 	set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
764 
765 	/* Reserve all the builtin and the syscall vector: */
766 	for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
767 		set_bit(i, used_vectors);
768 
769 #ifdef CONFIG_IA32_EMULATION
770 	set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
771 	set_bit(IA32_SYSCALL_VECTOR, used_vectors);
772 #endif
773 
774 #ifdef CONFIG_X86_32
775 	set_system_trap_gate(SYSCALL_VECTOR, &system_call);
776 	set_bit(SYSCALL_VECTOR, used_vectors);
777 #endif
778 
779 	/*
780 	 * Set the IDT descriptor to a fixed read-only location, so that the
781 	 * "sidt" instruction will not leak the location of the kernel, and
782 	 * to defend the IDT against arbitrary memory write vulnerabilities.
783 	 * It will be reloaded in cpu_init() */
784 	__set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
785 	idt_descr.address = fix_to_virt(FIX_RO_IDT);
786 
787 	/*
788 	 * Should be a barrier for any external CPU state:
789 	 */
790 	cpu_init();
791 
792 	x86_init.irqs.trap_init();
793 
794 #ifdef CONFIG_X86_64
795 	memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
796 	set_nmi_gate(X86_TRAP_DB, &debug);
797 	set_nmi_gate(X86_TRAP_BP, &int3);
798 #endif
799 }
800